Switch cabinet monitoring method and equipment based on Internet of Things technology, and medium

Abstract

The invention discloses a switch cabinet monitoring method and device based on the Internet of Things technology and a medium, and relates to the technical field of Internet of Things fusion, and the method comprises the steps: inputting a multi-modal time series data set into a multi-physics field coupling digital twin model, and generating a predictive instruction set; dynamically adjusting a discrimination probability threshold and aggressive degree parameters in the lightweight convolutional neural network based on the risk level quantized value in the predictive instruction set to obtain an adjusted lightweight convolutional neural network; performing pattern recognition on the partial discharge pulse waveform data by using the adjusted lightweight convolutional neural network to generate a fault early warning event identifier; by constructing a multi-modal time sequence data set, efficient fusion and time synchronization of multi-physical field information such as partial discharge, temperature and vibration of the switch cabinet are realized, data integrity and accuracy of state sensing are improved, and safety and stability of long-term operation of the switch cabinet are effectively guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet of Things fusion, and in particular to a switch cabinet monitoring method, device and medium based on Internet of Things technology. BACKGROUND

[0002] With the continuous evolution of the power system towards intelligence and digitization, the switch cabinet as a key device in the power distribution network, the operation state is directly related to the safety and reliability of the entire power grid, the rapid development of Internet of Things (Internet of Things, IoT) technology provides a new technical path for the state monitoring of power equipment, there has appeared a switch cabinet online monitoring system based on the combination of sensor network and wireless communication technology, which can realize the real-time collection and remote transmission of key parameters such as temperature, humidity, and partial discharge intensity, usually using independent sensing modules to collect single or limited dimensional data, and uploading data to the monitoring platform through ZigBee, LoRa or 4G communication mode, combining threshold comparison or simple statistical analysis method for abnormal judgment, to a certain extent, it improves the operation and maintenance efficiency, realizes the transition from "periodic maintenance" to "condition-based maintenance", and lays a foundation for the intelligent management of power equipment, the introduction of the concept of Digital Twin also initially realizes the data mapping and state synchronization between physical equipment and virtual model, and enhances the perception ability of the equipment operation situation; Under the existing technical framework, there is still optimization space for the deep fusion of multi-source heterogeneous monitoring data and dynamic response mechanism, and the current most monitoring systems process multi-physical field information such as partial discharge, temperature and vibration mostly using parallel independent analysis mode, lacking coupling modeling in time dimension and physical mechanism level, resulting in limited representation ability for complex fault evolution process, the fault discrimination logic generally depends on fixed threshold or static model, which is difficult to adapt to the dynamic operation characteristics of the switch cabinet under different loads, environments and aging states, thereby affecting the accuracy and timeliness of the early warning. SUMMARY

[0003] In view of the above existing problems, the present application is proposed.

[0004] Therefore, the present application provides a switch cabinet monitoring method based on Internet of Things technology to solve the problem of low early warning precision caused by insufficient fusion of multi-source monitoring data and static fault discrimination mechanism.

[0005] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides a switch cabinet monitoring method based on Internet of Things technology, which comprises collecting partial discharge pulse waveform data, temperature data and vibration data on the switch cabinet, and generating a multi-modal time series data set after processing by an Internet of Things gateway; inputting the multi-modal time series data set into a multi-physical field coupling digital twin model to generate a predictive instruction set; based on the risk level quantization value in the predictive instruction set, dynamically adjusting the discriminant probability threshold and the aggressiveness parameter in the lightweight convolutional neural network to obtain an adjusted lightweight convolutional neural network; using the adjusted lightweight convolutional neural network to perform pattern recognition on the partial discharge pulse waveform data to generate a fault warning event identifier; converting the fault warning event identifier into a voltage regulation instruction and sending it to a voltage regulator, and the voltage regulator reduces the line output voltage based on the voltage regulation instruction to suppress partial discharge; integrating the multi-modal time series data set, the predictive instruction set, the fault warning event identifier and the voltage regulation record to generate a switch cabinet health status report.

[0006] As a preferred scheme of the switch cabinet monitoring method based on Internet of Things technology, wherein: the collection of partial discharge pulse waveform data, temperature data and vibration data on the switch cabinet, and the generation of multi-modal time series data set after processing by the Internet of Things gateway, the specific steps are as follows, Capture the electromagnetic wave signal generated by partial discharge and convert it into partial discharge pulse waveform data, collect the temperature data of the bus connection point and the insulating surface, monitor the vibration data of the cabinet structure, and integrate them into the original switch cabinet data; Put a unified time stamp on the original switch cabinet data through the time protocol, and pre-process to obtain the pre-processed original switch cabinet data; Align the pre-processed original switch cabinet data according to the unified time stamp, and integrate it into a fixed time window data frame to obtain a multi-modal time series data set.

[0007] As a preferred scheme of the switch cabinet monitoring method based on Internet of Things technology, wherein: the multi-modal time series data set is input into the multi-physical field coupling digital twin model to generate a predictive instruction set, and the specific steps are as follows, The cloud server receives the multi-modal time series data set through the message middleware, and verifies the integrity and time stamp continuity of the multi-modal time series data set, removes the incomplete data caused by transmission errors, and obtains the verified multi-modal time series data set; The verified multi-modal time series data set is input into the solver of the multi-physical field coupling digital twin model, and the solver takes the verified multi-modal time series data set as the boundary condition and initial condition, and inversely calibrates the multi-physical field coupling digital twin model parameters to obtain a real-time digital twin of the switch cabinet; A forward transient simulation is performed on the real-time digital twin of the switch cabinet to simulate the evolution process of the switch cabinet under the future specified working condition and predict the change trajectory; The change trajectory is mapped to a risk level quantitative value, and the control parameter recommended value is generated with the lowest risk as the optimization target; The risk level quantitative value and the control parameter recommended value are packaged into a predictive instruction set through the instruction assembly service in the cloud server.

