Power transmission and transformation equipment state evaluation method and system based on digital twinning

Abstract

The invention discloses a power transmission and transformation equipment state evaluation method and system based on digital twinning. The method comprises the following steps: acquiring data of power transmission and transformation equipment and preprocessing the data; the LSTM algorithm model predicts the running state of the equipment; the fault tree analysis module evaluates possible fault types and fault probabilities; an interaction mechanism of the LSTM algorithm model and the fault tree analysis module is established, and when the LSTM algorithm model predicts that the operation state of the equipment is abnormal, related data and a prediction result are input into the fault tree analysis module, and a fault tree is guided to locate a fault; when the fault tree analysis module finds a potential fault risk, related information is fed back to the LSTM algorithm model; and performing weighted fusion on the LSTM algorithm prediction result and the fault tree analysis result according to the confidence of the LSTM algorithm prediction result and the fault tree analysis result to obtain an evaluation result. A model interaction mechanism is established, information sharing and cooperative work of the two are realized, and the capability of capturing the abnormal state and the potential fault risk of the equipment is enhanced.

Description

Technical Field

[0001] The present invention belongs to the field of power systems, relates to the state monitoring and management technology of power system transmission and transformation equipment, and particularly relates to a method and system for evaluating the state of transmission and transformation equipment based on digital twin. Background Art

[0002] Digital twin technology is a new technology that constructs a virtual model of a physical entity through digital means. In the field of transmission and transformation equipment management, by constructing a digital twin model of the transmission and transformation equipment, various operation data of the equipment can be mapped to the virtual model in real time, realizing real-time monitoring and comprehensive perception of the equipment operation state. Using advanced data analysis algorithms and simulation technologies, it is possible to accurately predict the future operation state of the equipment, discover potential fault risks in advance, and provide a scientific basis for the health management and maintenance decision-making of the equipment.

[0003] Currently, the common architecture for predicting transmission and transformation equipment based on digital twin technology includes a data acquisition layer, a data processing layer, a digital twin model layer, a state prediction layer, and a health management layer. The data acquisition layer is responsible for collecting in real time the electrical parameters (such as voltage, current, power, etc.), mechanical parameters (such as vibration, temperature, etc.), and environmental parameters (such as humidity, air pressure, etc.) of the equipment from various sensors on the transmission and transformation equipment. The collected data is transmitted to the data processing layer through wired or wireless communication methods. In the data processing layer, operations such as data cleaning and preprocessing are performed on the transmitted data to remove noise, outliers, and fill in missing data to improve data quality. The processed data is transmitted to the digital twin model layer, which constructs a digital twin model of the transmission and transformation equipment based on the cleaned data, combined with the physical structure, operation principle, and historical fault data of the transmission and transformation equipment, to realize the virtual mapping of the equipment operation state. The state prediction layer inputs the real-time data into the digital twin model, uses time series prediction algorithms to predict the operation state of the equipment for a period of time in the future, and combines the fault tree analysis method to evaluate the possible fault types and fault probabilities. The health management layer automatically generates a health status report of the equipment according to the state prediction results, conducts a hierarchical evaluation of the equipment health status, and provides maintenance suggestions and decision support for the operation and maintenance personnel when potential fault risks occur in the equipment. Data interaction is carried out between each layer through standard data interfaces and communication protocols to ensure smooth data transmission and coordinated operation of the system.

[0004] However, the current architecture has a problem of poor coordination between the time series prediction algorithm and the fault tree analysis method. The time series prediction algorithm (LSTM algorithm) focuses on capturing patterns from the time series of device operation data to predict future states, while the fault tree analysis method focuses on analyzing fault causes and probabilities based on logical relationships. There are differences between the two in terms of data processing logic and application scenario focus. In practical applications, it may occur that the LSTM predicts an abnormal device operation state, but the fault tree analysis cannot quickly locate the matching fault mode and cause, or the fault tree analysis identifies potential fault risks, but the LSTM model fails to reflect the relevant trends in the prediction, resulting in inconsistent state prediction and fault risk assessment results and affecting the accurate judgment of the device state. Summary of the Invention

[0005] Object of the Invention: To solve the problem of poor coordination between the time series prediction algorithm and the fault tree analysis method, a power transmission and transformation equipment state evaluation method and system based on digital twin are provided.

