Power transformation and distribution room equipment operation state evaluation method and system

Abstract

The invention relates to the technical field of equipment management, in particular to a power transformation and distribution room equipment operation state evaluation method and system. The method comprises the following steps: acquiring multi-dimensional sensing data of a power transformation and distribution room and identifying space-time abnormal features of equipment to obtain space-time feature tensors of the equipment; performing equipment operation trend deduction based on the equipment spatial-temporal feature tensor to obtain equipment operation trend prediction data; according to the equipment operation trend prediction data, executing direct current screen insulation degradation federated twinborn modeling to obtain an equipment credible twinborn verification chain; compiling an optimal switching instruction set of the spare power automatic switching device according to the equipment credible twin verification chain; and obtaining real-time operation data of the power transformation and distribution room equipment, implementing fault traceability analysis, and screening high-reliability fault early warning instructions to obtain a maintenance strategy of the power transformation and distribution room equipment. The accuracy and the timeliness of identifying the operation state of the equipment in the power transformation and distribution room can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment management, and particularly to a method and system for evaluating the operating status of substation equipment. Background Art

[0002] Traditional safety management of substations mainly relies on manual inspections, which are inefficient and difficult to achieve real-time monitoring: Manual inspections are easily affected by subjective factors of inspectors, such as fatigue and negligence, resulting in safety hazards being difficult to detect in a timely manner; Currently, simple electrical parameter monitoring devices, such as conventional current and voltage sensors, are used to assist manual inspections, but the collected data is relatively single and does not involve the analysis of associated changes among them, resulting in the inability to comprehensively grasp the operating status of the substation. For example, potential faults such as partial discharge and overheating of equipment cannot be effectively detected, and environmental factors (such as temperature, humidity, smoke, etc.) cannot be comprehensively analyzed;

[0003] In addition, traditional safety management methods lack intelligent early warning and decision support, mostly for after-the-fact alarms and lack the ability of early warning: When an accident occurs, it is impossible to quickly provide effective decision support to help managers take measures in a timely manner to reduce losses. Summary of the Invention

[0004] Based on this, it is necessary for the present invention to provide a method and system for evaluating the operating status of substation equipment to solve at least one of the above technical problems.

[0005] To achieve the above object, a method for evaluating the operating status of substation equipment includes the following steps:

[0006] Step S1: Obtain multi-dimensional sensing data of the substation and construct a dynamic spatio-temporal adjacency matrix; Identify the spatio-temporal abnormal characteristics of the equipment according to the dynamic spatio-temporal adjacency matrix to obtain the spatio-temporal feature tensor of the equipment;

[0007] Step S2: Deduce the operation trend of the equipment based on the spatio-temporal feature tensor of the equipment to obtain the predicted data of the operation trend of the equipment;

[0008] Step S3: Perform DC screen insulation degradation federated twin modeling according to the predicted data of the operation trend of the equipment, and perform dynamic characteristic diffeomorphic mapping to obtain the credible twin verification chain of the equipment;

[0009] Step S4: Reconstruct the thermal stress field of the bus connection point for the credible twin verification chain of the equipment, analyze the tap-changing strategy, and obtain the inrush current control strategy; Compile the optimal switching instruction set of the backup power supply automatic switching device based on the inrush current control strategy;

[0010] Step S5: Obtain the real-time operation data of the substation equipment, and perform fault evolution prediction on the real-time operation data of the substation equipment according to the optimal switching instruction set of the backup power supply automatic switching device to generate a global operation optimization evidence chain; perform fault traceability analysis based on the global operation evidence chain, and screen high-reliability fault warning instructions to obtain the maintenance strategy of the substation equipment.

[0011] Optionally, step S1 is specifically as follows:

[0012] Step S11: Obtain the multi-dimensional sensing data of the substation, and perform data preprocessing on the multi-dimensional sensing data of the substation to obtain the multi-dimensional sensing data to be analyzed;

[0013] Step S12: Divide the multi-dimensional sensing data to be analyzed to obtain equipment sensing data and environmental sensing data, and perform equipment operation environment perception based on the equipment sensing data and environmental sensing data to obtain equipment operation environment data;

[0014] Step S13: Extract equipment correlation features and interaction frequency features based on the equipment sensing data, and construct a preliminary spatio-temporal adjacency matrix according to the equipment correlation features and interaction frequency features;

[0015] Step S14: Perform spatio-temporal interaction feature mapping and adaptive aggregation on the preliminary spatio-temporal adjacency matrix according to the equipment operation environment data to obtain a dynamic spatio-temporal adjacency matrix;

[0016] Step S15: Identify equipment spatio-temporal anomalies according to the dynamic spatio-temporal adjacency matrix, and mark the spatio-temporal anomaly features of equipment anomalies to obtain an equipment spatio-temporal feature tensor.

[0017] Optionally, step S15 is specifically as follows:

[0018] Step S151: Perform auto-encoding of equipment operation state-environment change correlation features based on the dynamic spatio-temporal adjacency matrix, extract non-linear correlation features, and set the target dimension of dimensionality reduction to 16 for feature dimension reduction to generate an equipment operation state-environment change feature encoding tensor;

[0019] Step S152: Perform local outlier factor detection based on the equipment operation state-environment change feature encoding tensor, set the number of neighborhood samples to 20 to calculate the local anomaly factor score, set the anomaly determination threshold to 1.5 to screen out anomaly feature points, and obtain spatio-temporal anomaly detection data;

[0020] Step S153: Use the spatio-temporal anomaly detection data, combine with the preset equipment operation mode library for pattern matching and deviation metric analysis, calculate the equipment state drift deviation, and set the deviation threshold to 0.2 according to the distribution of the drift deviation to extract the equipment abnormal state features to obtain the equipment spatio-temporal anomaly feature data;

[0021] Step S154: Based on the device spatio-temporal anomaly feature data, perform adaptive spatio-temporal feature mapping in combination with the dynamic spatio-temporal adjacency matrix, execute backpropagation feature mapping weight adjustment, and set the feature embedding optimization threshold to 0.01 to optimize the feature embedding representation, obtaining the device abnormal spatio-temporal feature tensor and the device normal spatio-temporal feature tensor;

[0022] Step S155: Merge the device abnormal spatio-temporal feature tensor and the device normal spatio-temporal feature tensor for spatio-temporal feature tensor merging, construct a spatio-temporal feature expression, and obtain the device spatio-temporal feature tensor.

[0023] Optionally, step S2 is specifically as follows:

[0024] Step S21: Based on the device spatio-temporal feature tensor, perform variational autoencoder latent variable decomposition, extract the potential factors of the relay protection device operation state, and execute information bottleneck constraints to generate the device latent variable feature vector;

[0025] Step S22: Perform structured Bayesian network learning on the device latent variable feature vector, construct a device causal association graph model, and execute MAP device causal path weight optimization to obtain the device causal association weight matrix;

[0026] Step S23: Based on the device causal association weight matrix, execute distributed policy gradient optimization and perform proximal policy reinforcement learning iteration to obtain the optimal causal path decision parameters;

[0027] Step S24: Obtain historical device operation data, construct a device operation state transition model in combination with the optimal causal path decision parameters; perform causal reasoning on the device operation state evolution path according to the device operation state transition model, thereby generating a device causal decision graph;

[0028] Step S25: Perform device state prediction based on the device causal decision graph and execute particle filter prediction error correction to obtain the device operation trend prediction data.

[0029] Optionally, step S3 is specifically as follows:

[0030] Step S31: Based on the device causal decision graph and the device operation trend prediction data, identify the DC panel insulation deterioration variable, and execute multi-scale wavelet decomposition to extract the deterioration feature spectrum, generating the DC panel insulation deterioration feature space;

[0031] Step S32: Perform multi-site device distributed modeling based on the DC panel insulation deterioration feature space, and execute cross-site time series feature aggregation in combination with the preset adaptive gated recurrent unit to generate the DC panel insulation deterioration federated twin model;

[0032] Step S33: According to the DC screen insulation degradation federated twin model, differential homeomorphism manifold embedding mapping is performed, and manifold Laplace regularization optimization is performed to obtain a dynamic characteristic differential homeomorphism mapping result;

[0033] Step S34: Using the results of the dynamic characteristic differential homeomorphism mapping, local geometric consistency constraints are imposed on the equipment operation state transition model, and variational Bayesian inference is performed to obtain a trusted twin verification chain for the equipment.