[0008] As a preferred scheme of the switch cabinet monitoring method based on the Internet of Things technology, the specific steps of obtaining the adjusted lightweight convolutional neural network are as follows, The risk level quantitative value is extracted from the predictive instruction set, and the risk level quantitative value is mapped to different risk level positions, the risk level position is converted into a discriminant probability threshold adjustment coefficient, and the product of the discriminant probability threshold adjustment coefficient and the basic discriminant probability threshold in the lightweight convolutional neural network is taken as a new discriminant probability threshold; The voltage fluctuation rate and the load change rate of the switch cabinet are collected, and the response strategy aggressiveness parameter is derived in combination with the control strategy generator; The new discriminant probability threshold and the response strategy aggressiveness parameter are replaced into the lightweight convolutional neural network to obtain the adjusted lightweight convolutional neural network.

[0009] As a preferred scheme of the switch cabinet monitoring method based on the Internet of Things technology, the specific steps of obtaining the adjusted lightweight convolutional neural network are as follows, Based on the local discharge pulse waveform data, a fixed length waveform data segment is extracted and standardized and normalized to obtain a standard waveform data segment; The standard waveform data segment is input into the adjusted lightweight convolutional neural network, and continuous convolution and pooling operations are performed to extract time domain and frequency domain abstract features in the standard waveform data segment layer by layer to generate a probability distribution vector; Based on the probability distribution vector, a decision is made in combination with the new discriminant probability threshold, and when the probability value of any category in the probability distribution vector exceeds the discriminant probability threshold, the probability distribution vector and the new discriminant probability threshold are identified and events are generated to obtain a fault warning event identifier.

[0010] As a preferred scheme of the switch cabinet monitoring method based on the Internet of Things technology, the specific steps of obtaining the adjusted lightweight convolutional neural network are as follows, The discharge mode type and event severity level are extracted from the fault early warning event identification, and combined with real-time power grid load rate and environmental temperature parameters, and voltage regulation parameters are dynamically calculated through a multi-objective optimization algorithm; The voltage regulation parameters are structurally converted, and voltage regulation instructions are dynamically constructed; the voltage regulation instructions are safely packaged, and the voltage regulation instructions are transmitted to the voltage regulator through an encrypted channel.

[0011] As a preferred scheme of the switch cabinet monitoring method based on the Internet of Things technology, the voltage regulator reduces the line output voltage based on the voltage regulation instructions to suppress partial discharge, and the specific steps are as follows, The voltage regulation instructions are checked for integrity and legality to obtain the checked voltage regulation instructions; The checked voltage regulation instructions are decoded, control parameters are extracted, the control parameters are converted into corresponding PWM duty cycle values, and a pulse width modulation signal is generated in combination with the PWM generator hardware of the microcontroller; The conduction and turn-off of the power semiconductor switching device are controlled by the pulse width modulation signal driving circuit, the output voltage of the current conversion circuit is adjusted, and the voltage value is monitored in real time; The deviation between the monitored voltage value and the target voltage value is calculated by the inner loop closed loop control algorithm, the PWM signal duty cycle is corrected in real time by proportional-integral adjustment, the stable voltage is obtained, the internal electric field strength of the switch cabinet is reduced by the stable voltage, the partial discharge is suppressed, and the voltage regulation process is recorded.

[0012] As a preferred scheme of the switch cabinet monitoring method based on the Internet of Things technology, the switch cabinet health state report is generated, and the specific steps are as follows, The multi-modal time series data set, the predictive instruction set, the fault early warning event identification and the voltage regulation process are integrated to obtain a heterogeneous data set; Based on the heterogeneous data set, the comprehensive insulation state index of the current switch cabinet is calculated by an evaluation algorithm, and the equipment health state category is obtained according to the insulation state index and the risk level in the predictive instruction set; The comprehensive insulation state index and the equipment health state category are packaged to obtain the switch cabinet health state report.

[0013] In a second aspect, the present application provides a computer device comprising a memory and a processor, the memory storing a computer program, wherein the computer program is executed by the processor to realize any step of the switch cabinet monitoring method based on the Internet of Things technology according to the first aspect of the present application.

[0014] In a third aspect, the present application provides a computer readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements any step of the switchgear monitoring method based on Internet of Things technology according to the first aspect of the present application.

[0015] The present application has the beneficial effects that: by constructing a multi-modal time series dataset, efficient fusion and time synchronization of multi-physical field information such as switch cabinet partial discharge, temperature and vibration are realized, the data integrity and accuracy of state perception are improved, and high-quality input is provided for intelligent diagnosis; the adaptability and intelligent operation level under complex working conditions are also improved, and the safety and stability of long-term operation of the switch cabinet are effectively ensured. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Fig. 1 Flowchart of the switchgear monitoring method based on Internet of Things technology.

[0018] Fig. 2 Flowchart of generating a multi-modal time series dataset.

[0019] Fig. 3 Flowchart of generating a predictive instruction set for a multi-physical field coupled digital twin model.

[0020] Fig. 4 Adjustment and pattern recognition process of a lightweight convolutional neural network. DETAILED DESCRIPTION

[0021] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0022] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the scope of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0023] Second, the "one embodiment" or "an embodiment" referred to herein can include a particular feature, structure, or characteristic. The various embodiments appearing at different places in this specification are not necessarily all cumulative with one another. That is, an embodiment appearing at one place in this specification can or can not be combined with other embodiments appearing elsewhere in this specification.

[0024] Referring to Figs. 1-4 For one embodiment of the present application, the embodiment provides a switch cabinet monitoring method based on Internet of Things technology, comprising the following steps: S1. Collecting partial discharge pulse waveform data, temperature data and vibration data on the switch cabinet, and generating a multi-modal time series data set after processing through the Internet of Things gateway.

[0025] S1.1. Capture the electromagnetic wave signal generated by partial discharge and convert it into partial discharge pulse waveform data, collect the temperature data of the bus connection point and the insulating surface, monitor the vibration data of the cabinet structure, and integrate them into the original switch cabinet data.

