Converter station fire protection digital management and control system and method

Abstract

The embodiment of the invention provides a converter station fire-fighting digital management and control system, and the system comprises a converter station data processing module, a fire-fighting daily management module, and a fire-fighting emergency processing module, and the converter station data processing module is used for obtaining the fire-fighting internet-of-things data of a converter station, and transmitting the fire-fighting internet-of-things data to the fire-fighting daily management module. Performing abnormal data detection on the acquired fire-fighting internet-of-things data by using an abnormal data detection model; the fire-fighting daily management module is used for performing fire-fighting daily management on the converter station according to the fire-fighting internet-of-things data of the converter station; and the fire-fighting emergency processing module is used for determining the converter station with a dangerous case or a dangerous case trend according to at least one of the fire-fighting internet-of-things data of the converter station and the management result of the fire-fighting daily management module, and performing fire-fighting emergency processing on the converter station with the dangerous case or the dangerous case trend. Therefore, the digital fire-fighting management and control capability of the converter station is enhanced, and the accuracy and effectiveness of fire-fighting safety management and control of the converter station are improved.

Description

Technical Field

[0001] This application relates to the field of fire protection in converter stations, specifically to a digital fire protection control system and method for converter stations. Background Technology

[0002] As the core hub for the efficient and stable operation of the power system, the operating status of the converter station is directly related to the stability of the entire power system. Fire safety, as a key link in ensuring the smooth operation of the converter station, plays an important role in maintaining the stability of the power system. Therefore, efficient and reliable fire safety management of the converter station is essential for the power system.

[0003] With the rapid development of ultra-high voltage direct current (UHVDC) transmission technology, converter stations, as key nodes in power transmission projects for AC-DC or DC-AC conversion, are facing increasingly prominent fire safety issues. Currently, fire safety management in converter stations mainly relies on periodic manual inspections and early warnings. This passive, manual management method is inefficient and, due to its dependence on human experience and responsibility, may struggle to detect fire hazards in a timely manner. Furthermore, the lack of real-time monitoring and analysis of the numerous fire-related equipment within the converter station prevents accurate and efficient fire safety management.

[0004] In addition, during long-term use, the fire protection IoT devices in the converter station will inevitably experience aging, interference, and malfunctions. The detection data obtained through these fire protection IoT devices may change abruptly or drift, which may lead to false alarms and missed alarms of fire safety incidents (such as fire incidents), affecting the accuracy and effectiveness of fire safety management and control in the converter station. Summary of the Invention

[0005] This application provides a digital fire protection management and control system and method for converter stations, which is used to improve the digital fire protection capabilities of converter stations and enhance the accuracy and effectiveness of fire safety management and control.

[0006] Firstly, a digital fire protection control system for converter stations is provided, comprising a converter station data processing module, a daily fire protection management module, and a fire emergency response module, wherein:

[0007] Control the identification correspondence between the device identifier of the second camera and the device identifier of the first camera carried in the second video data, and verify the validity of the second video data according to the identification correspondence;

[0008] And / or,

[0009] The control includes carrying a priority transmission identifier in the second video data, which indicates that the second video data should be transmitted to the converter station digital fire control system with high priority.

[0010] In one possible implementation, the anomaly detection model is trained as follows:

[0011] Obtain normal timing data generated during the operation of the converter station;

[0012] Based on the sliding window and step size corresponding to the data type of the normal time series data, the normal time series data is converted into normal time series samples;

[0013] The abnormal time series samples are generated based on the normal time series samples;

[0014] The abnormal data detection model is obtained by training a hybrid expert model, which includes at least two sub-models, using the normal time series samples and the abnormal time series samples.

[0015] In one possible implementation, generating abnormal time series samples based on the normal time series samples includes:

[0016] The trend anomaly samples are generated based on the normal time series samples in the following manner:

[0017]

[0018] Where y1[i] represents the trend anomaly sample; x[i] represents the normal time series sample; trend is the added linear trend term, trend=a·(i-t1+1), a is the trend coefficient, a=(R / t1)·r1·(n / 10), R is the initial segment range, r1 is the trend direction adjustment factor, n is the anomaly intensity coefficient, t1 is the starting point of the anomaly interval, t1~Uniform{2,…l-1}, l is the sequence length of the original normal time series sample, and i is the sample in the anomaly interval; σ x This represents the initial standard deviation of the normal time series samples; For random perturbation,

[0019] In one possible implementation, generating abnormal time series samples based on the normal time series samples includes:

[0020] The fixed-bias anomaly sample is generated based on the normal time series sample in the following manner:

[0021] y2[i]=x[i]+x1

[0022] Where y2[i] is a fixed-bias abnormal sample; x[i] is a normal time series sample; and x1 is the injected fixed bias.

[0023] In one possible implementation, generating abnormal time series samples based on the normal time series samples includes:

[0024] The precision degradation anomaly samples are generated based on the normal time-series samples in the following manner:

[0025] y3[i]=x[i]+ε

[0026] Where y3[i] represents the anomalous sample with fixed bias; x[i] represents the normal time series sample; ε represents Gaussian white noise, ε ~ Normal(0,σ) 2 ), where σ is the noise amplitude, σ=β·r, β is the reference amplitude, and r is the random amplitude adjustment factor.

[0027] In one possible implementation, generating abnormal time series samples based on the normal time series samples includes:

[0028] The constant value anomaly sample is generated based on the normal time series sample in the following manner:

[0029] y4[t1:l]=x[t1]

[0030] Where y4[t1:l] represents the constant value abnormal sample; t1 is the random starting point, t1~Uniform{2,…l-1}, and l is the sequence length of the original normal time series sample; x[t1] indicates that the normal time series sample is set to a constant value starting from t1.

[0031] In one possible implementation, generating abnormal time series samples based on the normal time series samples includes:

[0032] The discrete point anomaly samples are generated based on the normal time series samples in the following manner:

[0033] y5[i]=x[i]+Δ

[0034] Where y5[i] is an outlier sample; x[i] is a normal time series sample; Δ is the outlier perturbation, Δ=(1+N)·μ·(n / 10 4 ), N is the standard normal perturbation, N ~ Normal(0,1), μ is the global mean benchmark of normal time series samples, and n is the anomaly intensity coefficient.

[0035] In one possible implementation, the hybrid expert model includes at least two sub-models selected from CNN sub-model, GRU sub-model, LSTM sub-model and Transformer sub-model;

[0036] The abnormal data detection model also includes a gating network, which is used to assign corresponding weights to each sub-model according to the data type of the input data;

[0037] Each sub-model is used to obtain the corresponding sub-model anomaly detection result based on the input data and the assigned weights; wherein, the anomaly detection result of the anomaly data detection model is obtained based on the anomaly detection results of each sub-model.

[0038] In one possible implementation, when training the hybrid expert model, the loss of each sub-model includes a binary cross-entropy loss for binary classification tasks and a multivariate cross-entropy loss for multi-anomaly classification tasks.

[0039] The total loss of the hybrid expert model is:

[0040]

[0041] Wherein, Weighted Output Loss represents the binary cross-entropy loss or multivariate cross-entropy loss of the weighted output of the hybrid expert model, and Expert Loss... k Let λ represent the binary cross-entropy loss or multivariate cross-entropy loss of the k-th sub-model, and let λ represent the weight coefficients of the sub-model loss.

[0042] In one possible implementation, the converter station data processing module is further configured to:

[0043] Determine the integrity score, validity score, and timeliness score of the real-time time series data corresponding to the fire protection IoT data of the converter station;

[0044] The comprehensive quality score of the real-time time series data is determined based on the completeness score, validity score, and timeliness score of the real-time time series data.

[0045] The comprehensive quality score is used to determine whether the fire protection IoT data meets the data access quality compliance requirements.

[0046] In one possible implementation, the fire protection daily management module includes a fire protection equipment control module, which is used for:

[0047] If the converter station data processing module determines that the fire protection IoT data meets the data access quality compliance conditions, it determines the quality level of the fire protection IoT data based on the comprehensive quality score, and obtains the first data quality distribution characteristics based on the quality level of the fire protection IoT data.

[0048] The first data quality distribution characteristics are mapped to a pre-determined health baseline to obtain the health status assessment results of the fire protection IoT devices corresponding to the fire protection IoT data;

[0049] The health baseline is obtained by clustering the second data quality distribution characteristics of the historical normal time series data of the fire protection IoT device in different time windows. The second data quality distribution characteristics are determined by extracting data of different quality levels within the time window through a sliding window based on the comprehensive quality score of the historical normal time series data.

[0050] In one possible implementation, the converter station data processing module is used to determine the integrity score, validity score, and timeliness score of the real-time time-series data corresponding to the fire protection IoT data of the converter station, including:

[0051] The importance weight of the real-time time series data is calculated using the analytic hierarchy process (AHP), and the integrity score of the real-time time series data is determined by combining the standardization check of the time series data content.

[0052] The effectiveness score of the real-time time series data is obtained by using either a method based on the accuracy of long-term and short-term historical time series data or an evaluation method based on time series reconstruction.

[0053] Based on the transmission interval of the real-time time series data, the timeliness score of the real-time time series data is obtained.

[0054] In one possible implementation, the converter station data processing module is used to calculate the importance weight of the real-time time series data using the analytic hierarchy process (AHP), and, in conjunction with a standardization check of the time series data content, determine the integrity score of the real-time time series data, including:

[0055] The importance weights of primary and secondary indicators in the real-time time-series data are calculated using the Analytic Hierarchy Process (AHP). The primary indicators include at least one of equipment information, monitoring data, equipment status, and communication information. The secondary indicators for equipment information include at least one of equipment identifier, equipment name, data code, equipment manufacturer, equipment model, and equipment installation location. The secondary indicators for monitoring data include at least one of data acquisition timestamp, sensor type, monitored value, and unit of monitored value. The secondary indicators for equipment status include at least one of equipment operating status, battery level, and signal strength. The secondary indicators for communication information include at least one of communication protocol type and network type.

[0056] The integrity score of the real-time time series data is determined based on the importance weights of the secondary indicators combined with the standardization check of the time series data content, using the following method:

[0057] y=∑ω i x i

[0058] Where y is the integrity score, ω iLet x be the weight of the i-th secondary indicator. i Let x be the score of the i-th secondary indicator. i The value of x is 1 or 0. If the content corresponding to the secondary indicator is missing, then x i If the value is 0, then x is zero if the content corresponding to the indicator is not missing. i The value is 1.

[0059] In one possible implementation, the converter station data processing module is used to obtain a validity score for the real-time time series data by combining the accuracy of long-term and short-term historical time series data when the real-time time series data is remote signaling data, including:

[0060] δ=ω1·x1+ω2·x2+ω3·x 13 +ω4·x4

[0061] Where δ is the data validity score; x1 is the historical accuracy of the device; x2 is the accuracy of the M most recent remote signaling signals sent by the device; x3 is the accuracy of the most recent M-5 remote signaling data; x4 is the accuracy of the last remote signaling signal sent; and ω1, ω2, ω3, and ω4 are the weights corresponding to x1, x2, x3, and x4.

[0062] In one possible implementation, when the real-time time-series data is telemetry data, the converter station data processing module is used to obtain a validity score for the real-time time-series data using an evaluation method based on time-series reconstruction, including:

[0063] The telemetry data was reconstructed using the Transformer reconstruction model to obtain the reconstructed sequence;

[0064] The reconstruction error of the reconstructed sequence was statistically analyzed using the POT threshold selection method to obtain the reconstruction error threshold. The initial threshold in the POT threshold selection method was determined using the three-standard-deviation method.

[0065] The validity score of data obtained based on the critical threshold in the reconstruction error threshold is:

[0066]

[0067] Where δ(i) is the effectiveness score for each time step i, e i It is the relative error of reconstruction at the time step. th F δ is the critical threshold used to determine data anomalies. l The validity score of telemetry data is set when the relative error of reconstruction at each time step is the anomaly judgment threshold.

[0068] In one possible implementation, based on the transmission interval of the real-time time series data, the converter station data processing module is used to obtain a timeliness score for the real-time time series data, including:

[0069] When the real-time time series data is remote signaling data, the timeliness score of the remote signaling data is obtained using a first evaluation function. The formula for the first evaluation function is as follows:

[0070] y = e -0.346x

[0071] Where y is the timeliness score of remote signaling data transmission, and x is the data transmission delay time;

[0072] When the real-time time series data is telemetry data, the timeliness score of the telemetry data is obtained using a second evaluation function. The formula for the second evaluation function is as follows:

[0073] y n =1-0.11I n 2

[0074] in, μ B Let σ be the mean of set B. B Let be the standard deviation of set B, and set B is [ΔT]. n-1 ,ΔT n-2 ,ΔT n-3 ,…,ΔT n-N ], ΔT n =T n -T n-1 T n Let ΔT be the time when the nth data point is transmitted to the monitoring center backend. n y represents the transmission time interval between the nth data and the previous data; n To score the timeliness of telemetry data acquisition.

[0075] In one possible implementation, the fire equipment management module is used to perform a nonlinear mapping between the first data quality distribution feature vector and a pre-determined health baseline to obtain the health status assessment result of the fire IoT equipment corresponding to the fire IoT data, including:

[0076] The overall feature weight is determined based on the importance weight and information entropy weight of the second data quality distribution feature;

[0077] Based on the aforementioned feature comprehensive weights, the distance between the first data quality distribution feature vector and the healthy baseline is evaluated using the W-ReLU nonlinear mapping model as the anomaly degree.

[0078] The health level of the fire protection IoT equipment is determined based on the anomaly level.

[0079] In one possible implementation, the formula for calculating the anomaly degree is:

[0080]

[0081] In the formula, δ n For anomaly degree, w′ i x is the feature comprehensive weight of feature i. 0k Let d be the eigenvalue of the k-th feature corresponding to the cluster center of the cluster to which the new sample belongs. kmin With d kmax The distance from the edge of the confidence range to the eigenvalue of the k-th feature at the cluster center is given by N, where N is the device scale parameter, and N = x. 01 +x 02 +x 03 +x 04 , [x 01 ,x 02 ,x 03 ,x 04 [x′1, x′2, x′3, x′4] is the cluster center, and [x′1, x′2, x′3, x′4] is the data quality distribution feature vector of the real-time time series data;

[0082] The formula for calculating the feature integration weight is as follows:

[0083]

[0084] In the formula, AW i For importance weights, RV i θ represents the information entropy weight, θ represents the assigned weight, and m represents the total number of features.

[0085] In one possible implementation, the fire protection daily management module includes a fire protection equipment control module, which is used for:

[0086] To obtain experimental data on temperature drop of fire-fighting pipelines containing electric heating cables under low-temperature conditions;

[0087] A numerical model of fire-fighting pipelines containing electric heat tracing cables was established, the conversion formula for the convective heat transfer coefficient was determined, and the numerical model was verified based on the temperature drop experimental data and the conversion formula for the convective heat transfer coefficient.

[0088] By changing the ambient temperature and the power of the electric heating tape, the antifreeze time of the fire-fighting pipeline containing the electric heating tape under different working conditions was obtained;

[0089] Based on the time of the different operating conditions, establish the power function of the electric heating cable with respect to the antifreeze time and ambient temperature;

[0090] Based on the power function of the electric heating cable and the given antifreeze time and ambient temperature, the optimal power of the electric heating cable for the fire-fighting pipeline containing the electric heating cable is determined under the required antifreeze time.

[0091] In one possible implementation, the formula for converting the convective heat transfer coefficient is: Where h is the convective heat transfer coefficient, β is determined by the temperature drop experimental data of the fire-fighting pipeline, R is the outer diameter of the fire-fighting pipeline, Pr and Gr are the Prandtl number and Grashof number of the ambient air, respectively, and A and m are constants.

[0092] In one possible implementation, the fire equipment control module is used to verify the numerical model based on the temperature drop experimental data and the convective heat transfer coefficient conversion formula, including:

[0093] Input the temperature, fire pipeline parameters, and electric heating cable parameters that are consistent with the temperature drop experimental data into the numerical model. Input the convective heat transfer coefficient for this working condition according to the convective heat transfer coefficient conversion formula. Set a temperature sensor at the symmetrical position of the fire pipeline heating cable about the center to obtain the pipe wall temperature. Change the convective heat transfer coefficient β until the pipe wall temperature of the numerical model is consistent with the temperature drop experimental data.

[0094] In one possible implementation, the fire equipment control module is used to obtain the antifreeze time of the fire pipeline containing electric heating cable under different operating conditions, including:

[0095] The liquid fraction obtained by installing a liquid fraction sensor on the outermost layer of the fire-fighting pipeline under different ambient temperatures and different electric heating tape powers is used to determine the antifreeze time under the corresponding working conditions from the start until the liquid fraction on the outermost layer of the fire-fighting pipeline becomes 0.02.

[0096] In one possible implementation, the power function of the electric heating tape is:

[0097] t = f(P,T) = a0 + a1P + a2T + a3P 2 +a4PT+a5T 2 ,

[0098] Where P is the power of the electric heating cable, T is the ambient temperature, t is the antifreeze time, and a0, a1, a2, a3, a4 and a5 are regression coefficients, which are obtained by fitting the power function using the Levenberg-Marquardt optimization algorithm.

[0099] In one possible implementation, the daily fire management module includes a fire equipment control module, or the fire emergency response module includes a fire equipment emergency monitoring module, wherein the fire equipment control module or the fire equipment emergency monitoring module is used for:

[0100] The first data set is obtained by collecting fluid data from the collection points of the fire-fighting foam pipeline;

[0101] Based on the first data set and the preset fluid identification model, the fluid type in the fire-fighting foam pipe is determined; the fluid type is either a Newtonian fluid or a non-Newtonian fluid.

[0102] Based on the leak identification model corresponding to the fluid type and the first data set, the leak identification result of the fire-fighting foam pipeline is determined.

[0103] In one possible implementation, the fire equipment management module or the fire equipment emergency monitoring module is used to determine the leak identification result of the fire foam pipeline based on the leak identification model corresponding to the fluid type and the first data set, including:

[0104] Based on the first data set, a second data set corresponding to Newtonian fluid is determined, and based on the second data set and the first leak identification model corresponding to Newtonian fluid, the leak identification result of the fire-fighting foam pipeline is determined.

[0105] or,

[0106] Based on the first data set, a third data set corresponding to the non-Newtonian fluid is determined, and based on the third data set and the second leak identification model corresponding to the non-Newtonian fluid, the leak identification result of the fire-fighting foam pipeline is determined.

[0107] In one possible implementation, the first data set includes: pressure gradient data, flow velocity profile data, shear rate-viscosity curves, temperature sensor data, and fluid density measurements.

[0108] Based on the first data set, a second data set corresponding to the Newtonian fluid is determined, including: determining the pressure gradient variation coefficient based on the pressure gradient data, determining the mass flow rate deviation based on the flow velocity profile data, determining the viscous dissipation power density based on the shear rate-viscosity curve and temperature sensor data, and determining the dynamic characteristics of the bulk modulus based on the fluid density measurement value.

[0109] Alternatively, based on the first data set, a third data set corresponding to the non-Newtonian fluid is determined, including: determining the shear rate-viscosity dynamic characteristics based on the shear rate-viscosity curve, determining the secondary flow intensity based on the flow velocity profile data, determining the viscous dissipation power density and the corresponding associated temperature based on the shear rate-viscosity curve and temperature sensor data, and determining the dynamic characteristics of the bulk modulus based on the fluid density measurement value.

[0110] In one possible implementation, the fire equipment control module or the fire equipment emergency monitoring module is further used for:

[0111] If, based on the second data set and the first leak identification model corresponding to Newtonian fluid, it is determined that the fire-fighting foam pipe is leaking, the pipe section with a leak is determined based on the collection point where the first deviation value is greater than a preset first threshold and the second deviation value is greater than a preset second threshold; wherein, the first deviation value is the difference between the pressure gradient variation coefficient corresponding to the collection point and the preset normal pressure gradient variation coefficient, and the second deviation value is the difference between the mass flow rate deviation corresponding to the collection point and the preset normal mass flow rate deviation.

[0112] or,

[0113] If a leak is determined in the fire-fighting foam pipeline based on the second leak identification model corresponding to the third data set and non-Newtonian fluid, a leak is determined in the target pipe section based on the pressure gradient decrease value of the target pipe section being greater than the third threshold and the viscous dissipation power density increase value of the target pipe section being greater than the fourth threshold; wherein, the target pipe section is the pipe section between two adjacent collection points.

[0114] In one possible implementation, the fire protection daily management module includes a station-side critical equipment management module, which is used for:

[0115] Under different fault or failure conditions inside the converter transformer tank, the temperature time series data inside the converter transformer tank in the real environment is obtained and recorded as the real dataset; and a proportional numerical simulation is carried out on the real environment to simulate the temperature evolution law inside the converter transformer tank under different fault or failure conditions, and the temperature time series data inside the converter transformer tank in the simulation environment is obtained and recorded as the simulation dataset.

[0116] The simulation dataset and the real dataset are compared to verify the reliability of the simulation dataset.

[0117] The simulation dataset after reliability verification is transformed into a temperature matrix with corresponding coordinates, and the time matrix is ​​determined based on the time in the simulation dataset after reliability verification as a variable; and the physical matrix with corresponding coordinates is determined based on the physical characteristics of the converter transformer tank.

[0118] The temperature matrix, the time matrix, and the physical matrix are input into a generative adversarial network to obtain the temperature field matrix inside the converter transformer tank.

[0119] Based on the temperature field matrix inside the converter transformer tank, the overpressure inside the converter transformer tank is determined. Then, based on the temperature field matrix inside the converter transformer tank and the influence of local high temperature on the strength of the tank material, the failure risk of the converter transformer tank is predicted.

[0120] In one possible implementation, the station-side critical equipment management module is used to verify the reliability of the simulation dataset, including: calculating the difference between the simulation dataset and the real dataset based on the average coefficient of determination, and comparing the difference comparison calculation result with a preset difference threshold; when the difference comparison calculation result is greater than or equal to the difference threshold, it indicates that the simulation dataset is reliable; if the difference comparison calculation result is less than the difference threshold, it indicates that the simulation dataset is unreliable, and the parameters of the simulation environment are readjusted, and the reliability of the simulation dataset after the adjustment of the simulation environment parameters is verified again with the real dataset, until the reliability meets the standard.

[0121] In one possible implementation, the key operating equipment control module on the station side is used to determine the physical matrix of corresponding coordinates based on the physical characteristics of the converter transformer tank, including:

[0122] The physical features are discretized based on the coordinates corresponding to the temperature matrix, and then divided into corresponding grids.

[0123] The internal structure, material strength, and thermal conductivity of the oil tank within the corresponding grid are taken as representative parameters of the local oil tank within that grid, and then matrixed to form the oil tank internal structure matrix, oil tank material strength matrix, and oil tank thermal conductivity matrix.

[0124] The internal structure matrix, material strength matrix, and thermal conductivity matrix of the fuel tank are used as physical matrices.

[0125] In one possible implementation, the station-side key operating equipment control module is used to acquire the temperature field matrix inside the converter transformer tank, including:

[0126] Generative adversarial networks consist of a generator and a discriminator;

[0127] The temperature matrix, the time matrix, and the physical matrix are merged to generate an input matrix for training the generative adversarial network.

[0128] The input matrix is ​​input as real data into the generator to obtain the corresponding fake data;

[0129] The discriminator distinguishes between real data and corresponding fake data, outputs the probability of being judged as real or fake, and performs backpropagation after calculating the loss to update the generator weights. The generative adversarial network is then iteratively trained until the generator can generate a high-quality temperature field matrix that is close to the real data.

[0130] Using a trained generative adversarial network and real data as input, the temperature field moment inside the converter transformer tank is obtained.

[0131] In one possible implementation, the station-side key operating equipment control module is used to determine the overpressure inside the converter transformer tank based on the temperature field matrix inside the converter transformer tank, including:

[0132] Based on the temperature field matrix inside the converter transformer tank, the vaporization amount is calculated based on the instantaneous temperature field and the thermal properties of the transformer oil, and the total vaporization amount of the transformer oil after the occurrence of potential risks is obtained by integration; the total vaporization amount of the converter transformer oil is introduced into the ideal gas state equation to calculate the overpressure amount inside the converter transformer tank.

[0133] In one possible implementation, the fire protection daily management module includes a converter station risk management module, which is used for:

[0134] Determine the importance weight of each fire protection subsystem in the converter station, wherein the fire protection subsystem includes at least one of the following: fire protection facility system, fire alarm system, fire extinguishing system, and fire evacuation system;

[0135] Based on the failure event tree model of each fire-fighting equipment in each fire-fighting subsystem, the failure probability of each fire-fighting equipment is determined.

[0136] Based on the equipment failure probability of each fire-fighting equipment in each fire-fighting subsystem, determine the system failure probability of each fire-fighting subsystem, and determine the subsystem health status of each fire-fighting subsystem based on the system failure probability of each fire-fighting subsystem.

[0137] The comprehensive fire status assessment result of the converter station is determined based on the importance weight and health status of each fire protection subsystem.

[0138] In one possible implementation, the fire protection daily management module includes a converter station risk management module, which is used for:

[0139] Identify fire control business events for each risk control object of the converter station, wherein the risk control object includes at least one of fire-fighting operations, fire-fighting equipment, and fire-fighting vehicles;

[0140] The risk level of each fire control business event is determined based on at least two of the severity, occurrence, and detectability of each fire control business event, wherein the risk level includes high risk, medium risk, and low risk.

[0141] Based on the risk level of each fire control business event, a risk level identifier (risk identifier code) matching the risk level is set for the corresponding fire control business event, and the display of each fire control business event and its corresponding risk level identifier is controlled.

[0142] In one possible implementation, the risk level of each fire control operational event is determined based on at least two of the following: severity, occurrence, and detectability:

[0143] The risk level of each fire control business event is determined by comparing the risk priority coefficient corresponding to the product of severity, occurrence, and detectability of each fire control business event with the preset coefficient threshold.

[0144] or,

[0145] A severity-occurrence matrix, a severity-detectability matrix, or an occurrence-detectability matrix is ​​formed based on any two of the severity, occurrence, and detectability of each fire control business event. A risk matrix ranking is obtained based on the severity-occurrence matrix, severity-detectability matrix, or occurrence-detectability matrix. The risk level of each fire control business event is then determined based on the risk matrix ranking.

[0146] In one possible implementation, the fire protection daily management module includes a converter station risk management module, which is used for:

[0147] The management indicator score of each primary management indicator is determined based on the score and weight of the subordinate management indicator corresponding to each primary management indicator in the converter station. The primary management indicators include at least one of the following: fire equipment management indicator, fire operation management indicator, fire vehicle management indicator, and fire personnel management indicator.

[0148] The fire protection management status of the converter station is determined based on the weight of each primary management indicator and the score of the management indicator.

[0149] In one possible implementation, the fire emergency response module includes a fire emergency decision-making module, which is used for:

[0150] The system acquires primary video data sent by a primary camera group associated with a fire in the converter station. The primary video data includes video data of the fire from multiple perspectives captured by primary cameras at multiple locations.

[0151] When the first video data of the fire object from the first perspective, determined based on the first-level video data, does not meet the emergency auxiliary decision-making conditions, a first camera that captures the first video data is determined from the first-level camera group, and a second camera corresponding to the position of the first camera is determined from the second-level camera group that has a positional mapping relationship with the first-level camera group; wherein, the second-level camera group and the first-level camera group have a positional mapping relationship with the fire object's position as the same perspective alignment center.

[0152] Obtain second video data of the fire object from the first perspective corresponding to the second camera;

[0153] Fire emergency response is carried out based on the first-level video data and the second-level video data.

[0154] In one possible implementation, the fire emergency decision-making module is used to determine, based on the primary video data, that the first video data from the first perspective of the fire object does not meet the emergency auxiliary decision-making conditions, including at least one of the following:

[0155] It is determined that the resolution of the first video data is lower than the resolution threshold;

[0156] It is determined that the resolution of the first video data is lower than a resolution threshold;

[0157] It is determined that the key target object in the first video data is occluded for a duration exceeding a predetermined duration;

[0158] It is determined that the contribution of the first video data to joint emergency decision-making with other video data in the first-level video data is less than the contribution threshold.