[0006] Technical Solution: To achieve the above object, the present invention provides a power transmission and transformation equipment state evaluation method based on digital twin, including the following steps:

[0007] S1: Collect power transmission and transformation equipment data and perform preprocessing;

[0008] S2: Input the preprocessed data into the LSTM algorithm model and the fault tree analysis module respectively;

[0009] Based on the relationship between the operation state learned from the device historical operation data and various parameters, the LSTM algorithm model predicts the operation state of the device for a period of time in the future, and outputs the change trends of each parameter and possible abnormal situations;

[0010] The fault tree analysis module combines the historical operation data and fault data of the device, determines the top event and bottom event of the fault tree by using the fault tree analysis method, and determines the probability of occurrence of each event in combination with expert experience and historical data. Calculate the probability of device failure through the logical relationship of the fault tree, analyze the influence degree of various fault causes on device failure, and evaluate possible fault types and fault probabilities;

[0011] S3: Establish an interaction mechanism between the LSTM algorithm model and the fault tree analysis module. When the LSTM algorithm model predicts an abnormal device operation state, input the relevant data and prediction results into the fault tree analysis module to guide the fault tree to quickly locate possible faults; when the fault tree analysis module discovers potential fault risks, feedback the relevant information to the LSTM algorithm model to help adjust the prediction direction;

[0012] S4: According to the confidence levels of the prediction results of the LSTM algorithm and the fault tree analysis results, perform weighted fusion on the two to obtain the evaluation results.

[0013] Further, the data collected in step S1 includes electrical parameters, mechanical parameters, and environmental parameters.

[0014] Further, the preprocessing in step S1 includes preliminary screening, outlier detection, noise removal, missing data filling, and constructing a unified data feature framework.

[0015] Further, in the preprocessing of step S1, constructing a unified data feature framework includes:

[0016] In the process of constructing electrical parameter features, in addition to using the time series data of each electrical parameter as the basic feature, analyze the correlation between the parameter and each fault event in the fault tree analysis;

[0017] Introduce the parameter - event correlation matrix, indicating the correlation degree between the i-th electrical parameter and the j-th fault event, calculated based on the frequency and intensity of electrical parameter anomalies and fault event occurrences in historical data, and the formula is

[0018]

[0019] where, represents the probability that the j-th fault event occurs under the condition that the i-th electrical parameter exceeds the threshold condition, represents the probability of occurrence of the j-th fault event in the overall sample;

[0020] Construct correlation features for each electrical parameter at time point t :

[0021]

[0022] where, represents the probability of occurrence of the j-th fault event at time point t;

[0023] In the process of constructing mechanical parameter and environmental parameter features, use the time series data of the parameter as the basic feature;

[0024] Use the short-time Fourier transform to convert the vibration signal from the time domain to the frequency domain; assume that the vibration signal of the parameter signal at time t is s(t), and its frequency domain transformation result is

[0025]

[0026] where, S(f) is the amplitude at frequency f;

[0027] In the frequency range within, the extracted features represents the signal energy within the frequency band, and the calculation formula is

[0028] .

[0029] Furthermore, the LSTM algorithm model in step S2 includes an input gate, a forget gate, and an output gate;

[0030] The calculation formula of the forget gate is:

[0031]

[0032] where f t represents the output of the forget gate, σ is the Sigmoid activation function, W f is the weight matrix of the forget gate, b f is the bias term of the forget gate, is the hidden state at the previous time step, and x t is the input at the current time step;

[0033] The calculation formula of the input gate is:

[0034]

[0035] where represents the output of the input gate, is the weight matrix of the input gate, represents the bias term of the input gate;

[0036] The candidate value of the new information is represented by the candidate state vector and its calculation formula is:

[0037]

[0038] where represents the weight matrix of the candidate state vector, represents the bias term of the candidate state vector;

[0039] The update formula of the cell state at the current time step is:

[0040]

[0041] where ⊙ represents the Hadamard product of element-by-element multiplication;

[0042] The calculation formula of the output gate is:

[0043]

[0044] where represents the output of the output gate, represents the weight matrix of the output gate, represents the bias term of the output gate;

[0045] The update formula for the hidden state is:

[0046] .

[0047] Furthermore, the operation of the fault tree analysis module in step S2 includes:

[0048] Let the top event be E, corresponding to a key fault of the device; let the set of bottom events be , representing each basic fault event that may cause the top event to occur; according to the logical relationship of the fault tree, if there is an "AND gate" logic between the top event and the bottom event, the probability of the top event occurring is:

[0049]

[0050] If there is an "OR gate" logic between the top event and the bottom event, the probability of the top event occurring is:

[0051]

[0052] Among them, P(E) represents the occurrence probability of the top event, represents the occurrence probability of the i-th bottom event.

[0053] Furthermore, when the LSTM algorithm model predicts that the device operating state is abnormal in step S3, at time point t, let the set of abnormal parameters predicted by the LSTM model be , where represents the observed value of the i-th parameter at time point t; for each abnormal parameter, the LSTM model gives the abnormal degree information, let the abnormal degree be , where represents the abnormal degree of the i-th parameter, and the calculation formula is

[0054]

[0055] Among them, represents the mean value of the parameter , represents the standard deviation of the parameter ;

[0056] After receiving the data, the fault tree analysis module locates the fault mode in the fault tree logical structure it has constructed according to the parameter abnormal information; let the top event of the fault tree be E, and its corresponding set of bottom events be , representing each basic event that may cause the device failure; define a mapping function , representing the set of basic events that each abnormal parameter may trigger; for an abnormal parameter , if the abnormality degree of the parameter exceeds the set threshold θ, the basic event associated with the parameter is regarded as a possible failure event, and its probability calculation formula is:

[0057]

[0058] The fault tree analysis module transfers the occurrence probabilities of all relevant basic events to the top event, and obtains the final probability of the top event through logical operations; if the final probability exceeds the set risk threshold, it is judged that the device has potential failure risks.