[0026] Specifically, the ultra-high frequency sensor deployed on the switch cabinet captures the electromagnetic wave signal generated by partial discharge, and converts the electromagnetic wave signal into a time-domain waveform signal through a signal conditioning circuit. After sampling and quantization by an analog-to-digital converter at a sampling rate of not less than 100MHz, the partial discharge pulse waveform data is formed. At the same time, the thermocouple temperature sensor collects the temperature data of the bus connection point and the insulating surface at a frequency of 1 per second, and the piezoelectric vibration sensor monitors the vibration data of the cabinet structure at a sampling frequency of 10kHz. The partial discharge pulse waveform data, temperature data and vibration data are packaged to form the original switch cabinet data.

[0027] S1.2. A uniform timestamp is stamped on the original switch cabinet data through a time protocol; and pre-processing is performed to obtain pre-processed original switch cabinet data.

[0028] Specifically, the Internet of Things gateway obtains standard time from a time server through a network time protocol, and stamps a uniform timestamp with millisecond-level precision on the partial discharge pulse waveform data, temperature data and vibration data in the original switch cabinet data. A Butterworth low-pass filter is used to filter out high-frequency noise from the partial discharge pulse waveform data, a sliding average algorithm is used to smooth the temperature data, and a notch filter is used to eliminate power frequency interference in the vibration data; the pre-processed original switch cabinet data is obtained.

[0029] S1.3. Align the pre-processed original switch cabinet data according to the uniform timestamp, and integrate it into a data frame with a fixed time window to obtain a multi-modal time series data set.

[0030] Specifically, the partial discharge pulse waveform data, temperature data and vibration data in the pre-processed original switch cabinet data are time-aligned according to the annotated millisecond-precision unified timestamp; a fixed-length sliding window mechanism (for example, the example value is 60 seconds) is used for segmentation, and the original switch cabinet data with different sampling rates in the fixed-length sliding window are resampled to keep the partial discharge pulse waveform data, temperature data and vibration data synchronized in the time dimension; the multi-source sensor data in each time window is packaged into a structured data frame to form a multi-modal time series data set.

[0031] It should be noted that each time window contains complete partial discharge pulse waveform data, temperature data and vibration data.

[0032] S2. Input the multi-modal time series data set into the multi-physical field coupled digital twin model to generate a set of predictive instructions.

[0033] S2.1. The cloud server receives the multi-modal time series data set through the message middleware and verifies the integrity and timestamp continuity of the multi-modal time series data set, removes incomplete data caused by transmission errors, and obtains the verified multi-modal time series data set.

[0034] Specifically, the cloud server receives the multi-modal time series data set from the Internet of Things gateway through the message middleware, verifies the data integrity by performing a cyclic redundancy check on the received multi-modal time series data set, checks whether the timestamps of the partial discharge pulse waveform data, temperature data and vibration data maintain a millisecond-level continuous increasing relationship, removes data frames that fail the check, incomplete data caused by timestamp discontinuity and transmission errors, and retains the verified multi-modal time series data set as the verified multi-modal time series data set.

[0035] S2.2. Input the verified multi-modal time series data set into the solver of the multi-physical field coupled digital twin model, and the solver takes the verified multi-modal time series data set as the boundary conditions and initial conditions to calibrate the parameters of the multi-physical field coupled digital twin model in reverse, obtaining a real-time digital twin of the switch cabinet.

[0036] Specifically, the verified multi-modal time series data set is input into the finite element solver of the multi-physical field coupled digital twin model, the temperature data in the verified multi-modal time series data set is taken as the thermal field boundary condition, the vibration data is taken as the structure field boundary condition, and the partial discharge pulse waveform data is taken as the electric field boundary condition, the material thermal conductivity coefficient parameter, the structure damping parameter and the dielectric constant parameter in the multi-physical field coupled digital twin model are calibrated by the back propagation algorithm, and the real-time digital twin of the switch cabinet that matches the real-time monitoring data is obtained.

[0037] It should be noted that the multi-physical field coupled digital twin model is trained, an initial finite element model is constructed based on the switch cabinet design drawings and material parameters, the temperature distribution data, vibration frequency spectrum data and partial discharge waveform data accumulated in the historical operation and maintenance database are used as multi-physical field training samples; the conjugate gradient method optimization algorithm is used, the mean square error between the simulation data and the measured data is used as the loss function, the physical field distribution is predicted through forward finite element simulation calculation, and then the material thermal conductivity coefficient, structure damping coefficient and dielectric constant and other physical parameters are iteratively adjusted through the back propagation algorithm; when the loss function of continuous 5 iterations is less than 0.1%, the training is terminated, and the multi-physical field coupled digital twin model reflecting the real physical characteristics of the switch cabinet is obtained.

[0038] S2.3. The real-time digital twin of the switch cabinet is subjected to forward transient simulation to simulate the evolution process of the switch cabinet under the future specified working condition and predict the change trajectory.

[0039] Specifically, based on the real-time digital twin of the switch cabinet, the future 72-hour specified working condition (for example, the example takes the environmental temperature of 40°C and the relative humidity of 85%) is set as the simulation condition, the explicit time integration method is used for transient simulation calculation of the multi-physical field coupled digital twin model, the dielectric constant decay trajectory, the partial discharge growth trajectory and the temperature rise distribution change trajectory in the future time period are obtained, the evolution process of the insulation state of the switch cabinet is predicted through the iterative operation of coupling electric field-thermal field-structure field, and the change prediction trajectory is output.

[0040] S2.4. The change trajectory is mapped to a risk level quantitative value, and the risk lowest is taken as the optimization target to generate a control parameter recommended value.

[0041] Specifically, the partial discharge growth trajectory, the dielectric constant decay trajectory and the temperature rise distribution change trajectory in the change prediction trajectory are fused in multiple dimensions, the risk quantitative values of three levels of high risk, medium risk and low risk are calculated through a weighted scoring algorithm according to the duration proportion of the partial discharge exceeding the standard threshold, the dielectric constant descending rate and the hot spot temperature rise value; taking the risk quantitative value lowest as the optimization target, the gradient descent method is used to search for the optimal solution in the parameter space to generate the control parameter recommended value.