[0159] In one possible implementation, the fire emergency decision-making module is used to determine a second camera corresponding to the position of the first camera based on a second-level camera group that has a positional mapping relationship with the first-level camera group, including:

[0160] Based on the location of the fire object, the secondary camera group is determined from the candidate camera group that has a positional mapping relationship with the primary camera group. A candidate camera group is determined with a fire monitoring position within the shared field of view of the primary camera group as the center of view. The shared field of view of the primary camera group includes at least one fire monitoring position, and the at least one fire monitoring position includes the location of the fire object.

[0161] The second camera corresponding to the position of the first camera is determined based on the secondary camera group.

[0162] In one possible implementation, the fire emergency decision-making module is used to determine a second camera corresponding to the position of the first camera based on a second-level camera group that has a positional mapping relationship with the first-level camera group, including:

[0163] Using the location of the fire object as the center of view, determine a reference camera that has a viewpoint candidate relationship with the cameras in the first-level camera group;

[0164] Acquire basic spatial data of the primary camera group and the reference camera, wherein the basic spatial data includes at least one of the following: camera position information, height information, angle information, field of view information, and environment information of the camera.

[0165] A graph neural network model is constructed based on the basic spatial data of the first-level camera group and the reference camera. Each camera in the first-level camera group and the reference camera serves as a node in the graph neural network model. The node features of each node include the basic spatial data of the corresponding camera. The edges between nodes indicate that there is a viewpoint candidate relationship between the cameras corresponding to that node. The edge weights are determined based on at least one of the following: viewpoint candidate range, relative distance, line-of-sight accessibility, and historical collaborative effectiveness score between the corresponding cameras.

[0166] Using the graph neural network model, a second-level camera group is formed by determining, based on the edge weights, cameras from the reference cameras that satisfy the position mapping relationship with each camera in the first-level camera group;

[0167] The second camera corresponding to the position of the first camera is determined based on the secondary camera group.

[0168] In one possible implementation, the fire emergency decision-making module is used to determine the second camera corresponding to the position of the first camera based on the secondary camera group, including:

[0169] The location of the first camera, the environmental information of the first camera, and the emergency auxiliary decision-making requirements are input into the graph neural network model to obtain at least one candidate camera that has a viewpoint candidate relationship with the first camera.

[0170] Determine the view complementarity score of each candidate camera relative to the first camera, wherein the view complementarity score is determined based on at least one of the following: the view candidate range of the candidate camera relative to the first camera, the missing information gain of the candidate camera under the environmental information, the predicted image quality of the candidate camera, and the cooperative feasibility of the candidate camera and the first camera.

[0171] The second camera corresponding to the position of the first camera is determined based on the complementarity score of the viewpoints of each candidate camera relative to the first camera.

[0172] In one possible implementation, the fire emergency decision-making module is used to determine a second camera corresponding to the position of the first camera based on the view complementarity score of each candidate camera relative to the first camera, including:

[0173] The candidate camera with the highest viewpoint complementarity score was selected as the second camera;

[0174] Alternatively, candidate cameras whose viewpoint complementarity scores are greater than the first score threshold can be identified as the second camera;

[0175] Alternatively, if the view complementarity scores of all candidate cameras are less than the second score threshold, the top N candidate cameras with the highest view complementarity scores are selected as the second camera. The second camera is used to synthesize a simulated viewpoint, which is used to synthesize the second video data, and N is a natural number.

[0176] In one possible implementation, the fire emergency decision-making module is used to acquire second video data of the fire object from the first perspective captured by the second camera, including:

[0177] Acquire at least one alternative video data of the fire target captured by the at least one second camera;

[0178] A third video data representing a first simulated perspective is synthesized based on the at least one candidate video data, and this third video data is used as the second video data, wherein the first simulated perspective is within the shared viewpoint range of the candidate video data captured by the at least one second camera; or...

[0179] A fourth video data of a second simulated perspective is synthesized based on the first video data and the at least one alternative video data, and the fourth video data is used as the second video data, wherein the second simulated perspective is within the shared perspective range of the alternative video data captured by the at least one second camera and the overlapping perspective range of the first perspective.

[0180] In one possible implementation, the fire emergency decision module is further used for:

[0181] Predict the development trend of the fire at the target location;

[0182] The key perspective is determined based on the fire development trend, wherein the key perspective is the perspective from which key video data needs to be captured relative to the fire object, or the key perspective is the perspective from the perspective of a primary camera that has failed relative to the fire object.

[0183] A backup camera corresponding to the key viewpoint is determined from the secondary camera group, and a preparation instruction is sent to the backup camera. The preparation instruction is used to instruct the backup camera to start or prepare to acquire video data of the fire object from the key viewpoint.

[0184] In one possible implementation, the fire emergency decision-making module is used to acquire second video data of the fire object from the first perspective captured by the second camera, including:

[0185] Control the identification correspondence between the device identifier of the second camera and the device identifier of the first camera carried in the second video data, and verify the validity of the second video data according to the identification correspondence;

[0186] And / or,

[0187] The control includes carrying a priority transmission identifier in the second video data, which indicates that the second video data should be transmitted to the converter station digital fire control system with high priority.

[0188] In one possible implementation, the converter station fire protection digital control system further includes a converter station inductance module, which is used for:

[0189] Based on the converter station's request for inclusion in the management system initiated by the converter station or based on the converter station's instruction to be included in the management system sent to the converter station, the target converter station that needs to be included in the management system and the site attribute information of the target converter station are determined, and the site identifier of the target converter station and the site attribute information are associated and stored in the list of managed converter stations.

[0190] Identify the equipment requiring pipe connection in the target converter station and the components requiring pipe connection within the equipment, and determine the data reporting mechanism for the equipment requiring pipe connection and / or the components requiring pipe connection;

[0191] A successful converter station connection notification is sent to the target converter station, and the device identifier of the device to be connected and the component identifier of the component to be connected are associated and stored with the station identifier; wherein, the successful converter station connection notification includes the identifiers of the device to be connected and the component to be connected, as well as the corresponding data reporting mechanism.

[0192] In one possible implementation, determining the equipment requiring pipe connection in the target converter station and the components requiring pipe connection within the equipment includes:

[0193] A configuration template for grid connection matching the site attribute information of the target converter station is determined, wherein the site attribute information includes at least one of site size, site distance, site priority, site attention, and site voltage level; the configuration template for grid connection includes the grid-connected equipment and grid-connected components in the configured converter station, as well as the data reporting mechanisms corresponding to the grid-connected equipment and / or the grid-connected components respectively.

[0194] The system determines the equipment and components requiring pipe connection in the target converter station according to the configuration in the pipe connection configuration template, and determines the data reporting mechanism corresponding to the equipment and / or components requiring pipe connection respectively; or, the system displays the pipe connection configuration template, responds to the template modification operation to modify the pipe connection configuration template, and determines the equipment and components requiring pipe connection in the target converter station according to the configuration in the modified pipe connection configuration template, and determines the data reporting mechanism corresponding to the equipment and / or components requiring pipe connection respectively.

[0195] In one possible implementation, the converter station fire protection digital control system further includes a converter station panoramic monitoring module, used for:

[0196] Identify the converter stations that have already been connected to the power grid, as well as the equipment and components within those stations that require connection to the power grid;

[0197] Based on the map data and site layout data of the existing converter stations, the equipment attribute data of the equipment to be connected to the pipeline, and the component attribute data of the components to be connected to the pipeline, digital twin models corresponding to the existing converter stations, the equipment to be connected to the pipeline, and the components to be connected to the pipeline are constructed respectively.

[0198] The panoramic monitoring data of the existing converter station is displayed based on the corresponding digital twin models of the existing converter station, the equipment to be connected to the pipeline, and the components to be connected to the pipeline.

[0199] Secondly, a digital fire protection control system for converter stations is provided, comprising a digital fire protection control platform for converter stations and at least one converter station, wherein:

[0200] The at least one converter station is used to acquire the fire protection IoT data of the converter station and send the acquired fire protection IoT data to the fire protection control platform of the converter station;

[0201] The digital platform for fire control of the converter station is used for: acquiring fire IoT data sent by the converter station; using an anomaly detection model to detect anomalies in the acquired fire IoT data; wherein the anomaly detection model is trained based on normal time-series samples generated during the operation of the converter station and anomaly time-series samples generated based on the normal time-series samples, the anomaly time-series samples including at least one of trend anomaly samples, fixed deviation anomaly samples, accuracy decline anomaly samples, constant value anomaly samples, and discrete point anomaly samples; performing daily fire management of the converter station based on the fire IoT data of the converter station, the daily fire management including at least one of fire equipment management, fire business management, fire personnel management, station-side key work equipment management, converter station risk management, and fire hazard management; and identifying converter stations that have experienced emergencies or have potential emergencies based on at least one of the fire IoT data of the converter station and the management results of the daily fire management module, and performing fire emergency response on the converter stations that have experienced emergencies or have potential emergencies.

[0202] Thirdly, a digital management and control method for fire protection in converter stations is provided, the method comprising:

[0203] The fire protection IoT data of the converter station is acquired, and an anomaly detection model is used to detect anomalies in the acquired fire protection IoT data. The anomaly detection model is trained based on normal time-series samples generated during the operation of the converter station and anomaly time-series samples generated based on the normal time-series samples. The anomaly time-series samples include at least one of the following: trend anomaly samples, fixed deviation anomaly samples, accuracy reduction anomaly samples, constant value anomaly samples, and discrete point anomaly samples.

[0204] The fire protection management of the converter station is carried out based on the fire protection IoT data of the converter station. The daily fire protection management includes at least one of the following: fire protection equipment management, fire protection business management, fire protection personnel management, station-side key work equipment management, converter station risk management, and fire hazard management.

[0205] Based on at least one of the converter station's fire protection IoT data and daily fire protection management results, identify converter stations that have experienced or are showing signs of potential danger, and carry out fire emergency response for these stations.

[0206] Fourthly, a computer storage medium is provided, wherein a computer program is stored in the computer storage medium, and when the computer program is executed, the steps of the converter station fire digital control method as described in the third aspect are implemented.

[0207] Fifthly, a computer program product is provided, the computer program product comprising a computer program, which, when executed, implements the steps of the converter station fire protection digital control method as described in the third aspect.

[0208] The converter station fire protection digital management and control system and method of this application embodiment can realize daily fire protection management and fire emergency response of the converter station by acquiring fire protection IoT data from the converter station. That is, it realizes digital fire safety management and control of the converter station, enhances the digital fire protection capabilities of the converter station, and improves the efficiency of fire safety management and control. Furthermore, the converter station fire protection digital management and control system can perform comprehensive fire safety management and control of the converter station, including fire equipment management, fire business management, fire personnel management, management of key station-side equipment, fire hazard management and control, and fire emergency response, enriching the fire safety management and control functions and capabilities of the converter station and further improving the digitalization of fire safety management and control of the converter station. Furthermore, the converter station's digital fire protection management and control system uses an anomaly detection model to detect abnormal data in the converter station's fire protection IoT data. This avoids risks such as errors in fire emergency decision-making or missed / false alarms in fire safety incidents caused by abnormal data, thus improving the accuracy and effectiveness of fire safety management and control at the converter station. Moreover, by generating various abnormal time-series samples that match the characteristics of the fire protection IoT devices in the converter station from normal time-series samples during operation, the system trains the anomaly detection model. This addresses the problem of insufficient abnormal samples and allows for targeted detection of different types of abnormal data, thereby improving the accuracy of anomaly detection and further enhancing the effectiveness of fire safety management and control at the converter station. Attached Figure Description

[0209] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0210] Figure 1 This is a schematic diagram illustrating an application scenario applicable to an embodiment of this application;

[0211] Figure 2 This is a schematic diagram of the converter station fire protection digital control system in the embodiments of this application;

[0212] Figure 3 This is a schematic diagram of the deployment of fire protection IoT devices in the converter station in this application embodiment;

[0213] Figure 4 This is a flowchart illustrating the process of obtaining the abnormal data detection model in an embodiment of this application;

[0214] Figure 5 This is a schematic diagram of the hybrid expert model in the embodiments of this application;

[0215] Figure 6 This is a schematic diagram of the process for assessing the health status of fire protection IoT devices in an embodiment of this application;

[0216] Figure 7 This is a schematic diagram of the converter station fire protection digital control system in the embodiments of this application;

[0217] Figure 8 This is a schematic diagram illustrating the process of determining the optimal power of the electric heating tape for fire-fighting pipelines in an embodiment of this application.

[0218] Figure 9 This is a schematic diagram of a numerical model of a fire-fighting pipeline containing an electric heating cable in an embodiment of this application;

[0219] Figure 10 This is a fitted graph of the power function of the electric heating cable in the embodiments of this application;

[0220] Figure 11 This is a flowchart illustrating the leak detection method for fire-fighting foam pipelines in converter stations according to an embodiment of this application.

[0221] Figure 12 This is a schematic diagram of the fire-fighting foam pipe model in the embodiments of this application;

[0222] Figure 13 This is a flowchart illustrating the converter transformer tank failure prediction method in the embodiments of this application;

[0223] Figure 14 This is a schematic diagram illustrating the acquisition of the temperature matrix, time matrix, and physical matrix in an embodiment of this application.

[0224] Figure 15 This is a schematic diagram of the failure fault tree of the fire protection system in the embodiments of this application;

[0225] Figure 16 This is a flowchart illustrating the video-based emergency response method for converter stations in this application.

[0226] Figure 17 This is a schematic diagram illustrating the deployment of the primary camera group and the secondary camera group in an embodiment of this application. Detailed Implementation

[0227] To make the technical solutions and advantages of the embodiments of this application clearer, the exemplary embodiments of this application are described in further detail below. Obviously, the described embodiments are only a part of the embodiments of this application, and not an exhaustive list of all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.

[0228] To better understand the technical solutions provided in the embodiments of this application, the application scenarios applicable to the technical solutions provided in the embodiments of this application are briefly introduced below. It should be noted that the application scenarios described below are only for illustrating the embodiments of this application and are not intended to limit the scope. In specific implementation, the technical solutions provided in the embodiments of this application can be flexibly applied according to actual needs.

[0229] Please see Figure 1 As shown, Figure 1 As an application scenario applicable to the technical solution of this application embodiment, the converter station fire protection digital control system in this application can remotely manage and control the fire safety of converter stations based on a three-level architecture of "headquarters-provincial-station". In this three-level architecture, "headquarters" can refer to the power grid system headquarters, "provincial" can refer to the provincial unit of the power grid system, and "station" can refer to converter stations deployed in various locations. Wherein, as... Figure 1 As shown, the "station side" is deployed with multiple converter stations, such as converter stations 1-n. Each converter station can connect to its respective provincial unit, for example, converter stations 1-2 connect to Province A, converter stations 3-4 connect to Province B, etc. On the other hand, each converter station can also directly connect to the converter station fire protection digital control system deployed at the "headquarters". In this way, each converter station can simultaneously report the fire protection IoT data in the converter station to its respective provincial unit and the converter station fire protection digital control system at the headquarters. In addition, the converter station's respective provincial unit and the converter station fire protection digital control system at the headquarters can also send relevant notifications and instructions to each converter station. For example, when a fire protection device in a converter station malfunctions or a fire hazard occurs in a converter station, the "headquarters" and the "provincial side" can send fire protection control-related instructions to the converter station, thereby realizing remote fire safety control of the converter station.

[0230] The "headquarters" in the above three-tier architecture of "headquarters-province side-station side" can also be understood as the "platform side". In this case, the corresponding converter station fire protection digital control system can also be called the converter station fire protection control platform.

[0231] Furthermore, the aforementioned three-tier architecture of "headquarters-provincial-station" can adopt a cloud-edge collaborative technical architecture. The "headquarters" and "provincial" can deploy the converter station's digital fire control system in the cloud, while the converter station in the "station" is deployed at the edge. Based on the converter station's cloud-edge collaborative technical architecture, multi-level fire emergency auxiliary decision-making can be achieved from the power grid headquarters to the provincial power grid units to the converter station. Fire data, fire resource data, and fire equipment status data from the converter station are transmitted from the converter station to the provincial units and headquarters. The provincial units and headquarters can make fire emergency decisions based on this data, realizing multi-level unified decision-making and online dispatch of digital fire emergency response at the converter station, thereby improving the efficiency of fire emergency response.

[0232] In another implementation, such as Figure 1 The digital fire control system for converter stations deployed at headquarters can also be deployed simultaneously in provincial units, for example... Figure 1 The converter station fire protection digital management and control system is deployed in provinces A, B and C. Each provincial unit can use the locally deployed converter station fire protection digital management and control system to manage the fire safety of its subordinate converter stations.

[0233] based on Figure 1 For the application scenarios applicable to the embodiments of this application shown, please refer to [the relevant documentation]. Figure 2 As shown, the converter station fire protection digital control system of this application includes a converter station data processing module, a fire protection daily management module, and a fire emergency response module. Based on these modules, the converter station fire protection digital control system can perform intelligent remote fire protection digital control of at least one converter station deployed at the end side, thereby improving the digital fire protection capabilities of the converter station and increasing the fire protection control efficiency of the converter station.

[0234] The converter station data processing module in the converter station fire protection digital control system is used to acquire fire protection IoT data of the converter station and use a data anomaly detection model to detect anomalies in the acquired fire protection IoT data. The data anomaly detection model is trained based on normal time-series samples generated during the operation of the converter station and abnormal time-series samples generated based on the normal time-series samples. The abnormal time-series samples include at least one of the following: trend anomaly samples, fixed deviation anomaly samples, accuracy reduction anomaly samples, constant value anomaly samples, and discrete point anomaly samples.

[0235] Fire protection IoT data primarily originates from data measured by monitoring sensors and data representing sensor status values. The health status of this data directly affects the safe and stable operation of the system. Standardized fire protection IoT data can be represented in JSON format, containing four parts: device information, monitoring data, device status, and communication information. Device information may include device identifier, device name, data code, device manufacturer, device model, and device installation location; monitoring data may include data acquisition timestamp, sensor type, monitored value, and unit of measurement; device status may include device operating status, battery level, and signal strength; and communication information may include communication protocol type and network type.

[0236] Currently, the types of telemetry and teleindication data in fire protection IoT are diverse. For example, teleindication data can include equipment status information, such as the working status information of fire doors, valves, sprinkler systems, exhaust fans, etc., as well as alarm information generated by detectors, control command execution status information, equipment maintenance and fault records, personnel access control information, system operation logs, etc. Telemetry data can include indoor and outdoor fire hydrant water pressure, sprinkler terminal water pressure, water level in water tanks and fire water tanks, gas cylinder pressure, smoke exhaust system flow data, temperature and smoke concentration data generated by heat and smoke detectors, spectral and radiation data detected by flame detectors, current and voltage of fire protection power supply systems, wind force, wind direction, air pressure and temperature data at converter stations, on-site images, and concentration data of toxic and hazardous substances, etc.

[0237] The daily fire management module within the converter station's digital fire protection management system is used to perform daily fire protection management of the converter station based on its fire protection IoT data. This daily fire protection management includes at least one of the following: fire equipment management, fire service management, fire personnel management, management of key station-side equipment, converter station risk management, and fire hazard management. In other words, the daily fire protection management module can manage and control fire equipment, fire services, fire personnel, key equipment, and fire hazard management within the converter station. During implementation, the corresponding management capabilities can be achieved through specific functional modules. For example, fire equipment management can be achieved through the fire equipment management module within the daily fire protection management module, and fire hazard management can be achieved through the fire hazard management module within the daily fire protection management system. The converter station's digital fire protection management system, through its daily fire protection management module, provides comprehensive daily fire protection management of the converter station, including fire equipment, fire personnel, and fire service personnel, enriching the fire safety management capabilities of the converter station and enhancing the digitalization of fire safety management.

[0238] The fire emergency response module in the converter station fire digital control system is used to identify converter stations that have experienced a hazard or have a potential hazard based on at least one of the converter station's fire IoT data and the management results of the fire daily management module, and to carry out fire emergency control and management for the converter stations that have experienced a hazard or have a potential hazard.

[0239] In this embodiment, the converter station can report its fire protection IoT data to the converter station's fire protection digital management system via a data interface. This fire protection IoT data includes data related to fire protection within the converter station, which can be collected by the fire protection IoT devices within the converter station, such as... Figure 3As shown, various fire-fighting IoT devices (such as sensors and cameras) can be deployed in converter stations. These devices can form a fire-fighting IoT network, collecting various types of fire-related data from the converter station. For example, fire-fighting IoT data includes status data of fire-fighting equipment in the converter station, such as the opening status of fire hydrants, water level data of fire pools, and pressure and temperature data of fire pipelines. Fire-fighting IoT data can also include fire-related environmental data in the converter station (such as environmental image data, environmental video data, environmental temperature, and smoke signal data). Furthermore, fire-fighting IoT data can include status data of critical operating equipment in the converter station. Critical operating equipment refers to electrical equipment used for power transmission in the converter station. Failures or abnormalities in this critical operating equipment may cause fires, so real-time monitoring is necessary. Critical operating equipment in the converter station includes, for example, converter valves and converter transformers.

[0240] Because the fire protection IoT devices in converter stations inevitably experience aging, interference, and malfunctions during long-term use, the fire protection IoT data collected and reported to the converter station's fire protection digital control system may also be abnormal, such as data mutations, drift, or failure of constant values. This can lead to errors in fire emergency decision-making or missed / false alarms of fire safety incidents when the converter station's fire protection digital control system relies on this fire protection IoT data for fire safety management, resulting in serious fire safety accidents. Therefore, in this embodiment of the application, after obtaining the fire IoT data reported by the converter station, the converter station fire protection digital control system can use a pre-trained data anomaly detection model to detect anomalies in the acquired fire IoT data. This allows for the identification of abnormal data within the fire IoT data. Furthermore, based on these anomalies, the system can determine whether the corresponding fire IoT equipment has malfunctioned or is abnormal. It can also analyze the equipment malfunction / abnormality type based on the characteristics of the anomalies. Alternatively, it can process the identified anomalies accordingly, such as removing them or restoring the data based on the anomalies and their types, to obtain valid fire IoT data. Finally, based on this valid fire IoT data, the system can perform corresponding fire safety control of the converter station. This improves the accuracy and effectiveness of fire safety control.

[0241] Furthermore, the data anomaly detection model in this embodiment is trained based on normal time-series samples generated during the operation of the converter station and anomaly time-series samples generated based on these normal time-series samples. The anomaly time-series samples generated from the normal time-series samples include one or more of the following: trend anomaly samples, fixed deviation anomaly samples, accuracy degradation anomaly samples, constant value anomaly samples, and discrete point samples. In this way, multiple types of anomaly time-series samples can be generated based on normal time-series samples to address the problem of insufficient anomaly samples in the actual training data. Moreover, the anomaly samples are generated according to the types of anomalies that are commonly encountered by the fire protection IoT equipment in the converter station, making them more targeted. This makes subsequent anomaly detection of IoT fire protection data in the converter station more targeted, thus improving the effectiveness of anomaly data detection.

[0242] Furthermore, the abnormal data detection model in this embodiment is a supervised deep learning model. Compared with other supervised deep learning models, by constructing an abnormal sample generation model, abnormal time series data samples such as constant value anomalies, trend anomalies, fixed deviations, outlier anomalies, and accuracy degradation anomalies are generated on the basis of normal time series data. These samples are used to train the abnormal data detection model, which reduces the pressure of supervised learning on the demand for abnormal samples and also reduces the time for collecting abnormal samples during the actual operation of the converter station. This improves the model training speed while ensuring the accuracy of model training.

[0243] The converter station fire protection digital management and control system of this application embodiment enables daily fire protection management and emergency response of the converter station by acquiring fire protection IoT data from the converter station. In other words, it achieves digital fire safety management and control of the converter station, enhancing its digital fire protection capabilities and improving fire safety management efficiency. Furthermore, the converter station fire protection digital management and control system can perform comprehensive fire safety management and control of the converter station, including fire equipment management, fire service management, fire personnel management, management of key station-side equipment, fire hazard management and control, and emergency response, enriching the fire safety management and control functions and capabilities of the converter station and further improving the digitalization of fire safety management and control. Furthermore, the converter station's digital fire protection management and control system uses an anomaly detection model to detect abnormal data in the converter station's fire protection IoT data. This avoids risks such as errors in fire emergency decision-making or missed / false alarms in fire safety incidents caused by abnormal data, thus improving the accuracy and effectiveness of fire safety management and control at the converter station. Moreover, by generating various abnormal time-series samples that match the characteristics of the fire protection IoT devices in the converter station from normal time-series samples during operation, the system trains the anomaly detection model. This addresses the problem of insufficient abnormal samples and allows for targeted detection of different types of abnormal data, thereby improving the accuracy of anomaly detection and further enhancing the effectiveness of fire safety management and control at the converter station.

[0244] See Figure 4 The abnormal data detection model in this application embodiment can be trained in the following way:

[0245] S401: Obtain normal timing data generated during the operation of the converter station;

[0246] S402: Based on the sliding window and step size corresponding to the data type of the obtained normal time series data, the normal time series data is converted into normal time series samples. That is, for different types of data, the corresponding sliding window and step size can be determined according to the data characteristics of the data type to improve the adaptability of the data type.

[0247] S403: Generate abnormal time series samples based on the obtained normal time series samples, wherein, as mentioned above, the abnormal time series samples include at least one of the following: trend abnormal samples, fixed deviation abnormal samples, accuracy reduction abnormal samples, constant value abnormal samples, and discrete point abnormal samples.

[0248] S404: Train a hybrid expert model, which includes at least two sub-models, using normal time-series samples and abnormal time-series samples to obtain a trained abnormal data detection model.

[0249] This application's embodiments design an anomaly generation algorithm with adjustable anomaly degree and start / end time, and then construct a hybrid expert model with strong generalization ability, ultimately achieving high-precision and high-robustness detection of fire-fighting time-series data anomalies.

[0250] For details, please refer to [link / reference]. Figure 5 The hybrid expert model in this embodiment includes at least two sub-models selected from CNN, GRU, LSTM, and Transformer sub-models; and the anomaly detection model further includes a gating network, which is used to assign corresponding weights to each sub-model according to the data type of the input data; each sub-model is used to obtain the corresponding anomaly detection result based on the input data and the assigned weights; wherein, the anomaly detection result of the anomaly detection model is obtained by weighting the anomaly detection results of each sub-model.

[0251] In this embodiment of the application, the above-mentioned Figure 4 In S403, the specific process of generating five types of abnormal time series samples based on normal time series samples—trend abnormal samples, fixed deviation abnormal samples, accuracy decline abnormal samples, constant value abnormal samples, and outlier abnormal samples—is described below.

[0252] (1) Samples with abnormal trends

[0253] Causes of trend anomalies include calibration errors and hardware aging. Trend anomalies can be defined as the sensor output telemetry data gradually increasing or decreasing at a constant rate, expressed as:

[0254]

[0255] In the formula, y1[i] represents the abnormal trend sample; x[i] represents the normal time series sample; trend is the added linear trend term, trend=a·(i-t1+1), a is the trend coefficient, a=(R / t1)·r1·(n / 10), R is the initial range, r1 is the trend direction adjustment factor, n is the abnormality intensity coefficient, t1 is the starting point of the abnormal interval, t1~Uniform{2,…l-1} means selecting any integer from it, l is the sequence length of the original normal time series sample, and i is the sample of the abnormal interval; σ x This represents the initial standard deviation of the normal time series samples; For random perturbation,

[0256] Specifically, the steps for generating trend anomaly samples are as follows:

[0257] Input: Original normal time series data x, anomaly intensity coefficient n;

[0258] Output: A sequence y containing trend anomalies;

[0259]

[0260] It should be noted that the trend anomaly generation step used in this embodiment, based on a mathematical model, allows for parameter adjustment to control the "degree" and starting point of the anomaly, thus covering trend anomalies of varying degrees and collection times. This is because trend a is jointly determined by the fluctuation degree of the normal time series and the anomaly intensity coefficient, and the final noise term is also determined by the fluctuation degree of the normal time series, making it more consistent with the normal time series pattern.