[0059] Furthermore, when the fault tree analysis module discovers potential failure risks in step S3, it feeds back the key basic events and their related parameter information that cause the failure risks to the LSTM algorithm model; let the set of basic events fed back be , and the corresponding parameter set is ; after receiving the feedback information, the LSTM algorithm model adjusts the weight parameters inside the model;

[0060] Let the hidden state of the LSTM algorithm model be h t , and its update formula is adjusted under the guidance of the feedback information as:

[0061]

[0062] Among them, represents the output of the output gate, α is an adjustment coefficient, represents the parameter weight in the model.

[0063] Furthermore, step S4 specifically includes:

[0064] Let the prediction result of the LSTM algorithm model be , and the result of the fault tree analysis be , and The confidence levels of and are respectively

[0065]

[0066]

[0067] Among them, represents the accuracy rate of the LSTM method for positive samples, represents the accuracy rate of the FTA method for positive samples, represents the recall rate of the LSTM method for positive samples, Indicates the recall rate of the FTA method for positive samples;

[0068] Based on the confidence level, a dynamic weight mechanism is introduced; let the operating time of the device be T, the cumulative number of faults be N, and the fault risk level at the current time point be R; define the dynamic weight and The calculation formula is:

[0069]

[0070]

[0071] The parameters β, γ, δ are adjustment coefficients, which are used to control the sensitivity of the weight to risk, operating time, and the number of faults respectively;

[0072] The final comprehensive evaluation result is the weighted sum of two parts, and the formula is:

[0073]

[0074] Among them, y represents the final result of the device status prediction and fault risk assessment.

[0075] The present invention also provides a power transmission and transformation equipment status evaluation system based on digital twin, including:

[0076] A data acquisition and preprocessing module, which is used to collect power transmission and transformation equipment data and perform preprocessing;

[0077] An LSTM algorithm module, which is used to learn the relationship between the operating state and various parameters according to the historical operating data of the device through the LSTM algorithm model, predict the operating state of the device in a future period of time, and output the change trend of each parameter and possible abnormal conditions;

[0078] A fault tree analysis module, which is used to determine the top event and bottom event of the fault tree by using the fault tree analysis method, and combine expert experience and historical data to determine the probability of each event occurring, calculate the probability of equipment failure through the logical relationship of the fault tree, analyze the influence degree of various fault causes on equipment failure, and evaluate the possible fault types and fault probabilities;

[0079] An interaction module, when the LSTM algorithm model predicts that the operating state of the device is abnormal, inputs the relevant data and prediction results into the fault tree analysis module to guide the fault tree to quickly locate possible faults; when the fault tree analysis module discovers potential fault risks, it feeds back the relevant information to the LSTM algorithm model to help adjust the prediction direction;

[0080] A comprehensive evaluation module, which is used to perform weighted fusion on the prediction results of the LSTM algorithm and the confidence of the fault tree analysis results to obtain the evaluation result.

[0081] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0082] Improve prediction accuracy: By constructing a unified data feature representation framework, the LSTM algorithm and the fault tree analysis method are analyzed based on similar data understanding, reducing prediction biases caused by differences in data processing logic. Establish a model interaction mechanism to achieve information sharing and collaborative work between the two, enhancing the ability to capture abnormal device states and potential fault risks. Design a comprehensive evaluation algorithm to weightedly fuse the results of the two, giving full play to the advantages of the two methods, improving the accuracy of device state prediction and fault risk assessment, and avoiding misjudgments caused by inconsistent prediction results.

[0083] Improve operation and maintenance efficiency: Accurate device state prediction and fault risk assessment can help operation and maintenance personnel promptly discover potential problems of the device, formulate maintenance plans in advance, and reasonably arrange maintenance resources. Reduce unnecessary device inspections and maintenance times, lower operation and maintenance costs, improve operation and maintenance efficiency, and ensure the stable operation of the power system.

[0084] Enhance system reliability: The timely and accurate early warning mechanism enables operation and maintenance personnel to quickly respond when the device shows abnormalities or faults, take effective measures to handle them, avoid the further expansion of device faults, reduce the occurrence of power outages, improve the reliability of power transmission and transformation equipment, and ensure the safe and stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 is the system framework diagram of the present invention;

[0086] Figure 2 is the interaction schematic diagram of the interaction module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0087] The present invention will be further clarified below with reference to the drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. After reading the present invention, various equivalent modifications made by those skilled in the art fall within the scope defined by the appended claims of this application.