[0042] It should be noted that the standard threshold is determined by analyzing the statistical distribution of the partial discharge, dielectric constant and temperature data in the historical operation and maintenance data, using a clustering algorithm (such as K-means) to identify the boundary values of normal and abnormal states, and combining the critical parameters in the actual fault cases.

[0043] S2.5. The risk level quantitative value and the control parameter recommended value are packaged into a predictive instruction set through the instruction assembly service in the cloud server.

[0044] Specifically, the instruction assembly service in the cloud server receives the risk quantification values and control parameter recommended values of three levels of high risk, medium risk and low risk from the risk quantification calculation process; the instruction assembly service maps the risk quantification values to standardized risk level codes and converts the control parameter recommended values to key-value pair data according to the general data class template in the IEC 61850 standard; and the key-value pair data, the uniform timestamp identifier and the valid time interval information are collectively encapsulated as a structured data message conforming to the JSON-LD specification to generate a set of predictive instructions.

[0045] S3. Based on the risk level quantification value in the set of predictive instructions, the discrimination probability threshold value and the aggressiveness parameter in the lightweight convolutional neural network are dynamically adjusted to obtain an adjusted lightweight convolutional neural network.

[0046] S3.1. The risk level quantification value is extracted from the set of predictive instructions, and the risk level quantification value is mapped to different risk level gears, the risk level gears are converted to discrimination probability threshold value adjustment coefficients, and the product of the discrimination probability threshold value adjustment coefficients and the basic discrimination probability threshold value in the lightweight convolutional neural network is taken as the new discrimination probability threshold value.

[0047] Specifically, the target field in the set of predictive instructions is located through the JSON-LD data architecture key name (for example, "riskLevelQuantValue") in the set of predictive instructions, the field value is extracted and converted to a numerical type, the specific risk level quantification value is obtained, the risk level quantification value is compared with the risk level gear mapping table, and the risk level quantification value is classified into a high risk gear, a medium risk gear or a low risk gear according to the numerical range specified in the risk level gear mapping table; the risk level gear is converted to a specific discrimination probability threshold value adjustment coefficient value according to the risk level gear and the discrimination probability threshold value adjustment coefficient correspondence table; and the product of the discrimination probability threshold value adjustment coefficient value and the pre-configured basic discrimination probability threshold value in the lightweight convolutional neural network is taken as the new discrimination probability threshold value of the lightweight convolutional neural network.

[0048] It should be noted that the risk level gear mapping table and the discrimination probability threshold value adjustment coefficient correspondence table collect the key parameter records such as the local discharge exceeding duration ratio, the dielectric constant decline rate and the hotspot temperature rise value in the historical operation and maintenance data of the switch cabinet, and the corresponding actual fault occurrence conditions, and extract the parameter safety threshold range specified in the power equipment safe operation specification; the historical data is divided into three data clusters of high risk, medium risk and low risk by using clustering analysis method, and the risk level gear mapping table is obtained according to the data distribution range of the three data clusters of high risk, medium risk and low risk. By statistically correlating the device failure probability under different risk levels with the original discriminant threshold false alarm rate, a discriminant probability threshold adjustment coefficient value is obtained that can guarantee device safety while avoiding excessive false alarms, and a discriminant probability threshold adjustment coefficient correspondence table is formed. It should also be noted that, in order to train the lightweight convolutional neural network, a labeled data set containing multiple typical partial discharge waveforms is collected, which contains partial discharge pulse waveform data and corresponding real discharge mode class labels; the stochastic gradient descent method is used as the optimization algorithm, and the cross-entropy loss function is used as the objective function; the forward propagation is used to calculate the prediction output of the lightweight convolutional neural network, and the back propagation algorithm is used to calculate the gradient of the loss function with respect to the convolution kernel weight and bias parameters in the lightweight convolutional neural network; the learning rate decay strategy is used to dynamically adjust the learning rate (for example, the initial learning rate is set to 0.01 and is decayed by 0.1 times every 10 training periods), and the Dropout regularization technique is applied to prevent overfitting; when the validation set accuracy does not improve for 5 consecutive training periods, the early stopping mechanism is triggered to terminate training, and the trained lightweight convolutional neural network model is obtained.

[0049] After the training of the lightweight convolutional neural network is completed, the prediction probability value distribution of all samples is calculated using the validation set, a best critical point that balances the false alarm rate and the false alarm rate is selected according to the probability distribution curve (for example, the probability value corresponding to a false alarm rate of 5%), and then the basic discriminant probability threshold is obtained by combining the partial discharge alarm threshold requirement specified in the power equipment safe operation specification DL / T 596.

[0050] S3.2. Collect the voltage fluctuation rate and load change rate of the switch cabinet, and combine the control strategy generator to derive the aggressiveness parameter of the response strategy.

[0051] Specifically, the voltage data and load current data of the switch cabinet are collected in real time by voltage sensors and current sensors to calculate the voltage fluctuation rate and load change rate; the voltage fluctuation rate and load change rate are input into the control strategy generator, which compares the voltage fluctuation rate and load change rate with the threshold range in the rule base based on the response strategy rule base, and selects the corresponding calculation rule according to the comparison result; for example, when the voltage fluctuation rate exceeds 5% and the load change rate is greater than 10% / min, the weighted average algorithm is used to set the voltage fluctuation rate weight to 0.6 and the load change rate weight to 0.4, and the aggressiveness parameter value is calculated to be 0.82; when the voltage fluctuation rate is between 2%-5% and the load change rate is less than 5% / min, the linear interpolation algorithm is used to map the fluctuation rate and change rate to the parameter interval of 0.3-0.6; when both indicators are below the threshold, the conservative strategy parameter 0.2 is directly output.