[0034] Optionally, step S31 is specifically:

[0035] Step S311: extract the DC panel operation trend from the equipment operation trend prediction data, and set the voltage fluctuation threshold to 2%, the current fluctuation threshold to 3%, and the abnormality judgment threshold to 5% to evaluate the DC panel operation performance and obtain the DC panel operation performance trend data;

[0036] Step S312: Screen the DC panel insulation performance degradation factors for the DC panel operating performance trend data according to the equipment causal decision map, set the causal decision factor screening threshold to 0.1, and perform contribution evaluation, select the top 5 causal decision factors in contribution ranking to construct the DC panel insulation state variable set;

[0037] Step S313: Based on the dynamic spatiotemporal adjacency matrix, a correlation analysis is performed on the set of DC screen insulation state variables, the Pearson correlation coefficient threshold is set to 0.6, and variables with a value less than the Pearson correlation coefficient threshold are eliminated, thereby extracting a subset of DC screen insulation degradation variables;

[0038] Step S314: perform empirical mode analysis on the DC panel insulation degradation variable subset, extract the intrinsic mode function, calculate the kurtosis and energy ratio, set the degradation feature threshold to 3, the energy proportion threshold to 5%, screen out the main degradation feature components, and generate the insulation degradation feature variable matrix;

[0039] Step S315: Set the decomposition level to 4, perform wavelet transform on the insulation degradation characteristic variable matrix for multi-scale time-frequency decomposition, extract the high-frequency abnormal component energy ratio greater than the high-frequency energy ratio threshold of 10%, and thus construct the DC screen insulation degradation feature space.

[0040] Optionally, the reconstruction of the thermal stress field of the busbar connection point described in step S4 is specifically as follows:

[0041] Based on the trusted twin verification chain of the device, the trusted operation status data and trusted environment parameter data of the device are obtained, and the busbar connection point status characteristics and on-load tap changer status characteristics in the trusted operation status data of the device are extracted;

[0042] Construct a thermal stress field model for the busbar connection point based on the state characteristics of the busbar connection point and the data of the trusted environment parameters;

[0043] Perform space-time coordinate transformation on the thermal stress field model of the busbar connection point, combine the data of the trusted operation state of the equipment, and set the initial scenario of the thermal stress field of the busbar connection point;

[0044] Calculate the thermal pressure field for the initial scenario of the thermal stress field of the busbar connection point through the thermodynamic model of the busbar connection point, and perform dynamic update according to the trusted twin verification chain of the equipment to obtain the stress field distribution map of the busbar connection point;

[0045] Identify the thermal stress concentration area according to the stress field distribution map of the busbar connection point to obtain the data of the thermal stress concentration area of the busbar;

[0046] Construct a non-exchange decision model based on the data of the thermal stress concentration area of the busbar and the state characteristics of the on-load tap-changer, input the initial scenario of the thermal stress field of the busbar connection point into the non-exchange decision model for non-change decision analysis, and optimize the equipment operation strategy according to the non-change decision analysis result to generate an inrush current control strategy.

[0047] Optionally, the compilation of the optimal switching instruction set for the backup power supply automatic switching device described in step S4 is specifically as follows:

[0048] Identify the power failure mode based on the data of the trusted operation state of the equipment, and establish a mapping model of equipment operation state - failure mode;

[0049] Perform fault source location analysis according to the mapping model of equipment operation state - failure mode, identify the faulty equipment and the power path, and generate preliminary fault source location data;

[0050] Obtain the historical switching data of the backup power supply automatic switching device, and perform switching mode recognition on the historical switching data of the backup power supply automatic switching device to obtain the switching mode data of the backup power supply automatic switching device;

[0051] Combine the preliminary fault source location data and the power failure mode, and perform backup power supply automatic switching device switching strategy matching on the switching mode data of the backup power supply automatic switching device to obtain the candidate switching strategies for the backup power supply automatic switching device;

[0052] Perform simulation tests on the candidate switching strategies for the backup power supply automatic switching device, evaluate the execution efficiency of the candidate switching strategies for the backup power supply automatic switching device and the power system restoration speed, and screen out the optimal switching strategy;

[0053] Perform instruction set compilation based on the optimal switching strategy to obtain the optimal switching instruction set for the backup power supply automatic switching device, and upload it to the instruction management platform of the backup power supply automatic switching device to implement instruction replacement.

[0054] Optionally, the fault evolution prediction described in step S5 is specifically as follows:

[0055] Extract real-time operation characteristics according to the real-time operation data of the substation equipment. Set the mother wavelet to the 4th order to extract the equipment operation stress characteristics, set the load fluctuation threshold to 0.05 pu to screen the equipment load characteristics, and set the equipment operation stress characteristics to 10 min to calculate the real-time operation temperature rise characteristics, so as to obtain the equipment operation stress characteristic data, equipment load characteristic data and real-time operation temperature rise characteristic data;

[0056] Based on the variable load fluctuation characteristics in the equipment load characteristic data, construct a load fluctuation viscous evolution model, perform Fourier transform to extract the load viscous component, calculate the viscous damping parameter in combination with the real-time operation temperature rise characteristic data, and set the viscous damping parameter calculation threshold to 0.02 to eliminate the invalid damping, so as to generate a load viscous characteristic matrix;

[0057] Screen historical equipment fault data based on the equipment operation mode library;

[0058] Perform dynamic evolution on the load viscous characteristic matrix, solve the fault-induced trend in combination with the equipment operation stress characteristic data and historical equipment fault data, set the viscous fault evolution threshold to 0.1, and obtain the load viscous fault evolution path;

[0059] Perform feature clustering according to the load viscous fault evolution path, divide the fault evolution stage, and perform fault matching based on the historical equipment fault data. Set the fault matching similarity threshold to identify the fault type and generate the fault evolution prediction result;

[0060] Match the switching instruction adaptively with the fault evolution prediction result according to the optimal switching instruction set of the backup power supply automatic switching device to obtain the load viscous fault switching instruction;

[0061] Based on the fault evolution prediction result, construct a viscous fault warning model, set the viscous instability threshold to 0.08, and adjust the dynamic warning strategy in combination with the equipment load characteristic data and the load viscous fault switching instruction. Use the trusted computing framework in the equipment trusted twin verification chain to construct a global operation optimization evidence chain.

[0062] The present invention obtains multi-dimensional sensing data of a variable substation and constructs a dynamic spatio-temporal adjacency matrix, making data acquisition more comprehensive and refined, capable of accurately identifying spatio-temporal abnormal characteristics of equipment, thereby enhancing the overall control ability of equipment operation status, and helping to identify potential safety hazards at an early stage. Based on the equipment spatio-temporal feature tensor, the equipment operation trend is deduced, enabling the system to predict the change trend of equipment health status, discover possible abnormal conditions in advance, and provide data support for subsequent fault warning. Combining the equipment causal decision graph and the equipment operation trend prediction data, the federal twin modeling of the DC panel insulation deterioration is realized, and the credible verification ability of the equipment health status is enhanced through dynamic characteristic diffeomorphic mapping, making the intelligent management of the variable substation change from passive response to active prediction. Further, by reconstructing the thermal stress field of the busbar connection point for the equipment credible twin verification chain and performing non-exchange decision analysis on the on-load tap changer, the system can optimize the inrush current control strategy specifically, thereby effectively reducing the impact of inrush current on the equipment, extending the service life of the equipment, and improving the operation stability at the same time. Compiling the optimal switching instruction set of the backup power supply automatic switching device based on the inrush current control strategy makes the equipment switching under fault conditions more accurate and efficient, ensuring the stable operation of the variable substation. On this basis, using the real-time operation data of the variable substation equipment and combining with the optimal switching instruction set of the backup power supply automatic switching device, the fault evolution prediction is realized, enabling the system to more accurately analyze the impact of load fluctuations on the equipment and improving the recognition accuracy of abnormal states. At the same time, according to the global operation optimization evidence chain, the fault tracing analysis is carried out, and the high-reliability fault warning instructions are screened, making the fault warning more accurate and reliable, avoiding false alarms or missed alarms, improving the scientificity and timeliness of equipment maintenance, and effectively enhancing the overall safety and management efficiency of the variable substation.