[0088] Embodiment 1:

[0089] As Figure 1 shown, this embodiment provides a method for evaluating the state of power transmission and transformation equipment based on digital twins, including the following steps:

[0090] S1: Collect power transmission and transformation equipment data and perform preprocessing:

[0091] The collected data includes electrical parameters, mechanical parameters, and environmental parameters; electrical parameters include voltage, current, frequency, phase angle, etc., mechanical parameters include vibration, displacement, rotational speed, temperature, etc., and environmental parameters include humidity, air pressure, wind speed, air temperature, etc.

[0092] Preprocessing includes preliminary screening, outlier detection, noise removal, missing data filling, and constructing a unified data feature framework;

[0093] Preliminary screening: Conduct preliminary screening on the collected data to remove obviously incorrect or unreasonable data. The screening criteria are based on physical constraints and the normal operating range of equipment parameters; for example, for voltage parameters, if the voltage value at a certain moment is much higher than the rated voltage of the equipment or significantly lower than the normal operating voltage range, it is determined that the data is obviously incorrect and is excluded.

[0094] Outlier detection includes: Using a statistics-based outlier detection method to identify and remove outlier data; determining the judgment range of outlier data based on the mean and standard deviation of the data; setting the data sample set as

[0095]

[0096] The sample mean is:

[0097]

[0098] The sample standard deviation is

[0099]

[0100] Then the range for determining outlier data is defined as:

[0101]

[0102] where k is the coefficient for outlier determination, usually with a value range between 2 and 3. This method has a strong ability to identify data in the dataset that significantly deviates from the average value;

[0103] Noise removal includes: Using the DBSCAN algorithm based on density-based spatial clustering of applications to remove noise points; the DBSCAN algorithm identifies data clusters with high density in the dataset by setting the radius parameter ϵ and the density threshold MinPts, and treats the data in isolated points or sparse regions as noise and eliminates them; for any data point p in the dataset, if the number of points contained in the neighborhood centered on p with a radius of ϵ is not less than MinPts, then the point p is marked as a core point, and all points that are directly or indirectly closely connected to the core point belong to the same cluster, and the unclassified points are regarded as noise points and eliminated;

[0104] Missing data filling includes: using the K-nearest neighbor interpolation algorithm based on machine learning for completion; setting the position of a missing value in a certain parameter of the dataset as x i , using the K-nearest neighbor algorithm to find x i The K sample points that are most similar in the feature space:

[0105]

[0106] Estimate the missing value by the method of weighted average, and the specific calculation formula is:

[0107]

[0108] Among them, w j is the reciprocal of the distance between the j-th sample point and x i , that is:

[0109]

[0110] Among them, d(x i ,x j ) represents the distance between the sample x i and x j , and the Euclidean distance is used for calculation:

[0111]

[0112] Among them, M is the feature dimension.

[0113] Constructing a unified data feature framework includes:

[0114] In the process of constructing electrical parameter features, in addition to using the time series data of each electrical parameter as the basic features, analyze the correlation between the parameters and each fault event in the fault tree analysis; set the observed value of the i-th electrical parameter at time point t as , and its historical observed data sequence is , and this sequence is used for trend feature extraction in the time series model;

[0115] At the same time, in order to further extract the features related to the fault event, introduce a parameter-event correlation matrix, where, represents the correlation degree between the i-th electrical parameter and the j-th fault event, and is calculated based on the frequency and intensity of the electrical parameter anomaly and the occurrence of the fault event in the historical data. The formula is

[0116]

[0117] Among them, represents the probability that the j-th fault event occurs under the condition that the i-th electrical parameter exceeds the threshold , Denote the occurrence probability of the j-th fault event in the overall sample; correlation degree The larger the value, the more likely the abnormality of the electrical parameter is to trigger the corresponding fault event;

[0118] Based on this correlation matrix, construct the correlation feature for each electrical parameter at time point t :

[0119]

[0120] Among them, Denote the occurrence probability of the j-th fault event at time point t; this probability is obtained according to the event deduction result in the fault tree analysis module; finally, the parameter The time series feature of and its correlation feature Are input into the model together to improve the accuracy of equipment status prediction and fault risk assessment;

[0121] In the process of constructing the features of mechanical parameters and environmental parameters, first use the time series data of the parameters as the basic features; let the observed value of the i-th parameter at time point t be And its historical observed data sequence is This sequence is used to extract the time trend feature;

[0122] On this basis, in order to further extract the correlation features between the parameters and specific fault events, it is necessary to combine the sensitivity of these parameters to relevant fault events in the fault tree analysis. Suppose a certain mechanical fault is associated with a specific frequency range of the vibration signal, and use the short-time Fourier transform to convert the vibration signal from the time domain to the frequency domain; let the vibration signal of the parameter signal at time t be s(t), and its frequency domain transformation result is

[0123]

[0124] Among them, S(f) is the amplitude at frequency f, indicating the signal strength at this frequency;

[0125] In the specific frequency range Extract the feature Indicates the signal energy within the frequency band, and the calculation formula is

[0126]

[0127] This feature is used to characterize the vibration characteristics related to fault events in mechanical parameters or environmental parameters. Finally, input the time series feature and the energy feature within the sensitive frequency range into the model at the same time to improve the accuracy of equipment status prediction and fault risk assessment.