[0052] It should be noted that the process involves collecting historical operation and maintenance data, including records of switchgear voltage fluctuation rate, load change rate, and corresponding successful handling strategies, and recommending control methods for different operating conditions. Cluster analysis is used to classify voltage fluctuation rate and load change rate into different levels of combined operating conditions. Association rule mining algorithms are then used to find the correspondence between each combined operating condition and the optimal handling strategy. The mined rules are then corrected and confirmed based on equipment safety operation specifications and actual operation and maintenance experience to form a response strategy rule base.

[0053] It should be noted that association rule mining algorithms are a method for automatically discovering frequently occurring strong association patterns of "if condition X occurs, then strategy Y is effective" from historical data. By calculating indicators such as support (frequency of pattern occurrence), confidence (probability of strategy effectiveness when condition occurs), and lift (relevance strength excluding randomness), the algorithm quantifies and evaluates the causal relationship between different combinations of working conditions and successful handling strategies, thereby providing data-driven decision-making basis for building a response strategy rule base.

[0054] It should be noted that the expression for calculating voltage fluctuation rate is: ; in, Indicates voltage fluctuation rate. The standard deviation of a voltage series is represented. This represents the arithmetic mean of the voltage series.

[0055] The expression for calculating the load change rate is: ; in, Indicates the rate of change of load. Indicates the current Current sampling value, Indicates the previous moment The current value, Indicates the sampling period.

[0056] S3.3. Replace the new discrimination probability threshold and the aggressiveness parameter of the response strategy in the lightweight convolutional neural network to obtain the adjusted lightweight convolutional neural network.

[0057] Specifically, the new discrimination probability threshold and the aggressiveness parameter of the response strategy are input into the parameter configuration interface of the lightweight convolutional neural network. The new discrimination probability threshold directly covers the discrimination probability threshold parameter of the original output layer of the lightweight convolutional neural network, and the aggressiveness parameter of the response strategy is written into the strategy control parameter storage area of the lightweight convolutional neural network model. After the parameter replacement is completed, the lightweight convolutional neural network will use the new discrimination probability threshold to make classification decisions on the partial discharge pulse waveform data in the subsequent inference process, and simultaneously adjust the response sensitivity and action trigger condition of the classification output according to the aggressiveness parameter of the response strategy, thereby forming an adjusted lightweight convolutional neural network.

[0058] S4. Utilize the adjusted lightweight convolutional neural network to perform pattern recognition on the partial discharge pulse waveform data, and generate a fault warning event identifier.

[0059] S4.1. Based on the partial discharge pulse waveform data, extract a fixed-length waveform data segment, and perform standardization and normalization processing to obtain a standard waveform data segment.

[0060] Specifically, based on the real-time collected partial discharge pulse waveform data, a fixed-length waveform data segment (for example, an example value of 256 sampling points) is extracted using a fixed time window extraction algorithm. The extracted waveform data segment is first subjected to direct current component elimination processing, and a Butterworth band-pass filter is used to retain signal components in a specific frequency range (for example, an example value of 30 kHz-1 MHz). Standardization processing is performed, the root mean square value of the waveform data segment is calculated, and the amplitudes of all sampling points are divided by the root mean square value. Finally, normalization processing is performed, and the amplitude values are mapped to a specific range (for example, an example value of [-1, 1]) through linear transformation, thereby obtaining a standard waveform data segment.

[0061] S4.2. The standard waveform data segment is input into the adjusted lightweight convolutional neural network, and continuous convolution and pooling operations are performed to extract time domain and frequency domain abstract features in the standard waveform data segment layer by layer, thereby generating a probability distribution vector.

[0062] Specifically, the standard waveform data segment is input into the adjusted lightweight convolutional neural network, a set of learnable convolution kernels (for example, 32 5x1 size convolution kernels) are used in the first convolution layer to perform one-dimensional convolution operation on the standard waveform data segment, to extract the primary time domain features in the partial discharge pulse waveform data and generate a feature map; the feature map is nonlinearly transformed using a rectified linear unit activation function, and the feature map is dimensionally reduced by a maximum pooling algorithm (for example, a 2x1 pooling window) in the first pooling layer; the convolution-activation-pooling operation is repeated in the subsequent layers of the adjusted lightweight convolutional neural network (for example, repeated 3 times), to extract higher-level time domain and frequency domain abstract features layer by layer; the feature map is converted into a feature vector by a global average pooling layer, and the feature vector is converted into a probability distribution vector by a fully connected layer combined with a Softmax function.

[0063] S4.3. Based on the probability distribution vector, a new discriminant probability threshold is combined to make a decision, when the probability value of any class in the probability distribution vector exceeds the discriminant probability threshold, the probability distribution vector and the new discriminant probability threshold are pattern recognized and event generated, to obtain the fault warning event identification.

[0064] Specifically, each probability value in the probability distribution vector is compared with the new discriminant probability threshold; when it is detected that the probability value of any class in the probability distribution vector exceeds the new discriminant probability threshold, the class number and the corresponding probability value are recorded, and the corresponding discharge mode type description (such as "insulator surface discharge", "bus connection floating potential discharge", "cable terminal internal air gap discharge") is obtained; the discharge mode type description, the excessive probability value, the current timestamp and the new discriminant probability threshold value are combined to generate a standardized event description string; and encapsulated into a fault warning event identification conforming to the IEC 61850 event format specification.

[0065] S5. The fault warning event identification is converted into a voltage regulation instruction and sent to a voltage regulator, the voltage regulator reduces the line output voltage based on the voltage regulation instruction to suppress the partial discharge.

[0066] S5.1. The discharge mode type and event severity level are extracted from the fault warning event identification, and combined with real-time power grid load rate and environmental temperature parameters, to dynamically calculate the voltage regulation parameters by a multi-objective optimization algorithm.