[0063] Optionally, this specification also provides a system for evaluating the operation status of variable substation equipment, which is used to execute the method for evaluating the operation status of variable substation equipment as described above. The system for evaluating the operation status of variable substation equipment includes:

[0064] A data acquisition module, configured to obtain multi-dimensional sensing data of a variable substation and construct a dynamic spatio-temporal adjacency matrix; identify spatio-temporal abnormal characteristics of equipment according to the dynamic spatio-temporal adjacency matrix to obtain an equipment spatio-temporal feature tensor;

[0065] An equipment operation trend deduction module, configured to deduce the equipment operation trend based on the equipment spatio-temporal feature tensor to obtain equipment operation trend prediction data;

[0066] A DC panel insulation performance analysis module, configured to perform federal twin modeling of DC panel insulation deterioration according to the equipment operation trend prediction data and perform dynamic characteristic diffeomorphic mapping to obtain an equipment credible twin verification chain;

[0067] An instruction compilation module, which is used to reconstruct the thermal stress field of the busbar connection points of the device trusted twin verification chain, analyze the tap-changing strategy, and obtain an inrush current control strategy; compile the optimal switching instruction set of the backup power supply automatic switching device based on the inrush current control strategy;

[0068] A fault tracing module, which is used to obtain the real-time operation data of the substation equipment, and perform fault evolution prediction on the real-time operation data of the substation equipment according to the optimal switching instruction set of the backup power supply automatic switching device, and generate a global operation optimization evidence chain; perform fault tracing analysis according to the global operation evidence chain, and screen high-reliability fault warning instructions to obtain the maintenance strategy of the substation equipment.

[0069] The operation state evaluation system of the substation equipment of the present invention can implement any operation state evaluation method of the present invention, and is used as a medium for the operation and signal transmission between each module to complete the operation state evaluation method of the substation equipment. The internal modules of the system cooperate with each other, thereby improving the scientificity and timeliness of equipment maintenance, and effectively improving the overall safety and management efficiency of the substation. Brief Description of the Drawings

[0070] By reading the detailed description of the non-restrictive embodiments with reference to the following drawings, other features, purposes and advantages of the present invention will become more obvious:

[0071] Figure 1 It is a schematic diagram of the step flow of the operation state evaluation method of the substation equipment of the present invention;

[0072] Figure 2 It is a detailed schematic diagram of the step flow of step S1 in the present invention;

[0073] The realization, functional characteristics and advantages of the purpose of the present invention will be further described with reference to the embodiments and the drawings. Detailed Embodiments

[0074] The technical method of the present invention will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative work belong to the scope of protection of the present invention.

[0075] In addition, the attached drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus their repeated description will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities may be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0076] It should be understood that although terms such as "first" and "second" may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed related items.

[0077] To achieve the above object, please refer to Figures 1 to 2 , the present invention provides a method for evaluating the operation state of substation equipment, and the method includes the following steps:

[0078] Step S1: Obtain multi-dimensional sensing data of the substation and construct a dynamic spatio-temporal adjacency matrix; identify the spatio-temporal abnormal characteristics of the equipment according to the dynamic spatio-temporal adjacency matrix to obtain the spatio-temporal feature tensor of the equipment;

[0079] In this embodiment, high-precision sensors such as current, voltage, and power factor are deployed in the substation to collect the operation parameters of the electrical equipment in the substation in real time. Temperature and humidity sensors, smoke sensors, gas sensors, etc. are installed in the substation to monitor the environmental conditions of the substation in real time. Vibration sensors, infrared thermometers and other devices installed on the substation equipment are used to monitor the mechanical vibration and temperature changes of the electrical equipment. The data collection is preprocessed by an edge computing node (MEC), and the time synchronization threshold is set to 10 ms for data time series alignment. A dynamic spatio-temporal adjacency matrix is constructed, and the dynamic time warping (DTW) is used to calculate the spatio-temporal similarity between devices, and the dynamic adjacency weight threshold is set to 0.7 for edge screening. The constructed adjacency matrix is subjected to a graph convolutional neural network (GCN) for feature extraction, low-pass filtered features are screened, and the local outlier factor (LOF) is used to calculate the outlier factor score, and the outlier feature points are screened. The outlier factor threshold is set to 1.5. Through the feature embedding method, the screened abnormal features are mapped to a high-dimensional feature space to construct a spatio-temporal feature tensor of the equipment, which is stored in the edge computing node for subsequent analysis.

[0080] Step S2: Based on the device spatio-temporal feature tensor, deduce the device operation trend to obtain the device operation trend prediction data;

[0081] In this embodiment, based on the device spatio-temporal feature tensor, a long short-term memory network (LSTM) is used for time series modeling, and the historical data window is set to 7 days for trend prediction. The LSTM network adopts a multi-scale attention mechanism, sets the attention weight threshold to 0.85, screens key feature variables, and conducts recursive prediction. During the trend deduction process, an autoregressive integrated moving average model (ARIMA, p = 3, d = 1, q = 2) is introduced for short-term fluctuation correction to ensure the stability of the prediction results. The error of the prediction results is evaluated, the mean square error (MSE) is used to calculate the error range, and the error threshold is set to 0.02 for prediction correction. The exponential smoothing method (with a smoothing coefficient of 0.6) is combined with the trend extrapolation method to adjust the long-term operation trend of the device, output the device operation trend prediction data, and provide it to the subsequent twin modeling module.

[0082] Step S3: Execute the DC panel insulation degradation federated twin modeling according to the device operation trend prediction data, and perform dynamic characteristic differential homeomorphic mapping to obtain the device trusted twin verification chain;

[0083] In this embodiment, based on the device operation trend prediction data and the device causal decision graph, federated learning (FL) is used for DC panel insulation degradation modeling. Each substation node performs data pre-training based on local edge computing, and the local iteration number is set to 5. Each node uses the federated average (FedAvg) method for global model aggregation, and the global learning rate is set to 0.01 for gradient update. The differential homeomorphic mapping (DHM) of dynamic characteristics is used to perform non-linear transformation on the device state variables, and the transformation order is set to 3 to ensure the preservation of the local topological structure. Finally, a trusted execution environment (TEE) is used for encrypted data storage, and a device trusted twin verification chain is generated in combination with blockchain smart contracts for subsequent fault prediction and strategy optimization.

[0084] Step S4: Reconstruct the thermal stress field of the bus connection point for the device trusted twin verification chain, analyze the tap-changing strategy, and obtain the inrush current control strategy; Compile the optimal switching instruction set of the backup power supply automatic switching device based on the inrush current control strategy;

[0085] In this embodiment, based on the device trusted twin verification chain, the thermal stress field of the busbar connection point is reconstructed. The finite element analysis (FEM) is used to calculate the thermal stress distribution, and the mesh division step size is set to 0.5 mm to ensure the calculation accuracy. The thermal-fluid coupling model is used to calculate the change trend of thermal stress, and the stress threshold is set to 150 MPa for abnormal point identification. For the non-switching decision of the on-load tap changer, the Bayesian optimization (BO) is used for strategy search. The number of iterations is set to 50, and the Gaussian process regression (GPR) is selected as the surrogate model to optimize the tap changer switching scheme. Based on the optimization results and combined with the historical fault data, the reinforcement learning (RL) is used to train the inrush current control strategy, and the discount factor is set to 0.95 to optimize the decision-making effect. According to the control strategy, the optimal switching instruction set of the backup power supply automatic switching device is compiled to ensure the stability and safety of load switching.

[0086] Step S5: Obtain the real-time operation data of the substation equipment, and perform fault evolution prediction on the real-time operation data of the substation equipment according to the optimal switching instruction set of the backup power supply automatic switching device to generate a global operation optimization evidence chain; perform fault traceability analysis according to the global operation evidence chain, and screen out high-reliability fault warning instructions to obtain the maintenance strategy of the substation equipment.

[0087] In this embodiment, the real-time operation data of the substation equipment is obtained through the sensors deployed in the substation, and the load fluctuation viscous solution fault evolution prediction is performed based on the optimal switching instruction set of the backup power supply automatic switching device. The viscoelastic damping model is used to calculate the load viscous damping coefficient, and the damping threshold is set to 0.03 for fault screening. The hidden Markov model (HMM, number of states N = 4) is used for trend speculation in the fault evolution prediction, and the principal component analysis (PCA, retaining the variance ratio of 95%) is combined for dimensionality reduction to improve the calculation efficiency. Based on the prediction results, the blockchain trusted computing framework is used to generate a global operation optimization evidence chain for historical data comparison and fault traceability. Combining the fault traceability analysis, high-reliability fault warning instructions and switching instructions are screened out. The support vector machine (SVM, kernel function: RBF) is used for fault classification, and the misjudgment rate threshold δth = 0.05 is set to ensure the classification accuracy. Finally, combining the switching instructions, the equipment operation trend and the fault warning information, the maintenance strategy of the substation equipment is generated, and the maintenance optimization plan is automatically executed through the smart contract.