[0128] S2: Input the preprocessed data into the LSTM algorithm model and the fault tree analysis module respectively;

[0129] The LSTM algorithm model learns the relationship between the operating state and various parameters from the historical operating data of the device, predicts the operating state of the device in the future for a period of time, and outputs the change trends of each parameter and possible abnormal situations;

[0130] The fault tree analysis module combines the historical operating data and fault data of the device, determines the top event and bottom event of the fault tree using the fault tree analysis method, and determines the probability of each event occurring by combining expert experience and historical data. It calculates the probability of device failure through the logical relationship of the fault tree, analyzes the influence degree of various fault causes on device failure, and evaluates the possible fault types and fault probabilities;

[0131] The LSTM algorithm model includes an input gate, a forget gate, and an output gate;

[0132] The calculation formula of the forget gate is:

[0133]

[0134] where, f t represents the output of the forget gate, σ is the Sigmoid activation function, W f is the weight matrix of the forget gate, b f is the bias term of the forget gate, is the hidden state of the previous time step, and x t is the input of the current time step;

[0135] The calculation formula of the input gate is:

[0136]

[0137] where, represents the output of the input gate, is the weight matrix of the input gate, represents the bias term of the input gate;

[0138] The candidate value of the new information is represented by the candidate state vector and its calculation formula is:

[0139]

[0140] where, represents the weight matrix of the candidate state vector, represents the bias term of the candidate state vector;

[0141] The update formula of the cell state at the current time step is:

[0142]

[0143] Among them, ⊙ represents the Hadamard product of element-by-element multiplication;

[0144] The calculation formula of the output gate is:

[0145]

[0146] Among them, represents the output of the output gate, represents the weight matrix of the output gate, represents the bias term of the output gate;

[0147] The update formula of the hidden state is:

[0148]

[0149] The operation of the fault tree analysis module includes:

[0150] Let the top event be E, corresponding to a key fault of the device; let the set of bottom events be , representing each basic fault event that may cause the top event to occur; according to the logical relationship of the fault tree, if there is an "AND gate" logic between the top event and the bottom event, the probability of the top event occurring is:

[0151]

[0152] If there is an "OR gate" logic between the top event and the bottom event, the probability of the top event occurring is:

[0153]

[0154] Among them, P(E) represents the occurrence probability of the top event, represents the occurrence probability of the i-th bottom event.

[0155] S3: As Figure 2 shown, establish an interaction mechanism between the LSTM algorithm model and the fault tree analysis module. When the LSTM algorithm model predicts that the device operation state is abnormal, input the relevant data and prediction results into the fault tree analysis module to guide the fault tree to quickly locate possible faults; when the fault tree analysis module discovers potential fault risks, feedback the relevant information to the LSTM algorithm model to help it adjust the prediction direction and enhance the ability to capture specific fault trends;

[0156] When the LSTM algorithm model predicts that the device operation state is abnormal, at time point t, let the set of abnormal parameters predicted by the LSTM model be , among which represents the observed value of the i-th parameter at time point t; for each abnormal parameter, the LSTM model gives the abnormal degree information, let the abnormal degree be , among which Represents the degree of abnormality of the i-th parameter, which can be measured by the amplitude of the parameter deviating from the normal range, and its calculation formula is

[0157]

[0158] Among them, Represents the mean value of the parameter , Represents the parameter The standard deviation of; the larger the value of the degree of abnormality, the greater the amplitude of the parameter deviating from the normal range;

[0159] After receiving the data, the fault tree analysis module quickly locates the possible fault modes in the fault tree logical structure it has constructed according to the parameter abnormality information; let the top event of the fault tree be E, and its corresponding set of bottom events be , indicating the basic events that may cause equipment failures; to quickly match the correlation between abnormal parameters and fault events, a mapping function is defined , representing the set of bottom events that each abnormal parameter may trigger; for the abnormal parameter , if the degree of abnormality of the parameter exceeds the set threshold θ, the bottom event associated with the parameter is regarded as a possible fault event, and its probability calculation formula is:

[0160]

[0161] The fault tree analysis module transmits the occurrence probabilities of all relevant bottom events to the top event, and obtains the final probability of the top event through logical operations; if the final probability exceeds the set risk threshold, it is determined that the equipment has potential fault risks.