[0261] (2) Abnormal samples with fixed bias

[0262] The causes of fixed data deviation include short circuits in the sensing unit or calibration errors. Fixed deviation is defined as a constant deviation in the sensor's output telemetry signal, expressed as:

[0263] y2[i]=x[i]+x1

[0264] In the formula, y2[i] represents the fixed deviation abnormal sample; x[i] represents the normal time series sample; and x1 represents the injected fixed deviation.

[0265] Specifically, the steps for generating fixed-bias outlier samples are as follows:

[0266] Input: Original data sequence x, anomaly intensity coefficient n;

[0267] Output: A sequence y containing a fixed bias anomaly;

[0268]

[0269] It should be noted that the constant term of the fixed bias is determined by both the normal time series and the anomaly intensity coefficient. The maximum fixed bias can be determined manually based on the parameters. In addition, a random adjustment factor is introduced to augment the data and prevent sample duplication. Moreover, the algorithm also randomizes the start and end of anomaly generation to increase sample diversity. After repeatedly running the algorithm, fixed bias anomaly time series with different start and end times and different "degrees" can be obtained.

[0270] (3) Abnormal samples with decreased accuracy

[0271] Causes of accuracy degradation include poor circuit connections, high-frequency noise in the system, or physical damage to the sensor. Accuracy degradation is defined as the addition of zero-mean and high-variance noise to the output telemetry data, expressed by the following formula:

[0272] y3[i]=x[i]+ε

[0273] In the formula, y3[i] represents abnormal samples with decreased accuracy; x[i] represents normal time-series samples; ε represents Gaussian white noise, ε ~ Normal(0,σ) 2 ) represents selecting any composite uniformly distributed value (non-integer) from the range, where σ is the noise amplitude, σ=β·r, β is the reference amplitude, and r is the random amplitude adjustment factor.

[0274] Specifically, the steps for generating anomalous samples with decreased accuracy are as follows:

[0275] Input: Original data sequence x, anomaly intensity coefficient n;

[0276] Output: The sequence y containing the anomaly of decreased precision;

[0277]

[0278] in, Let L represent the time series dimension and L represent the time series length. In single-dimensional time series anomaly detection... =1, L is the timing window length, This means that the shape of the generated ε is the temporal dimension multiplied by the temporal length. Since the dimension is 1, the shape is 1 multiplied by the temporal length.

[0279] It should be noted that the essence of accuracy degradation is that, even when the physical quantity itself does not fluctuate significantly, sensor malfunctions cause large fluctuations in readings under the same environmental conditions. Therefore, the noise that can be controlled is a characteristic of the accuracy degradation fault itself. The anomaly generation algorithm used in this embodiment is based on a mathematical model, and is implemented automatically with controllable "degree" of degradation, introducing randomness for data augmentation.

[0280] (4) Abnormal samples of constant values

[0281] The causes of constant value anomalies include sensor short circuits, circuit breaks, or sensor component damage, which completely malfunction the IoT telemetry data, locking it at a fixed value. This anomaly may occur temporarily or permanently, but both are detrimental to the system. A constant value anomaly is defined as follows:

[0282] y4[t1:l]=x[t1]

[0283] In the formula, y4[t1:l] represents the constant value abnormal sample; t1 is the random starting point, t1~Uniform{2,…l-1} means selecting any integer from it, l is the sequence length of the original normal time series sample; x[t1] means that the normal time series sample is set to a constant value starting from t1.

[0284] The steps for generating constant value outlier samples are as follows:

[0285] Input: Original data sequence x;

[0286] Output: A sequence y containing a constant anomaly;

[0287]

[0288]

[0289] It should be noted that, considering that anomaly identification should provide early warning, this embodiment selects the first half as the normal time sequence and the second half as the abnormal form. At the same time, the location of constant value anomalies can be randomly selected to enhance data diversity.

[0290] (5) Outlier samples

[0291] For outlier anomalies, a large spike in IoT telemetry data occurring at a constant or non-constant time interval is defined as:

[0292] y5[i]=x[i]+Δ

[0293] In the formula, y5[i] represents outlier samples; x[i] represents normal time series samples; Δ represents outlier perturbation, Δ=(1+N)·μ·(n / 10 4N is the standard normal perturbation, N~Normal(0,1) represents any composite uniform distribution value (non-integer) selected from the range, μ is the global mean benchmark of normal time series samples, and n is the anomaly intensity coefficient.

[0294] The steps for generating outlier samples are as follows:

[0295] Input: Original data sequence x, anomaly intensity coefficient n, anomaly probability coefficient m;

[0296] Output: The sequence y containing outlier anomalies;

[0297]

[0298] It should be noted that the value of the anomaly intensity coefficient depends on the normal time series itself. The minimum value of the intensity coefficient requires that there are anomalous features that are significantly different from the characteristics of the normal time series itself, while the maximum value has almost no limit; it's just that the larger it is, the more obvious it becomes. If it is large enough, it will exceed the detector threshold, and non-deep learning algorithms, such as thresholding methods and three-standard-deviation methods, can accurately identify it. In this experiment, all anomaly intensity coefficients were set to 100.

[0299] Because different densities of outliers may be found during the fire protection IoT data collection process, such as a single outlier or continuous jumps, this embodiment enhances the diversity of the algorithm's generated abnormal time sequences by controlling the degree, number, and location of outliers, and by randomly selecting the degree and location of outliers.

[0300] It should be noted that this embodiment selects these five anomaly types based on the characteristics of telemetry signals from fire protection equipment. By observing the telemetry data of fire protection IoT sensors used in converter stations, such as those for temperature, humidity, water pressure, and water level, the abnormal time series that can be collected can generally be included in these categories. First, outlier anomalies are the most common anomalies. Decreased accuracy can also occur in various time series sensors. Fixed deviations and trend anomalies will significantly deviate from the sensor's change pattern. For example, temperature sensors and water pressure sensors will show obvious daily periodic changes. Deviating from this daily trend can be identified as abnormal time series, even though their changes will still be continuous. As for constant value anomalies, observing the telemetry time series of fire protection IoT equipment in converter stations reveals that although changes in water level and other parameters are sometimes very small, they will fluctuate slightly within a certain range, and long-term constant values ​​are rare. Therefore, constant values ​​are not normal time series either.

[0301] See also Figure 5Among them, Convolutional Neural Networks (CNNs) are deep learning models specifically designed for processing data with a grid structure (such as images and time series). Their basic structure typically includes convolutional layers, pooling layers, and fully connected layers. In the process of temporal anomaly detection, CNNs excel at capturing local features and spatial dependencies.

[0302] CNN model parameters: sub-expert learning rate of 0.001, two convolutional layers, kernel size of 3, padding of 1, pooling layer size of 2, stride of 2.

[0303] The Gated Recurrent Unit (GRU) is an improved recurrent neural network (RNN) architecture specifically designed for processing sequential data (such as time series and text). By introducing a gating mechanism, GRU addresses the vanishing and exploding gradient problems that traditional RNNs often encounter during long sequence training, while simultaneously reducing model parameters and improving training efficiency. In the process of temporal anomaly detection, GRU effectively handles the temporal dependencies in long sequence data.

[0304] GRU model parameters: sub-expert learning rate of 0.001, hidden layer dimension of 128, number of hidden layers of 2.

[0305] LSTM, first proposed by Hochreiter and Schmidhuber, is an excellent variant of recurrent neural networks. It not only inherits most of the advantages of recurrent neural networks but also solves the gradient vanishing problem caused by gradual reduction during gradient backpropagation. It is very suitable for processing time-series data, such as speech, text, video, and sensor data. In the process of time-series anomaly detection, it can effectively handle the temporal dependencies in time-series data, just like GRU.

[0306] LSTM model parameters: sub-expert learning rate of 0.001, hidden layer dimension of 128, number of hidden layers of 2.

[0307] The Transformer model uses a multi-head attention mechanism instead of the traditional recurrent neural network (RNN) structure. Compared to RNNs, it can better solve long-range dependency problems, and due to the parallelism of the attention mechanism, it outperforms many RNN-based models in terms of runtime efficiency. In the temporal anomaly detection process, the Transformer captures global dependencies through a self-attention mechanism.

[0308] Transformer model parameters: sub-expert learning rate of 0.001, includes an embedding layer to map input features to a 64-dimensional space, includes a positional encoding layer to add positional information to the input sequence, and a Transformer encoder consisting of two encoder layers, each with eight attention heads.

[0309] It's important to note that Mixture of Experts (MoE) is an ensemble learning method that solves complex tasks by combining multiple sub-models (called "experts"). Each expert model focuses on processing a specific part or feature of the input data, while a gating network dynamically allocates input data to the most suitable expert model. For example, the gating network can assign corresponding weights to each sub-model based on the data type. The core idea of ​​MoE is to leverage the strengths of each sub-model through division of labor and collaboration, thereby improving the overall model's performance and generalization ability. It should be understood that in the hybrid expert-based fire protection IoT data anomaly detection model, sub-experts can be added to or replaced with other neural networks, such as RNNs, RBFs, BP neural networks, and support vector machines.

[0310] Expert sub-models: Each expert is an independent deep learning network used to extract features from different patterns in time series data. The "features" mentioned above are abstract concepts, such as local anomaly features and temporal variation features. The output of each sub-model is an unnormalized class score. The output of the hybrid expert model is a weighted probability distribution of the sub-model class scores, normalized using softmax. The class with the highest probability distribution value is used as the classification result. In this embodiment, the expert sub-models include LSTM, CNN, GRU, and Transformer, each used to capture different feature patterns in the time series data.

[0311] Gating Network: A gating network dynamically assigns weights to each expert model based on the characteristics of the input data. Its output is a probability distribution representing the importance of each expert model to the current input. Each sub-model (i.e., each expert model) obtains its corresponding anomaly detection result based on how the weights are allocated to the data.

[0312] Weighted Output: The final output is a weighted sum of the outputs of each expert model. In other words, the anomaly detection results of each sub-model are summed in weights to obtain the anomaly detection result of the anomaly detection model. As mentioned above, the weights are determined by the gating network. This approach flexibly combines the advantages of multiple expert models, adapts to different input data distributions, and has strong universality.

[0313] During training, the gating network has a learning rate of 0.001 (a higher learning rate allows it to adapt to weight allocation tasks more quickly), and the number of training steps is 2000. The 2000 training steps allow each deep learning model to be fully trained in various abnormal training scenarios.

[0314] The hybrid expert model MoE constructed in this embodiment can handle complex input data distributions through the collaborative work of multiple expert sub-models. Each expert sub-model focuses on specific features or patterns, thereby improving the model's expressive power. The gating network can dynamically adjust the weights of each expert sub-model according to the characteristics of the input data, enabling the model to adaptively handle different data distributions. In addition, the MoE structure has good scalability, allowing the number of expert models to be easily increased or decreased to adapt to different task requirements.

[0315] As a further technical solution, when training the anomaly detection model, the loss function of each sub-model includes a binary cross-entropy loss for binary classification tasks and a multivariate cross-entropy loss for multi-anomaly classification tasks; specifically:

[0316] In binary classification tasks, the loss function for all sub-models is the binary cross-entropy loss (BCE Loss), the formula of which is:

[0317]

[0318] Among them, y i For real labels, σ represents the model's original output (logits), and σ is the Sigmoid function.

[0319] In multi-anomaly classification tasks, normal samples are labeled 0, and anomaly samples are labeled 1 to 5. Therefore, deep learning is a 6-class classification task, and the loss function is the multivariate cross-entropy loss, with the following formula:

[0320]

[0321] Where, ω c The preset category weights are typically the reciprocal of the category frequencies.

[0322] As a further technical solution, in the MoE model, the total loss is:

[0323]

[0324] Where Weighted Output Loss is the BCE or CE loss of the MoE weighted output, and Expert Loss is...k Let λ be the BCE or CE loss of the k-th sub-model, and λ be the weight coefficient of the sub-model loss, which is 0.1 in this embodiment.

[0325] This loss function design allows each sub-model to approach the optimal solution while maintaining the contribution of the overall MoE model, thus achieving excellent performance in the task of detecting anomalies in fire protection IoT data in converter stations.

[0326] Furthermore, this embodiment uses CNN, LSTM, GRU, and Transformer networks as sub-models to construct a hybrid expert (MoE)-based abnormal data detection model for fire protection IoT data. The parameters of each sub-network, the loss weights of each sub-network, and the temperature parameters of the gating network are set, and the model is trained. Then, the effectiveness of the model is verified using real-world collected normal and abnormal samples. In subsequent use, real-world collected abnormal samples can be added to the training set to improve the accuracy of anomaly detection.

[0327] It's important to note that during model training, it's easy to get stuck in local optima. Therefore, a temperature parameter is added to the gating network. A parameter greater than 1 results in a more even distribution of weights in the gating network, while a parameter closer to 0 results in a more concentrated weight distribution. During model training, the initial temperature parameter is set to 2 to ensure even exploration of the performance of each model. The final temperature parameter is set to 0.5, allowing the model to select the optimal sub-model for output after inputting time-series data. The temperature parameter decreases linearly with each epoch during training.

[0328] Furthermore, after the hybrid expert model is trained, an effectiveness test can be conducted. If the test is successful, the real-time data obtained by the converter station fire protection digital control system can be input into the model to detect abnormal data, thereby improving the accuracy and efficiency of data anomaly detection.

[0329] In one embodiment of this application, the data access quality compliance condition of the fire protection IoT data of the converter station can be judged, and then, when the fire protection IoT data meets the data access quality compliance condition, the health status of the fire protection IoT equipment corresponding to the fire protection IoT data can be evaluated. Specifically, as shown in the following example... Figure 6 As shown:

[0330] S601. Obtain real-time time-series data during the operation of fire protection IoT equipment;

[0331] It should be noted that the time-series data includes telemetry time-series data and / or remote signaling data. In this embodiment, the time-series data generated during the operation of the fire protection IoT equipment can be obtained through the converter station's fire protection digital management and control system. Physical quantities are collected at the process layer and uploaded to the converter station's fire protection digital management and control system via different protocols and formats. The telemetry time-series data and remote signaling signals sent by the converter station's fire protection IoT equipment are extracted from the converter station's fire protection digital management and control system. The collected telemetry / remote signaling data forms the data basis for evaluating the health status of the fire protection IoT equipment in this embodiment.

[0332] S602. Determine the integrity score, validity score, and timeliness score of the real-time time series data corresponding to the fire protection IoT data of the converter station;

[0333] S603. Based on the integrity score, validity score, and timeliness score of the real-time time series data, determine the comprehensive quality score of the real-time time series data. For example, the comprehensive quality score of the real-time time series data is obtained by weighted summation of the integrity score, validity score, and timeliness score.

[0334] S604. Determine whether the fire protection IoT data meets the data access quality compliance conditions based on the comprehensive quality score of the real-time time series data.

[0335] It should be noted that the real-time time series samples (which consist of the latest time step data and historical time step data that meet the sample window length) are evaluated by the comprehensive quality of each time step in the sample through completeness, validity (calculated by the Transformer reconstruction model after training on historical data), and timeliness. Then, real-time time series samples are constructed based on the number of data of different quality levels.

[0336] S605. If the fire protection IoT data meets the data access quality compliance conditions, the quality level of the fire protection IoT data is determined based on the comprehensive quality score, and the first data quality distribution characteristics are obtained based on the quality level of the fire protection IoT data.

[0337] S606. Map the first data quality distribution feature vector to a predetermined health baseline, for example, by performing a nonlinear mapping, to obtain the health status assessment result of the fire protection IoT equipment corresponding to the fire protection IoT data.

[0338] Among them, the health baseline is obtained by clustering the second data quality distribution feature vector of the historical normal time series data of fire protection IoT equipment in different time windows. The second data quality distribution feature is determined by extracting data of different quality levels in the time window through a sliding window based on the comprehensive quality score of the historical normal time series data.

[0339] The above-described S601-S604 can be executed by the converter station data processing module in the converter station fire protection digital control system. By judging whether the data access quality compliance conditions are met through three dimensions—data integrity, validity, and timeliness—the accuracy of data access quality can be improved. The above-described S605-S606 can be executed by other functional modules (such as the fire equipment control module) in the converter station fire protection digital control system, and this application embodiment does not impose any restrictions on this.

[0340] It should be noted that this embodiment adopts an unsupervised method, extracts feature vectors based on the temporal distribution of data quality, and only uses normal time-series data generated during the normal operation of fire protection IoT equipment for training. The most representative healthy sample points are obtained through clustering as the health baseline, which is used to measure the distance between real-time time-series samples and healthy samples to assess the health status of fire protection IoT equipment. This solves the problems of poor transferability and weak interpretability of fire protection IoT equipment fault detection and health status assessment models.

[0341] In one possible implementation, the quality level of normal time series data is determined based on the comprehensive quality score of normal time series data, and the number of normal time series data of different quality levels within a time window is extracted by a sliding window to obtain the second data quality distribution feature vector within different time windows. Then, the second data quality distribution feature vector within different time windows is clustered, and the cluster center obtained by clustering is used as the health baseline.

[0342] It should be noted that this embodiment is based on an unsupervised deep learning method to construct a data quality assessment model for time-series data of fire protection IoT devices. The data quality is comprehensively evaluated based on the evaluation methods of the integrity, timeliness, and effectiveness of telemetry data and remote signaling data. The data quality assessment process for real-time time-series data and historical normal time-series data is the same.

[0343] In one possible implementation, this embodiment uses the importance weights obtained from the analytic hierarchy process (AHP) and the results of standardization checks to obtain a data integrity score for remote signaling data; it also obtains a data validity score based on a remote signaling signal validity evaluation model that combines long-term and short-term data transmission delays; and it obtains a data timeliness score based on data transmission latency. The specific implementation process is as follows:

[0344] (1) Calculate the integrity score of remote information data

[0345] The data uploaded to the converter station's digital fire protection control system has been standardized in format and includes four primary indicators: equipment information, monitoring data, equipment status, and communication information. Equipment information includes: equipment identifier, equipment name, data code, equipment manufacturer, equipment model, and equipment installation location. Monitoring data includes: data acquisition timestamp, sensor type, monitored value, and unit of measurement. Equipment status includes: equipment operating status, battery level, and signal strength. Communication information includes: communication protocol type and network type. Therefore, the completeness quantification indicators are divided into 4 primary indicators and 15 secondary indicators.

[0346] This embodiment uses the Analytic Hierarchy Process (AHP) to obtain importance weights for data information, which are used to measure the degree of importance of content. Specifically:

[0347] In the Analytic Hierarchy Process (AHP), the judgment matrix A is used to measure the importance of indicators, and it is obtained as follows:

[0348]

[0349] The elements in the judgment matrix A satisfy: a ij >0; a ii =1; Determine the value a of an element in the matrix. ij According to Table 1:

[0350] Table 1 shows the values ​​of the judgment matrix.

[0351]

[0352] The weight vector W of the indicator can be obtained by the following formula:

[0353]

[0354] After calculating the indicator weights, a consistency check is required. The consistency check first requires calculating the largest eigenvalue λ of the matrix. max :

[0355]

[0356] Secondly, the consistency index CI is calculated using the largest eigenvalue:

[0357]

[0358] Where CI = 0 indicates that the judgment matrices are completely consistent, and the larger CI is, the more serious the inconsistency of the judgment matrices.

[0359] Then, based on CI and RI, the CR value is calculated to determine whether the consistency is passed. RI is obtained by looking up Table 2.

[0360] Table 2. Relationship between R.I. values ​​and matrix order

[0361]

[0362] Finally, the CR calculation method is as follows:

[0363]

[0364] When CR < 0.1, it indicates that the consistency of the judgment matrix A is considered to be within the acceptable range, and the weight vector can be calculated using the eigenvectors of A. If CR ≥ 0.1, then the judgment matrix A should be modified.

[0365] The complete primary indicator judgment matrix is ​​shown in Table 3:

[0366] Table 3. Integrity Level 1 Indicator Judgment Matrix

[0367] Equipment Information Monitoring data Equipment status Communication Information Equipment Information 1.00 0.50 1.00 2.00 Monitoring data 2.00 1.00 2.00 3.00 Equipment status 1.00 0.50 1.00 2.00 Communication Information 0.50 0.33 0.50 1.00

[0368] The judgment matrices for each primary indicator are shown in Tables 4 to 7:

[0369] Table 4 Equipment Information Judgment Matrix

[0370]

[0371] Table 5 Monitoring Data Judgment Matrix

[0372]

[0373] Table 6 Equipment Status Judgment Matrix

[0374]

[0375]

[0376] Table 7 Communication Information Judgment Matrix

[0377]

[0378] All the evaluation matrices passed the consistency test of the analytic hierarchy process, with CR values ​​of 0.028, 0.0257, 0, 0, and 0, respectively.

[0379] Furthermore, the weights of the secondary indicators are obtained by multiplying the weights of the primary indicators by the sub-weights of the secondary indicators, ultimately yielding the completeness indicator weights, as shown in Table 8:

[0380] Table 8. Weights of Integrity Indicators

[0381]

[0382] After calculating the importance weights of time series data using the analytic hierarchy process, and combining this with a standardization check of the time series data content, a data integrity score is determined. Specifically, during the data evaluation process, an automated program is used to extract the data content, compare whether the content is accurate or does not conform to the filling specifications, and obtain a remote information data integrity score based on the weighted scores of the non-compliant content.

[0383] The integrity evaluation function is:

[0384] y=∑ω i x i

[0385] Where, ω i Let x be the weight of the i-th secondary indicator. i Let x be the score of the i-th secondary indicator. i The value of is 1 or 0. If the content corresponding to the secondary indicator is missing, the evaluation score for the completeness indicator is 0; if the content corresponding to the indicator is not missing, the evaluation score for the indicator value is 1. The final data completeness evaluation score ranges from a minimum of 0 to a maximum of 1.

[0386] (2) Calculate the validity score of remote information data

[0387] This embodiment adopts a combination of long-term and short-term methods, using the accuracy of historical remote signaling signals to measure the effectiveness of real-time remote signaling data. The accuracy of historical remote signaling data is determined by on-site processing of alarm signals to determine whether they are false alarms.

[0388] The validity score of the remote information data is:

[0389] δ=ω1·x1+ω2·x2+ω3·x3+ω4·x4

[0390] Where δ represents the accuracy score of the real-time remote signaling signal; x1 represents the historical accuracy rate of the device, which is the overall performance over the entire historical period; x2 represents the accuracy rate of the 10 most recently transmitted remote signaling signals, and this window helps to capture the accuracy performance over a longer period of time; x3 represents the accuracy rate of the 5 most recent remote signaling data; x4 represents the accuracy rate of the last transmitted remote signaling signal (0 or 100%); ω i The corresponding weights.

[0391] This embodiment uses the historical performance of the device to measure the potential accuracy of real-time remote signaling data. The largest possible value for ω4 ensures the model's adaptability, and signals closer to the current signal are more accurately reflected in the device's current performance. Therefore, ω1 = 0.1, ω2 = 0.2, ω3 = 0.3, and ω4 = 0.4 are chosen.

[0392] (3) Calculate the timeliness score of remote information data

[0393] The converter station's fire protection system requires real-time alarm signal response. Therefore, in the timeliness assessment, a 15-second transmission delay is rated as 0, and a 0-second transmission delay is rated as 1. Furthermore, considering that the minimum wireless communication latency of IoT alarm devices is typically 5 to 50 milliseconds, and wireless communication latency is generally controlled within 300 milliseconds, while wired networks typically have even lower communication latency, an exponential function is used to regress the expected evaluation value to obtain the evaluation function:

[0394] y = e -0.346x

[0395] Where y is the evaluation result of the timeliness of remote signal transmission, and x is the data transmission delay time. The transmission delay time is determined by the acquisition timestamp of the bottom-level sensors of the converter station and the reception time of the converter station's fire protection digital control system.

[0396] In one possible implementation, this embodiment uses the importance weights obtained from the analytic hierarchy process (AHP) and the results of standardization checks to obtain a data integrity score for telemetry time-series data; it uses an automatic threshold acquisition method combining the Transformer reconstruction model with three standard deviations and POT extreme value theory to obtain a data validity score; and it uses the time-series data transmission interval to obtain a data timeliness score. The specific implementation process is as follows:

[0397] (1) Calculate the integrity score of telemetry data

[0398] In this embodiment, the telemetry data integrity assessment is the same as that for telemetry data. The data uploaded is parsed from the station control layer of the converter station fire protection IoT system, compared with the standard format, and combined with the importance weights obtained in Table 8 to obtain the telemetry data integrity assessment results.

[0399] (2) Calculate the validity score of telemetry data

[0400] This embodiment employs a time-series reconstruction-based evaluation method for assessing the effectiveness of telemetry data. The model used is a Transformer reconstruction model, and the automatic threshold acquisition method is based on the POT (Point of Interest) extreme value theory. Figure 4 As shown, the specific implementation steps are as follows:

[0401] 1) The telemetry data was reconstructed using the Transformer reconstruction model to obtain the reconstructed sequence;

[0402] Specifically, the Transformer model structure is as follows: Figure 5 As shown, the Transformer reconstruction model is as follows: Figure 6 As shown, the training data for the Transformer model comes from the telemetry time-series data collected by the equipment at the station control layer of the converter station's fire protection system.

[0403] Transformer model input: The input sequence before reconstruction, with a length of L. i- The input sequence after reconstruction has a length of L. i+ This setting is L. i- =L i+ =N, therefore the time step length of the samples in this input sample dataset is N, and the number of features is 2, which is the time series data before and after the reconstruction sequence. The training set is constructed by obtaining the original time series through a sliding window, and the number of samples is also determined by the original time series length.

[0404] Transformer model output: Reconstructed sequence, length L i Here we take L i =1.

[0405] 2) The reconstruction error of the reconstructed sequence is statistically analyzed using the POT threshold selection method to obtain the reconstruction error threshold. The initial threshold in the POT threshold selection method is determined using the three-standard-deviation method.

[0406] It should be noted that this embodiment uses the POT (Peaks-Over-Threshold) threshold selection method to perform statistical analysis on the reconstruction error of the time series and obtain the threshold for the reconstruction error that is judged as abnormal. For example, if the reconstruction error of a batch of time series is between 0.001 and 0.5, after statistical analysis and maximum likelihood estimation based on extreme value theory, the reconstruction error threshold is determined to be 0.3. Then, the reconstruction error of the data at a certain moment is 0.4, which exceeds the threshold and is judged as an outlier.

[0407] The POT (Progressive Thresholding) method is an automatic threshold selection method based on extremum theory. This theory posits that while different things may have different data distributions, their extreme events mathematically exhibit a consistent distribution; that is, the portion of their extreme values ​​exceeding a certain threshold conforms to a Pareto distribution. Since extreme values ​​are typically located at the tail of the probability distribution, the basic idea of ​​the POT algorithm is to fit the tail of the probability distribution using a parametric generalized Pareto distribution, as shown in the formula below:

[0408]

[0409] Where S = {S1, S2, ..., S} N′ Let} represent the anomaly score calculated for time series data with N′ feature dimensions. The anomaly score also has N′ dimensions, with each dimension corresponding to a specific feature. th represents the initial threshold for the anomaly score, S-th represents the portion exceeding the threshold, and γ and β are the shape and scale parameters of the generalized Pareto distribution, respectively. Both parameters are calculated using the maximum likelihood estimation method. The results are expressed as and , respectively. Then, the final critical threshold th... F The calculation formula is:

[0410]

[0411] where q represents the expected probability of observing S > th, N' is the characteristic dimension of the anomaly score S, and N' th is the characteristic dimension that satisfies S i < th.

[0412] In this embodiment, the main parameters involved are the number of windows N and the initial threshold th. The larger the value of the number of windows N, the more it can reflect the actual operating conditions. The larger the initial threshold th, the fewer points will be finally determined as anomalies, and vice versa, the more points will be determined as anomalies.

[0413] Furthermore, the determination of the initial threshold th combines the three - standard - deviation method. First, use the three - standard - deviation method to obtain the number of potential anomaly points:

[0414] ε = μ(e i ) + zσ(e i )

[0415] The reconstructed error sequence e i = y i - y i ′, z = 3. The number of time series exceeding the dynamic threshold ε in the time - series reconstruction is n. So the initial threshold th in the POT extreme - value theory is:

[0416]

[0417] This method retains the high precision of the POT extreme - value theory when the data distribution is relatively complex and makes the initial threshold th highly self - adaptive.