[0088] Optionally, step S1 is specifically:

[0089] Step S11: Obtain the multi-dimensional sensing data of the substation, and perform data preprocessing on the multi-dimensional sensing data of the substation to obtain the multi-dimensional sensing data to be analyzed;

[0090] In this embodiment, high-precision current, voltage, power factor and other sensors are deployed in the substation to collect the operating parameters of the electrical equipment in the substation in real time. Temperature and humidity sensors, smoke sensors, gas sensors, etc. are installed in the substation to monitor the environmental conditions of the substation in real time. Vibration sensors, infrared thermometers and other equipment installed in the substation equipment are used to monitor the mechanical vibration and temperature changes of electrical equipment. The collected multidimensional sensor data is preprocessed. First, data cleaning is performed to eliminate missing values and abnormal data. Linear interpolation is used to fill in missing values, and the maximum interpolation offset is set to 1.5 seconds. Subsequently, each sensor data is mapped to the interval of 0 to 1 through standardization processing, and the data is normalized through z-score standardization. Finally, timing alignment is performed, and the maximum allowable time difference is set to 5ms to ensure the consistency of data timestamps. The preprocessed data forms multidimensional sensor data to be analyzed.

[0091] Step S12: performing sensor data division on the multi-dimensional sensor data to be analyzed to obtain device sensor data and environment sensor data, and performing device operating environment perception based on the device sensor data and environment sensor data to obtain device operating environment data;

[0092] In this embodiment, after the multidimensional sensor data to be analyzed is processed, the data is first divided into two categories according to the sensor type: device sensor data and environmental sensor data. Device sensor data includes parameters such as current, voltage, load, vibration, etc. that directly reflect the operating status of the equipment; environmental sensor data includes external factors such as temperature, humidity, and gas concentration that affect the operating environment of the equipment. Through correlation analysis, the Pearson correlation coefficient and the mutual information method are combined to perform correlation analysis on the device sensor data and the environmental data, and the key interaction features between the device and the environment are screened out. Combined with the environmental sensor data, the device operating environment is perceived, and the spatiotemporal correlation features are extracted through the spatiotemporal clustering algorithm to analyze the interaction mode between the device and the environment. The spatiotemporal window size is set to 30 minutes, and the environmental factors are combined with the device status to obtain the device operating environment data for subsequent modeling.

[0093] Step S13: extracting device correlation features and interaction frequency features based on device sensor data, and constructing a preliminary spatiotemporal adjacency matrix according to the device correlation features and interaction frequency features;

[0094] In this embodiment, based on the device sensing data, the autocorrelation analysis method and cross-correlation analysis are used to extract the device correlation features, with a focus on the non-linear relationships between device parameters, such as the correlation between current and load, the interaction relationship between voltage and temperature, etc. The sliding window method is used, with the window size set to 10 minutes and the window moving 1 minute each time, to extract the time series features within the window. Through this method, the regular changes during the device operation can be identified. In addition, by calculating the interaction frequency features between devices, the interaction intensity and frequency between each device and the environment are analyzed. The interaction frequency features are calculated based on the signal transmission time and frequency between devices, and time-domain and frequency-domain analysis methods are used to extract the periodic features through Fourier transform. The frequency threshold is set to 0.5 Hz to identify the device state changes of high-frequency and low-frequency interaction signals. Based on the extracted device correlation features and interaction frequency features, a preliminary spatio-temporal adjacency matrix is constructed. Each element of the adjacency matrix represents the correlation and interaction frequency between devices. Assuming there are N devices, the preliminary spatio-temporal adjacency matrix is an N×N matrix, where each element A(i,j) represents the relationship strength between device i and device j. Specifically: by calculating the Pearson correlation coefficient or mutual information value between device i and device j, the preliminary correlation part of matrix A is obtained. By calculating the frequency synchronization between device i and device j, the interaction frequency part of matrix A is obtained. The correlation matrix and the interaction frequency matrix are weighted and synthesized, with the weights of correlation and interaction frequency set to 0.7 and 0.3 respectively, to obtain the comprehensive spatio-temporal relationship between devices. Finally, through the above comprehensive features, a dynamic preliminary spatio-temporal adjacency matrix is obtained, which can reflect the correlation and interaction frequency between devices and provide a basis for subsequent spatio-temporal feature mapping and anomaly recognition.

[0095] Step S14: Perform spatio-temporal interaction feature mapping and adaptive aggregation on the preliminary spatio-temporal adjacency matrix according to the device operation environment data to obtain a dynamic spatio-temporal adjacency matrix;

[0096] In this embodiment, according to the preliminarily constructed spatio-temporal adjacency matrix, spatio-temporal interaction feature mapping is first performed. This process uses a graph convolutional neural network (GCN) for data propagation and feature transformation to effectively capture the complex relationships between devices and the environment in the graph space. During the mapping process, the adjacency matrix aggregation weight is set to 0.8 to ensure that the core correlation features between devices can be preferentially processed. Then, an adaptive aggregation algorithm is used for weighted aggregation of the adjacency matrix, and dynamic weighting is performed according to the device state and environmental factors to enhance the accuracy and flexibility of the spatio-temporal features. During the aggregation process, an aggregation algorithm based on the self-attention mechanism is used, and the self-attention weight threshold is set to 0.7 to optimize the expression ability of the adjacency matrix. Through this process, a dynamic spatio-temporal adjacency matrix is finally generated, which can reflect the real-time interaction relationship between devices and the environment.

[0097] Step S15: Identify device spatio-temporal anomalies based on the dynamic spatio-temporal adjacency matrix, mark the spatio-temporal anomaly features of the device anomalies, and obtain the device spatio-temporal feature tensor.

[0098] In this embodiment, based on the dynamic spatio-temporal adjacency matrix, a deep learning anomaly detection model (such as a deep autoencoder or LSTM-based anomaly detection) is used to identify device spatio-temporal anomalies. The time range of the training set is set to 7 days, and the hyperparameters of the model are optimized by the cross-validation method to ensure detection accuracy. The device operation is monitored in real time through spatio-temporal anomaly scores. The anomaly score threshold is set to 0.85. When the score exceeds the threshold, it can be marked as an abnormal event. When an anomaly is identified, combined with the time series features and spatial correlation features, the abnormal spatio-temporal features of the device are marked, and the root cause analysis of the abnormal event is carried out. The marking of abnormal features includes information such as the amplitude of the change in the device operation state, the moment when the abnormal point occurs, and the correlation with environmental changes. Finally, a device spatio-temporal feature tensor is generated through the marking of device spatio-temporal features, and this tensor will be stored and used for subsequent analysis, fault prediction, or other intelligent decision-making tasks.

[0099] Optionally, step S15 is specifically as follows:

[0100] Step S151: Perform auto-encoding of the correlation features between the device operation state and environmental changes based on the dynamic spatio-temporal adjacency matrix, extract non-linear correlation features, and set the target dimension for dimensionality reduction to 16 for feature dimension reduction, generating a device operation state - environmental change feature encoding tensor;

[0101] In this embodiment, based on the dynamic spatio-temporal adjacency matrix, auto-encoding of the correlation features between the device operation state and environmental changes is performed to extract the non-linear correlation features between the device and the environment. In the specific implementation process, first, the dynamic spatio-temporal adjacency matrix is input into the autoencoder model. The autoencoder model consists of an encoder and a decoder. The encoder is responsible for compressing the input high-dimensional features into low-dimensional representations, and the decoder restores the original features. To effectively extract non-linear correlation features, the ReLU activation function is used to enhance the non-linear expression of the features. Then, the target dimensionality reduction dimension is set to 16, and the high-dimensional features are compressed to 16 dimensions through the middle layer of the autoencoder. This dimensionality reduction operation can reduce redundant information while retaining the most representative features, generating a device operation state - environmental change feature encoding tensor as the input for subsequent analysis.

[0102] Step S152: Perform local outlier factor detection based on the device operation state - environmental change feature encoding tensor, set the number of neighborhood samples to 20 to calculate the local anomaly factor score, set the anomaly determination threshold to 1.5 to screen out abnormal feature points, and obtain spatio-temporal anomaly detection data;

[0103] In this embodiment, based on the feature encoding tensor of the device operating state and environmental changes, local outlier factor (LOF) detection is performed to identify abnormal feature points. First, the number of neighborhood samples is set to 20, that is, when calculating the local outlier factor of each sample, 20 surrounding sample points are considered. This parameter ensures the accuracy of local outlier detection. Then, by calculating the local outlier factor score, which reflects the density difference between a data point and its neighborhood samples. The higher the score, the more likely the point is an outlier. To screen out obvious abnormal feature points, the threshold for anomaly determination is set to 1.5, that is, when the local outlier factor score of a data point is greater than 1.5, the point is considered abnormal. Finally, the selected abnormal data points form a spatio-temporal anomaly detection data set for use in subsequent steps. For example, when performing anomaly detection on the equipment in the substation, if the temperature sensor data of equipment A shows high fluctuations during a certain period, while the temperature data of other surrounding equipment fluctuates less, and the calculated local outlier factor score of equipment A is 2.0, exceeding the threshold of 1.5, then equipment A is marked as abnormal during this period.