[0162] When the fault tree analysis module discovers potential fault risks, it feeds back the key bottom events and their related parameter information that cause the fault risks to the LSTM algorithm model; let the set of bottom events fed back be , and the corresponding parameter set be ; After receiving the feedback information, the LSTM algorithm model adjusts the weight parameters inside the model according to the historical change trends of these parameters to enhance the attention to specific fault trends;

[0163] Let the hidden state of the LSTM algorithm model be h t , and its update formula is adjusted under the guidance of the feedback information to:

[0164]

[0165] Among them, Represents the output of the output gate, α is the adjustment coefficient, Represents the parameter The weights in the model; by introducing the feedback parameter and its weight, the LSTM algorithm model can focus on the change trend of potential risk parameters, further improving the prediction accuracy of specific fault trends.

[0166] S4: According to the confidence levels of the LSTM algorithm prediction results and the fault tree analysis results, perform weighted fusion on the two to obtain the evaluation result;

[0167] Let the prediction result of the LSTM algorithm model be , and the result of the fault tree analysis be , and their respective confidence levels be and , and the confidence levels are calculated based on the accuracy and recall rates of each method in historical predictions and evaluations. The calculation formula is:

[0168]

[0169]

[0170] Among them, represents the accuracy of the LSTM method for positive samples, represents the accuracy of the FTA method for positive samples, represents the recall rate of the LSTM method for positive samples, represents the recall rate of the FTA method for positive samples; the harmonic mean of the two is used as the confidence level value to measure the reliability of each method;

[0171] Based on the confidence level, introduce a dynamic weight mechanism, and the setting of the weight is related to the operating state of the device; let the operating time of the device be T, the cumulative number of faults be N, and the fault risk level at the current time point be R; define the dynamic weights and The calculation formulas are:

[0172]

[0173]

[0174] The parameters β, γ, δ are adjustment coefficients, which are used to control the sensitivity of the weight to risk, operating time, and the number of faults respectively; as the fault risk R increases, rises, falls; when the device has a short operating time and a small cumulative number of faults, wLSTM is larger;

[0175] The final comprehensive evaluation result is obtained by weighted summation of two parts, and the formula is:

[0176]

[0177] Among them, y represents the final result of equipment status prediction and fault risk assessment, which can lean towards the prediction result of the LSTM algorithm when the equipment is operating stably, and focus more on the fault tree analysis result when the equipment is approaching the high-fault period, so as to obtain more accurate and consistent prediction and evaluation conclusions.

[0178] S5: According to the equipment status prediction and fault risk assessment results after comprehensive evaluation, the system automatically generates an equipment status prediction report and a fault risk assessment report, which detail information such as the predicted operating status of the equipment, possible fault types, and fault probabilities; if the evaluation results show that the equipment is in an abnormal or faulty state, the system issues a warning signal in a timely manner.

[0179] Embodiment 2:

[0180] Based on the evaluation method provided in Embodiment 1, this embodiment provides a digital-twin-based power transmission and transformation equipment status evaluation system, including:

[0181] A data acquisition and preprocessing module, which is used to collect power transmission and transformation equipment data and perform preprocessing;

[0182] An LSTM algorithm module, which is used to predict the operating status of the equipment in the next period of time through the relationship between the operating status learned by the LSTM algorithm model from the equipment historical operation data and various parameters, and output the change trends of each parameter and possible abnormal situations;

[0183] A fault tree analysis module, which is used to determine the top event and bottom event of the fault tree by using the fault tree analysis method, and determine the probability of each event occurring in combination with expert experience and historical data, calculate the probability of equipment failure through the logical relationship of the fault tree, analyze the influence degree of various fault causes on equipment failure, and evaluate possible fault types and fault probabilities;

[0184] An interaction module, when the LSTM algorithm model predicts that the equipment operating status is abnormal, it inputs relevant data and prediction results into the fault tree analysis module to guide the fault tree to quickly locate possible faults; when the fault tree analysis module discovers potential fault risks, it feeds back relevant information to the LSTM algorithm model to help it adjust the prediction direction and enhance the ability to capture specific fault trends;

[0185] A comprehensive evaluation module, which is used to perform weighted fusion on the two according to the confidence levels of the LSTM algorithm prediction result and the fault tree analysis result to obtain the evaluation result.

[0186] Embodiment 3:

[0187] In order to verify the effectiveness of the solution of the present invention, large-scale experimental tests were carried out in this embodiment. The experimental data was collected from the actual operation data of a power transmission and transformation station in a certain area, and different working conditions were simulated to evaluate the prediction accuracy and fault warning ability of the system.

[0188] I. Data collection situation

[0189] Collection equipment: transformers, circuit breakers, power cables, lightning arresters, instrument transformers;

[0190] Monitoring parameters:

[0191] Electrical parameters: voltage (kV), current (A), power factor, frequency (Hz), harmonic content of power quality;

[0192] Mechanical parameters: vibration frequency (Hz), equipment temperature (°C), equipment noise (dB), oil level change (mm);

[0193] Environmental parameters: humidity (%RH), air temperature (°C), wind speed (m / s), air pressure (hPa);

[0194] Collection period: once every 1 minute, continuous operation for 180 days;

[0195] Data storage capacity: over 15 TB;

[0196] II. State prediction experiment

[0197] The LSTM model is used for time series prediction to evaluate the prediction error, as shown in Table 1 specifically.