[0418] 3) Obtain the validity score of the data based on the critical threshold in the reconstruction error threshold, including:

[0419] Obtain the validity score of the telemetry data based on the critical threshold: According to the threshold, the evaluation result of the data validity can be obtained. The evaluation function meets the following requirements:

[0420]

[0421] In the formula, e i is the relative reconstruction error at the time step, th F is the critical threshold, δ(i) is the validity score at the time step i, and δ l is the validity score set for the telemetry data when the relative reconstruction error at the time step is the anomaly - judgment critical threshold. To make the formula meet the above requirements, the specific form of the function can be set as follows:

[0422]

[0423] In the formula After setting the validity score when telemetry data is judged to be abnormal, the validity evaluation results for each time step can be obtained through δ(i).

[0424] (3) Calculate the timeliness score of telemetry data

[0425] Unlike remote signaling data, telemetry data is typically collected periodically, resulting in larger data volumes. This involves data buffering, processing, and transmission, potentially leading to significant delays between data acquisition and receipt at the monitoring center. Furthermore, since telemetry signals are often used for deeper analysis, the system prefers as uniform a signal as possible. Therefore, the timeliness of telemetry data is measured based on the data upload interval. Definition:

[0426] ΔT n =T n -T n-1

[0427] Among them, T n Let ΔT be the time when the nth data point is transmitted to the monitoring center backend. n The time interval between the nth data and the previous data.

[0428] The window size N is set, and the timeliness index T of the telemetry access data is related to the mean and standard deviation of the first N data transmission time intervals. The expression for the timeliness index I of the nth data is:

[0429]

[0430] Where, μ B Let σ be the mean of set B. B Let be the standard deviation of set B, and set B is [ΔT]. n-1 ,ΔT n-2 ,ΔT n-3 ,…,ΔT n-N The larger N is, the greater μ is. B The closer the model is to the rated acquisition interval of the telemetry equipment, the more adaptive it becomes, making it transferable to various types of telemetry equipment. The timeliness index I is affected by the time window N. Using real, long-term acquired fire telemetry time-series data from converter stations, the relationship between different time window numbers and the timeliness index was analyzed. When N is 100, gradient descent is more reasonable and exhibits suitable stability, adaptability, and reasonable computational resource consumption.

[0431] Timeliness evaluation function: Obtain timeliness index I n After that, the evaluation function also needs to be obtained:

[0432] y n =f(I n )

[0433] I n =Timeliness evaluation when y = 0 n =1. I n <1 represents the majority of the timeliness index value distribution. In the actual operation of the converter station, the time interval fluctuations of most uploaded data are within an acceptable range; therefore, I is considered appropriate. n A timeliness score of <1 is considered excellent. (I) n A value greater than three standard deviations indicates an outlier, resulting in a score of 0. Multiple regression models were considered, including linear, multinomial, exponential, and logarithmic regression. The multinomial regression model yielded the smallest R-value and was the simplest to express. The regression equation is as follows:

[0434] y n =1-0.11I n 2

[0435] After obtaining scores for the completeness, validity, and timeliness of telemetry / telecommunications data, a weighted sum is used to obtain the overall data quality score.

[0436] As a further technical solution, the clustering of the second data quality distribution feature vectors within different time windows, and the use of the cluster centers as the health baseline, specifically includes the following steps:

[0437] The optimal number of clusters for the second data quality distribution feature vector within different time windows is obtained using the silhouette coefficient.

[0438] The K-means++ clustering method was used to obtain the four-dimensional feature positions of the centers of each cluster, which served as the health baseline.

[0439] Specifically, this embodiment first calculates data integrity, timeliness, and validity scores for the collected historical normal time-series data and real-time time-series data of equipment operation, respectively. Based on the data integrity, timeliness, and validity scores, a comprehensive data quality score is obtained. According to the comprehensive data quality score corresponding to the historical normal time-series data, the number of data of different quality levels within the time window is extracted by a sliding window as a data quality distribution feature and clustered to obtain a health baseline. Combining feature importance weight and information entropy weight, a highly interpretable equipment anomaly measurement index is obtained to measure the distance between real-time time-series samples and healthy samples, thus obtaining the health status assessment result of the fire protection IoT equipment.

[0440] Specifically, by setting the time-series sliding window length and sliding step size, and based on the comprehensive data quality score, the feature vector of device health status within the time window is extracted as a sample. The extraction rules are shown in Table 9.

[0441] Table 9. Classification of Data Quality Types

[0442]

[0443] Specifically, the overall quality of the aforementioned historical normal time series data will be transformed into a data quality distribution feature vector in the health status assessment. First, after selecting the sliding window length, as shown in Table 9, the number of data points whose quality falls within the scoring range of the first column is used as the feature vector variable. For example, if the window length is 450, and the overall quality of 415 data points is between 0.8 and 1.0, then the value of the first variable of the feature vector is 415. The overall quality of the remaining 35 data points falls within the range of 0.6-0.8, so the second variable is 35. Therefore, the feature vector of this window sample is [415, 35, 0, 0].

[0444] The reason for generating the feature vector in this embodiment is that the converter station's data itself fluctuates. For example, the water pressure data may fluctuate and rise daily. During daily use, there may be data with a quality between 0.6 and 0.8. The feature vector representing health may be [430,20,0,0]. If the feature vector of the final sample is [450,0,0,0], it means that the data is continuous and not operating normally. Or if the feature vector is [0,0,0,0], it means that no data was received within the specified time window. The water pressure sensor will be judged as unhealthy.

[0445] Furthermore, in this embodiment, after obtaining the sample set composed of the data quality distribution feature vectors corresponding to each time window, the optimal number of clusters is automatically obtained using the silhouette coefficient, and the four-dimensional feature position of each cluster center is obtained using the K-means++ clustering method.

[0446] The silhouette coefficient takes into account both the closeness of the data point to other points in the same cluster (average intra-cluster distance a(i)) and the separation of the data point from the nearest cluster (inter-cluster distance b(i)). The calculation formula is as follows:

[0447]

[0448] The silhouette coefficient ranges from -1 to 1. A value closer to 1 indicates better clustering, while a value closer to -1 suggests that data points may be misclassified. By calculating the average silhouette coefficient of all data points, the overall performance of the clustering model can be comprehensively evaluated, and the optimal number of clusters k can be selected.

[0449] In the K-means++ clustering initialization phase, the first cluster center is randomly selected from the dataset. Subsequently, other cluster centers are selected based on a probability P(x), which is proportional to the squared distance from the data point to the nearest selected cluster center.

[0450]

[0451] Here, D(x) is the squared distance from data point x to the nearest cluster center. This strategy ensures a more reasonable distribution of initial cluster centers, thus improving the algorithm's clustering performance.

[0452] The distance used in K-means++ clustering and silhouette coefficient calculation is Euclidean distance, a common method for calculating the straight-line distance between two points. n ) and Q(y1,y2,…,y n The Euclidean distance formula is:

[0453]

[0454] Where d(P,Q) is the Euclidean distance between the two points.

[0455] During the normal operation of fire protection IoT equipment, different working conditions may arise, and the representative feature vectors under each condition may differ significantly. Therefore, the cluster centers generated by k=1 clustering cannot effectively distinguish between different working conditions. Introducing silhouette coefficients can capture the representative feature vectors of each working condition in the high-dimensional feature vector, i.e., the feature vectors represented by the cluster centers, and can further improve the interpretability of the model. Ultimately, the cluster centers of each cluster are the most representative feature vectors of healthy samples, serving as the health baseline.

[0456] As a further implementation, the aforementioned nonlinear mapping of the first data quality distribution feature vector to a predetermined health baseline to obtain the health status assessment result of the fire protection IoT equipment specifically includes the following steps: combining the importance weight and information entropy weight of the second data quality distribution feature vector to determine the feature comprehensive weight; based on the feature comprehensive weight, using the W-ReLU nonlinear mapping model to evaluate the distance between the first data quality distribution feature vector and the health baseline as the anomaly degree; and determining the health level of the converter station fire protection IoT equipment based on the anomaly degree.

[0457] Specifically, the overall feature weight is determined by combining the importance weight and information entropy weight of the second data quality distribution feature vector. After obtaining the healthy baseline, the W-ReLU nonlinear mapping model is used to evaluate the distance between the data quality distribution feature vector of the real-time time series samples and the healthy baseline. The distance is affected by the differences between each feature of the feature vector and each feature of the healthy baseline, as well as the feature weights. The feature weight consists of two parts: importance weight and feature distribution weight. Regarding the importance weight, data quality levels naturally have different levels of importance; therefore, the importance weight AW... i The values ​​are AW1 = 0.1, AW2 = 0.2, AW3 = 0.3, and AW4 = 0.4.

[0458] The distribution weights are obtained using the entropy weight method, assuming the index X... ij This represents the j-th index value of the i-th sample. In the entropy weight method, the data is first standardized.

[0459]

[0460] According to the definition of information entropy in information theory, the information entropy E of each feature j in a set of data is... j The calculation formula is:

[0461]

[0462] Where k is a constant, usually set to 1. Where n is the total number of feature vector samples obtained under normal operating conditions for clustering. The information entropy of each index is calculated as E1, E2, ..., E k The weights RV of each indicator are calculated using information entropy. i :

[0463]

[0464] Where m is the total number of features, i = 1, 2, ..., m, and m takes the value 4. When the information is more concentrated, the information entropy is smaller, and the weight is larger.

[0465] Therefore, the feature integration weight ω′ i The value can be:

[0466]

[0467] In the formula, AW i For importance weights, RV i Let θ be the information entropy weight, θ be the assigned weight, and m be the total number of features.

[0468] Specifically, based on the comprehensive feature weights, the W-ReLU nonlinear mapping model is used to evaluate the distance between the first data quality distribution feature vector and the healthy baseline as the anomaly degree. It should be noted that after obtaining the feature weights, the degree of deviation of the feature vector within a certain time window from the representative feature vector generated under normal equipment operation can be obtained.

[0469] The number of clusters and the set of clusters C = {C1, C2, ..., C} were automatically obtained through clustering. n Let Z represent the i-th cluster C, where n is the total number of clusters. i All samples, Z = {z1, z2, ..., z} m} indicates that the number of samples in the i-th cluster is m. A single feature vector sample z a =[xa1 ,x a2 ,x a3 ,x a4 It contains four features, with the cluster center identified as z0. Its feature vector is obtained through clustering and is z0 = [x 01 ,x 02 ,x 03 ,x 04 ].

[0470] This invention assumes that the feature vector data used for clustering are all collected during the normal operation of healthy devices. Therefore, it is assumed that each feature vector falls within the interval representing health. For feature k of this cluster, the confidence range for determining it as normal can be defined as [x...]. 0k -d kmin ,x 0k +d kmax ], and d kmin With d kmax The distance from the edge of the confidence range to the eigenvalue of the k-th feature at the cluster center is expressed as:

[0471]

[0472] Here, α is the confidence coefficient, used to broaden the range, and is generally greater than 1.

[0473] In calculating the degree of anomaly, it is assumed that the anomaly can only be calculated when it exceeds the confidence range. Therefore, the ReLU function, which is commonly used in deep learning, is introduced. When the input is greater than 0, the output of the ReLU function is equal to the input itself; when the input is less than or equal to 0, the output of the ReLU function is 0.

[0474] For a new feature vector sample point n, its feature vector is [x′1,x′2,x′3,x′4], and the anomaly degree δ n The calculation formula is as follows:

[0475]

[0476] Among them, w i x is the combined weight of feature i. 0k Let d be the eigenvalue of the k-th feature corresponding to the cluster center of the cluster to which the new sample belongs. kmin With d kmax As shown above, for a cluster, each feature has a unique d. kmin With d kmax N is the equipment dimensional parameter, N = x 01 +x 02 +x 03 +x 04, which represents the total number of telemetry and teleinformation data points collected over a period of time, as represented by the cluster center. The larger the amount of data collected by a device over a period of time, the more likely it is to produce greater feature deviations in various features. Therefore, this method is used as a standardization method.

[0477] It should be noted that, after obtaining the anomaly level, this embodiment can determine the equipment health level based on the anomaly level. According to the commonly used health level classification, the health level of the converter station fire protection IoT equipment is divided into four categories: healthy, sub-healthy, abnormal, and faulty. The relevant characteristics corresponding to each level are shown in Table 10.

[0478] Table 10 Health Level Classification of Fire Protection IoT Equipment

[0479]

[0480]

[0481] Where, δ max The value of δ is related to the cluster center to which the eigenvector belongs. For telemetry equipment, δ max The value of δ is the minimum anomaly score of the feature vectors [0,0,0,0] and [0.5N,0,0,0.5N]. The corresponding physical characteristics are communication interruptions, hardware failures, and other signal reception failures, as well as low-quality data failures such as sensor aging and environmental interference. For remote signaling equipment, δ max The value is the feature vector [ax] 01 ,ax 02 ,ax 03 ,ax 04 The minimum anomaly score of [x] and [0.5N, 0, 0, 0.5N], where a is the excess ratio, [x] is the anomaly score of [x] and [0.5N, 0, 0, 0.5N]. 01 ,x 02 ,x 03 ,x 04 ] is the health baseline feature vector, representing the number of alarm signals that greatly exceed the number of alarm signals generated during normal operation, as well as low-quality alarm data failures.

[0482] This embodiment first calculates data integrity, timeliness, and validity scores on the collected historical normal time-series data and real-time time-series data of equipment operation, respectively. Based on these scores, a comprehensive data quality score is obtained. Using the comprehensive data quality score corresponding to the historical normal time-series data, a sliding window is used to extract the quantity of data at different quality levels within a time window as a data quality distribution feature, and clustering is performed to obtain a health baseline. This baseline is then compared with the data quality distribution feature vector corresponding to the real-time time-series data to obtain the health status assessment result of the fire protection IoT equipment. This embodiment employs an unsupervised method, extracting feature vectors based on the temporal distribution of data quality. Training is performed using only normal time-series data generated during the normal operation of the IoT equipment. Clustering is used to obtain the most representative healthy sample points as the health baseline, measuring the distance between real-time time-series samples and healthy samples to assess the health status of the fire protection IoT equipment. This solves the problems of poor model transferability and weak interpretability in fire protection IoT equipment fault detection and health status assessment.

[0483] The embodiments of this application employ the Transformer reconstruction model, which has strong adaptability in evaluating the effectiveness of data and is not limited by the difference between equipment and observed physical quantities. By extending the goal from identifying anomalies to quantitative evaluation of the degree of anomalies, that is, obtaining the effectiveness evaluation results of each time step, it provides a data foundation for equipment status evaluation. Compared with existing comprehensive data quality evaluation methods, the quantitative acquisition method of the index values ​​of this invention is better able to mine the temporal information of the data.

[0484] This application combines feature importance weight and information entropy weight to obtain a highly interpretable device anomaly measurement index, which is used to measure the distance between real-time time series samples and healthy samples, thereby improving the accuracy of device health status judgment.

[0485] Figure 7 Other embodiments of the converter station fire protection digital control system described in this application are illustrated, such as... Figure 7As shown, the converter station fire protection digital control system may also include a converter station integration module, a converter station panoramic monitoring module, and an access control module. The converter station integration module is used to integrate one or more converter stations requiring fire protection control. The converter station panoramic monitoring module is used to perform panoramic dynamic monitoring of the integrated converter stations. The access control module is used to manage and control the access rights of personnel using the converter station fire protection digital control system. For example, after users with different permissions log in to the converter station fire protection digital control system, the system displays information in a mode matching their permissions. Specifically, different display modes have different viewing / editing permissions for certain data, and different display options have different display levels. For example, for display option a, high-permission users have four display levels, while low-permission users only have two. This allows different users with different permissions to view different levels of data.

[0486] In one implementation, the converter grid connection module can be used to: determine the target converter station to be connected and its site attribute information based on a converter station's request for grid connection initiated by the converter station to the converter station's fire protection digital management system, or a grid connection instruction sent by the converter station's fire protection digital management system to the converter station to be connected; and associate and store the site identifier of the target converter station and its corresponding site attribute information in a list of grid-connected converter stations. The site attribute information includes site size, site location, site distance, site priority, and site attention level. The system identifies at least one of the following: the voltage level of the site; then it identifies the equipment requiring pipe connection and the components within the equipment requiring pipe connection in the target converter station, and determines the data reporting mechanism for the equipment requiring pipe connection and / or the components requiring pipe connection; then it sends a successful pipe connection notification to the target converter station, and associates and stores the equipment identifier of the equipment requiring pipe connection and the component identifier of the components requiring pipe connection with the site identifier of the corresponding converter station, wherein the successful pipe connection notification includes the identifiers of the equipment requiring pipe connection and the components requiring pipe connection determined by the converter station fire protection digital control system, as well as their corresponding data reporting mechanism.

[0487] In other words, converter stations requiring network management can proactively request inclusion by initiating a request to the converter station fire protection digital control system, or they can be notified of inclusion by receiving a network management instruction from the system. Furthermore, the process of including converter stations in network management achieves fine-grained management at three levels: station, equipment within the station, and components within the equipment. This improves the accuracy of station inclusion and better meets the differentiated inclusion needs of various converter stations. Moreover, for some converter stations that do not require the inclusion of all equipment or all components within the equipment, data transmission at the station can be reduced, saving network resources. Furthermore, when managing converter stations and their internal equipment and components, corresponding data reporting mechanisms are specified for different internal equipment and / or internal components. That is, the data reporting base stations of internal equipment and / or internal components are also managed. For example, a certain internal equipment is required to report data according to a specified cycle, while a certain internal component is required to report data in real time. In this way, the flexibility and timeliness of data reporting can be improved through differentiated data reporting mechanisms. For example, some data can be reported off-peak to avoid network congestion, while some data needs to be reported in real time to ensure the timeliness of important data reporting.

[0488] In one implementation, the equipment requiring pipe connection and the components within the equipment requiring pipe connection in the target converter station can be determined in the following manner:

[0489] Determine a duct configuration template that matches the site attribute information of the target converter station. The duct configuration template includes the duct-receiving equipment and duct-receiving components in the converter station that have been configured, as well as the data reporting mechanisms corresponding to the duct-receiving equipment and / or duct-receiving components respectively.

[0490] According to the configuration in the configuration template, determine the equipment and components that need to be included in the pipeline in the target converter station, and determine the data reporting mechanism corresponding to the equipment and / or components that need to be included in the pipeline respectively;

[0491] Alternatively, display the duct configuration template, respond to user modification requests for the duct configuration template, modify the duct configuration template, determine the duct-required equipment and components in the target converter station according to the configuration in the modified duct configuration template, and determine the data reporting mechanism corresponding to the duct-required equipment and / or components respectively.

[0492] In other words, converter stations can be quickly connected to the fire control system by pre-configuring a fire control configuration template. Specifically, corresponding fire control configuration templates can be configured for converter stations of different scales, levels, priorities, and voltage levels (e.g., UHV, extra-high voltage, high voltage). For example, the equipment and components within the same equipment in converter stations of the same scale that need to be connected to the fire control system are generally the same. Similarly, the working equipment and fire-fighting equipment in various UHV converter stations are also basically similar. Therefore, for UHV converter stations, a fire control configuration template adapted to UHV converter stations can be used for automatic fire control. Thus, the automatic fire control method using the fire control configuration template can improve fire control efficiency and achieve differentiated and flexible fire control for converter stations with different attributes, making it easier to conduct targeted and differentiated fire control for each converter station and enhancing the digital fire control capabilities of the converter station's digital fire control system.

[0493] Furthermore, to improve the accuracy and effectiveness of grid connection, the system can also display a grid connection configuration template matched with the site attribute information of the converter station to be connected to the grid. Users can personalize and improve the grid connection configuration according to their actual needs. That is, users can modify the grid connection configuration template, and the system can modify the grid connection configuration template accordingly in response to the user's modification operation. Then, according to the configuration in the modified grid connection configuration template, the system determines the equipment and components in the target converter station that need to be connected to the grid, and determines the data reporting mechanism corresponding to the equipment and / or components that need to be connected to the grid.

[0494] In one embodiment, the panoramic monitoring module of the converter station in the converter station fire protection digital control system can be used to: identify the converter stations already under management, as well as the equipment and components within those stations that require management; construct digital twin models corresponding to the converter stations, the equipment requiring management, and the components requiring management, based on the identified map data and internal layout data of the converter stations, the equipment attribute data of the equipment requiring management, and the component attribute data of the components requiring management; and then display the panoramic monitoring data of the converter stations based on the constructed digital twin models. Thus, through the panoramic monitoring method using digital twin models, the relevant status and data of the managed equipment and components in each converter station can be viewed in real time and dynamically, enhancing the visualization effect of the converter station fire protection digital control system and facilitating visualized fire safety monitoring and management of each converter station by management personnel.

[0495] Furthermore, such as Figure 7As shown, the daily fire management module in the converter station fire digital management and control system in this application embodiment may include a converter station risk management and control module, a fire equipment management and control module, a fire business management and control module, a fire personnel management and control module, a fire hazard management module, and a station-side key work equipment management and control module, etc., while the fire emergency response module of the converter station fire digital management and control system may include a fire equipment emergency management and control module, a fire emergency plan management module, a video conferencing module, a fire emergency decision-making module, and a fire emergency drill module, etc.

[0496] It should be noted that, as Figure 7 The converter station grid connection module, converter station panoramic monitoring module, access control module, daily fire management module, and fire emergency response module shown in the diagram are all optional functional modules. Therefore, in Figure 7 All are represented by dashed boxes. In the specific implementation process, one or more modules can be selected according to actual needs. This application embodiment does not impose any restrictions.

[0497] To facilitate a further understanding of the converter station fire protection digital control system in the embodiments of this application, the following is combined with Figure 7 The above functional modules will be introduced in detail.

[0498] The fire protection equipment management module is used to manage and control the fire protection equipment in the converter station that is connected to the pipeline. This means that the managed fire protection equipment is also on the list of connected equipment. Management and control includes monitoring the status of fire protection equipment, detecting / predicting fire protection equipment faults, detecting leaks in the fire protection equipment's piping / pipeline network / wiring system, protecting and controlling fire protection equipment in extreme environments (e.g., using electric heat tracing for fire protection pipelines in low-temperature environments), and assessing fire protection equipment risks. This module allows for equipment-level management and control of the fire protection equipment in the converter station, providing a highly targeted approach. Close monitoring and control of the fire protection equipment allows for the early detection of potential hazards, preventing negative impacts on the fire safety management of the converter station.

[0499] Because malfunctions or abnormalities in some critical equipment in the converter station may lead to fires, in order to achieve comprehensive fire safety management of the converter station, the station-side critical equipment management module can be used to monitor and control some critical equipment in the converter station. For example, the converter valves, converter transformers and other critical equipment in the converter station can be dynamically monitored and controlled to reduce the impact of these devices on the fire safety of the converter station.

[0500] The fire service management module in the converter station's digital fire protection management system is mainly used for the scheduling, dispatching, and management of fire protection services. The fire personnel management module is mainly used for managing fire-related personnel, including registration, attendance, attendance statistics, and management of station maintenance personnel. The fire hazard management module is mainly used to control the dispatching of identified fire hazard tasks within the converter station, record the hazard execution process, and update the hazard task status, so as to quickly resolve fire hazards within the converter station and reduce their impact on the station's fire safety.

[0501] The converter station risk management module in the digital fire protection management system is mainly used for site-level risk management of converter stations connected to the power grid. This allows for a comprehensive fire risk assessment of the entire converter station, providing a holistic view of the station's fire risk compared to assessing the fire risk of individual devices. For example, it evaluates the overall fire safety status of the converter station based on the condition of each fire protection subsystem, determining whether the station is healthy or at risk. It also detects and statistically analyzes fire-related management events that pose a risk throughout the station, marking these events with red, yellow, or green codes based on their risk level.

[0502] The fire equipment emergency management module in the converter station fire digital management and control system can be used to dispatch and monitor fire equipment during fire emergencies. For example, it can determine which types of fire equipment to dispatch to perform fire extinguishing tasks for a certain fire situation. Also, it can dynamically monitor whether there is foam fluid leakage in the compressed-air foam extinguishing system (CAFS) during the fire extinguishing task in a fire emergency scenario.

[0503] The fire emergency decision-making module in the converter station's digital fire control system can be used to make emergency decisions based on acquired fire scene data (including video data and various sensor data). These decisions include determining the fire emergency plan and whether to conduct video conferences with station and provincial personnel. The video conferencing module in the system provides online video conferencing capabilities. System personnel can use this function to conduct timely online video chats with station and provincial personnel. Furthermore, video data from the fire scene can be dynamically displayed in the video chat room, allowing personnel at all levels to understand the fire situation promptly and communicate on-site to quickly determine emergency response plans.

[0504] The fire emergency plan management module in the converter station's digital fire protection control system is used to formulate, store, and update fire emergency plans, thereby enabling the management of these plans. The fire emergency drill module in the same system is used to perform operations related to emergency drills, such as simulation and execution.

[0505] By deploying multiple functional modules in the converter station's digital fire control system, various specific fire control functions can be realized, improving the comprehensiveness of fire control in the converter station and further enhancing the digital fire control capabilities of the system.

[0506] In one embodiment, the fire equipment control module can be configured as described above. Figure 6 The corresponding approach involves assessing the health status of fire-fighting equipment based on the IoT data of the equipment. This allows for timely identification of fire-fighting equipment in poor condition, enabling timely warnings or repairs to prevent any impact on the fire safety of the converter station.

[0507] In some embodiments, certain fire-fighting equipment requires additional protection and management in extreme environments. For example, in areas with extremely low winter temperatures, outdoor fire-fighting pipelines often face the threat of freezing. If freezing occurs, the pipelines will become blocked, preventing the normal use of fire-fighting water sources. In the event of a fire, these frozen fire-fighting equipment will be unusable, seriously affecting fire safety. Therefore, in extremely cold / low-temperature environments, electric heat tracing can be used to raise the pipeline temperature to prevent freezing. However, to save power consumption, the heat tracing power of the electric heat tracing tape needs to be precisely controlled. Therefore, how to reasonably select the power of the electric heat tracing tape for fire-fighting pipelines is a problem that needs to be considered in the electric heat tracing of fire-fighting pipelines. In this regard, embodiments of this application provide a method for determining the optimal power of the electric heat tracing tape for fire-fighting pipelines. This method can be implemented by the aforementioned fire-fighting equipment management module, such as... Figure 8 As shown, the method includes:

[0508] S801: Obtain temperature drop test data of fire-fighting pipelines with electric heating cables under low temperature conditions. The temperature drop test data mainly includes fire-fighting pipeline specifications, electric heating cable specifications and constant power, ambient temperature, and test ambient temperature and pipe wall temperature data that change over time.

[0509] S802: Establish a numerical model for fire-fighting pipelines containing electric heat tracing cables, determine the conversion formula for the convective heat transfer coefficient, and verify the numerical model based on temperature drop experimental data and the convective heat transfer coefficient conversion formula. The specific process is as follows:

[0510] Based on the actual dimensions and physical properties of the fire protection pipeline, establish as follows: Figure 9 The numerical model of a fire-fighting pipeline containing an electric heating cable is shown.

[0511] The parameters of the numerical model for the fire protection pipeline in this embodiment are shown in Table 11 below:

[0512] Table 11 Numerical Model Parameters for Fire Protection Piping

[0513]

[0514] The parameters of the electric tracing cable in this embodiment are shown in Table 12 below:

[0515] Table 12 Parameters of Electric Tracing Cable

[0516]

[0517] The formula for converting the convective heat transfer coefficient is as follows: Where h is the convective heat transfer coefficient, β is determined from the temperature drop experimental data of the fire-fighting pipeline, R is the outer diameter of the fire-fighting pipeline, Pr and Gr are the Prandtl number and Grashof number of ambient air, respectively, and A and m are constants, taken as 0.48 and 0.25, respectively. Input the temperature, steel pipe, and electric heating cable parameters consistent with the experimental data into the numerical model. Input the convective heat transfer coefficient for this experimental condition according to the convective heat transfer coefficient conversion formula. Set temperature sensors at the symmetrical position of the fire-fighting pipeline heating cable about the center to obtain the pipe wall temperature, such as... Figure 9 Thermocouple #2 is shown. The convective heat transfer coefficient β was varied using a trial-and-error method until the pipe wall temperature in the numerical model matched the experimental data.