[0104] Step S153: Use the spatio-temporal anomaly detection data, combine with the preset device operation mode library for pattern matching and deviation metric analysis, calculate the device state drift deviation, and set the deviation threshold to 0.2 according to the distribution of the drift deviation to extract the device abnormal state characteristics, and obtain the device spatio-temporal anomaly characteristic data;

[0105] In this embodiment, the spatio-temporal anomaly detection data is used in combination with the device operation mode library for pattern matching and deviation metric analysis. First, using the historical data in the device operation mode library, a standard template of the device operation mode is constructed. The device operation mode library includes the historical device operation data of the equipment in the substation under different physical locations or operating environments. By matching with the current device operation state data, the drift deviation of the device state is calculated, that is, the deviation degree between the current device state and the standard mode. The drift deviation reflects whether the current operating state of the device has changed significantly, which is crucial for the warning system. According to the distribution of the drift deviation, the deviation threshold is set to 0.2, that is, when the drift deviation of the device state exceeds 0.2, the device is considered to have entered an abnormal state. At this time, by extracting the device state characteristics with larger drift deviations, the spatio-temporal anomaly characteristic data of the device is obtained, which serves as the basis for fault prediction. For example, when analyzing equipment B in the substation, it is found that the deviation of its voltage data from the historical operation mode is 0.25, exceeding the set deviation threshold of 0.2, so equipment B is considered to be in an abnormal state, and the spatio-temporal anomaly characteristic data of this equipment is timely extracted to provide a basis for subsequent fault diagnosis.

[0106] Step S154: Based on the device spatio-temporal anomaly feature data, combined with the dynamic spatio-temporal adjacency matrix, perform adaptive spatio-temporal feature mapping, execute backpropagation feature mapping weight adjustment, and set the feature embedding optimization threshold to 0.01 to optimize the feature embedding representation, obtaining the device abnormal spatio-temporal feature tensor and the device normal spatio-temporal feature tensor;

[0107] In this embodiment, based on the device spatio-temporal anomaly feature data, combined with the dynamic spatio-temporal adjacency matrix, adaptive spatio-temporal feature mapping is performed. This mapping adjusts the mapping weights of features according to the current operating state of the device through the adaptive weight adjustment mechanism in the neural network model, so that abnormal features and non-abnormal features can be expressed more accurately. During the feature mapping process, the backpropagation algorithm is used to optimize the weights and adjust the parameters of each layer in the model, enabling the model to better capture the relationships between spatio-temporal features. During the feature mapping process, the abnormal state of the device is marked and compared with the feature tensor of the normal state of the device. By optimizing the feature embedding weights, when the model reaches the optimization threshold (set to 0.01), tensors containing normal state features and abnormal state features can be obtained. These tensors represent the normal operating state and abnormal operating state of the device respectively. That is, when the adjustment amplitude of the model parameters is less than 0.01, it is considered that the feature embedding has converged and the optimization process ends. Finally, the device abnormal spatio-temporal feature tensor and the device normal spatio-temporal feature tensor are obtained, providing a more accurate feature representation for subsequent fault diagnosis. When processing device C, through adaptive spatio-temporal feature mapping, it is found that the relationship between the temperature feature and the load feature of device C has changed over a period of time. By optimizing the feature embedding, the abnormal spatio-temporal feature tensor of device C is finally obtained, which successfully captures the key information of the device anomaly.

[0108] Step S155: Merge the device abnormal spatio-temporal feature tensor and the device normal spatio-temporal feature tensor for spatio-temporal feature tensor merging, construct a spatio-temporal feature expression, and obtain the device spatio-temporal feature tensor.

[0109] In this embodiment, the spatio-temporal feature tensor of the abnormal device is merged with the spatio-temporal feature tensor of the normal device. The merged feature tensor can contain the spatio-temporal information of both the normal and abnormal device states. This spatio-temporal feature tensor provides comprehensive information for device state analysis and can effectively support subsequent fault prediction and intelligent decision-making processes. The merged spatio-temporal feature tensor uses a weighted synthesis method to fuse the features of the normal and abnormal states according to certain weights to ensure that the model can accurately identify potential faults when dealing with abnormal situations. The finally obtained device spatio-temporal feature tensor is a multi-dimensional tensor that contains all-round information about the device state. When analyzing the operating state of the power distribution room equipment, the normal operating feature tensor of device A is merged with the abnormal operating feature tensor of device B to generate a device spatio-temporal feature tensor, which can be used as the input for device health assessment to guide subsequent fault warning and maintenance decisions.

[0110] Optionally, step S2 is specifically as follows:

[0111] Step S21: Based on the device spatio-temporal feature tensor, perform variational autoencoder latent variable decomposition, extract the potential factors of the relay protection device operating state, and execute information bottleneck constraints to generate a device latent variable feature vector;

[0112] In this embodiment, based on the device spatio-temporal feature tensor, a variational autoencoder (VAE) is used for latent variable decomposition. The device spatio-temporal feature tensor contains various operating states and environmental data of the device at different time and space nodes. By encoding the data with VAE, high-dimensional data can be compressed into low-dimensional potential factors. To enhance the generalization ability of the model, information bottleneck constraints are imposed to limit the amount of information in the potential factors and avoid overfitting. During this process, the dimension of the potential factors is set to 32, and the Kullback-Leibler divergence is used to measure the difference between the distribution of the latent variables and the standard normal distribution to ensure the effective transmission and constraint of information. For example, for the relay protection device in the power distribution room, based on the spatio-temporal data of the device (such as current, voltage, load changes, etc.), after being processed by VAE, the potential factors of the device can be extracted, and these factors represent the implicit features of the device in different operating states, such as load fluctuations, fault triggers, etc. Through the information bottleneck constraint, the influence of noise can be effectively reduced, and a more concise and predictive latent variable feature vector can be extracted.

[0113] Step S22: Perform structured Bayesian network learning on the device latent variable feature vector, construct a device causal association graph model, and execute MAP device causal path weight optimization to obtain a device causal association weight matrix;

[0114] In this embodiment, a Bayesian network is constructed based on the device latent variable feature vector to learn the causal relationships between various features of the device. The Bayesian network describes the uncertainty in the device system through a graphical model, where nodes represent device states or features, and edges represent the probabilistic dependence relationships between these states or features. Through the structured Bayesian network learning algorithm, potential causal paths are automatically identified from the data. To optimize the performance of the network, the maximum a posteriori (MAP) estimation method is used to optimize the weights of the device causal paths, obtaining the optimal device causal association weight matrix. For example, when analyzing the causes of device failures, the Bayesian network can connect device operation data (such as temperature, voltage) with failure modes (such as current overload), learn the causal relationships between device states, and thus infer which factors are the main causes of the failures. Through MAP optimization, the accuracy of the causal paths can be improved, providing more reliable support for subsequent decision-making.

[0115] Step S23: Based on the device causal association weight matrix, perform distributed policy gradient optimization and proximal policy reinforcement learning iteration to obtain the optimal causal path decision parameters;

[0116] In this embodiment, the optimal causal path decision parameters are obtained through the distributed policy gradient optimization algorithm and proximal policy reinforcement learning (PPO) iteration. In the distributed policy gradient optimization, multiple nodes calculate gradients in parallel and update the policy synchronously, thus accelerating the convergence of the policy. Proximal policy reinforcement learning avoids over-updating by constraining the policy update step size to ensure policy stability. Through multiple rounds of iteration, the optimal causal path decision parameters that can maximize the reward function are obtained. Specifically, the exploration degree of each decision step is set (for example, epsilon = 0.2), and the device operation decision is optimized through the algorithm, enabling the device to select the most appropriate action path when facing complex operating states. For example, in the case of large fluctuations in device load, PPO can optimize the device scheduling strategy to keep the system running efficiently while minimizing the failure risk.