[0198] Table 1

[0199]

[0200] III. Fault warning accuracy

[0201] By comparing the system warning results with the actually occurred fault events, the warning accuracy is calculated.

[0202] Total number of equipment faults: 50 times;

[0203] Number of times the system successfully warned: 46 times;

[0204] Warning accuracy: 92%;

[0205] False alarm rate: 3.2%;

[0206] Missed alarm rate: 4.8%;

[0207] The experimental data and equipment test results show that the solution of the present invention can effectively improve the accuracy of equipment condition monitoring, can identify potential fault risks in advance, improve the reliability of power transmission and transformation equipment, reduce the operation and maintenance costs, and provide important technical support for smart grid management.

Claims

1. A method for evaluating the state of power transmission and transformation equipment based on digital twins, characterized in that: The steps include: S1: Collect data from power transmission and transformation equipment and perform preprocessing; S2: Input the preprocessed data into the LSTM algorithm model and the fault tree analysis module respectively; The LSTM algorithm model predicts the operating status of the equipment based on the relationship between the operating status and various parameters learned from the historical operating data of the equipment, and outputs the changing trend of each parameter and possible abnormal situations; The fault tree analysis module combines the historical operation data and fault data of the equipment, uses the fault tree analysis method to determine the top event and bottom event of the fault tree, and combines expert experience with historical data to determine the probability of each event. It calculates the probability of equipment failure through the logical relationship of the fault tree, analyzes the impact of various fault causes on equipment failure, and evaluates possible fault types and fault probabilities; S3: Establish an interactive mechanism between the LSTM algorithm model and the fault tree analysis module. When the LSTM algorithm model predicts that the equipment is in an abnormal operating state, the relevant data and prediction results are input into the fault tree analysis module to guide the fault tree to locate the fault. When the fault tree analysis module finds a potential fault risk, it feeds back the relevant information to the LSTM algorithm model to help adjust the prediction direction. S4: Based on the confidence of the LSTM algorithm prediction results and the fault tree analysis results, the two are weightedly fused to obtain the evaluation results.

2. According to claim 1, a method for evaluating the state of power transmission and transformation equipment based on digital twins is characterized in that: The data collected in step S1 include electrical parameters, mechanical parameters and environmental parameters.

3. According to claim 2, a method for evaluating the state of power transmission and transformation equipment based on digital twins is characterized in that: The preprocessing in step S1 includes preliminary screening, outlier detection, noise removal, missing data filling, and building a unified data feature framework.

4. According to claim 3, a method for evaluating the state of power transmission and transformation equipment based on digital twins is characterized in that: In the preprocessing of step S1, building a unified data feature framework includes: In the process of constructing electrical parameter features, in addition to using the time series data of each electrical parameter as the basic feature, the correlation between the parameters and each fault event in the fault tree analysis is analyzed; Introducing the parameter-event correlation matrix, It represents the correlation between the i-th electrical parameter and the j-th fault event, which is calculated based on the frequency and intensity of electrical parameter anomalies and fault events in historical data. The formula is: ； in, Indicates that the i-th electrical parameter exceeds the threshold Under the condition, the probability of the jth fault event occurring is: represents the probability of occurrence of the jth fault event in the overall sample; Construct an associated feature for each electrical parameter at time point t : ； in, represents the probability of occurrence of the jth fault event at time point t; In the process of constructing the features of mechanical parameters and environmental parameters, the time series data of the parameters are used as the basic features; The vibration signal is converted from the time domain to the frequency domain using short-time Fourier transform. Assume that the vibration signal of the parameter signal at time t is s(t), and the frequency domain transformation result is ； Where S(f) is the amplitude at frequency f; In the frequency range The extracted features Represents the signal energy within the frequency band, and the calculation formula is: 。 5. The method for evaluating the state of power transmission and transformation equipment based on digital twin according to claim 1, characterized in that: The LSTM algorithm model in step S2 includes an input gate, a forget gate and an output gate; The calculation formula of the forget gate is: ； Among them, f t represents the output of the forget gate, σ is the Sigmoid activation function, W f is the weight matrix of the forget gate, b f is the bias term of the forget gate, is the hidden state of the previous time step, x t is the input of the current time step; The calculation formula of the input gate is: ； in, Represented as the input gate output, is the input gate weight matrix, represents the input gate bias term; The candidate value of the new information is given by the candidate state vector It means that the calculation formula is: ； in, represents the weight matrix of the candidate state vector, Represents the bias term of the candidate state vector; The update formula of the cell state at the current time step is: ； Where ⊙ represents the element-wise Hadamard product; The calculation formula of the output gate is: ； in, represents the output of the output gate, represents the weight matrix of the output gate, Represents the bias term of the output gate; The update formula of the hidden state is: 。 6. A method for evaluating the state of power transmission and transformation equipment based on digital twins according to claim 4, characterized in that: The operation of the fault tree analysis module in step S2 includes: Let the top event be E, corresponding to a critical failure of the equipment; let the bottom event set be , represents each basic fault event that may cause the top event to occur; according to the logical relationship of the fault tree, if there is an "AND gate" logic between the top event and the bottom event, the probability of the top event occurring is: ； If there is an "OR gate" logic between the top event and the bottom event, the probability of the top event occurring is: ； Among them, P(E) represents the probability of occurrence of the top event, represents the probability of occurrence of the i-th bottom event.