[0518] Example: The conversion formula for the convective heat transfer coefficient in this embodiment is:

[0519] S803: By changing the ambient temperature and the power of the electric heating tape, the antifreeze time of the fire-fighting pipeline containing the electric heating tape under different operating conditions is obtained; wherein, the process of obtaining the antifreeze time is as follows: a liquid fraction sensor is set on the outermost layer of the fire-fighting pipeline to obtain the liquid fraction, and the time from the start until the liquid fraction on the outermost layer of the fire-fighting pipeline becomes 0.02 is taken as the antifreeze time; then, the ambient temperature and the power of the electric heating tape are changed, and the convective heat transfer coefficient of the numerical model under different operating conditions is obtained according to the convective heat transfer coefficient conversion formula determined in S802.

[0520] The operating conditions for numerical simulation in this embodiment are shown in Table 13 below:

[0521] Table 13 Operating conditions in numerical simulation

[0522]

[0523] A liquid fraction sensor is installed on the outermost layer of the pipeline to obtain the liquid fraction. The time from the start to the outermost node when the liquid fraction becomes 0.02 is defined as the time to maintain non-freezing. The time to maintain non-freezing under different operating conditions is obtained.

[0524] S804: Based on the antifreeze time obtained under different operating conditions in S803, establish a power function for the electric heating cable related to the antifreeze time and ambient temperature, specifically including:

[0525] The power function of the electric heating cable regarding antifreeze time and ambient temperature is as follows:

[0526] t = f(P,T) = a0 + a1P + a2T + a3P 2 +a4PT+a5T 2

[0527] Where P is the power of the electric heating cable, T is the ambient temperature, t is the antifreeze time, and a0, a1, a2, a3, a4 and a5 are regression coefficients to be determined.

[0528] The Levenberg-Marquardt optimization algorithm was used to fit the power function. The fitting result in this embodiment is as follows: Figure 10 As shown, the fitting function is t = 30133 + 981P + 4150T - 5P 2 +6PT+72T 2 The coefficient of determination is 0.90.

[0529] S805: Calculate the optimal power of the fire-fighting pipeline under the required antifreeze time using the fitted electric heating cable power function. That is, based on the electric heating cable power function and the given antifreeze time and ambient temperature, determine the optimal electric heating cable power of the fire-fighting pipeline with electric heating cable under the required antifreeze time. Specifically, input the ambient temperature and antifreeze time into the fitted electric heating cable power function to calculate the optimal power of the heating cable of the fire-fighting pipeline under the required antifreeze time.

[0530] This embodiment considers the influence of ambient temperature and fire-fighting pipeline properties on the convective heat transfer coefficient, takes into account the deviation between the theoretical convective heat transfer coefficient and the actual value, and proposes a conversion formula to improve the accuracy of numerical simulation. It adopts a method combining numerical simulation and function fitting, which reduces the difficulty of determining the optimal power of the electric heating cable. At the same time, the calculation speed of the fitting function is extremely fast, which not only ensures the calculation accuracy, but also greatly reduces the calculation time and difficulty.

[0531] In one embodiment, compressed-air foam systems (CAFS) are a commonly used fire-fighting equipment in converter stations. CAFS uses foam as the extinguishing medium. Due to the different pressure and flow characteristics of foam compared to water, leak detection in fire-fighting pipelines differs from that of water. Therefore, existing leak detection methods for water pipelines are difficult to apply to pipelines using foam as the extinguishing medium, such as CAFS. In view of this, this application provides a leak detection method for fire-fighting foam pipelines in converter stations, which can quickly and accurately detect leaks in foam pipelines. In specific implementation, this method can be implemented by the fire-fighting equipment management module, or it can be implemented by the fire-fighting equipment emergency management module within the fire emergency response module.

[0532] The leak detection method for fire-fighting foam pipelines in converter stations according to this application embodiment can be applied in various emergency use scenarios of CAFS. For example, when a fire occurs in the converter station, due to the complexity of the operating environment, users cannot promptly monitor the pipeline status of the CAFS. In this case, the foam pipeline can be monitored in real time using the embodiment of this application during the use of the CAFS, and an early warning can be issued when a foam leak is detected. This method can be implemented, for example, by the fire equipment emergency management module in the fire emergency response module. Furthermore, the embodiment of this application can also be used for routine inspections during the daily management and maintenance of the CAFS. Specifically, the CAFS can be briefly activated to fill the pipeline with foam fluid, and the foam leak can be quickly detected using the embodiment of this application to meet the needs of routine maintenance or testing of the CAFS.

[0533] like Figure 11 As shown in the embodiment of this application, a method for detecting leaks in fire-fighting foam pipelines in converter stations is proposed. This method can detect foam leaks in CAFS (Converter Air System) pipes connected to the converter station. The method includes the following steps:

[0534] S111: Obtain the first data set based on fluid data collected from the collection point of the fire-fighting foam pipeline;

[0535] S112: Based on the first data set and the preset fluid identification model, determine the fluid type; the fluid type includes Newtonian fluid or non-Newtonian fluid;

[0536] S113: Determine the leak identification result based on the leak identification model corresponding to the fluid type and the first data set.

[0537] In this embodiment, fluid data collected from the sampling points of the foam pipeline is processed to obtain a first data set. This first data set is then input into a preset fluid identification model for identification, determining whether the flow type is Newtonian or non-Newtonian. Subsequently, the corresponding leak identification model is called to process the first data set for different fluid types, performing leak identification and obtaining the leak identification result. The specific implementation of each step is described in detail below.

[0538] In S111, data collection points can be set at certain intervals on various pipes of the fire-fighting foam pipe network. In this embodiment, it mainly targets pipes with a diameter of DN200, but can also be reasonably extended to pipes of other diameters. For example, in a non-ring network (where the beginning and end are not connected), with a pipe diameter of DN200, pressure sensors, temperature sensors, ultrasonic Doppler flow meters, and Coriolis mass flow meters are installed approximately every 100 meters. Other alternative sensor tools can also be used. Due to practical factors, there may be acceptable errors at each installation location. Each installation location is used as a data collection point for the pipeline to collect fluid data.

[0539] After preliminary processing of the collected fluid data, the first data set can be obtained. For example, fluid pressure parameters, velocity profile data, temperature sensor data, fluid density measurements, and shear stress can be acquired. For instance, based on the fluid pressure parameters and pipe length, the pressure gradient (ΔP / ΔL) can be calculated; the shear rate-viscosity curve can be obtained from the velocity profile data and shear stress. The preliminary processing involves processing the sensor-collected data, which is well-known to those skilled in the art and will not be elaborated upon here. The first data set specifically includes pressure gradient data, velocity profile data, shear rate-viscosity curve, temperature sensor data, and fluid density measurements; these data are time-series data.

[0540] In S112, based on the first data set and the preset fluid identification model, the fluid type is determined, which is either a Newtonian fluid or a non-Newtonian fluid.

[0541] The fluid recognition model employs a Bidirectional Long Short-Term Memory (LSTM)-Residual Fully Connected Network (BiLSTM-ResDNN) model. The Bidirectional LSTM layer processes time-series data, such as continuously sampled flow velocity profile data sequences, effectively capturing the temporal dependencies of fluid behavior and extracting the dynamic dependencies of various fluid data points over time, such as viscosity hysteresis. The first data set can be processed into multivariate parameter vectors for each time step, including pressure gradient data (ΔP / ΔL), flow velocity profile data, shear rate-viscosity curves, temperature sensor data, and fluid density measurements. After processing by the Bidirectional LSTM layer, temporal feature encoding is obtained and then input to the next processing module, namely the residual fully connected layer.

[0542] The residual fully connected network of the fluid recognition model can consist of two layers. The first layer, L1, can use 64 neurons and employs the Swish activation function to handle the ratio of shear rate to viscosity. The second layer, L2, can use 32 neurons and combines the Batch Normalization method to handle the interaction effect of temperature and pressure. Through the stacking of residual fully connected layers, the complex mapping relationship between shear rate and viscosity is learned. Furthermore, for Newtonian fluids, the network learns that the weights of shear rate input and viscosity output approach zero, meaning that changes in shear rate have no effect on changes in viscosity. A dropout layer is introduced when calculating the output of each layer during forward propagation. For example, with a dropout rate of 0.3, 30% of the neuron outputs are randomly masked (set to zero) during forward propagation to prevent overfitting to single fluid data.

[0543] Optionally, L1 / L2 joint regularization can be introduced when calculating the loss function during forward propagation. Specifically, a penalty term is added to the loss function to constrain the weights of the fully connected layers and ensure physical rationality. The implementation is as follows: the loss function of the residual fully connected network is:

[0544]

[0545] Among them, L 正则化项 Used to reset weights unrelated to rheological properties to zero, L 物理约束 The viscosity gradient used to make the Newtonian fluid prediction result approach zero; λ1 is the L1 regularization coefficient (in this embodiment, it can be on the order of 1e-4), λ2 is the L2 regularization coefficient (in this embodiment, it can be on the order of 1e-3), ∑||W||1 is the L1 norm, and W is the weight. Weights unrelated to rheological properties can be cleared to zero. The L2 norm can suppress abnormal fluctuations, ensuring physical rationality while preventing overfitting; λ phy Eγ is the physical constraint strength coefficient. ˙To obtain the desired value for the shear rate distribution, μ pred γ is the fluid viscosity value predicted by the neural network. ˙ The shear rate is used. Physical constraints can be applied to make the viscosity gradient in the Newtonian fluid prediction results approach zero, ensuring the accuracy of the Newtonian fluid prediction.

[0546] The output layer of the fluid identification model can use the Sigmoid function to output the probability value P of belonging to a non-Newtonian fluid. A dynamic threshold is then used to determine whether the fluid in the pipe is a Newtonian or non-Newtonian fluid. For example, when P < 0.4, it is judged as a Newtonian fluid; when P is greater than 0.6, it is judged as a non-Newtonian fluid; other P values ​​are judged as "uncertain".

[0547] Furthermore, fuzzy probabilities can be set to improve the accuracy of fluid type identification. A fuzzy probability indicates that the fluid type cannot be determined as either a Newtonian or non-Newtonian fluid. When the fluid identification model outputs fuzzy probabilities, an auxiliary judgment can be initiated. For example, when the model probability is in the fuzzy range (0.4 ≤ P ≤ 0.6), an auxiliary judgment can be initiated. The auxiliary judgment can be any of the following:

[0548] When the pipe frequency is greater than a preset frequency threshold, the fluid type is determined to be a Newtonian fluid;

[0549] When the flow rate in the pipeline exceeds the predicted value for a theoretical Newtonian fluid, the fluid type is determined to be a non-Newtonian fluid.

[0550] When the model probability is within the fuzzy range (0.4 ≤ P ≤ 0.6), auxiliary judgment can be initiated. Because non-Newtonian fluids exhibit elastic memory effects, stress relaxation occurs during pipe flow, leading to pressure fluctuations. Therefore, in one possible implementation, auxiliary judgment can be based on pipe frequency and flow rate. For example, based on abnormally high-frequency fluctuations detected by multiple (e.g., three or more, without limitation) pressure sensors, if the pipe frequency exceeds a preset frequency threshold, it is determined to be a non-Newtonian fluid. The frequency threshold can be determined through testing under various possible operating conditions simulating a Newtonian fluid in the pipe. Based on the deviation of multiple (e.g., three or more, without limitation) flowmeter readings from the theoretical Newtonian fluid prediction exceeding a preset deviation threshold, such as greater than 15%, it can be determined to be a non-Newtonian fluid. The theoretical Newtonian fluid prediction value can be determined by passing a Newtonian fluid through the pipeline and conducting multiple sets of tests under different operating conditions of the compressed air foam fire extinguishing system. Each operating condition corresponds to a specific theoretical Newtonian fluid prediction value. Simultaneously, a non-Newtonian fluid is passed through the pipeline, and multiple sets of tests are conducted under different operating conditions of the compressed air foam fire extinguishing system to determine the flow meter readings. The difference between the two fluids under the corresponding operating conditions is then calculated as a deviation threshold. When determining the fluid type, the corresponding theoretical Newtonian fluid prediction value is selected for the specific operating condition to calculate the specific deviation, which is then compared with the deviation threshold to determine whether it is a non-Newtonian fluid. In other words, when the model cannot identify the fluid type, auxiliary judgment methods can be used to further identify the fluid type, thereby improving the accuracy and effectiveness of fluid type identification.

[0551] After identifying the fluid type using the BiLSTM-ResDNN model, step S113 is further executed to identify leaks in the foam pipe. In S113, the leak identification result is determined based on the leak identification model corresponding to the fluid type and the first data set.

[0552] S113 can specifically include two cases: one is the leak identification process when the leak identification result is a Newtonian fluid; the other is the leak identification process when the leak identification result is a non-Newtonian fluid.

[0553] When the fluid type is a Newtonian fluid, S111 includes:

[0554] Step A: Based on the first data set, determine the second data set corresponding to the Newtonian fluid. The second data set is used to characterize the pressure gradient data fluctuation in the pipeline. The second data set includes parameters used to characterize the pressure gradient data fluctuation in the pipeline. For example, the second data set includes parameters used to characterize the pressure gradient data fluctuation in the pipeline.

[0555] Step B: Based on the second data set and the first leak identification model corresponding to Newtonian fluid, determine the leak identification result.

[0556] The first data set includes: pressure gradient data, flow velocity profile data, shear rate-viscosity curves, temperature sensor data, and fluid density measurements; the second data set includes pressure gradient data, pressure gradient coefficient of variation, mass flow rate deviation, viscous dissipation power density, and dynamic characteristics of bulk modulus. The methods for determining some parameters in the second data set are as follows:

[0557] The pressure gradient data (ΔP / ΔL) is processed into the pressure gradient coefficient of variation (CV). The CV measures the relative fluctuation of the pressure gradient data. Pipeline leakage will significantly increase the CV. The CV is obtained as follows:

[0558]

[0559] Where ΔP / ΔL is the pressure gradient (unit: Pa / m), representing the pressure change per unit pipe length; std() is the standard deviation, reflecting the degree of fluctuation of the pressure gradient; mean() is the mean, representing the long-term stable value of the pressure gradient.

[0560] The mass flow rate deviation R_mass is obtained based on the flow velocity profile data, and the method is as follows;

[0561]

[0562] Among them, Q in Q represents the mass flow rate (kg / s) at the previous measurement point in the pipeline flow direction; out The initial Q is the mass flow rate (kg / s) at the current measurement point. in Provides the initial mass flow rate of the pipeline; if there is a bifurcation or pipe diameter change between two sampling points, the parameters at the sampling point in the same pipe section shall be used to avoid data distortion.

[0563] Based on the viscous dissipation power density q˙ determined through shear rate-viscosity curves and temperature sensor data, pipeline leakage will cause this viscous dissipation power density q˙ value to increase. The method for obtaining the viscous dissipation power density q˙ is as follows:

[0564]

[0565] Where q˙ is the viscous dissipation power density (W / m³) 3 The temperature-dependent dynamic viscosity (Pa·s) μ(T) can be obtained by fitting a viscosity-temperature curve. Radial velocity gradient (s) -1 ), which can be measured and determined.

[0566] Based on fluid density measurements, the dynamic characteristic K of the bulk modulus is determined.eff A leak in the pipeline will cause the dynamic characteristic K of the bulk modulus to change. eff Value fluctuation, dynamic characteristics of bulk modulus K eff The methods for obtaining it are as follows:

[0567]

[0568] Where ρ is the fluid density (kg / m³) 3 ), This is the partial derivative of pressure with respect to density under isothermal conditions.

[0569] In step A, different operating conditions exist during pipeline use. Therefore, the above parameters, such as pressure gradient data, pressure gradient coefficient of variation (CV), and mass flow rate deviation (R), can be used. mass Viscous dissipation power density q ˙ Dynamic characteristics of bulk modulus K eff The input neural network model learns the correlations between various parameters and further explores these deeper relationships to achieve more accurate leak detection of Newtonian fluids. This first leak detection model can employ a commonly used LSTM+CNN+Attention neural network structure, or other neural network models; there are no restrictions. Finally, the leak detection result is obtained, which can indicate whether the pipe is leaking or not.

[0570] Following step B, the process further includes: when the leak identification result indicates a leak in the pipeline, determining the leaking pipe section based on sampling points where a first deviation value is greater than a preset first threshold and a second deviation value is greater than a preset second threshold; wherein, the first deviation value is the difference between the pressure gradient variation coefficient corresponding to the sampling point and a preset normal pressure gradient variation coefficient, and the second deviation value is the difference between the mass flow rate deviation corresponding to the sampling point and a preset normal mass flow rate deviation. The first and second thresholds can be determined through experimental verification in actual application scenarios. After determining that a pipeline leak has occurred, the above process can quickly locate the leak point, effectively improving detection accuracy and avoiding the manual investigation costs caused by detection errors.

[0571] When the fluid type is a non-Newtonian fluid, S111 includes:

[0572] Step A': Based on the first data set, determine the third data set corresponding to the non-Newtonian fluid;

[0573] Step B': Based on the third data set and the second leak identification model corresponding to non-Newtonian fluids, determine the leak identification result.

[0574] In step A', the first data set includes: pressure gradient data, flow velocity profile data, shear rate-viscosity curve, temperature sensor data, and fluid density measurement. The third data set is used to ensure the degree of flow separation in the pipeline. This third data set includes parameters characterizing the degree of flow separation, such as shear rate-viscosity dynamics, the secondary flow intensity, viscous dissipation power density and the corresponding associated temperature, and dynamic characteristics of the bulk modulus. The third data set also includes pressure gradient, pressure gradient coefficient of variation, mass flow rate deviation, viscous dissipation power, and dynamic characteristics of the bulk modulus. The method for determining some parameters of the third data set is as follows:

[0575] Based on the shear rate-viscosity curve, the dynamic characteristic of shear rate-viscosity (apparent viscosity) μ is determined as follows:

[0576] μ(γ ˙ )=K·γ ˙n-1

[0577] Where μ is the shear rate-viscosity dynamic characteristic (Pa·s), reflecting the fluid flow resistance; γ˙ is the shear rate (s -1 The velocity gradient characterizes the fluid deformation rate; K is the consistency coefficient (Pa·s). n The apparent viscosity is denoted by , where n is the power-law exponent. A sudden increase in shear rate near the leak point can cause a significant decrease in apparent viscosity, deviating from the normal rheological curve.

[0578] The secondary flow intensity Γ is determined based on velocity profile data, as follows:

[0579]

[0580] Wherein, Γ is the secondary flow intensity (dimensionless), reflecting the ratio of the radial and tangential velocity components to the axial velocity, and can be used to characterize the degree of flow separation. v_r, v_θ, and v_z are the radial, tangential, and axial velocity components (m / s), respectively. Pipeline leakage can cause flow separation, leading to an abnormal increase in the secondary flow intensity Γ.

[0581] Based on shear rate-viscosity curves and temperature sensor data, the viscous dissipation power density and corresponding associated temperature are determined as follows:

[0582]

[0583] Where q˙ is the viscous dissipation power density (W / m³) 3 Heat production rate per unit volume; The radial gradient of the axial flow velocity (s) -1 ); ΔT is the temperature rise (K), which can represent the abnormal temperature near the leak point; ρ is the fluid density (kg / m³).3 );C p Specific heat capacity (J / (kg·K)). Shear rate gradient near the leak point. Increasing this will lead to q ˙ As the temperature rises, the local temperature ΔT will increase significantly.

[0584] Based on fluid density measurements, the dynamic characteristic K of the bulk modulus is determined. eff The leak will lead to K eff The value fluctuates, and the method for obtaining it has been explained above, so it will not be repeated here.

[0585] The above pressure gradient data, shear rate-viscosity dynamics (apparent viscosity) μ, secondary flow intensity Γ, and viscous dissipation power density q are used to... ˙ and the dynamic characteristics of temperature ΔT and bulk modulus K eff As input data, the data is fed into a neural network model for learning, studying the correlations between various parameters, and further exploring their deeper relationships to achieve more accurate non-Newtonian fluid leak detection. This second leak detection model can employ a commonly used LSTM+CNN+Attention neural network structure, or other neural network models; there are no restrictions. Finally, the leak detection result is obtained, which can indicate whether the pipe is leaking or not.

[0586] Following step B', the method further includes: when the leak identification result indicates a pipe leak, based on the pressure gradient decrease value of the target pipe segment being greater than a third threshold and the viscous dissipation power density increase value of the target pipe segment being greater than a fourth threshold, a leak is determined in the target pipe segment; wherein, the target pipe segment is the pipe segment between two adjacent acquisition points. Specifically, when the obtained detection result is a leak, the pipe segment with the leak can be located using the pressure gradient data (ΔP / ΔL) and viscous dissipation power density q˙ corresponding to each detection point. Combined with ΔT verification, false alarms due to pressure fluctuations (such as pump and valve operation interference) are eliminated, and the obtained The data obtained during pump and valve operation are compared and verified with ΔT. When the deviation is within the set threshold range, it can be considered as pump and valve operation interference. Furthermore, when determining the pipe section where a leak occurs, taking sampling points Pi and Pi+1 as an example, if the pressure gradient data (ΔP / ΔL) of pipe section Pi→Pi+1 decreases by more than a preset third threshold (e.g., 15%), and the viscous dissipation power density q˙ in that pipe section simultaneously increases by more than a preset fourth threshold (e.g., 20%), then the leak is determined to occur in that pipe section, thus quickly identifying the leak point. This process effectively improves detection accuracy and avoids the manual investigation costs caused by detection errors.

[0587] In this embodiment, the fluid identification model, the first leak identification model, and the second leak identification model are all pre-trained models. To train the models, a pipeline model corresponding to the actual application scenario can be constructed. For example... Figure 12 As shown, in this embodiment, when constructing the pipeline model, an outdoor experimental pipeline of approximately 400 meters (non-ring network, not connected end to end) with a diameter of DN200 was selected. Leakage holes were machined on the pipeline to simulate leakage. Pressure sensors, temperature sensors, ultrasonic Doppler flow meters, and Coriolis mass flow meters were installed approximately every 100 meters. Other alternative sensor tools could also be used (due to practical factors, the actual interval between installation positions P1 and P2 is 107 meters, between P2 and P3 is 86 meters, between P3 and P4 is 105 meters, and between P4 and P5 is 70 meters). These installation positions correspond to the fluid data acquisition points. Theoretically, the pipeline is divided into four sections, with leakage points set. Holes are drilled in increments of 5 meters, 10 meters, and 15 meters within each section. The leakage opening is a module consisting of an electromagnetic flow meter, a needle valve, and a gravity sensor. Taking the pressure sensor as an example, the arrangement is shown in the table below:

[0588] Table 14 Actual Distance of Pressure Sensor

[0589]

[0590] The aforementioned pipeline model effectively simulates leakage, normal use, and non-use conditions, allowing for the collection of fluid data at various installation locations at different times. This data includes fluid pressure parameters, velocity profile data, temperature sensor data, fluid density measurements, and shear stress. Based on the fluid pressure parameters and pipeline length, pressure gradient data can be calculated. Shear rate-viscosity can be obtained from the velocity profile data and shear stress, resulting in a shear rate-viscosity curve. The collected data can be processed into the following time-series data: pressure gradient data (ΔP / ΔL), velocity profile data, shear rate-viscosity curve, temperature sensor data, and fluid density measurements. Each data set is labeled with two categories: first, Newtonian fluids and non-Newtonian fluids; second, leakage, normal use, and non-use. This data can then be used as the raw data for model training. Existing methods such as interpolation and Generative Adversarial Networks (GANs) are then used to expand the sample. The BiLSTM-ResDNN model is trained using the sample data collected and processed by the above experimental model to obtain the fluid recognition model.

[0591] Furthermore, fluid pressure parameters, velocity profile data, temperature sensor data, fluid density measurements, and shear stress can be processed into pressure gradient data, pressure gradient variation coefficient, mass flow rate deviation, viscous dissipation power density, and dynamic characteristics of bulk modulus. Data annotation and sample expansion are then performed to obtain samples for training the first leak detection model. By training the model using these samples and learning the correlations between parameters, further exploring their deeper relationships, a well-trained first leak detection model can be obtained.

[0592] Furthermore, fluid pressure parameters, velocity profile data, temperature sensor data, fluid density measurements, and shear stress can be processed into pressure gradient data, shear rate-viscosity dynamic characteristics, secondary flow intensity, viscous dissipation power density, and associated temperature and bulk modulus dynamic characteristics. Data annotation and sample expansion are then performed to obtain samples for training the second leak detection model. By training the model using these samples, the correlations between various parameters are learned, and further exploration of these deeper correlations yields a well-trained second leak detection model.

[0593] In summary, the leak detection method for foam pipes in this embodiment has the following advantages:

[0594] 1) The overall architecture of the leak detection method for fire-fighting foam pipelines in the converter station in this embodiment is to identify the fluid type by using a bidirectional long short-term memory network-residual fully connected network model (BiLSTM-ResDNN), and then call the first / second leak identification model corresponding to the fluid type to identify leaks of Newtonian / non-Newtonian fluids. In the application scenario of foam pipelines, it can significantly improve the leak identification accuracy and has good timeliness.

[0595] 2) The loss function of the BiLSTM-ResDNN model adds a regularization term, which can clear weights unrelated to rheological properties to zero and suppress abnormal fluctuations, ensuring physical rationality and preventing overfitting. At the same time, the loss function also adds a physical constraint term, which can make the viscosity gradient of the Newtonian fluid prediction result approach zero, ensuring the accuracy of the Newtonian fluid prediction.

[0596] 3) When using the first / second leak identification model, different secondary preprocessing is performed on the pressure gradient data, flow velocity profile data, shear rate-viscosity curve, temperature sensor data, and fluid density measurement values ​​to adapt to the characteristics of Newtonian / non-Newtonian fluids. This allows the first / second leak identification model to learn the correlation between various parameters more accurately, further explore their deep correlation, improve the leak detection accuracy, and the use of the same set of collected data for two different preprocessing does not increase the data acquisition cost.

[0597] 4) By learning the deep correlation between parameters such as pressure gradient data, pressure gradient variation coefficient, mass flow rate deviation, viscous dissipation power density, and dynamic characteristics of bulk modulus through the first leak identification model, more accurate Newtonian fluid leak detection is achieved. By learning the deep correlation between pressure gradient data, shear rate-viscosity dynamic characteristics (apparent viscosity), secondary flow intensity, viscous dissipation power density, and associated temperature and bulk modulus dynamic characteristics through the second leak identification model, more accurate non-Newtonian fluid leak detection is achieved.

[0598] 5) For leaks of Newtonian fluids, the leaking pipe section can be determined by the pressure gradient variation coefficient data and mass flow deviation corresponding to each detection point, enabling rapid location of the leak; for leaks of non-Newtonian fluids, the leak location can be rapidly located by the pressure gradient and viscous dissipation power density corresponding to each detection point, ensuring the efficiency of pipeline fault diagnosis.

[0599] In one embodiment, malfunctions in some critical equipment within the converter station can easily lead to fires / incidents. These critical equipment include converter valves and converter transformers. For example, the converter transformer, as a core component of the power transmission system, has a large amount of insulating oil and operates at high temperatures, posing a significant safety hazard. Oil-immersed converter transformers have complex internal structures and high potential failure risks. Thermal runaway within the tank can rapidly cause localized high temperatures and pressures. If this cannot be effectively predicted and warned of, it could lead to tank rupture and eventually a fire or explosion, severely threatening life and property. Therefore, this application provides a method for predicting the failure risk of converter transformer tanks. This method can be implemented by the station-side critical equipment control module within the fire safety daily management module. Please refer to [link to relevant documentation]. Figure 13 The method specifically includes:

[0600] S131. Under different fault or failure conditions inside the transformer tank, obtain the time series data of the internal temperature of the converter transformer tank in the real environment, and record it as the real dataset; and carry out a proportional numerical simulation of the real environment to simulate the temperature evolution law inside the converter transformer tank under different fault or failure conditions, and obtain the time series data of the internal temperature of the converter transformer tank in the simulation environment, and record it as the simulation dataset.