[0117] Step S24: Obtain historical device operation data, construct a device operation state transition model in combination with the optimal causal path decision parameters; perform causal reasoning on the device operation state evolution path according to the device operation state transition model, thereby generating a device causal decision graph;

[0118] In this embodiment, based on historical device operation data, combined with the optimal causal path decision parameters, and using the structured Bayesian network learning algorithm, a state transition model of the device is constructed. This model establishes the conditional dependence relationship between the various states of the device through the causal path decision parameters, and learns and quantifies the state transition probability of the device under different working conditions. The device operation state transition model describes the transition probability of the device in different states, and can reflect the possibility of the device transitioning from one operation state to another. Through the causal inference algorithm, the possible future states of the device can be inferred from historical data, and these inferences are used as the basis for device fault diagnosis and decision-making. This process involves the generation of a causal graph and path reasoning to infer the state change process of the device under specific conditions. For example, when analyzing the state transition of a certain relay protection device in a substation, by collecting the operation data of the device (such as circuit breaker status, current change, etc.) and combining the optimal decision parameters, a transition graph of the device state change can be constructed. Through causal inference, the transition path between different states of the device before and after a fault occurs is inferred, providing data support for the next prediction and optimization.

[0119] Step S25: Based on the device causal decision graph, perform device state prediction, and execute particle filter prediction error correction to obtain device operation trend prediction data.

[0120] In this embodiment, the device causal decision graph is used to predict the device state, and the particle filter (Particle Filter) technology is used to correct the prediction error of the prediction result. The particle filter is a recursive Bayesian filtering technology based on the Monte Carlo method. By generating a large number of particles to represent different possible states of the device, and then obtaining the optimal state prediction of the device through weighted averaging. This method can handle device systems with high nonlinearity and uncertainty, and is especially suitable for the dynamic prediction of device operation states. For example, for substation equipment, based on historical device operation data and causal graphs, the particle filter can effectively predict the future state of the device. When the device state changes, the particle filter continuously adjusts the weights of the particles to correct the prediction error and ensure the accuracy of the device state prediction. Through multiple iterations, more accurate device operation trend prediction data is obtained, providing real-time and reliable information for the subsequent decision support system.

[0121] Optionally, step S3 is specifically as follows:

[0122] Step S31: Based on the device causal decision graph and the device operation trend prediction data, identify the DC screen insulation deterioration variable, and perform multi-scale wavelet decomposition to extract the deterioration feature spectrum, generating the DC screen insulation deterioration feature space;

[0123] In this embodiment, based on the device causal decision-making graph and the device operation trend prediction data, variables related to the insulation deterioration of the DC power supply panel are identified, such as voltage, current, and temperature. Subsequently, the multi-scale wavelet decomposition technology is used to analyze the current waveform signal of the DC power supply panel insulation, and the deterioration characteristic spectra are extracted from multiple frequency scales. These spectral characteristics can effectively reflect the deterioration state of the insulating material, and the DC power supply panel insulation deterioration characteristic space is constructed through the calculated characteristic spectral matrix. The set operation parameters include the scale setting of wavelet decomposition, such as selecting high-frequency and low-frequency wavelet scales (such as 4-scale wavelet decomposition) to capture details and trend changes.

[0124] Step S32: Perform multi-site device distributed modeling based on the DC power supply panel insulation deterioration characteristic space, and combine the preset adaptive gated recurrent unit to perform cross-site time series feature aggregation to generate a DC power supply panel insulation deterioration federated twin model;

[0125] In this embodiment, based on the DC panel insulation deterioration feature space, the deterioration data of the DC panels at multiple device sites are modeled. The device operation mode library contains device data from multiple sites (the data for each site can include various metrics, such as the operating status of the device, environmental conditions (such as temperature, humidity), load changes, etc.). This data represents device data from different physical locations or operating environments, and all this data is docked with the insulation deterioration features extracted previously through multi-scale wavelet decomposition. An Adaptive Gated Recurrent Unit (GRU) network is used to model the temporal features of each site. The number of hidden units in the GRU is set to 128 to improve the network's ability to remember and capture features in long time series. First, an independent GRU model is trained for each site separately by processing the historical device operation data of each site to learn the temporal features of each site. Next, through the design of the adaptive gated recurrent unit, the weights of the model are adjusted in real time to ensure enhanced adaptability to time changes. After the temporal feature vectors independently generated by each site, through the cross-site temporal feature aggregation algorithm, the features of different sites are fused, and the weighted average method is used to handle the data differences from different sites, and finally the fused device feature data is obtained. To ensure data privacy and avoid centralized storage, a federated learning framework is adopted for feature learning and sharing. In federated learning, each site independently trains a local model and only shares the updated model parameters instead of the original data. The federated learning algorithm combines the model parameters from each site in an iterative manner and uses methods such as weighted average to ensure the effective integration of the features of each site. Finally, based on the aggregation of cross-site data, a DC panel insulation deterioration federated twin model is generated. This model can simultaneously reflect the operating status of different device sites and can use global information for optimized prediction. Through this process, the obtained DC panel insulation deterioration federated twin model can consider the deterioration features of each site, optimize the global device management strategy, and ensure data privacy at the same time.

[0126] Step S33: According to the DC panel insulation deterioration federated twin model, perform a diffeomorphic manifold embedding mapping and execute a manifold Laplacian regularization optimization to obtain a dynamic characteristic diffeomorphic mapping result;

[0127] In this embodiment, a differential homeomorphic manifold embedding mapping is performed using the DC power supply insulation degradation federated twin model. Through this process, the high-dimensional features of the device operating state can be embedded into the low-dimensional manifold space to extract its inherent geometric structure. When performing manifold embedding, first, a differential homeomorphic method is used for embedding mapping to simplify and accurately model the complex device state. Then, manifold Laplacian regularization optimization is performed to ensure that the manifold structure of the device state has good smoothness and consistency, thereby reducing the influence of noise. This optimization process regularizes the Laplacian matrix of neighboring points in the manifold to avoid overfitting and enhance the robustness of the mapping result.

[0128] Step S34: Use the differential homeomorphic mapping result of the dynamic characteristics to perform local geometric consistency constraints on the device operating state transition model, and perform variational Bayesian inference to obtain the device credible twin verification chain.

[0129] In this embodiment, the dynamic characteristic diffeomorphic mapping results are used to optimize the operation state transition model of the device to ensure the generation of the device trusted twin verification chain. First, the dynamic characteristic diffeomorphic mapping results provide a low-dimensional embedding space for the device state transition, and these embedding spaces can accurately capture the state change characteristics of the device at different time points. Based on these mapping results, local geometric consistency constraints are used to optimize the device state transition to ensure the stability of the geometric structure of the state transition, thereby reducing errors caused by data noise or device differences. Next, the variational Bayesian inference method is used to refine the device state transition model to infer the potential states of the device at different time points and their transition paths. The variational Bayesian inference ensures that the model parameters can be gradually updated and finally converge to a stable solution during the model training process by setting the maximum number of iterations to 1000 and the learning rate to 0.01. To optimize the inference process, the Laplace approximation is used to calculate the posterior probability, thereby ensuring high-precision inference of the model under different conditions. Through these steps, each link of the device state transition is constrained by the posterior distribution in the probability space, thereby avoiding overfitting and improving the generalization ability of the model. Then, by combining the trusted computing framework in blockchain technology with the constrained device operation state transition model, the device trusted twin verification chain is generated. This verification chain not only contains the device state transition information, but also protects the integrity and privacy of the data through encryption. The verification chain can provide reliable proof for each state transition of the device and prevent data tampering or counterfeiting. During the construction of the device trusted twin verification chain, blockchain technology is used to record and verify the transition process of each device state to ensure that each step of the transition can be traced and provide credible proof. Through these steps, a highly credible device state verification chain is finally obtained, providing a strong guarantee for the state evaluation and management of the device. The generated device trusted twin verification chain can provide highly reliable verification for the credibility of the device state, help evaluate the operation state of the device, ensure the accuracy of model prediction, and provide a basis for subsequent device management and maintenance decisions.

[0130] Optionally, step S31 is specifically as follows:

[0131] Step S311: Extract the DC panel operation trend from the device operation trend prediction data, and set the voltage fluctuation threshold to 2%, the current fluctuation threshold to 3%, and the abnormal determination threshold to 5% to evaluate the DC panel operation performance, and obtain the DC panel operation performance trend data;

[0132] In this embodiment, the operation trend of the DC panel is extracted from the equipment operation trend prediction data, which specifically involves real-time data monitoring of voltage and current. The voltage fluctuation threshold is set to 2%, that is, when the voltage fluctuation exceeds 2%, it is considered that there is an abnormal fluctuation; the current fluctuation threshold is set to 3%, that is, when the current fluctuation exceeds 3%, it is considered that there is a current anomaly. These data are trend extracted and predicted by a prediction model such as a long short-term memory network LSTM or an ARIMA model. By comparing the predicted data such as voltage, current, and temperature with the actual data, the error value is calculated. When the error value exceeds 5%, it is considered that the operating state of the DC panel is abnormal. At this time, it is necessary to further investigate the changes in its operating performance so that detailed analysis can be performed in subsequent steps. Based on data processing and trend prediction, the operating state of the DC panel is evaluated. If the operating data of the equipment (such as current and voltage fluctuations) is within the set threshold, the equipment is considered to be in a normal state; if it exceeds the set threshold, it is considered that the equipment may have performance degradation or is about to fail.