7. The method for evaluating the state of power transmission and transformation equipment based on digital twin according to claim 6 is characterized in that: In step S3, when the LSTM algorithm model predicts that the equipment operation status is abnormal, at time point t, the abnormal parameter set predicted by the LSTM model is ,in represents the observed value of the i-th parameter at time point t; for each abnormal parameter, the LSTM model gives the abnormal degree information, assuming the abnormal degree is ,in Indicates the abnormality of the i-th parameter, and the calculation formula is ； in, Representation parameters The mean of Representation parameters The standard deviation of After receiving the data, the fault tree analysis module locates the fault mode in the fault tree logic structure it has built based on the parameter abnormality information; let the top event of the fault tree be E, and its corresponding bottom event set be , represents the basic events that may cause equipment failure; define a mapping function , represents the bottom event set that each abnormal parameter may cause; for abnormal parameters , if the abnormal degree of the parameter exceeds the set threshold θ, the bottom event associated with the parameter is regarded as a possible fault event, and its probability calculation formula is: ； The fault tree analysis module transfers the occurrence probabilities of all related bottom events to the top event, and obtains the final probability of the top event through logical operations; if the final probability exceeds the set risk threshold, it is judged that the equipment has a potential failure risk.

8. The method for evaluating the state of power transmission and transformation equipment based on digital twin according to claim 7 is characterized in that: In step S3, when the fault tree analysis module finds a potential fault risk, the key bottom events and their related parameter information that cause the fault risk are fed back to the LSTM algorithm model; the bottom event set fed back is , the corresponding parameter set is ;After receiving feedback information, the LSTM algorithm model adjusts the weight parameters within the model; Assume the hidden state of the LSTM algorithm model is h t , its update formula is adjusted to: ； in, represents the output of the output gate, α is the adjustment coefficient, Representation parameters The weights in the model.

9. The method for evaluating the state of power transmission and transformation equipment based on digital twin according to claim 7, characterized in that: The step S4 specifically includes: Assume that the prediction result of the LSTM algorithm model is The result of the fault tree analysis is , and The confidence levels are and , the calculation formula is: ； ； in, Indicates the accuracy of the LSTM method for positive samples, It represents the accuracy of the FTA method for positive samples. Represents the recall rate of the LSTM method for positive samples, Represents the recall rate of the FTA method for positive samples; Based on the confidence level, a dynamic weight mechanism is introduced; let the running time of the equipment be T, the cumulative number of failures be N, and the failure risk level at the current time point be R; define the dynamic weight and The calculation formula is: ； ； The parameters β, γ, and δ are adjustment coefficients, which are used to control the sensitivity of the weight to risk, operating time, and number of failures, respectively; The final comprehensive evaluation result is the weighted sum of the two parts, and the formula is: ； Among them, y represents the final result of equipment status prediction and failure risk assessment.

10. A power transmission and transformation equipment status assessment system based on digital twins, characterized in that: include: Data collection and preprocessing module, used to collect and preprocess data of power transmission and transformation equipment; LSTM algorithm module, which is used to predict the operating status of the equipment based on the relationship between the operating status and various parameters learned from the historical operating data of the equipment through the LSTM algorithm model, and output the change trend of each parameter and possible abnormal conditions; Fault tree analysis module, which is used to determine the top and bottom events of the fault tree using the fault tree analysis method, and determine the probability of each event by combining expert experience and historical data. It calculates the probability of equipment failure through the logical relationship of the fault tree, analyzes the impact of various fault causes on equipment failure, and evaluates possible fault types and fault probabilities; Interaction module: When the LSTM algorithm model predicts that the equipment is in an abnormal operating state, the relevant data and prediction results are input into the fault tree analysis module to guide the fault tree to locate the fault; when the fault tree analysis module finds potential fault risks, the relevant information is fed back to the LSTM algorithm model to help adjust the prediction direction; The comprehensive evaluation module is used to perform weighted fusion of the LSTM algorithm prediction results and the fault tree analysis results according to their confidence levels to obtain the evaluation results.

Images