[0601] During implementation, a temperature acquisition system can be deployed inside the converter transformer tank to obtain the aforementioned real and simulation datasets. Specifically, the temperature acquisition system inside the converter transformer tank is arranged as follows: several temperature acquisition devices are placed horizontally and several are arranged vertically inside the converter transformer tank. The temperature acquisition devices are thermocouples.

[0602] In this embodiment, in a real environment, it can be understood as using a transformer oil tank with a pre-arranged temperature acquisition system placed inside a converter transformer oil tank fire test platform, and acquiring time-series data of transformer oil tank temperature under different fault or failure conditions inside the oil tank, which is recorded as a real dataset.

[0603] In this embodiment, the simulation environment can be understood as a transformer oil tank simulation model established in simulation software, such as COMSOL, that is identical to the transformer oil tank in a real environment, including a transformer oil tank fire test platform and a pre-arranged temperature acquisition system. In the simulation environment, by changing different fault or failure conditions inside the oil tank, time-series data of the transformer oil tank temperature is obtained and recorded as the simulation dataset.

[0604] In this embodiment, the failure boundary conditions inside the transformer tank are the same in both real and simulated environments. However, the failure boundary conditions inside the transformer tank can be altered, including but not limited to the initial oil temperature, the initial location of the ignition source, and the heating intensity of the heat source.

[0605] In this embodiment, several temperature acquisition devices in the temperature acquisition system are numbered and grouped before being connected to the data processing module, and the experimental data is transmitted to the computer in real time for processing and storage. For example, the temperature data acquisition refresh time of the temperature acquisition system is set to 1 second. The internal temperature of the converter transformer tank generally goes through an initial combustion stage, a full development stage, and a quenching stage. The experimental time for each operating condition is recorded until the end of the quenching stage.

[0606] In this embodiment, specifically, acquiring the timing data of the converter transformer tank temperature includes:

[0607] Establish a coordinate system for the temperature acquisition system inside the converter transformer tank. Under different fault or failure conditions inside the tank, record the temperature time series data corresponding to the position coordinates of each temperature acquisition device to form the transformer tank temperature time series data.

[0608] S132, compare the simulation dataset with the real dataset to verify the reliability of the simulation dataset.

[0609] In this implementation, the difference between the simulated dataset and the real dataset is calculated based on the average coefficient of determination, and the result is compared with a preset difference threshold. If the difference comparison result is greater than or equal to the threshold, the simulated dataset is considered reliable. If the difference comparison result is less than the threshold, the simulated dataset is considered unreliable. In this case, the simulation environment parameters are readjusted, and the reliability of the adjusted simulated dataset is verified again against the real dataset until the reliability meets the standard.

[0610] In this embodiment, the average coefficient of determination is calculated in the range of -1 to 1. A result close to 1 indicates complete data overlap, while a result far below 1 indicates a mismatch in the characteristics of the two sets of data. A gap threshold of 0.9 is set, and simulation datasets with a gap comparison result greater than or equal to 0.9 are considered reliable simulation datasets for subsequent analysis.

[0611] S133, transform the simulation dataset after reliability verification into a temperature matrix with corresponding coordinates, and determine the time matrix based on the time in the simulation dataset after reliability verification as a variable; and determine the physical matrix with corresponding coordinates based on the physical characteristics of the converter transformer tank, that is, determine the physical matrix with corresponding coordinates based on the physical characteristics of the converter transformer tank.

[0612] Please see Figure 14 As shown, in the temperature matrix, the temperature at each coordinate is the same as the setting position in the real environment and the simulated environment.

[0613] The physical characteristics of the converter transformer tank are introduced and transformed into a physical matrix with corresponding coordinates, including:

[0614] First, the physical features are discretized based on the coordinates corresponding to the temperature matrix, and then divided into corresponding grids.

[0615] In this embodiment, the continuous physical features based on the transformer tank need to be matrixed to accommodate computer matrix operations. Therefore, the physical features are discretized based on the coordinates corresponding to the temperature matrix, and a corresponding grid is created.

[0616] In this embodiment, the physical characteristics of the converter transformer tank include its internal structure, material strength, thermal conductivity, etc.

[0617] Secondly, the internal structure, material strength, and thermal conductivity of the oil tank within the corresponding grid are taken as representative parameters of the local oil tank within that grid, and then matrixed to form the oil tank internal structure matrix, oil tank material strength matrix, and oil tank thermal conductivity matrix.

[0618] In this embodiment, considering that four iron cores are distributed inside the fuel tank, and that the areas shielded by these iron cores significantly reduce the rate of temperature rise in a fire, different numbers are defined for the local positions near the iron cores and the remaining positions in the fuel tank's internal structure matrix. This can be understood as setting a distance threshold; positions within this threshold range, with the iron cores as fixed points, are considered local positions near the iron cores. More specifically, in this implementation, local positions near the iron cores are defined as 1, and the remaining positions are defined as 0.

[0619] In this embodiment, the material strength of the fuel tank mainly depends on its local thickness, reflecting the strength differences of the steel at various locations within the tank. These strength differences are primarily caused by processes such as welding and flange connections. Different numerical values ​​are defined for the local material strength at the flange connection and welding locations, the local material strength of the grid adjacent to the flange connection and welding locations, and the local material strength away from the flange connection and welding locations, forming a fuel tank material strength matrix. More specifically, the local material strength at the flange connection and welding locations is defined as 0; the local material strength of the grid adjacent to the flange connection and welding locations is defined as 1; and the local material strength away from the flange connection and welding locations, or in other words, the local material strength at locations other than the flange connection and welding locations and their corresponding adjacent grid locations, is defined as 2.

[0620] In this embodiment, the thermal conductivity of the oil tank mainly depends on the material type, thickness, and structure. The main body of the oil tank is made of steel with uniform thickness, while the material type and thickness vary at the flange and weld locations. The local thermal conductivity at the flange connection, the weld location, and other locations are defined as different numbers, forming an oil tank thermal conductivity matrix. Specifically, the local thermal conductivity at the flange connection is defined as 0, the local thermal conductivity at the weld location is defined as 1, and the local thermal conductivity at other locations is defined as 2.

[0621] In this embodiment, generally, to better assess the risk, the main parameters of the matrixing depend primarily on the most dangerous situation within the corresponding mesh. For local material strength and thermal conductivity, lower values ​​imply a higher probability of fracture, i.e., relatively more dangerous. For internal structures, proximity to the iron core implies a higher probability of fracture, i.e., relatively more dangerous.

[0622] Finally, the internal structure matrix of the fuel tank, the material strength matrix of the fuel tank, and the thermal conductivity matrix of the fuel tank are used as physical matrices.

[0623] S134: Input the temperature matrix, time matrix, and physics matrix into the generative adversarial network to obtain the internal temperature field matrix of the converter transformer tank.

[0624] In this embodiment, the temperature matrix, time matrix, and physics matrix can be preprocessed, such as data pruning, before being input into the generative adversarial network.

[0625] In this embodiment, the generative adversarial network (GAN) is based on the Pix2Pix (Conditional Generative Adversarial Networks) architecture, including a generator and a discriminator. The generator, built using a UNet (Convolutional Neural Network) structure, is responsible for generating false temperature outputs that can fool the discriminator. The discriminator is primarily responsible for distinguishing between genuine and false generated data, thereby enhancing the generator's performance. The mean absolute error (MAE), defined as the difference between the input and predicted output data, is used as the generator's loss function, which can be minimized through a certain number of training iterations. The coefficient of determination (COD) is used to represent the accuracy of the prediction.

[0626] In this embodiment, obtaining the internal temperature field matrix of the converter transformer tank includes:

[0627] The temperature matrix, time matrix, and physics matrix are merged to form an input matrix that facilitates the training of generative adversarial networks.

[0628] In this implementation, the temperature matrix, time matrix, and physical matrix are one-dimensional numerical values. To facilitate generative adversarial network (GAN) computation, they can be expanded to 128*128 matrices. After dimensionality expansion, the temperature matrix, time matrix, and physical matrix are merged to form a 128*128*5 input matrix for GAN training. The GAN output is designed as temperature data at a specific time location, presented as a 128*128*1 output for GAN training.

[0629] The input matrix is ​​fed into the generator as real data to obtain the corresponding fake data.

[0630] The discriminator distinguishes between real data and corresponding fake data, outputting the probability of being real or fake. After calculating the loss, backpropagation is performed to update the generator weights. The generative adversarial network is iteratively trained until the generator can generate a high-quality temperature field matrix that is close to the real data.

[0631] Using a trained generative adversarial network and real data as input, the internal temperature field matrix of the converter transformer tank is obtained.

[0632] In this embodiment, in the field of heat and mass transfer, generative adversarial networks (GANs) rely too heavily on data patterns, which has significant drawbacks: GANs cannot understand physical laws, leading to generated images that do not conform to existing heat transfer and energy conservation laws. Therefore, in this embodiment, the physical characteristics of the transformer tank are used as matrix variables in the dataset. This helps the computer understand the initial physical information and better predict the time series matrix based on this information, effectively and reasonably predicting and improving accuracy.

[0633] S135, based on the temperature field matrix inside the converter transformer tank, determines the overpressure inside the transformer tank, and then predicts the failure location of the converter transformer tank based on the temperature field matrix inside the converter transformer tank and the influence of local high temperature on the strength of the tank material.

[0634] In this embodiment, the internal pressure of the transformer is assessed based on the generated internal temperature field matrix of the converter transformer tank. The vaporization rate is calculated based on the instantaneous temperature field and the thermal properties of the transformer oil. The total vaporization rate of the transformer oil after a potential risk occurs is obtained through integration. This total vaporization rate is then incorporated into the ideal gas law to calculate the overpressure inside the transformer tank.

[0635] In this embodiment, in existing accidents, potential rupture points in the oil tank are mainly concentrated at flange connections, welds, and the center of the steel. Whether the metal or flange connection ruptures depends primarily on whether the local stress exceeds a critical condition. Studies show that the critical condition mainly depends on material strength and local temperature, while the magnitude of the local stress mainly depends on the internal pressure of the oil tank. Next, a semi-quantitative analysis of the rupture probability from the perspectives of pressure and temperature will be conducted based on the internal temperature field matrix of the transformer oil tank and physical derivations.

[0636] In this embodiment, the amount of gas produced by the transformer oil mainly depends on the gas production efficiency and the gas production time. Transformer oil will undergo cracking and evaporation when exposed to high temperatures. The main products of high-temperature cracking include ethane and hydrogen, and the gas production rate increases with increasing temperature. According to the ideal gas law:

[0637] PV = nRT;

[0638] Where P, V, n, R, and T represent the gas state parameters: pressure, volume, number of particles, Boltzmann constant, and temperature, respectively.

[0639] It can be seen that, under the condition that the internal volume of the transformer oil tank remains constant (without rupture), an increase in the number of gas moles and temperature will cause a significant increase in the internal pressure of the oil tank. It should be noted that an increase in temperature will increase the gas production efficiency of the transformer oil.

[0640] P∝nT;

[0641]

[0642] Furthermore, differentiating the gas production time and coupling it with the temperature field can estimate the total gas production. The local gas production mainly depends on the local temperature range and gas production, and the following equation can be derived:

[0643]

[0644] Therefore, the overpressure inside the transformer tank is:

[0645]

[0646] In the formula, P t The overpressure inside the transformer tank is given by δ, a constant characterizing the gas production efficiency of the transformer oil, and P0 is the initial ambient pressure. t ΔT represents the total vaporization amount of transformer oil after a potential risk occurs, ΔT represents the temperature change, T0 represents the initial ambient temperature, n represents the number of gas particles, t represents time, XYZ represents the spatial coordinates, and n0 represents the vaporization amount.

[0647] In this embodiment, it should be noted that the pressure on each surface corresponding to the grid can be discretized and integrated to form an overpressure matrix for each surface.

[0648] In the embodiments, regarding the influence of localized high temperatures on the strength of the oil tank material, the oil tank material of oil-immersed transformers is typically high-quality steel. Taking the more common G250 and G550 as examples, the ultimate bearing capacity of the steel decreases with increasing temperature. Studies have shown that when the temperature exceeds 350℃, the steel strength decreases significantly. In research, 400℃ is usually considered as the critical temperature for the decrease in steel strength. Furthermore, experimental results of the mechanical properties of Q345 steel at different temperatures show that the steel strength generally decreases almost linearly with increasing temperature. At 450℃, the mechanical properties have already decreased to half of those at room temperature. Similarly, the mesh is discretized and integrated according to the pressure on each surface to form a matrix of the decrease in mechanical properties (temperature) for each surface.

[0649] In this embodiment, based on the above analysis, the internal overpressure and steel strength reduction caused by high internal temperature during a fault are directly proportional and inversely proportional to the local temperature, respectively. Combining the characteristics of objects inside the tank with the temperature field generated by a generative adversarial network, a semi-quantitative evaluation of localized ruptures and ejections in the transformer tank, or at potentially hazardous locations, can be performed.

[0650]

[0651] In the formula, B 失效矩阵 Let λ be the predicted transformer tank failure risk matrix, where λ represents the impact of temperature rise and gas production on B. 失效矩阵 The constant of influence, This is the physical matrix in the temperature field matrix inside the transformer tank. This represents the time-step-based temperature matrix within the internal temperature field matrix of the transformer tank; P t This refers to the overpressure inside the transformer tank. This represents the inverse relationship between temperature and the strength of the fuel tank material.

[0652] This application embodiment uses a physically constrained generative adversarial neural network to learn the internal temperature rise data of the converter transformer tank and predict the internal temperature field of the tank after a certain period of time. The prediction results are combined with the gas pressure balance equation and the failure threshold of the tank steel to obtain the potential hazards and dangerous areas at a specific future time. In this way, it can effectively avoid practitioners relying solely on experience to judge the hazards. The prediction results based on the physical model can serve as a reference for real-time monitoring and early warning of fire risks. This application embodiment is based on matrix calculation and can complete the calculation and risk identification within seconds, effectively assisting relevant personnel to make decisions more quickly and accurately, and to more efficiently assess the risk of converter transformer tank rupture, thereby ensuring the fire safety of the converter station.

[0653] In this embodiment, a failure heatmap for future moments can be depicted based on the predicted risk matrix of the converter transformer tank, and the potential location and magnitude of the highest probability of rupture can be inferred from the calculation results. Based on this, rapid treatment of localized areas can be prioritized to reduce the risk of rupture and fire, thereby providing strong guidance for emergency response and rescue strategies.

[0654] As mentioned above, the converter station risk management module in this application embodiment can be used to perform site-level risk management on the converter station connected to the pipeline, so as to realize the fire risk assessment of the entire converter station. Compared with the fire risk assessment of a single device, the fire risk situation of the converter station can be clearly defined from a macro perspective.

[0655] In one possible embodiment, the converter station risk management module can be used to: determine the importance weight of each fire protection subsystem in the converter station, wherein the fire protection subsystem includes at least one of a fire protection facility system, a fire alarm system, a fire extinguishing system, and a fire evacuation system; determine the equipment failure probability of each fire protection equipment based on the failure event tree model of each fire protection equipment in each fire protection subsystem; determine the system failure probability of each fire protection subsystem based on the equipment failure probability of each fire protection equipment in each fire protection subsystem, and determine the subsystem health status of each fire protection subsystem based on the system failure probability of each fire protection subsystem; and then determine the comprehensive fire protection status assessment result of the converter station based on the importance weight and subsystem health status of each fire protection subsystem.

[0656] This application combines the Analytic Hierarchy Process (AHP) and fault tree analysis, using equipment health status as model input to obtain the comprehensive fire safety assessment results of the converter station. Fire protection systems, fire alarm systems, fire extinguishing systems, and safe evacuation systems are used as indicators to measure the comprehensive fire safety status. The AHP is employed to measure the importance of different indicators and obtain their weights, i.e., the importance weights of each fire protection subsystem. Specifically, the fire alarm system is crucial for ensuring timely fire detection and triggering other emergency responses, serving as the first step in fire safety; the effectiveness of the fire extinguishing system directly affects fire suppression, enabling rapid extinguishing of fire sources and minimizing losses; the fire protection system is vital in preventing fire spread and providing structural protection during a fire; although the safe evacuation system is crucial for personnel safety, it is typically an emergency measure implemented after a fire occurs, and its priority is slightly lower than the aforementioned systems. Based on this consideration, for example, the importance weights of the fire protection system, fire alarm system, fire extinguishing system, and safe evacuation system are 0.14, 0.5, 0.28, and 0.08, respectively.

[0657] For each fire protection subsystem, the health status of each fire protection IoT device contained in each subsystem is used as the basis for evaluating the health status of the subsystem. For example, the health status of IoT devices such as fire doors, fire shutters, and fire dampers is used to measure the health status of the fire protection facility system. The health status assessment of the subsystem will be combined with the fault tree analysis method, which can obtain the mutual influence between devices. For example, in the fire extinguishing system, the fire water tank and the elevated fire water tank can support each other. For example, the failure of the water supply system will lead to the failure of the fine water mist fire extinguishing system and the automatic sprinkler fire extinguishing system.

[0658] For basic events, i.e., the failure probability of various types of fire protection IoT devices, the total number of telemetry and teleindication data types contained in the fire protection device and the total number of fire protection IoT devices of that type will be considered. For individual fire protection IoT devices, the telemetry and teleindication sensor with the worst health status will be used as the health status assessment result for that device. For example, a fire pump device includes operational status telemetry data and two telemetry signal collection sensors for flow and pressure; the worst health status among these three will be used as the health status assessment result for that fire pump device. For basic events in the fault tree, the total number of devices of a certain type needs to be considered. For example, if a fire pump fails, the health status of all fire pumps needs to be measured. The probability of a single basic event can be expressed as:

[0659]

[0660] Where P is the failure probability of the basic event, N is the total number of fire-fighting equipment of the type corresponding to the basic event, and p i p represents the failure probability of a single device under this type, where the failure probability of a single device is transformed by the health status. When the health status is healthy, p i =0, when the health state is sub-healthy, pi =0.1, when the health status is abnormal, p i =0.5, when the health status is faulty, p i =1.0.

[0661] For example, regarding fire protection systems, please refer to [link / reference]. Figure 15 The diagram shows a failure fault tree for a fire protection system. The fault tree involves four basic events for four types of fire protection IoT devices, specifically: fire door failure, fire shutter failure, fire damper failure, and oil storage and containment facility failure. Taking the fire door as an example, the probability of the basic event of fire door failure is determined by the total number of telemetry and teleindication data types contained in the fire protection IoT device (i.e., the fire door) and the total number of fire protection IoT devices of that type.

[0662] When all the equipment in a fire protection subsystem is faulty, the probability of failure of the subsystem is the same as the probability of failure of any single type of equipment it contains, which is 100%. Therefore, when all the equipment in a fire protection subsystem is faulty, the subsystem is determined to be faulty and receives the lowest fault score. When all the equipment in a fire protection subsystem is abnormal, the subsystem is determined to be abnormal and receives the lowest abnormal score. When all the equipment in a fire protection subsystem is in a sub-healthy state, the subsystem is determined to be in a sub-healthy state and receives the lowest abnormal score. When all the equipment in a fire protection subsystem is healthy, the subsystem is determined to be healthy and receives a perfect score.

[0663] In this embodiment, the health status of each fire protection subsystem in the converter station is determined based on the failure event tree model of each fire protection equipment in each fire protection subsystem, thereby obtaining the comprehensive fire protection status assessment result of the converter station. This achieves comprehensive fire protection status monitoring at the station level and improves the overall fire protection control capability of the entire converter station. Furthermore, since each fire protection subsystem in the converter station is particularly important for fire protection control, assessing the fire protection status of each subsystem can largely characterize the overall fire protection status of the converter station.

[0664] In one possible embodiment, the converter station risk management module can be used to: determine fire control business events for each risk management object of the converter station, wherein the risk management object includes at least one of fire fighting operations, fire fighting equipment, and fire trucks; determine the risk level of each fire control business event based on at least two of the severity, occurrence, and detectability, wherein the risk level includes high risk, medium risk, and low risk; set a risk level identifier matching the risk level for the corresponding fire control business event based on the risk level of each fire control business event, and control the association and display of each fire control business event and its corresponding risk level identifier.

[0665] Among them, fire control business events corresponding to fire fighting operations include fire patrols, fire maintenance, fire testing, and fire assessments; fire control business events corresponding to fire equipment include fire water supply systems, water mist extinguishing systems, and compressed air foam extinguishing systems; and fire control business events corresponding to fire trucks include on-site fire personnel obtaining certificates, fire personnel being dispatched, and maintenance personnel obtaining certificates.

[0666] Furthermore, the risk level of each fire control business event can be calculated based on at least two of the severity (S), occurrence (O), and detectability (D) of the fire control business event corresponding to each risk control object.

[0667] In one implementation, the risk level of each fire control event can be determined by comparing the risk priority coefficient corresponding to the product of severity (S), occurrence (O), and detectability (D) with a preset threshold coefficient. Specifically, for a fire control event, the risk priority coefficient (RPN) is calculated as S × O × D, where RPN represents the risk priority coefficient. The RPN is then compared with a preset threshold coefficient. For example, the preset threshold coefficient may include multiple levels, such as 160 and 40. If the RPN is greater than 160, it is determined to be high risk; if the RPN is less than 40, it is determined to be low risk; and if the RPN is between 40 and 160, it is determined to be medium risk.

[0668] In another implementation, a severity-occurrence matrix (S×O matrix), a severity-detectability matrix (S×D matrix), or an occurrence-detectability matrix (O×D matrix) can be formed based on any two of the severity (S), occurrence (O), and detectability (D) corresponding to each fire control business event. The risk matrix is ​​then sorted based on the obtained severity-occurrence matrix, severity-detectability matrix, or occurrence-detectability matrix. The risk level of each fire control business event is then determined based on the obtained risk matrix sorting. For example, an RMR matrix can be obtained based on the S×O matrix, S×D matrix, and O×D matrix, and the risk level of the corresponding fire control business event can then be determined based on the RMR matrix.

[0669] Furthermore, risk level identifiers matching the risk level of each fire control business event are set. These risk level identifiers can be special color markings, such as red, yellow, and green codes to represent high risk, medium risk, and low risk, respectively. In other words, each fire control business event is assigned a red, yellow, or green code based on its risk level. This explicit code marking can clearly indicate the risk level, so as to provide a noticeable reminder to the relevant personnel.

[0670] In one possible embodiment, the converter station risk management module can be used to: determine the management indicator score of each primary management indicator based on the score and weight of the lower-level management indicators corresponding to each primary management indicator in the converter station, wherein the primary management indicators include at least one of fire equipment management indicators, fire operation management indicators, fire vehicle management indicators, and fire personnel management indicators; and then determine the comprehensive fire management status of the converter station based on the weight and score of each primary management indicator.

[0671] The lower-level management indicators for firefighting operations can include, for example, daily maintenance and emergency response; the lower-level management indicators for firefighting equipment can include, for example, status assessment; the lower-level management indicators for fire trucks can include, for example, vehicle management; and the lower-level management indicators for firefighters can include, for example, their performance capabilities and on-site firefighting status. In other words, the converter station can be comprehensively managed from four dimensions: firefighting equipment, firefighting operations, fire trucks, and firefighters. This comprehensive management index reflects the overall management level and effectiveness of the fire station, thereby enhancing the overall management capabilities of the converter station.

[0672] As described above, the fire emergency decision-making module in the converter station's digital fire control system can be used to make emergency decisions based on acquired fire scene data (including video data and various sensor data) during a fire emergency. In actual fire emergency decision-making processes, the video data transmitted from the converter station's cameras to the digital fire control system may not meet the current fire emergency decision-making needs, leading to an inability to make timely and effective fire emergency decisions or resulting in erroneous fire emergency decisions. Therefore, this application provides a video-based fire emergency handling method for converter stations. This method can be executed by the fire emergency decision-making module within the fire emergency handling module, or it can be directly executed by the fire emergency handling module itself. That is, either the fire emergency handling module or the fire emergency decision-making module can be used to execute this video-based fire emergency handling method for converter stations. Please refer to... Figure 16 As shown, the method includes S161-S164, which are described below in conjunction with... Figure 16 The fire emergency response method for converter stations based on video recall, according to an embodiment of this application, will be described.

[0673] S161: Obtain the primary video data sent by the primary camera group associated with the fire object in the converter station. The primary video data includes video data of the fire object from multiple perspectives captured by primary cameras at multiple locations.

[0674] Multiple cameras are deployed in various areas of the converter station to provide video surveillance of the station's various areas and equipment. In this embodiment, to enable comprehensive video surveillance of objects where a fire may occur, multi-level cameras can be pre-deployed for these objects. Specifically, the equipment within the converter station prone to fire can be identified, such as valve towers, converter valves, converter transformers, DC field equipment, and cable trenches. Multi-level camera groups can then be deployed for these devices. For example, two-level camera groups can be deployed for some conventional equipment, while three-level camera groups can be deployed for some critical equipment. The spatial deployment positions of each level of camera group differ, including the distance and height from the equipment.

[0675] For example, see Figure 17 The diagram shows a two-tiered camera group deployed around equipment within a station. The first-tier camera group includes cameras 1-1, 1-2, 1-3, and 1-4, all positioned around the equipment. The second-tier camera group includes cameras 2-1, 2-2, 2-3, 2-4, 2-5, 2-6, 2-7, and 2-8, also positioned around the equipment. The distance between the first-tier camera group and the equipment is less than the distance between the second-tier camera group and the equipment. This can be understood as the distance between the multi-tiered camera group and the equipment increasing progressively, and the cameras in each tier are deployed as close as possible to the equipment. For example... Figure 3 The circular deployment shown can also be replaced with a square or triangular deployment. This allows cameras at each level to capture video data from multiple angles within the station, providing a comprehensive view of the equipment's condition from various perspectives. This enables timely and effective fire emergency decisions in the event of a fire. Furthermore, camera groups at different levels can have different or the same deployment height. It's important to understand that "different or the same" here refers to approximate heights. Figure 3 For example, cameras 1-2 in the primary camera group are deployed at a height of approximately 4 meters, while cameras 2-3, 2-4, and 2-5 in the corresponding secondary camera group are deployed at heights of approximately 4.8 meters, 4 meters, and 3.3 meters, respectively. In this way, cameras 1-2 and cameras 2-3, 2-4, and 2-5 can capture video from different heights within the station from the same perspective, so as to obtain video data of the equipment at different heights from the same perspective.

[0676] In this embodiment, each device within a station (e.g., a fire-prone object) can have a corresponding primary camera group and a secondary camera group. That is, the primary and secondary camera groups associated with each device within a station can be different. In other embodiments, if two devices within a station are located close to each other, they can share the same pair of primary and secondary camera groups. The secondary camera group and primary camera group of a certain device within a station have a positional mapping relationship with the fire-prone object's location as the same viewing angle center. In other words, each secondary camera in the secondary camera group has a positional mapping relationship with a certain primary camera in the primary camera group. This positional mapping relationship uses the location of the device within the station as a reference point, ensuring that the secondary and primary cameras have the same viewing angle range with the device's location as the center. In other words, the secondary camera and its corresponding primary camera have the same shooting viewing angle range with the device's location as the reference point. That is, there is a viewpoint backup / substitute relationship between multiple camera groups. Thus, if a primary camera malfunctions or the captured video data does not meet the conditions for fire emergency use, a secondary camera with the same viewing angle corresponding to the primary camera can be used as a substitute, thereby improving the camera's fault tolerance. Continuing with... Figure 3 For example, the secondary cameras that correspond to the primary camera 1-1 are cameras 2-1 and 2-2. It can be seen that the primary camera 1-1 and the secondary cameras 2-1 and 2-2 have roughly the same shooting angle range with the location of the fire-affected device in the picture as the center of alignment. For example, they all shoot video from the "lower left angle" of the fire object. When the primary camera 1-1 malfunctions or the video data captured by the primary camera 1-1 is unavailable, the secondary cameras 2-1 and / or 2-2 can be quickly selected as substitutes for the primary camera 1-1 to continue shooting video data from the "lower left angle" of the fire object.