[0067] Specifically, the discharge mode type field and the event severity level value are parsed from the fault early warning event identification, real-time power grid load rate data is obtained from SCADA (Supervisory Control And Data Acquisition), and environmental temperature parameters are obtained from temperature sensors; the discharge mode type, event severity level, real-time power grid load rate, and environmental temperature parameters are input into a multi-objective optimization algorithm, a weighted cost function including three optimization objectives is established, voltage adjustment range, load rate limitation, and environmental temperature influence are set as constraint conditions according to the safe operating voltage range of equipment specified in the power operation regulation, the load capacity limit value calibrated on the nameplate of the switch cabinet, and the temperature characteristic curve of the insulating material; a Pareto optimization algorithm with constraints is used to solve the multi-objective problem, and the voltage adjustment parameter is output.

[0068] It should be noted that the voltage adjustment parameter includes a voltage adjustment amplitude (a percentage deviation of a target voltage from a rated voltage), an adjustment duration (a maintenance time of a voltage adjustment operation), an adjustment rate (a voltage change rate per unit time), and an optimal execution time point (an optimal action timing based on load prediction).

[0069] It should be noted that the voltage adjustment parameter expression is: ; Wherein, represents the voltage adjustment parameter, represents the weight of the partial discharge suppression term, represents the weight of the power supply reliability term, represents the weight of the energy consumption loss term, represents the event severity level, represents the real-time power grid load rate, represents the environmental temperature, represents the partial discharge suppression utility function, represents the power supply continuity influence function, represents the energy consumption loss function.

[0070] It should be noted that, an exponential decay form is used to describe the suppression effect of voltage reduction on discharge, based on the physical law that the discharge intensity and the electric field intensity usually have an exponential relationship (similar to Paschen's law), and multiple factors are introduced through linear weighting; a quadratic penalty term is directly used to punish the degree of deviation of the state variable (voltage, load) from the rated value; linear and quadratic terms are combined, the linear term represents the loss (such as Ohmic loss) that is proportional to the adjustment amplitude and the load, and the quadratic term represents the nonlinear loss in the switching device (such as the switching loss of IGBT).

[0071] S5.2. Structurally convert the voltage regulation parameters, and dynamically build the voltage regulation instruction; securely package the voltage regulation instruction, and transmit the voltage regulation instruction to the voltage regulator through an encrypted channel.

[0072] Specifically, the voltage regulation amplitude, regulation duration, and execution time point in the voltage regulation parameters are structurally converted according to the controllable device data device in the IEC61850 standard, to generate standardized data conforming to the general substation event format; the voltage regulation instruction containing the target voltage value, regulation rate, and execution timestamp is dynamically built based on the standardized data; a security message header is added to the voltage regulation instruction, and a cyclic redundancy check code is calculated, and the complete instruction is encrypted by using the AES-256 encryption algorithm; the encrypted voltage regulation instruction is transmitted to the control interface of the intelligent voltage regulator through a TLS 1.3 encryption channel based on certificate authentication.

[0073] S5.3. Verify the integrity and legality of the voltage regulation instruction, to obtain the verified voltage regulation instruction.

[0074] Specifically, the received encrypted voltage regulation instruction is decrypted by using the AES-256 algorithm to obtain the original voltage regulation instruction data; the decrypted voltage regulation instruction is calculated for a cyclic redundancy check code, and the original cyclic redundancy check code carried in the voltage regulation instruction is compared to verify the integrity; the target voltage value, regulation rate, and execution timestamp fields in the voltage regulation instruction are analyzed to verify whether the target voltage value is within the allowed regulation range (for example, the example value is 90% to 110% of the rated voltage), whether the regulation rate exceeds the maximum allowed change rate, and whether the execution timestamp is after the current time and within the effective time window; the voltage regulation instruction that passes all verifications is marked as the verified voltage regulation instruction.

[0075] It should be noted that the target voltage value regulation range is determined by analyzing the dielectric strength characteristics of the insulation materials of the switch cabinet device, the voltage withstand margin design parameters of the power electronic device, and the statistical distribution of the voltage safe operation interval in the historical operation and maintenance data; The maximum allowed change rate of the regulation rate is determined by referring to the voltage fluctuation rate limit in the power dispatching regulations, to prevent protection misoperation, abnormal load device, or transient oscillation of the regional power grid caused by excessive voltage regulation; The effective time window is set considering the measured statistical value of the communication network transmission delay (such as the substation optical fiber network delay is usually less than 50 ms) and the voltage regulator response time (usually 100-500 ms).

[0076] S5.4. Decode the checked voltage regulation instruction, extract the control parameters, convert the control parameters into corresponding PWM duty cycle values, and combine the microcontroller PWM generator hardware to generate a pulse width modulation signal.

[0077] Specifically, the checked voltage regulation instruction is protocol decoded to extract the target voltage value, regulation rate and execution timestamp control parameters; according to the ratio relationship between the target voltage value and the rated voltage, the corresponding PWM duty cycle value is obtained; the PWM duty cycle value is written into the duty cycle control register of the PWM generator hardware of the microcontroller, and at the same time, the frequency control register value of the PWM generator hardware is set according to the regulation rate parameter; the PWM generator hardware automatically generates a pulse width modulation signal with corresponding duty cycle and frequency according to the register configuration.

[0078] S5.5. The on and off of the power semiconductor switching device is controlled by the pulse width modulation signal driving circuit to adjust the output voltage of the current conversion circuit and monitor the voltage value in real time.

[0079] Specifically, the pulse width modulation signal is amplified by the gate drive circuit to control the gate voltage of the insulated gate bipolar transistor, and the on and off time ratio of the insulated gate bipolar transistor is accurately controlled; the switching state change of the insulated gate bipolar transistor adjusts the energy storage and release process of the inductor element in the Buck-Boost current conversion circuit, thereby adjusting the output voltage of the current conversion circuit; during the voltage adjustment process, the voltage sensor monitors the output voltage value of the current conversion circuit in real time at a sampling frequency of 1000 times per second.