[0133] Step S312: Screen the DC panel insulation performance degradation factors for the DC panel operating performance trend data according to the equipment causal decision map, set the causal decision factor screening threshold to 0.1, and perform contribution evaluation, select the top 5 causal decision factors in contribution ranking to construct the DC panel insulation state variable set;

[0134] In this embodiment, the operating performance trend data of the DC panel is analyzed based on the equipment causal decision map. By setting the causal decision factor screening threshold to 0.1, the factors that have a significant impact on the degradation of the insulation performance of the DC panel are screened out. The DC panel operating performance trend data is used to reflect the abnormal fluctuations that may occur during the operation of the equipment, such as excessive fluctuations in voltage and current or abnormal temperature, which may cause degradation of the insulation layer. Then, the causal relationship between each operating parameter and the insulation performance is analyzed through the equipment causal decision map to identify which factors have a significant impact on the insulation performance. Then, the contribution is evaluated using regression analysis and feature importance evaluation methods (for example, based on Shapley value or LIME algorithm). The evaluation results show the specific impact of each factor on the operating performance of the equipment, and finally the top 5 causal decision factors with the highest contribution are selected. These factors can be important factors that affect the insulation state of the equipment, such as temperature, humidity, and load fluctuations, and then a set of DC panel insulation state variables is constructed to provide a basis for subsequent fault prediction and performance optimization.

[0135] Step S313: Based on the dynamic spatiotemporal adjacency matrix, a correlation analysis is performed on the set of DC screen insulation state variables, the Pearson correlation coefficient threshold is set to 0.6, and variables with a value less than the Pearson correlation coefficient threshold are eliminated, thereby extracting a subset of DC screen insulation degradation variables;

[0136] In this embodiment, based on the dynamic spatio-temporal adjacency matrix, through the correlation analysis of time and space, the correlation between variables in the set of DC panel insulation state variables is evaluated. The Pearson correlation coefficient is used to measure the linear relationship between variables, and the correlation coefficient threshold is set to 0.6, which means that when the correlation coefficient between two variables is lower than 0.6, the two variables are considered to have insufficient correlation and are thus excluded. The key to this step is to quantify the correlation relationship between variables through the construction of the spatio-temporal adjacency matrix, and finally extract the core variable subset of DC panel insulation deterioration. These variables can truly reflect the insulation deterioration process of the equipment, thereby improving the accuracy of subsequent analysis and model prediction.

[0137] Step S314: Conduct empirical mode analysis on the subset of DC panel insulation deterioration variables, extract intrinsic mode functions, calculate kurtosis and energy ratio, set the deterioration feature threshold to 3, the energy proportion threshold to 5%, screen out the main deterioration feature components, and generate an insulation deterioration feature variable matrix;

[0138] In this embodiment, empirical mode decomposition (EMD) is performed on the extracted subset of DC panel insulation deterioration variables. This method can effectively extract intrinsic mode functions (IMFs) from complex signals, and these functions represent important signal components in the data. For each intrinsic mode function, its contribution to the deterioration of the DC panel insulation performance is evaluated by calculating its kurtosis and energy ratio. The deterioration feature threshold is set to 3. When the kurtosis value is greater than 3, it is considered that the mode function strongly reflects the equipment deterioration characteristics. At the same time, the energy proportion threshold is set to 5% to screen out the main deterioration features with large energy proportions. Finally, the deterioration feature components with a larger contribution degree obtained by combining kurtosis and energy proportion are screened out, and a DC panel insulation deterioration feature variable matrix is constructed to provide high-quality feature data for the subsequent model.

[0139] Step S315: Set the decomposition level to 4, perform wavelet transform on the insulation deterioration feature variable matrix for multi-scale time-frequency decomposition, extract the high-frequency abnormal component energy ratio greater than the high-frequency energy ratio threshold of 10%, thereby constructing the DC panel insulation deterioration feature space.

[0140] In this embodiment, wavelet transform is used to perform multi-scale time-frequency decomposition on the insulation degradation characteristic variable matrix obtained in the previous step. The number of wavelet decomposition layers is set to 4, that is, the signal is decomposed into four frequency bands with different scales. Through time-frequency analysis, high-frequency components in the signal can be extracted to reveal abnormal changes in the device. The high-frequency energy ratio threshold is set to 10%, that is, if the energy ratio of the high-frequency component exceeds 10%, it is considered that this component may reflect abnormal behavior or degradation of the device. High-frequency abnormal components with an energy ratio greater than 10% are extracted through wavelet transform, and finally the insulation degradation characteristic space of the DC power supply panel is constructed. This characteristic space contains abnormal characteristics of the device at different frequencies, can accurately reflect changes in the insulation state of the device, and thus provides efficient support for subsequent fault diagnosis and prediction.

[0141] Optionally, the reconstruction of the thermal stress field at the busbar connection point described in step S4 is specifically as follows:

[0142] Based on the device trusted twin verification chain, obtain the device trusted operating state data and trusted environment parameter data, and extract the state characteristics of the busbar connection point and the on-load tap-changer state characteristics in the device trusted operating state data;

[0143] In this embodiment, the device operating data and environment parameter data are obtained through the device trusted twin verification chain. The device operating state data includes information such as the current, voltage, temperature, and vibration of the busbar connection point, while the environment parameter data includes the external temperature, humidity, and power grid load, etc. The credibility of these data is verified by combining variational Bayesian inference and the trusted computing framework. The device operating state data includes key parameters such as the current, voltage, temperature, vibration, and power load of the busbar connection point, which are collected in real time by sensors and uploaded to the trusted computing platform. At the same time, the environment parameter data includes external conditions such as the external temperature, humidity, air quality, and load fluctuation at the location of the device. While these data are encrypted and securely stored through the trusted twin verification chain, the privacy and security of the information during transmission are ensured. During this process, the device trusted twin verification chain provides the data traceability ability, enabling each data point to be traced back to its source to ensure the authenticity and integrity of the data. On this basis, the state characteristics of the busbar connection point are extracted from the device operating state data, including the temperature change of the busbar connection point, the current fluctuation at the connection point, and the change in contact resistance, etc., while the on-load tap-changer state characteristics include the current load, contact point temperature, and current fluctuation characteristics of the tap-changer. Through the extraction of these characteristics, it can provide important inputs for subsequent thermal stress field modeling and analysis.

[0144] Construct a thermal stress field model for the busbar connection point according to the state characteristics of the busbar connection point and the trusted environment parameter data;

[0145] In this embodiment, based on the state characteristic data and environmental parameter data of the busbar connection points extracted from the previous stage, a thermal stress field model of the busbar connection points is constructed. The thermal stress field model simulates the thermal stress distribution of the busbar connection points during operation according to the current load, temperature gradient, and heat conduction characteristics of the busbar connection points. Specifically, the finite element analysis (FEA) method is used to model the structure of the busbar connection points, set physical properties such as the coefficient of thermal expansion, thermal conductivity, and specific heat capacity of the material, and combine the current fluctuations and temperature changes during the operation of the equipment to calculate the thermal stress distribution model of the busbar connection points under different environmental conditions. This model can reveal the variation law of the thermal stress of the busbar connection points during actual operation and provide a necessary data basis for subsequent space-time coordinate transformation and thermal pressure field calculation.

[0146] Perform space-time coordinate transformation on the thermal stress field model of the busbar connection points, and combine the credible operation state data of the equipment to set the initial scenario of the thermal stress field of the busbar connection points;

[0147] In this embodiment, the purpose of performing space-time coordinate transformation on the established thermal stress field model of the busbar connection points is to map the thermal stress field in the model to the actual working environment. By combining the credible operation state data of the equipment, the initial scenario of the thermal stress field of the busbar connection points is set, that is, the initial thermal stress state of the busbar connection points is defined. During this process, considering the actual operation conditions of the busbar connection points, such as current changes, working temperature, and environmental parameters (such as humidity, external temperature, etc.), and incorporating these factors into the thermal stress field model. During the space-time coordinate transformation process, the spatial interpolation method and the dynamic time adjustment algorithm are used to connect the idealized data in the model with the actual equipment operation environment in the equipment operation mode library, so as to more accurately reflect the thermal stress field distribution of the equipment during actual operation.