[0677] For example, the equipment or area in a converter station where a fire occurs is referred to as the fire target. To promptly handle fire emergencies related to the fire target, primary video data sent by the primary camera group associated with the fire target can be acquired. This primary video data can include video data of the fire target captured from multiple perspectives by multiple primary cameras deployed in multiple locations within the primary camera group. Continuing with... Figure 3 For example, Level 1 video data may include multi-view video data captured by Level 1 cameras 1-1, 1-2, 1-3, and 1-4 from their respective perspectives. Level 1 video data can be used to view and analyze the real-time situation of the fire from multiple angles, such as determining the area where the fire is located, the type of fire, the range of the fire, the base of the flames, and other equipment, personnel, or objects around the fire, so as to promptly determine fire emergency plans or fire emergency rescue schemes based on this information and improve fire emergency efficiency.

[0678] S162: When the first video data of the first perspective of the fire object determined based on the first-level video data does not meet the emergency auxiliary decision-making conditions, the first camera that captures the first video data is determined from the first-level camera group, and the second camera corresponding to the position of the first camera is determined from the second-level camera group that has a positional mapping relationship with the first-level camera group; wherein, the second-level camera group and the first-level camera group have a positional mapping relationship with the same perspective center as the position of the fire object.

[0679] After obtaining the primary video data captured by the primary camera group, the video data corresponding to each viewpoint in the primary video data is analyzed. If the video data of each viewpoint meets the emergency auxiliary decision-making conditions, then fire emergency decisions are made based on this video data. If the video data of one or more viewpoints (e.g., referred to as the first viewpoint) does not meet the emergency auxiliary decision-making conditions, then secondary cameras with the first viewpoint can be quickly linked for video replacement. First, the primary camera capturing the first video data from the first viewpoint (e.g., referred to as the first camera) is determined from the primary camera group. Then, based on the positional mapping relationship between the primary and secondary camera groups, a secondary camera with the same field of view as the first camera (e.g., referred to as the second camera) is determined from the secondary camera group. It can be understood that there can be one or more secondary cameras. Continuing... Figure 3 For example, if the first video data captured from a first-person perspective does not meet the emergency auxiliary decision-making conditions, then the second-level cameras with the same perspective include second-level cameras 2-3, 2-4, and 2-5. In this case, the second-level cameras (i.e., the second cameras) used as backup videos can be one or more of the second-level cameras 2-3, 2-4, and 2-5. That is, some or all of the second-level cameras with the same backup perspective as the first camera are selected to participate in the subsequent video-based fire emergency response.

[0680] In specific implementation, determining that the first video data does not meet the emergency auxiliary decision-making conditions may include at least one of the following: determining that the clarity of the first video data is lower than a clarity threshold, indicating that the clarity of the first video data is too low to meet the video analysis requirements and therefore cannot be used for fire emergency decision-making; or determining that the resolution of the first video data is lower than a resolution threshold, indicating that the resolution of the first video data is too low to meet the video analysis requirements and therefore cannot be used for fire emergency decision-making; or determining that the duration of obstruction of a key target object (such as a key component of equipment or the ignition point of equipment) in the first video data exceeds a predetermined duration, which would seriously affect the assessment of the key target. The emergency response judgment of the target object is therefore deemed unsuitable for fire emergency decision-making; or, if the contribution of the first video data in joint emergency decision-making with other video data in the first-level video data is less than the contribution threshold, for example, if the contribution of the first video data in joint emergency decision-making with other video data in the first-level video data other than the first video data is low in the entire joint emergency decision-making task to determine the location of the fire object, analyze the fire environment, and predict the fire trend, then it is deemed unsuitable for this fire emergency decision-making. That is, joint emergency decision-making includes at least one of the following: fire object location, fire environment detection, and fire trend prediction.

[0681] Among these, at least one of the aforementioned sharpness threshold, resolution threshold, predetermined duration, and viewpoint contribution threshold is determined based on the attribute information of the fire object and / or the fire scene in which the fire object is located. In other words, the corresponding threshold can be determined according to the specific fire scene conditions to adapt to the current fire situation and improve the accuracy of judging whether the emergency auxiliary decision-making conditions are met.

[0682] Before selecting the second camera, it is necessary to accurately select the second-level camera group that has a positional mapping relationship with the first-level camera. In this regard, the embodiments of this application provide the following two implementation methods.

[0683] In the first embodiment, the secondary camera group can be pre-configured, that is, a pre-configured static camera group. In this case, the secondary camera group can be determined as follows: based on the location of the fire object, the secondary camera group is determined from candidate camera groups that have a positional mapping relationship with the primary camera group. A candidate camera group is determined with a fire monitoring position within the shared field of view of the primary cameras in the primary camera group as the azimuth center. The shared field of view of the primary camera group includes at least one fire monitoring position, which includes the location of the fire object. In other words, within the shared field of view of all primary cameras in the primary camera group, multiple on-site devices that need to monitor the fire can be included. The location of each on-site device that needs to be monitored corresponds to a fire monitoring position. For each of the multiple fire monitoring positions, one or more secondary cameras corresponding to each primary camera position can be determined with that fire monitoring position as the same azimuth center, thereby obtaining the secondary camera group corresponding to the primary camera group of that fire monitoring position. For each fire monitoring location, a corresponding primary camera group and a secondary camera group are pre-configured. When the secondary camera is linked based on the primary camera, the secondary camera that needs to be replaced can be quickly located, resulting in high linkage efficiency.

[0684] In the second implementation, the secondary camera group can be dynamically determined by the system. For example, the secondary camera group can be dynamically determined for the fire target in the following way: using the position of the fire target as the center of the same viewing angle, determine reference cameras that have a viewpoint candidate relationship with the cameras in the primary camera group. Specifically, for each primary camera in the primary camera group, determine a reference camera with a viewpoint candidate relationship with it; then acquire the basic spatial data of the primary camera group and the determined reference cameras. The basic spatial data of the cameras includes at least one of the following: camera position information, height information, angle information, viewing angle range information, and environmental information of the camera; construct a graph based on the basic spatial data of the primary camera group and the reference cameras. The network model uses each primary camera and each reference camera in the primary camera group as a node in the graph neural network model. The node features of each node include the basic spatial data of the corresponding camera. The edges connecting nodes indicate that the node has a viewpoint candidate relationship with the corresponding camera. That is, only two nodes with viewpoint candidate relationships are connected by an edge. The edge weight is determined based on at least one of the following: viewpoint candidate range, relative distance, line-of-sight accessibility, and historical collaborative effectiveness score between the corresponding cameras. Then, using the constructed graph neural network model, the reference cameras that satisfy the position mapping relationship with each primary camera in the primary camera group are selected from the reference cameras to form the secondary camera group based on the edge weights.

[0685] In other words, the secondary camera group can be dynamically determined through graph neural networks. This method can flexibly and accurately determine the secondary camera group based on the fire scene conditions and the relative spatial relationship and historical coordination of each primary camera and each reference camera, so as to achieve effective linkage and complementarity with the primary cameras.

[0686] After dynamically determining the secondary camera group based on the second method described above, in one implementation, the second camera corresponding to the first camera in the primary camera group can be determined from the secondary camera group in the following manner: the position of the first camera, the environmental information of the first camera, and the emergency auxiliary decision-making requirement information are input into a graph neural network model to obtain at least one candidate camera that has a viewpoint complementarity relationship with the first camera; then, the viewpoint complementarity score of each candidate camera relative to the first camera is determined, wherein the viewpoint complementarity score is determined based on at least one of the following: the viewpoint complementarity range of the candidate camera relative to the first camera, the missing information gain of the candidate camera under the environmental information, the predicted image quality of the candidate camera, and the collaborative feasibility between the candidate camera and the first camera; the second camera corresponding to the position of the first camera is determined based on the viewpoint complementarity score of each candidate camera relative to the first camera.

[0687] For the scoring factors used to determine the complementarity score, the candidate camera's viewpoint range relative to the first camera indicates the overlap between the candidate camera's viewpoint relative to the fire object and the first camera's viewpoint relative to the fire object. A larger overlap indicates that the candidate camera can provide a wider field of view for the angles missing by the first camera, completely or partially covering the first camera's viewpoint range. The missing information gain of the candidate camera under its environmental information indicates how much key information missing from the first-level camera can the candidate camera provide under the current environment (smoke, lighting) and its own condition, such as the clarity of the fire base, the location of the fire front, and the situation of people in a specific area. The more key information provided, the greater the corresponding missing information gain. The predicted image quality of the candidate camera indicates the predicted video quality of the candidate camera under the current environment, such as clarity and occlusion level. The feasibility of collaboration between the candidate camera and the first camera can be determined by considering factors such as network conditions, computing resources, and physical line-of-sight accessibility (whether there is dynamic occlusion between the candidate camera and the fire object).

[0688] In other words, the second camera with the best view complementarity can be determined from at least one candidate camera with view complementarity relationship with the first camera using a graph neural network. In specific implementation, the corresponding weights can be assigned to each scoring factor for determining the view complementarity score according to the current emergency auxiliary decision-making needs information, and then the view complementarity score of each candidate camera can be calculated by weighted summation. Finally, the second camera is selected based on the obtained view complementarity score.

[0689] The embodiments of this application use a graph neural network model to determine the optimal secondary camera for the first camera. By taking into account the current emergency auxiliary decision-making needs and the scoring factors of various influencing perspective complementarity scores, the accuracy and effectiveness of secondary camera selection can be improved, thereby enhancing the reliability of fire emergency response.

[0690] In this embodiment, a second camera corresponding to the position of the first camera is determined based on the complementarity score of the viewpoints of each candidate camera relative to the first camera. Specifically, the following implementation methods may be included.

[0691] In the first implementation, the candidate camera with the highest view complementarity score is determined as the second camera. In this case, there can be one or more second cameras.

[0692] In the second implementation, candidate cameras whose viewpoint complementarity score is greater than the first score threshold are identified as second cameras. In this case, there can be one or more second cameras.

[0693] In the third embodiment, when the view complementarity scores of each candidate camera are all less than the second score threshold, the candidate cameras with the top N view complementarity scores are selected as the second cameras, where N is a natural number. In this case, there can be one or more second cameras, and the second cameras selected in this way are used to synthesize simulated viewpoints, which are used to synthesize the second video data.

[0694] S203: Obtain the second video data of the fire object from the first-person perspective corresponding to the second camera.

[0695] The second cameras identified using the first and second methods above both have high scores for complementary viewpoints, indicating that they can effectively replace the first camera. In this case, the video data captured by the second camera can be directly used as second video data in fire emergency decision-making, which is highly efficient.

[0696] According to the third method described above, the complementary viewpoint score of the second camera is relatively low. In this case: a third video data of the first simulated viewpoint can be synthesized based on the video data obtained by the second camera (e.g., alternative video data), and the synthesized third video data can be used as the second video data of the fire object in the first viewpoint, wherein the first simulated viewpoint is within the shared viewpoint range of the alternative video data corresponding to at least one second camera; or, a fourth video data of the second simulated viewpoint can be synthesized based on the first video data and at least one alternative video data, and the fourth video data can be used as the second video data of the fire object in the first viewpoint, wherein the second simulated viewpoint is within the shared viewpoint range of the alternative video data corresponding to at least one alternative second camera and the overlapping viewpoint range of the first viewpoint. In other words, for example, if the fire environment (such as smoke at the fire scene, or displacement or damage to buildings or cameras at the fire scene) affects the candidate cameras, resulting in low visual complementarity scores for each candidate camera, the video data captured by these candidate cameras is not directly used as a substitute video for the first camera. Instead, video synthesis is performed based on the video data captured by these secondary cameras with low visual complementarity scores, using specific simulated perspectives. This better meets the video requirements of the fire target from the first perspective, improving the effectiveness and reliability of fire emergency decision-making. Furthermore, to improve emergency response efficiency and reduce data processing volume, keyframe images from the candidate video data can be selected for synthesis during video synthesis.

[0697] S204: Conduct fire emergency response based on primary and secondary video data.

[0698] After obtaining the primary video data collected by the primary camera group and the secondary video data corresponding to the secondary cameras, the primary and secondary video data can be visualized and displayed. This allows for further fire emergency response and timely, reliable fire emergency decision-making. For example, video data from different perspectives can be spatially aligned and fused, and then mapped into a view in a unified coordinate system based on precise location, such as 3D reconstruction, panoramic stitching, and bird's-eye view generation, to help determine effective fire emergency decision-making solutions.

[0699] In this embodiment, as the fire continues to develop, the fire development trend corresponding to the fire object can be predicted. Then, based on the predicted fire development trend, a key perspective is determined. This key perspective is either the perspective from which key video data needs to be captured relative to the fire object or the perspective of a primary camera that has failed relative to the fire object. Next, a backup camera corresponding to the key perspective is determined from the secondary camera group, and a preparation command is sent to the backup camera. This preparation command instructs the backup camera to begin acquiring or prepare to acquire video data of the fire object from the key perspective. In other words, based on changes in the fire trend, key perspective prediction and secondary camera preloading can be proactively performed, enabling predictive calling of secondary cameras. This reduces the delay in calling secondary cameras when they are indeed needed, improves video transmission efficiency, and ultimately enhances the efficiency of fire emergency decision-making.

[0700] In addition, after obtaining the second video data of the fire object from the first perspective corresponding to the second camera, the second video data needs to be sent to the system as soon as possible.

[0701] In one implementation, a priority transmission identifier can be included in the second video data. This priority transmission identifier indicates that the second video data should be transmitted to the system with high priority, so that the video data from the secondary camera can be transmitted to the system as soon as possible to assist in fire emergency decision-making and improve the efficiency of fire emergency decision-making.

[0702] In another implementation, the identifier mapping between the device identifier of the second camera and the device identifier of the first camera in the second video data can be controlled, so that the system can verify the validity of the second video data based on this identifier mapping. Figure 3 For example, if the device identifier of the first camera in a primary camera group is "1-2" and the device identifier of the determined second camera is "2-3", then the reported second video data can carry the correspondence between "1-2" and "2-3". By carrying the correspondence between the identifiers of the secondary camera and its corresponding primary camera in the reported second video data, the system can directly verify the second video data after receiving it by checking the correspondence between the device identifiers of the second and first cameras. For example, it can verify whether there is a positional mapping relationship between these two device identifiers, and whether the secondary camera previously indicated by the system as a substitute for the first camera is indeed the secondary camera with the device identifier "2-3". This is equivalent to performing a preliminary check on the second video data through this correspondence, thereby improving the validity of the second video data.

[0703] The video-based emergency response method for converter stations in this application first acquires primary video data sent by a primary camera group associated with a fire in the converter station. This primary video data includes video data from multiple perspectives of the fire captured by primary cameras at multiple locations within the primary camera group. This multi-perspective video data allows for a more comprehensive understanding of the fire's situation, aiding in timely and effective fire emergency decisions. If, based on the primary video data, the first video data from the first perspective of the fire does not meet the conditions for emergency decision support, the method identifies the first camera capturing the first video data from the primary camera group and the second camera corresponding to the first camera's position from a secondary camera group that has a positional mapping relationship with the primary camera group. The secondary camera group and the primary camera group have a positional mapping relationship with the fire's position as the center of alignment. Then, the method acquires the second video data of the fire from the first perspective corresponding to the second camera, enabling fire emergency decision processing based on both the primary and secondary video data. In other words, when video data captured by a camera in the primary camera group (i.e., the first camera) from its corresponding shooting angle (i.e., the first-person perspective) is insufficient to assist in fire emergency decision-making, a secondary camera (i.e., the second camera) at its corresponding location can be quickly linked as a backup / alternate perspective for the first-person perspective. This allows for the rapid acquisition of alternative video data (i.e., second video data) from the primary video data that does not meet the requirements of the first-person perspective. Timely and effective fire emergency decisions are then made based on the alternative / alternate video data from the first-person perspective and the primary video data from other perspectives. Thus, by rapidly linking primary and secondary cameras, alternative video data from a specific perspective of the fire can be used for emergency response, improving the timeliness, effectiveness, and reliability of fire emergency response at the converter station.

[0704] Based on the same inventive concept, this application also provides a digital fire protection management and control method for converter stations, the method comprising:

[0705] The fire protection IoT data of the converter station is acquired, and an anomaly detection model is used to detect anomalies in the acquired fire protection IoT data. The anomaly detection model is trained based on normal time-series samples generated during the operation of the converter station and anomaly time-series samples generated based on the normal time-series samples. The anomaly time-series samples include at least one of the following: trend anomaly samples, fixed deviation anomaly samples, accuracy reduction anomaly samples, constant value anomaly samples, and discrete point anomaly samples.

[0706] The fire protection management of the converter station is carried out based on the fire protection IoT data of the converter station. The daily fire protection management includes at least one of the following: fire protection equipment management, fire protection business management, fire protection personnel management, station-side key work equipment management, converter station risk management, and fire hazard management.

[0707] Based on at least one of the converter station's fire protection IoT data and daily fire protection management results, identify converter stations that have experienced or are showing signs of potential danger, and carry out fire emergency response for these stations.

[0708] The embodiments of the above-mentioned digital fire protection management and control method for converter stations can be found in the embodiments of the above-mentioned digital fire protection management and control system for converter stations.

[0709] Based on the same inventive concept, this application also provides a converter station fire protection digital control system, which includes a converter station fire protection digital control platform and at least one converter station, wherein:

[0710] The at least one converter station is used to acquire fire protection IoT data of the converter station and send the acquired fire protection IoT data to the converter station fire protection digital management and control platform;

[0711] The converter station fire protection digital management and control platform is used for: acquiring fire protection IoT data sent by the converter station; using an anomaly detection model to detect anomalies in the acquired fire protection IoT data; wherein, the anomaly detection model is trained based on normal time-series samples generated during the operation of the converter station and anomaly time-series samples generated based on the normal time-series samples, the anomaly time-series samples including at least one of trend anomaly samples, fixed deviation anomaly samples, accuracy reduction anomaly samples, constant value anomaly samples, and discrete point anomaly samples; performing daily fire protection management of the converter station based on the fire protection IoT data of the converter station, the daily fire protection management including at least one of fire equipment management, fire business management, fire personnel management, station-side key work equipment management, converter station risk management, and fire hazard management; and identifying converter stations that have experienced emergencies or have potential emergencies based on at least one of the fire protection IoT data of the converter station and the management results of the daily fire protection management module, and performing fire emergency response on the converter stations that have experienced emergencies or have potential emergencies.

[0712] The converter station fire control platform can correspond to the above embodiments (e.g.) Figure 2 The converter station fire protection digital control system in this application is described above. The specific implementation of the converter station fire protection digital control system in this embodiment can refer to the above embodiments and will not be repeated here.

[0713] Furthermore, based on the same inventive concept, embodiments of this application provide a computer storage medium storing a computer program, which, when executed, implements the steps of the aforementioned converter station fire protection digital control method.

[0714] Furthermore, based on the same inventive concept, this application provides a computer program product, which includes a computer program that, when executed, implements the steps of the aforementioned converter station fire protection digital control method.

[0715] The module division in this embodiment is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in each embodiment of this application can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0716] In some possible implementations, various aspects of the converter station fire protection digital control method provided in the embodiments of this application can also be implemented in the form of a program product, which includes program code. When the program product is run on a computer, the program code is used to cause the computer to perform the steps in the converter station fire protection digital control method according to the various exemplary embodiments of this application described above.

[0717] In some possible implementations, various aspects of the converter station fire protection digital control method provided in the embodiments of this application can also be implemented in the form of a program product, which includes program code. When the program product is run on a computer, the program code is used to cause the computer to perform the steps in the converter station fire protection digital control method according to the various exemplary embodiments of this application described above.

[0718] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0719] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0720] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0721] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0722] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0723] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A digital fire protection control system for converter stations, characterized in that, The converter station's digital fire control system includes a converter station data processing module, a daily fire management module, and a fire emergency response module, wherein: The converter station data processing module is used to acquire fire protection IoT data of the converter station and use an anomaly data detection model to detect anomalies in the acquired fire protection IoT data. The anomaly data detection model is trained based on normal time-series samples generated during the operation of the converter station and anomaly time-series samples generated based on the normal time-series samples. The anomaly time-series samples include at least one of the following: trend anomaly samples, fixed deviation anomaly samples, accuracy reduction anomaly samples, constant value anomaly samples, and discrete point anomaly samples. The fire protection daily management module is used to perform daily fire protection management of the converter station based on the fire protection IoT data of the converter station. The daily fire protection management includes at least one of the following: fire protection equipment management, fire protection business management, fire protection personnel management, station-side key work equipment management, converter station risk management, and fire hazard management. The fire emergency response module is used to identify converter stations that have experienced a fire or have a potential fire based on at least one of the fire IoT data of the converter station and the management results of the fire daily management module, and to carry out fire emergency response for the converter stations that have experienced a fire or have a potential fire.

2. The converter station fire protection digital control system according to claim 1, characterized in that, The abnormal data detection model was trained in the following manner: Obtain normal timing data generated during the operation of the converter station; Based on the sliding window and step size corresponding to the data type of the normal time series data, the normal time series data is converted into normal time series samples; The abnormal time series samples are generated based on the normal time series samples; The abnormal data detection model is obtained by training a hybrid expert model, which includes at least two sub-models, using the normal time series samples and the abnormal time series samples.

3. The converter station fire protection digital control system according to claim 2, characterized in that, Generating abnormal time series samples based on the normal time series samples includes: The trend anomaly samples are generated based on the normal time series samples in the following manner: Where y1[i] represents the trend anomaly sample; x[i] represents the normal time series sample; trend is the added linear trend term, trend=a·(i-t1+1), a is the trend coefficient, a=(R / t1)·r1·(n / 10), R is the initial segment range, r1 is the trend direction adjustment factor, n is the anomaly intensity coefficient, t1 is the starting point of the anomaly interval, t1~Uniform{2,…l-1}, l is the sequence length of the original normal time series sample, and i is the sample in the anomaly interval; σ x This represents the initial standard deviation of the normal time series samples; For random perturbation, 4. The converter station fire protection digital control system according to claim 2, characterized in that, Generating abnormal time series samples based on the normal time series samples includes: The fixed-bias anomaly sample is generated based on the normal time series sample in the following manner: y2[i]=x[i]+x1 Where y2[i] is a fixed-bias abnormal sample; x[i] is a normal time series sample; and x1 is the injected fixed bias.

5. The converter station fire protection digital control system according to claim 2, characterized in that, Generating abnormal time series samples based on the normal time series samples includes: The precision degradation anomaly samples are generated based on the normal time-series samples in the following manner: y3[i]=x[i]+ε Where y3[i] represents the anomalous sample with fixed bias; x[i] represents the normal time series sample; ε represents Gaussian white noise, ε ~ Normal(0,σ) 2 ), where σ is the noise amplitude, σ=β·r, β is the reference amplitude, and r is the random amplitude adjustment factor.

6. The converter station fire protection digital control system according to claim 2, characterized in that, Generating abnormal time series samples based on the normal time series samples includes: The constant value anomaly sample is generated based on the normal time series sample in the following manner: y4[t1:l]=x[t1] Where u4[t1:l] represents the constant value abnormal sample; t1 is the random starting point, t1~Uniform{2,…l-1}, and l is the sequence length of the original normal time series sample; x[t1] indicates that the normal time series sample is set to a constant value starting from t1.

7. The converter station fire protection digital control system according to claim 2, characterized in that, Generating abnormal time series samples based on the normal time series samples includes: The discrete point anomaly samples are generated based on the normal time series samples in the following manner: y5[i]=x[i]+Δ Where y5[i] is an outlier sample; x[i] is a normal time series sample; Δ is the outlier perturbation, Δ=(1+N)·μ·(n / 10 4 ), N is the standard normal perturbation, N ~ Normal(0,1), μ is the global mean benchmark of normal time series samples, and n is the anomaly intensity coefficient.

8. The converter station fire protection digital control system according to claim 2, characterized in that, The hybrid expert model includes at least two sub-models from CNN sub-model, GRU sub-model, LSTM sub-model and Transformer sub-model; The abnormal data detection model also includes a gating network, which is used to assign corresponding weights to each sub-model according to the data type of the input data; Each sub-model is used to obtain the corresponding sub-model anomaly detection result based on the input data and the assigned weights; wherein, the anomaly detection result of the anomaly data detection model is obtained based on the anomaly detection results of each sub-model.

9. The converter station fire protection digital control system according to claim 8, characterized in that, When training a hybrid expert model, the loss of each sub-model includes binary cross-entropy loss for binary classification tasks and multivariate cross-entropy loss for multi-anomaly classification tasks. The total loss of the hybrid expert model is: Wherein, Weighted Output Loss represents the binary cross-entropy loss or multivariate cross-entropy loss of the weighted output of the hybrid expert model, and Expert Loss... k Let λ represent the binary cross-entropy loss or multivariate cross-entropy loss of the k-th sub-model, and let λ represent the weight coefficients of the sub-model loss.

10. The converter station fire protection digital control system according to claim 1, characterized in that, The converter station data processing module is also used for: Determine the integrity score, validity score, and timeliness score of the real-time time series data corresponding to the fire protection IoT data of the converter station; The comprehensive quality score of the real-time time series data is determined based on the completeness score, validity score, and timeliness score of the real-time time series data. The comprehensive quality score is used to determine whether the fire protection IoT data meets the data access quality compliance requirements.

11. The converter station fire protection digital control system according to claim 10, characterized in that, The fire protection daily management module includes a fire equipment control module, which is used for: If the converter station data processing module determines that the fire protection IoT data meets the data access quality compliance conditions, it determines the quality level of the fire protection IoT data based on the comprehensive quality score, and obtains the first data quality distribution characteristics based on the quality level of the fire protection IoT data. The first data quality distribution characteristics are mapped to a pre-determined health baseline to obtain the health status assessment results of the fire protection IoT devices corresponding to the fire protection IoT data; The health baseline is obtained by clustering the second data quality distribution characteristics of the historical normal time series data of the fire protection IoT device in different time windows. The second data quality distribution characteristics are determined by extracting data of different quality levels within the time window through a sliding window based on the comprehensive quality score of the historical normal time series data.

12. The converter station fire protection digital control system according to claim 10, characterized in that, The converter station data processing module is used to determine the integrity score, validity score, and timeliness score of the real-time time-series data corresponding to the fire protection IoT data of the converter station, including: The importance weight of the real-time time series data is calculated using the analytic hierarchy process (AHP), and the integrity score of the real-time time series data is determined by combining the standardization check of the time series data content. The effectiveness score of the real-time time series data is obtained by using either a method based on the accuracy of long-term and short-term historical time series data or an evaluation method based on time series reconstruction. Based on the transmission interval of the real-time time series data, the timeliness score of the real-time time series data is obtained.

13. The converter station fire protection digital control system according to claim 12, characterized in that, The converter station data processing module is used to calculate the importance weights of the real-time time series data using the analytic hierarchy process (AHP), and, in conjunction with a standardization check of the time series data content, determine the integrity score of the real-time time series data, including: The importance weights of primary and secondary indicators in the real-time time-series data are calculated using the Analytic Hierarchy Process (AHP). The primary indicators include at least one of equipment information, monitoring data, equipment status, and communication information. The secondary indicators for equipment information include at least one of equipment identifier, equipment name, data code, equipment manufacturer, equipment model, and equipment installation location. The secondary indicators for monitoring data include at least one of data acquisition timestamp, sensor type, monitored value, and unit of monitored value. The secondary indicators for equipment status include at least one of equipment operating status, battery level, and signal strength. The secondary indicators for communication information include at least one of communication protocol type and network type. The integrity score of the real-time time series data is determined based on the importance weights of the secondary indicators combined with the standardization check of the time series data content, using the following method: y=∑ω i x i Where y is the integrity score, ω i Let x be the weight of the i-th secondary indicator. i Let x be the score of the i-th secondary indicator. i The value of x is 1 or 0. If the content corresponding to the secondary indicator is missing, then x i If the value is 0, then x is zero if the content corresponding to the indicator is not missing. i The value is 1.