[0080] S5.6. Calculate the deviation of the monitored voltage value and the target voltage value through the inner loop closed loop control algorithm, real-time correct the PWM signal duty cycle through proportional-integral regulation, get stable voltage, reduce the electric field strength inside the switch cabinet through stable voltage, suppress local discharge, and record the voltage regulation process.

[0081] Specifically, the proportional-integral regulator in the inner loop closed-loop control algorithm calculates the deviation between the voltage sensor monitored voltage value and the target voltage value in the voltage regulation instruction in real time, and generates a PWM duty cycle correction amount according to the deviation and an integral time constant; based on the sum of the PWM duty cycle correction amount and the current PWM duty cycle value, a new PWM duty cycle value is obtained and immediately written into the duty cycle control register of the PWM generator hardware; the PWM generator hardware generates a corrected pulse width modulation signal according to the updated duty cycle value, adjusts the output of the current conversion circuit through the driving circuit, and stabilizes the monitored voltage value around the target voltage value; after the stable voltage is applied to the bus of the switch cabinet, the electric field strength on the surface of the insulating material is reduced, and the suppression of local discharge activity is realized according to the Paschen law principle; at the same time, the recorder records the stable voltage value, the PWM duty cycle value and the monitored voltage value at a frequency of 100 times per second, forming a voltage regulation process.

[0082] S6. Integrating the multi-modal time series data set, the predictive instruction set, the fault early warning event identification and the voltage regulation record to generate a switch cabinet health status report.

[0083] S6.1. Integrating the multi-modal time series data set, the predictive instruction set, the fault early warning event identification and the voltage regulation process to obtain a heterogeneous data set.

[0084] Specifically, the local discharge pulse waveform data, temperature data and vibration data in the multi-modal time series data set, the risk level code and control parameter key-value pair in the predictive instruction set, the discharge mode type description and over-standard probability value in the fault early warning event identification, and the stable voltage value, PWM duty cycle value and monitored voltage value in the voltage regulation process record are time series aligned and associated through a unified timestamp; different structures of data are converted into a unified binary format using the Apache Avro data serialization framework, and all time-aligned and format-unified data packets are organized according to time series to generate a heterogeneous data set.

[0085] S6.2. Based on the heterogeneous data set, the comprehensive insulation state index of the current switch cabinet is calculated by an evaluation algorithm, and the equipment health status category is obtained according to the insulation state index and the risk level in the predictive instruction set.

[0086] Specifically, based on the partial discharge pulse waveform data, temperature data, vibration data, risk level code, control parameter key-value pair, discharge mode type description, over-standard probability value, stable voltage value, PWM duty cycle value and monitoring voltage value in the heterogeneous data set, the comprehensive insulation state index of the current switch cabinet is calculated by the dielectric loss factor (tan delta) calculation method. The discharge phase distribution spectrum is extracted from the partial discharge pulse waveform data, and the temperature data is combined with the temperature characteristic curve of the insulation medium capacitance value to correct and compensate the capacitance change caused by mechanical stress, and the dielectric loss factor is calculated by the standard capacitance current vector model. The partial discharge quantity, temperature rise and other parameters are fused into a comprehensive insulation state index in the range of 0-100 by a weighted arithmetic mean method; the comprehensive insulation state index and the risk level code in the predictive instruction set are input into the health state classifier, and the health state classifier outputs the device health state category according to the threshold range (for example, the comprehensive insulation state index is greater than or equal to 80, the health state is 50-79, the attention state is less than 50, and the early warning state is the device health state category).

[0087] S6.3. Package the comprehensive insulation state index and the device health state category to obtain a switch cabinet health state report.

[0088] Specifically, the comprehensive insulation state index and the device health state category are taken as core data elements, and are structured and packaged according to the IEC62361 standard to create a standardized data object containing a timestamp field, a comprehensive insulation state index value field and a device health state category code field; a data source identifier is added to point to the insulation state evaluation algorithm and the health state classifier, and an evaluation time interval parameter is added to explain the validity range; the standardized data object is converted into a machine-readable document in the JSON-LD serialization mode to generate a switch cabinet health state report.

[0089] The embodiment also provides a computer device suitable for the switch cabinet monitoring method based on the Internet of Things technology, which comprises a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the switch cabinet monitoring method based on the Internet of Things technology proposed in the above embodiment. The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved by WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0090] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the switch cabinet monitoring method based on the Internet of Things technology as described in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk.

[0091] To sum up, the application realizes efficient fusion and time synchronization of multi-physical field information such as switch cabinet partial discharge, temperature and vibration by constructing a multi-modal time series data set, improves the data integrity and accuracy of state perception, and provides high-quality input for intelligent diagnosis. In addition, the adaptability and intelligent operation level under complex working conditions are improved, and the safety and stability of long-term operation of the switch cabinet are effectively ensured.

[0092] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A switch cabinet monitoring method based on Internet of Things technology, characterized in that: The method comprises the following steps: Collecting partial discharge pulse waveform data, temperature data and vibration data on the switch cabinet, and generating a multi-modal time series data set after processing through an Internet of Things gateway; Inputting the multi-modal time series data set into a multi-physics field coupled digital twin model to generate a predictive instruction set; Based on the risk level quantification value in the predictive instruction set, dynamically adjusting the discriminant probability threshold and the aggressiveness parameter in the lightweight convolutional neural network to obtain an adjusted lightweight convolutional neural network; Using the adjusted lightweight convolutional neural network to perform pattern recognition on the partial discharge pulse waveform data to generate a fault early warning event identifier; Converting the fault early warning event identifier into a voltage regulation instruction and sending it to a voltage regulator, and the voltage regulator reduces the line output voltage based on the voltage regulation instruction to suppress partial discharge; Integrating the multi-modal time series data set, the predictive instruction set, the fault early warning event identifier and the voltage regulation record to generate a switch cabinet health status report.