14. The converter station fire protection digital control system as described in claim 12, characterized in that, The converter station data processing module is used to obtain a validity score for the real-time time series data when the real-time time series data is remote signaling data, by using a method based on the accuracy of long-term and short-term historical time series data, including: δ=ω1·x1+ω2·x2+ω3·x3+ω4·x4 Where δ is the data validity score; x1 is the historical accuracy of the device; x2 is the accuracy of the M most recent remote signaling signals sent by the device; x3 is the accuracy of the most recent M-5 remote signaling data; x4 is the accuracy of the last remote signaling signal sent; and ω1, ω2, ω3, and ω4 are the weights corresponding to x1, x2, x3, and x4.

15. The converter station fire protection digital control system as described in claim 12, characterized in that, When the real-time time series data is telemetry data, the converter station data processing module is used to obtain a validity score for the real-time time series data using an evaluation method based on time series reconstruction, including: The telemetry data was reconstructed using the Transformer reconstruction model to obtain the reconstructed sequence; The reconstruction error of the reconstructed sequence was statistically analyzed using the POT threshold selection method to obtain the reconstruction error threshold. The initial threshold in the POT threshold selection method was determined using the three-standard-deviation method. The validity score of data obtained based on the critical threshold in the reconstruction error threshold is: Where δ(i) is the effectiveness score for each time step i, e i It is the relative error of reconstruction at the time step. th F δ is the critical threshold used to determine data anomalies. l The validity score of telemetry data is set when the relative error of reconstruction at each time step is the anomaly judgment threshold.

16. The converter station fire protection digital control system as described in claim 12, characterized in that, The converter station data processing module is used to obtain a timeliness score of the real-time time series data based on the transmission interval of the real-time time series data, including: When the real-time time series data is remote signaling data, the timeliness score of the remote signaling data is obtained using a first evaluation function. The formula for the first evaluation function is as follows: y; -0.346x Where y is the timeliness score of remote signaling data transmission, and x is the data transmission delay time; When the real-time time series data is telemetry data, the timeliness score of the telemetry data is obtained using a second evaluation function. The formula for the second evaluation function is as follows: y n =1-0.11I n 2 in, μ B Let σ be the mean of set B. B Let be the standard deviation of set B, and set B is [ΔT]. n-1 ,ΔT n-2 ,ΔT n-3 ,…,ΔT n-N ], ΔT n =T n -T n-1 T n Let ΔT be the time when the nth data point is transmitted to the monitoring center backend. n y represents the transmission time interval between the nth data and the previous data; n To score the timeliness of telemetry data acquisition.

17. The converter station fire protection digital control system as described in claim 11, characterized in that, The fire equipment management module is used to perform a nonlinear mapping between the first data quality distribution feature vector and a pre-determined health baseline to obtain the health status assessment results of the fire IoT equipment corresponding to the fire IoT data, including: The overall feature weight is determined based on the importance weight and information entropy weight of the second data quality distribution feature; Based on the aforementioned feature comprehensive weights, the distance between the first data quality distribution feature vector and the healthy baseline is evaluated using the W-ReLU nonlinear mapping model as the anomaly degree. The health level of the fire protection IoT equipment is determined based on the anomaly level.

18. The converter station fire protection digital control system as described in claim 17, characterized in that, The formula for calculating the anomaly degree is: In the formula, δ n For anomaly degree, w′ i x is the feature comprehensive weight of feature i. 0k Let d be the eigenvalue of the k-th feature corresponding to the cluster center of the cluster to which the new sample belongs. kmin With d kmax The distance from the edge of the confidence range to the eigenvalue of the k-th feature at the cluster center is given by N, where N is the device scale parameter, and N = x. 01 +x 02 +x 03 +x 04 , [x 01 ,x 02 ,x 03 ,x 04 [x′1, x′2, x′3, x′4] is the cluster center, and [x′1, x′2, x′3, x′4] is the data quality distribution feature vector of the real-time time series data; The formula for calculating the feature integration weight is as follows: In the formula, AW i For importance weights, RV i θ represents the information entropy weight, θ represents the assigned weight, and m represents the total number of features.

19. The converter station fire protection digital control system according to claim 1, characterized in that, The fire protection daily management module includes a fire equipment control module, which is used for: To obtain experimental data on temperature drop of fire-fighting pipelines containing electric heating cables under low-temperature conditions; A numerical model of fire-fighting pipelines containing electric heat tracing cables was established, the conversion formula for the convective heat transfer coefficient was determined, and the numerical model was verified based on the temperature drop experimental data and the conversion formula for the convective heat transfer coefficient. By changing the ambient temperature and the power of the electric heating tape, the antifreeze time of the fire-fighting pipeline containing the electric heating tape under different working conditions was obtained; Based on the time of the different operating conditions, establish the power function of the electric heating cable with respect to the antifreeze time and ambient temperature; Based on the power function of the electric heating cable and the given antifreeze time and ambient temperature, the optimal power of the electric heating cable for the fire-fighting pipeline containing the electric heating cable is determined under the required antifreeze time.

20. The converter station fire protection digital control system according to claim 19, characterized in that, The formula for converting the convective heat transfer coefficient is as follows: Where h is the convective heat transfer coefficient, β is determined by the temperature drop experimental data of the fire-fighting pipeline, R is the outer diameter of the fire-fighting pipeline, Pr and Gr are the Prandtl number and Grashof number of the ambient air, respectively, and A and m are constants.

21. The converter station fire protection digital control system according to claim 19, characterized in that, The fire-fighting equipment control module is used to verify the numerical model based on the temperature drop experimental data and the convective heat transfer coefficient conversion formula, including: Input the temperature, fire pipeline parameters, and electric heating cable parameters that are consistent with the temperature drop experimental data into the numerical model. Input the convective heat transfer coefficient for this working condition according to the convective heat transfer coefficient conversion formula. Set a temperature sensor at the symmetrical position of the fire pipeline heating cable about the center to obtain the pipe wall temperature. Change the convective heat transfer coefficient β until the pipe wall temperature of the numerical model is consistent with the temperature drop experimental data.

22. The converter station fire protection digital control system according to claim 19, characterized in that, The fire equipment control module is used to obtain the antifreeze time of the fire pipeline containing electric heating cable under different operating conditions, including: The liquid fraction obtained by the liquid fraction sensor installed on the outermost layer of the fire-fighting pipeline under different ambient temperatures and different electric heating tape powers is used to determine the antifreeze time under the corresponding working conditions from the start until the liquid fraction on the outermost layer of the fire-fighting pipeline becomes 0.

02.

23. The converter station fire protection digital control system according to claim 19, characterized in that, The power function of the electric tracing cable is: t=f(P,T)=a0+a1P+a2T+a3P 2 +a4PT+a5T 2 , Where P is the power of the electric heating cable, T is the ambient temperature, t is the antifreeze time, and a0, a1, a2, a3, a4 and a5 are regression coefficients, which are obtained by fitting the power function using the Levenberg-Marquardt optimization algorithm.

24. The converter station fire protection digital control system according to claim 1, characterized in that, The fire protection daily management module includes a fire equipment control module, or the fire emergency response module includes a fire equipment emergency monitoring module. The fire equipment control module or the fire equipment emergency monitoring module is used for: The first data set is obtained by collecting fluid data from the collection points of the fire-fighting foam pipeline; Based on the first data set and the preset fluid identification model, the fluid type in the fire-fighting foam pipe is determined; the fluid type is either a Newtonian fluid or a non-Newtonian fluid. Based on the leak identification model corresponding to the fluid type and the first data set, the leak identification result of the fire-fighting foam pipeline is determined.

25. The converter station fire protection digital control system according to claim 24, characterized in that, The fire equipment management module or the fire equipment emergency monitoring module is used to determine the leak identification result of the fire foam pipeline based on the leak identification model corresponding to the fluid type and the first data set, including: Based on the first data set, a second data set corresponding to the Newtonian fluid is determined, and based on the second data set and the first leak identification model corresponding to the Newtonian fluid, the leak identification result of the fire-fighting foam pipeline is determined, wherein the second data set includes parameters used to characterize the pressure gradient data fluctuation in the pipeline; or, Based on the first data set, a third data set corresponding to the non-Newtonian fluid is determined, and based on the third data set and the second leak identification model corresponding to the non-Newtonian fluid, the leak identification result of the fire-fighting foam pipeline is determined, wherein the third data set includes parameters used to characterize the degree of flow separation.

26. The converter station fire protection digital control system according to claim 25, characterized in that, The first data set includes: pressure gradient data, flow velocity profile data, shear rate-viscosity curves, temperature sensor data, and fluid density measurements; Based on the first data set, a second data set corresponding to the Newtonian fluid is determined, including: determining the pressure gradient variation coefficient based on the pressure gradient data, determining the mass flow rate deviation based on the flow velocity profile data, determining the viscous dissipation power density based on the shear rate-viscosity curve and temperature sensor data, and determining the dynamic characteristics of the bulk modulus based on the fluid density measurement value. Alternatively, based on the first data set, a third data set corresponding to the non-Newtonian fluid is determined, including: determining the shear rate-viscosity dynamic characteristics based on the shear rate-viscosity curve, determining the secondary flow intensity based on the flow velocity profile data, determining the viscous dissipation power density and the corresponding associated temperature based on the shear rate-viscosity curve and temperature sensor data, and determining the dynamic characteristics of the bulk modulus based on the fluid density measurement value.

27. The converter station fire protection digital control system according to claim 25, characterized in that, The fire equipment control module or the fire equipment emergency monitoring module is also used for: If a leak is determined in the fire-fighting foam pipeline based on the second data set and the first leak identification model corresponding to Newtonian fluid, the pipeline segment with a leak is determined based on the collection point where the first deviation value is greater than a preset first threshold and the second deviation value is greater than a preset second threshold; wherein, the first deviation value is the difference between the pressure gradient variation coefficient corresponding to the collection point and the preset normal pressure gradient variation coefficient, and the second deviation value is the difference between the mass flow rate deviation corresponding to the collection point and the preset normal mass flow rate deviation. or, If a leak is determined in the fire-fighting foam pipeline based on the second leak identification model corresponding to the third data set and non-Newtonian fluid, a leak is determined in the target pipe section based on the pressure gradient decrease value of the target pipe section being greater than the third threshold and the viscous dissipation power density increase value of the target pipe section being greater than the fourth threshold; wherein, the target pipe section is the pipe section between two adjacent collection points.

28. The converter station fire protection digital control system according to claim 1, characterized in that, The fire protection daily management module includes a station-side key equipment management module, which is used for: Under different fault or failure conditions inside the converter transformer tank, the temperature time series data inside the converter transformer tank in the real environment is obtained and recorded as the real dataset; and a proportional numerical simulation is carried out on the real environment to simulate the temperature evolution law inside the converter transformer tank under different fault or failure conditions, and the temperature time series data inside the converter transformer tank in the simulation environment is obtained and recorded as the simulation dataset. The simulation dataset and the real dataset are compared to verify the reliability of the simulation dataset. The simulation dataset after reliability verification is transformed into a temperature matrix with corresponding coordinates, and the time matrix is ​​determined based on the time in the simulation dataset after reliability verification as a variable; and the physical matrix with corresponding coordinates is determined based on the physical characteristics of the converter transformer tank. The temperature matrix, the time matrix, and the physical matrix are input into a generative adversarial network to obtain the temperature field matrix inside the converter transformer tank. Based on the temperature field matrix inside the converter transformer tank, the overpressure inside the converter transformer tank is determined. Then, based on the temperature field matrix inside the converter transformer tank and the influence of local high temperature on the strength of the tank material, the failure risk of the converter transformer tank is predicted.

29. The converter station fire protection digital control system according to claim 28, characterized in that, The station-side critical equipment management module is used to verify the reliability of the simulation dataset, including: calculating the difference between the simulation dataset and the real dataset based on the average coefficient of determination, and comparing the difference comparison calculation result with a preset difference threshold; when the difference comparison calculation result is greater than or equal to the difference threshold, it indicates that the simulation dataset is reliable; if the difference comparison calculation result is less than the difference threshold, it indicates that the simulation dataset is unreliable, and the parameters of the simulation environment are readjusted, and the reliability of the simulation dataset after the adjustment of the simulation environment parameters is verified again with the real dataset until the reliability meets the standard.

30. The converter station fire protection digital control system according to claim 28, characterized in that, The key operating equipment control module at the substation side is used to determine the physical matrix of corresponding coordinates based on the physical characteristics of the converter transformer tank, including: The physical features are discretized based on the coordinates corresponding to the temperature matrix, and then divided into corresponding grids. The internal structure, material strength, and thermal conductivity of the oil tank within the corresponding grid are taken as representative parameters of the local oil tank within that grid, and then matrixed to form the oil tank internal structure matrix, oil tank material strength matrix, and oil tank thermal conductivity matrix. The internal structure matrix, material strength matrix, and thermal conductivity matrix of the fuel tank are used as physical matrices.

31. The converter station fire protection digital control system according to claim 28, characterized in that, The station-side key operating equipment control module is used to obtain the temperature field matrix inside the transformer tank, including: Generative adversarial networks consist of a generator and a discriminator; The temperature matrix, the time matrix, and the physical matrix are merged to generate an input matrix for training the generative adversarial network. The input matrix is ​​input as real data into the generator to obtain the corresponding fake data; The discriminator distinguishes between real data and corresponding fake data, outputs the probability of being judged as real or fake, and performs backpropagation after calculating the loss to update the generator weights. The generative adversarial network is then iteratively trained until the generator can generate a high-quality temperature field matrix that is close to the real data. Using a trained generative adversarial network and real data as input, the temperature field matrix inside the converter transformer tank is obtained.

32. The converter station fire protection digital control system according to claim 28, characterized in that, The station-side key operating equipment control module is used to determine the overpressure inside the converter transformer tank based on the temperature field matrix inside the converter transformer tank, including: Based on the temperature field matrix inside the converter transformer tank, the vaporization amount is calculated based on the instantaneous temperature field and the thermal properties of the transformer oil, and the total vaporization amount of the transformer oil after the occurrence of potential risks is obtained by integration; the total vaporization amount of the converter transformer oil is introduced into the ideal gas state equation to calculate the overpressure amount inside the converter transformer tank.

33. The converter station fire protection digital control system according to claim 1, characterized in that, The fire protection daily management module includes a converter station risk control module, which is used for: Determine the importance weight of each fire protection subsystem in the converter station, wherein the fire protection subsystem includes at least one of the fire protection facility system, fire alarm system, fire extinguishing system, and fire evacuation system; Based on the failure event tree model of each fire-fighting equipment in each fire-fighting subsystem, the failure probability of each fire-fighting equipment is determined. Based on the equipment failure probability of each fire-fighting equipment in each fire-fighting subsystem, determine the system failure probability of each fire-fighting subsystem, and determine the subsystem health status of each fire-fighting subsystem based on the system failure probability of each fire-fighting subsystem. The comprehensive fire status assessment result of the converter station is determined based on the importance weight and health status of each fire protection subsystem.

34. The converter station fire protection digital control system according to claim 1, characterized in that, The fire protection daily management module includes a converter station risk control module, which is used for: Identify fire control business events for each risk control object of the converter station, wherein the risk control object includes at least one of fire-fighting operations, fire-fighting equipment, and fire-fighting vehicles; The risk level of each fire control business event is determined based on at least two of the severity, occurrence, and detectability of each fire control business event, wherein the risk level includes high risk, medium risk, and low risk. Based on the risk level of each fire control business event, a risk level identifier matching the risk level is set for the corresponding fire control business event, and the display of each fire control business event and its corresponding risk level identifier is controlled.

35. The converter station fire protection digital control system according to claim 34, characterized in that, The risk level of each fire control operation event is determined based on at least two of the following: severity, occurrence, and detectability: The risk level of each fire control business event is determined by comparing the risk priority coefficient corresponding to the product of severity, occurrence, and detectability of each fire control business event with the preset coefficient threshold. or, A severity-occurrence matrix, a severity-detectability matrix, or an occurrence-detectability matrix is ​​formed based on any two of the severity, occurrence, and detectability of each fire control business event. A risk matrix ranking is obtained based on the severity-occurrence matrix, severity-detectability matrix, or occurrence-detectability matrix. The risk level of each fire control business event is then determined based on the risk matrix ranking.

36. The converter station fire protection digital control system according to claim 1, characterized in that, The fire protection daily management module includes a converter station risk control module, which is used for: The management indicator score of each primary management indicator is determined based on the score and weight of the subordinate management indicator corresponding to each primary management indicator in the converter station. The primary management indicators include at least one of the following: fire equipment management indicator, fire operation management indicator, fire vehicle management indicator, and fire personnel management indicator. The fire protection management status of the converter station is determined based on the weight of each primary management indicator and the score of the management indicator.

37. The converter station fire protection digital control system according to claim 1, characterized in that, The fire emergency response module includes a fire emergency decision-making module, which is used for: The system acquires primary video data sent by a primary camera group associated with a fire in the converter station. The primary video data includes video data of the fire from multiple perspectives captured by primary cameras at multiple locations. When the first video data of the fire object from the first perspective, determined based on the first-level video data, does not meet the emergency auxiliary decision-making conditions, a first camera that captures the first video data is determined from the first-level camera group, and a second camera corresponding to the position of the first camera is determined from the second-level camera group that has a positional mapping relationship with the first-level camera group; wherein, the second-level camera group and the first-level camera group have a positional mapping relationship with the fire object's position as the same perspective alignment center. Obtain second video data of the fire object from the first perspective corresponding to the second camera; Fire emergency response is carried out based on the first-level video data and the second-level video data.

38. The converter station fire protection digital control system according to claim 37, characterized in that, The fire emergency decision-making module is used to determine, based on the primary video data, that the first-view video data of the fire object does not meet the emergency auxiliary decision-making conditions, including at least one of the following: It is determined that the resolution of the first video data is lower than the resolution threshold; It is determined that the resolution of the first video data is lower than a resolution threshold; It is determined that the key target object in the first video data is occluded for a duration exceeding a predetermined duration; It is determined that the contribution of the first video data to joint emergency decision-making with other video data in the first-level video data is less than the contribution threshold.

39. The converter station fire protection digital control system according to claim 37, characterized in that, The fire emergency decision-making module is used to determine the second camera corresponding to the position of the first camera based on the second-level camera group which has a position mapping relationship with the first-level camera group, including: Based on the location of the fire object, the secondary camera group is determined from the candidate camera group that has a positional mapping relationship with the primary camera group. A candidate camera group is determined with a fire monitoring position within the shared field of view of the primary camera group as the center of view. The shared field of view of the primary camera group includes at least one fire monitoring position, and the at least one fire monitoring position includes the location of the fire object. The second camera corresponding to the position of the first camera is determined based on the secondary camera group.

40. The converter station fire protection digital control system according to claim 37, characterized in that, The fire emergency decision-making module is used to determine the second camera corresponding to the position of the first camera based on the second-level camera group which has a position mapping relationship with the first-level camera group, including: Using the location of the fire object as the center of view, determine a reference camera that has a viewpoint candidate relationship with the cameras in the first-level camera group; Acquire basic spatial data of the primary camera group and the reference camera, wherein the basic spatial data includes at least one of the following: camera position information, height information, angle information, field of view information, and environment information of the camera. A graph neural network model is constructed based on the basic spatial data of the first-level camera group and the reference camera. Each camera in the first-level camera group and the reference camera serves as a node in the graph neural network model. The node features of each node include the basic spatial data of the corresponding camera. The edges between nodes indicate that there is a viewpoint candidate relationship between the cameras corresponding to that node. The edge weights are determined based on at least one of the following: viewpoint candidate range, relative distance, line-of-sight accessibility, and historical collaborative effectiveness score between the corresponding cameras. Using the graph neural network model, a second-level camera group is formed by determining, based on the edge weights, cameras from the reference cameras that satisfy the position mapping relationship with each camera in the first-level camera group; The second camera corresponding to the position of the first camera is determined based on the secondary camera group.

41. The converter station fire protection digital control system according to claim 40, characterized in that, The fire emergency decision-making module is used to determine the second camera corresponding to the position of the first camera based on the secondary camera group, including: The location of the first camera, the environmental information of the first camera, and the emergency auxiliary decision-making requirements are input into the graph neural network model to obtain at least one candidate camera that has a viewpoint candidate relationship with the first camera. Determine the view complementarity score of each candidate camera relative to the first camera, wherein the view complementarity score is determined based on at least one of the following: the view candidate range of the candidate camera relative to the first camera, the missing information gain of the candidate camera under the environmental information, the predicted image quality of the candidate camera, and the cooperative feasibility of the candidate camera and the first camera. The second camera corresponding to the position of the first camera is determined based on the complementarity score of the viewpoints of each candidate camera relative to the first camera.

42. The converter station fire protection digital control system according to claim 41, characterized in that, The fire emergency decision-making module is used to determine the second camera corresponding to the position of the first camera based on the view complementarity score of each candidate camera relative to the first camera, including: The candidate camera with the highest viewpoint complementarity score was selected as the second camera; Alternatively, candidate cameras whose viewpoint complementarity scores are greater than the first score threshold can be identified as the second camera; Alternatively, if the view complementarity scores of all candidate cameras are less than the second score threshold, the top N candidate cameras with the highest view complementarity scores are selected as the second camera. The second camera is used to synthesize a simulated viewpoint, which is used to synthesize the second video data, and N is a natural number.

43. The converter station fire protection digital control system according to claim 42, characterized in that, The fire emergency decision-making module is used to acquire second video data of the fire object from the first perspective captured by the second camera, including: Acquire at least one alternative video data of the fire target captured by the at least one second camera; A third video data representing a first simulated perspective is synthesized based on the at least one candidate video data, and this third video data is used as the second video data, wherein the first simulated perspective is within the shared viewpoint range of the candidate video data captured by the at least one second camera; or... A fourth video data of a second simulated perspective is synthesized based on the first video data and the at least one alternative video data, and the fourth video data is used as the second video data, wherein the second simulated perspective is within the shared perspective range of the alternative video data captured by the at least one second camera and the overlapping perspective range of the first perspective.

44. The converter station fire protection digital control system according to claim 37, characterized in that, The fire emergency decision-making module is also used for: Predict the development trend of the fire at the target location; The key perspective is determined based on the fire development trend, wherein the key perspective is the perspective from which key video data needs to be captured relative to the fire object, or the key perspective is the perspective from the perspective of a primary camera that has failed relative to the fire object. A backup camera corresponding to the key viewpoint is determined from the secondary camera group, and a preparation instruction is sent to the backup camera. The preparation instruction is used to instruct the backup camera to start or prepare to acquire video data of the fire object from the key viewpoint.

45. The converter station fire protection digital control system according to claim 37, characterized in that, The fire emergency decision-making module is used to acquire second video data of the fire object from the first perspective captured by the second camera, including: Control the identification correspondence between the device identifier of the second camera and the device identifier of the first camera carried in the second video data, and verify the validity of the second video data according to the identification correspondence; And / or, The control includes carrying a priority transmission identifier in the second video data, which indicates that the second video data should be transmitted to the converter station digital fire control system with high priority.

46. ​​The converter station fire protection digital control system according to claim 1, characterized in that, The converter station fire protection digital control system also includes a converter station pipeline connection module, which is used for: Based on the converter station's request for inclusion in the management system initiated by the converter station or based on the converter station's instruction to be included in the management system sent to the converter station, the target converter station that needs to be included in the management system and the site attribute information of the target converter station are determined, and the site identifier of the target converter station and the site attribute information are associated and stored in the list of managed converter stations. Identify the equipment requiring pipe connection in the target converter station and the components requiring pipe connection within the equipment, and determine the data reporting mechanism for the equipment requiring pipe connection and / or the components requiring pipe connection; A successful converter station connection notification is sent to the target converter station, and the device identifier of the device to be connected and the component identifier of the component to be connected are associated and stored with the station identifier; wherein, the successful converter station connection notification includes the identifiers of the device to be connected and the component to be connected, as well as the corresponding data reporting mechanism.

47. The converter station fire protection digital control system according to claim 46, characterized in that, Identifying the equipment requiring pipe connection in the target converter station and the components within that equipment requiring pipe connection, including: A configuration template for grid connection matching the site attribute information of the target converter station is determined, wherein the site attribute information includes at least one of site size, site location, site distance, site priority, site attention level, and site voltage level; the configuration template for grid connection includes the grid-connected equipment and grid-connected components in the configured converter station, as well as the data reporting mechanisms corresponding to the grid-connected equipment and / or the grid-connected components respectively. The system determines the equipment and components requiring pipe connection in the target converter station according to the configuration in the pipe connection configuration template, and determines the data reporting mechanism corresponding to the equipment and / or components requiring pipe connection respectively; or, the system displays the pipe connection configuration template, responds to the template modification operation to modify the pipe connection configuration template, and determines the equipment and components requiring pipe connection in the target converter station according to the configuration in the modified pipe connection configuration template, and determines the data reporting mechanism corresponding to the equipment and / or components requiring pipe connection respectively.

48. The converter station fire protection digital control system according to claim 46, characterized in that, The converter station fire protection digital control system also includes a converter station panoramic monitoring module, used for: Identify the converter stations that have already been connected to the power grid, as well as the equipment and components within those stations that require connection to the power grid; Based on the map data and site layout data of the existing converter stations, the equipment attribute data of the equipment to be connected to the pipeline, and the component attribute data of the components to be connected to the pipeline, digital twin models corresponding to the existing converter stations, the equipment to be connected to the pipeline, and the components to be connected to the pipeline are constructed respectively. The panoramic monitoring data of the existing converter station is displayed based on the corresponding digital twin models of the existing converter station, the equipment to be connected to the pipeline, and the components to be connected to the pipeline.

49. A digital fire protection control system for converter stations, characterized in that, This includes a digital fire control platform for converter stations and at least one converter station, wherein: The at least one converter station is used to acquire the fire protection IoT data of the converter station and send the acquired fire protection IoT data to the fire protection control platform of the converter station; The converter station fire protection digital management and control platform is used for: acquiring fire protection IoT data sent by the converter station; using an anomaly detection model to detect anomalies in the acquired fire protection IoT data; wherein, the anomaly detection model is trained based on normal time-series samples generated during the operation of the converter station and anomaly time-series samples generated based on the normal time-series samples, the anomaly time-series samples including at least one of trend anomaly samples, fixed deviation anomaly samples, accuracy decline anomaly samples, constant value anomaly samples, and discrete point anomaly samples; performing daily fire protection management of the converter station based on the fire protection IoT data of the converter station, the daily fire protection management including at least one of fire equipment management, fire business management, fire personnel management, station-side key work equipment management, converter station risk management, and fire hazard management; and identifying converter stations that have experienced emergencies or have potential emergencies based on at least one of the fire protection IoT data of the converter station and the management results of the daily fire protection management module, and performing fire emergency response on the converter stations that have experienced emergencies or have potential emergencies.

50. A digital management and control method for fire protection in converter stations, characterized in that, The method includes: The fire protection IoT data of the converter station is acquired, and an anomaly detection model is used to detect anomalies in the acquired fire protection IoT data. The anomaly detection model is trained based on normal time-series samples generated during the operation of the converter station and anomaly time-series samples generated based on the normal time-series samples. The anomaly time-series samples include at least one of the following: trend anomaly samples, fixed deviation anomaly samples, accuracy reduction anomaly samples, constant value anomaly samples, and discrete point anomaly samples. The fire protection management of the converter station is carried out based on the fire protection IoT data of the converter station. The daily fire protection management includes at least one of the following: fire protection equipment management, fire protection business management, fire protection personnel management, station-side key work equipment management, converter station risk management, and fire hazard management. Based on at least one of the converter station's fire protection IoT data and daily fire protection management results, identify converter stations that have experienced or are showing signs of potential danger, and carry out fire emergency response for these stations.

51. A computer storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed, implements the steps of the converter station fire protection digital control method as described in claim 50.

52. A computer program product, characterized in that, The computer program product includes a computer program that, when executed, implements the steps of the converter station fire protection digital control method as described in claim 50.

Images