2. The switch cabinet monitoring method based on Internet of Things technology according to claim 1, characterized in that: The method comprises the following steps of collecting partial discharge pulse waveform data, temperature data and vibration data on the switch cabinet, and generating a multi-modal time series data set after processing through an Internet of Things gateway, Capturing electromagnetic wave signals generated by partial discharge and converting them into partial discharge pulse waveform data, collecting temperature data at bus connection points and insulating surfaces, monitoring vibration data of the cabinet structure, and integrating them into raw switch cabinet data; Applying a unified timestamp to the raw switch cabinet data through a time protocol and pre-processing to obtain pre-processed raw switch cabinet data; Aligning the pre-processed raw switch cabinet data according to the unified timestamp and integrating them into data frames with fixed time windows to obtain a multi-modal time series data set. 3.The switch cabinet monitoring method based on Internet of Things technology according to claim 2, characterized in that: The method comprises the following steps of inputting the multi-modal time series data set into a multi-physics field coupled digital twin model to generate a predictive instruction set, The cloud server receives the multi-modal time series data set through a message middleware and verifies the integrity and timestamp continuity of the multi-modal time series data set, removes incomplete data caused by transmission errors, and obtains verified multi-modal time series data set; Inputting the verified multi-modal time series data set into the solver of the multi-physics field coupled digital twin model, the solver taking the verified multi-modal time series data set as boundary conditions and initial conditions, inversely calibrating the multi-physics field coupled digital twin model parameters, and obtaining a real-time digital twin of the switch cabinet; Performing forward-looking transient simulation on the real-time digital twin of the switch cabinet to simulate the evolution process of the switch cabinet under future specified operating conditions and predict the change trajectory; Mapping the change trajectory to a risk level quantification value and taking the lowest risk as the optimization target to generate a control parameter recommendation value; The cloud server's instruction assembly service encapsulates the risk level quantification value and the control parameter recommendation value into a predictive instruction set.

4. The switch cabinet monitoring method based on Internet of Things technology according to claim 1, characterized in that: The method comprises the following steps of obtaining an adjusted lightweight convolutional neural network, Extracting the risk level quantification value from the predictive instruction set and mapping it to different risk level gears, converting the risk level gears to discriminant probability threshold adjustment coefficients, and taking the product of the discriminant probability threshold adjustment coefficients and the basic discriminant probability threshold in the lightweight convolutional neural network as a new discriminant probability threshold; The voltage fluctuation rate and load change rate of the switch cabinet are collected, and the control strategy generator is combined to derive the aggressiveness parameter of the response strategy; The new discrimination probability threshold and the aggressiveness parameter of the response strategy are replaced into the lightweight convolutional neural network to obtain an adjusted lightweight convolutional neural network.

5. The switch cabinet monitoring method based on Internet of Things technology according to claim 4, characterized in that: The pattern recognition of the local discharge pulse waveform data is performed by using the adjusted lightweight convolutional neural network to generate a fault warning event identifier, and the specific steps are as follows, Based on the local discharge pulse waveform data, a fixed-length waveform data segment is extracted and standardized and normalized to obtain a standard waveform data segment; The standard waveform data segment is input into the adjusted lightweight convolutional neural network, and continuous convolution and pooling operations are performed to extract time domain and frequency domain abstract features in the standard waveform data segment layer by layer to generate a probability distribution vector; Based on the probability distribution vector, a decision is made in combination with the new discrimination probability threshold, when the probability value of any category in the probability distribution vector exceeds the new discrimination probability threshold, the probability distribution vector and the new discrimination probability threshold are pattern recognized and event generated to obtain a fault warning event identifier.

6. The switchgear monitoring method based on Internet of Things technology according to claim 5, characterized in that: The fault warning event identifier is converted into a voltage regulation instruction and sent to a voltage regulator, and the specific steps are as follows, The discharge mode type and event severity level are extracted from the fault warning event identifier, and the real-time power grid load rate and environmental temperature parameters are combined to dynamically calculate the voltage regulation parameter through a multi-objective optimization algorithm; The voltage regulation parameter is structured and converted, and the voltage regulation instruction is dynamically constructed; the voltage regulation instruction is safely packaged, and the voltage regulation instruction is transmitted to the voltage regulator through an encrypted channel.

7. The switch cabinet monitoring method based on Internet of Things technology according to claim 6, characterized in that: The voltage regulator reduces the line output voltage based on the voltage regulation instruction to suppress the partial discharge, and the specific steps are as follows, The integrity and legality of the voltage regulation instruction are checked to obtain a checked voltage regulation instruction; The checked voltage regulation instruction is decoded, the control parameter is extracted, the control parameter is converted into a corresponding PWM duty cycle value, and a pulse width modulation signal is generated in combination with the PWM generator hardware of the microcontroller; The conduction and turn-off of the power semiconductor switching device are controlled through the pulse width modulation signal driving circuit to adjust the output voltage of the current conversion circuit and monitor the voltage value in real time; The deviation between the monitored voltage value and the target voltage value is calculated through the inner loop closed loop control algorithm, the PWM signal duty cycle is corrected in real time through proportional-integral adjustment to obtain a stable voltage, the internal electric field strength of the switch cabinet is reduced through the stable voltage to suppress the partial discharge, and the voltage regulation process is recorded. 8.The switch cabinet monitoring method based on Internet of Things technology according to claim 1, characterized in that: The switch cabinet health status report is generated, and the specific steps are as follows, The multi-modal time series data set, the predictive instruction set, the fault warning event identifier and the voltage regulation process are integrated to obtain a heterogeneous data set; Based on the heterogeneous data set, the comprehensive insulation state index of the current switch cabinet is calculated through an evaluation algorithm, and the equipment health status category is obtained according to the insulation state index and the risk level in the predictive instruction set; The comprehensive insulation state index and the equipment health status category are packaged to obtain the switch cabinet health status report. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The computer program is executed by the processor to realize the steps of the switch cabinet monitoring method based on the Internet of Things technology in any one of claims 1-8.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the switch cabinet monitoring method based on the Internet of Things technology in any one of claims 1-8.

Images