[0148] Perform thermal pressure field calculation on the initial scenario of the thermal stress field of the busbar connection points through the thermodynamic model of the busbar connection points, and perform dynamic update according to the credible twin verification chain of the equipment to obtain the stress field distribution map of the busbar connection points;

[0149] In this embodiment, through the thermodynamic model of the busbar connection points, a detailed calculation of the thermal stress field in the initial scenario is carried out to obtain the thermal pressure field distribution map of the busbar connection points. During the calculation process, the thermodynamic transfer equation is combined with the finite element method to calculate the thermal stress field, considering temperature, heat conduction, heat convection, and the load change of the busbar connection points in the dynamic working environment. The calculation results of the thermal pressure field will reveal the thermal stress distribution of the busbar connection points under different working conditions. To ensure the timeliness and accuracy of the results, the real-time operation data obtained from the credible twin verification chain of the equipment is used to dynamically update the calculation results, ensuring that the impact of the equipment operation state change on the thermal stress field is timely feedback. The finally obtained stress field distribution map of the busbar connection points can provide reliable data support for subsequent identification of thermal stress concentration areas and equipment optimization.

[0150] Identify the thermal stress concentration area based on the stress field distribution map of the busbar connection point, and obtain the data of the busbar thermal stress concentration area;

[0151] In this embodiment, based on the stress field distribution map of the busbar connection point obtained in Step 4, the identification of the thermal stress concentration area is carried out. Specifically, using image processing technology and pattern recognition algorithms, the stress field distribution map is analyzed to identify the areas on the busbar connection point where the thermal stress is most concentrated. These areas may cause equipment damage or failures due to overheating, so the timely identification of the thermal stress concentration area is crucial. During the identification process, algorithms such as gradient-based region detection, thermal stress threshold setting, and clustering analysis are adopted to ensure the accurate positioning of the thermal stress concentration area. The finally generated data of the busbar thermal stress concentration area can provide an important basis for optimizing the equipment operation strategy.

[0152] Construct a non-exchange decision model based on the data of the busbar thermal stress concentration area and the state characteristics of the on-load tap-changer, and input the initial scenario of the thermal stress field of the busbar connection point into the non-exchange decision model for non-varying decision analysis, and optimize the equipment operation strategy according to the results of the non-varying decision analysis to generate an inrush current control strategy.

[0153] In this embodiment, based on the data of the busbar thermal stress concentration area and the state characteristics of the on-load tap-changer, a non-exchange decision model is constructed. This model adopts non-exchange game theory and multi-objective optimization methods to evaluate the optimal configuration of the equipment operation strategy under different working conditions. By inputting the initial scenario of the thermal stress field of the busbar connection point into this model for decision analysis, it is possible to evaluate the impact of different operation modes on the thermal stress distribution and performance of the equipment in a specific environment. The model will evaluate the impact of various operation strategies (such as adjusting the load, optimizing the voltage regulation settings, adjusting the busbar current, etc.) on the equipment thermal stress according to the current thermal stress concentration area, equipment state characteristics, and load information, and select the most suitable operation strategy for the current environment and load conditions. The optimized strategy can reduce the equipment thermal stress and avoid overheating, thereby improving the operation efficiency and safety of the equipment. Finally, the generated inrush current control strategy will be used to real-time control the thermal stress distribution of the busbar connection point to ensure the stability and reliability of the equipment operation.

[0154] Optionally, the optimal switching instruction set of the compiled standby power supply automatic switching device described in Step S4 is specifically:

[0155] Identify the power supply fault mode based on the equipment trusted operation state data, and establish an equipment operation state - fault mode mapping model;

[0156] In this embodiment, the power network topology structure is obtained from the power distribution and transformation equipment management platform. The power distribution and transformation equipment management platform is connected to various components in the equipment and the power network (such as transformers, switches, busbars, etc.), and can monitor and obtain the structural information of the power network in real time. The topology structure includes key information such as the connection relationship between power supply devices, line status, and fault conditions. The system obtains the latest network topology diagram by calling the API interface of the management platform and converts it into a standardized data structure. Through the trusted operating state data of the equipment, including parameters such as voltage, current, load, and temperature, a machine learning model is used to analyze the operating state of the equipment and identify power fault modes. The specific operation is as follows: First, various operating parameters are extracted from the trusted operating state data of the equipment. Then, classification algorithms such as support vector machine (SVM) or decision tree are used to establish a fault mode recognition model based on the operating state of the equipment. For example, when the current and voltage fluctuations are greater than 3%, it is marked as the "power overload" mode; when the voltage is lower than the specified threshold (such as 180V), it is marked as the "voltage fault" mode. In this way, different power fault modes can be identified and corresponding mapping models can be established, providing accurate basic data for subsequent fault source location. Combining the power network topology structure and the equipment operating state data, potential power fault modes will be further identified. Using machine learning algorithms (such as decision trees, support vector machines, etc.), based on historical equipment fault data and real-time operating data, the system can automatically identify different fault modes. Once the equipment operating state data and the fault mode recognition results are available, the system will use machine learning methods (such as random forests, support vector machines, etc.) to establish a mapping model between the equipment operating state and the fault mode. This model will map to the corresponding fault mode according to the operating state characteristics of different equipment (such as voltage, current, temperature, etc.). For example, if the current of the equipment suddenly increases to 1.5 times the normal value and the temperature begins to rise rapidly, the mapping model may identify this state as an "overload fault" or "power instability". At the same time, it will be trained and optimized through historical data to improve the accuracy of the model. After establishing the equipment operating state - fault mode mapping model, a validation set will be used to test the model to ensure that it can accurately identify equipment fault modes. During the validation process, new equipment operating data will be input and the predicted fault mode will be output. If the fault mode prediction is accurate, the model will be officially put into use.

[0157] According to the equipment operating state - fault mode mapping model, perform fault source location analysis, identify the faulty equipment and the power path, and generate preliminary fault source location data;

[0158] In this embodiment, by using the established mapping model of equipment operation status - fault mode, when a fault occurs, the real - time equipment data is compared with the fault mode to locate the fault source. For example, when the current fluctuation is greater than the preset 5%, it is judged as a power supply fault. Combining with the power network topology structure, a fault location algorithm (such as the shortest path algorithm) is used to accurately locate the fault source. If the voltage drops below 180V and the fault mode matches "overload" or "short - circuit", the faulty equipment will be located, such as "Equipment 2" or "Power path 3". The entire power path and the affected equipment are gradually identified to generate preliminary fault source location data, which includes the equipment ID of the fault source, the location and time of the fault occurrence, and the specific situation of the power path. This data provides a necessary basis for formulating subsequent fault recovery strategies.

[0159] Obtain the historical switching data of the backup power supply automatic switching device, and perform switching mode recognition on the historical switching data of the backup power supply automatic switching device to obtain the switching mode data of the backup power supply automatic switching device;

[0160] In this embodiment, historical switching data is extracted from the management system of the backup power supply automatic switching device. These data include the time of each switching operation, the power path, the switching device ID, and the subsequent power recovery situation, etc. Using pattern recognition algorithms, such as K - means clustering or time - series analysis, analyze the historical switching data to identify different switching modes. For example, when a power path fault occurs, the historical data may show that some switching modes successfully restored the system in similar past faults, while other switching modes had poor restoration effects. Through pattern recognition of the historical data, effective switching modes are extracted, marked as "overload mode switching" or "voltage anomaly mode switching", etc., to form the switching mode data of the backup power supply automatic switching device. These mode data will help the system select appropriate switching strategies for future fault scenarios.

[0161] Combine the preliminary fault source location data and the power supply fault mode to perform backup power supply automatic switching device switching strategy matching on the switching mode data of the backup power supply automatic switching device to obtain the candidate switching strategies of the backup power supply automatic switching device;

[0162] In this embodiment, according to the generated preliminary fault source location data and the obtained switching mode data of the backup power supply automatic switching device, switching strategy matching will be carried out. First, by comparing the matching degree between the fault mode and the historical switching data, select the switching strategy of the backup power supply automatic switching device that matches the current fault scenario. For example, if an overload fault occurs in power path 3, the strategy of "switching path 2 during overload fault" shown in the historical switching data is the optimal solution. Therefore, the switching strategy that best matches the fault source location (such as Equipment 2) will be selected. At this time, combining the equipment working parameters, switching time requirements, and recovery goals, multiple candidate switching strategies are screened out, and these strategies are used as the input for subsequent simulation tests.

[0163] Perform simulation tests on the candidate switching strategies of the backup power supply automatic switching device, evaluate the execution efficiency of the candidate switching strategies of the backup power supply automatic switching device and the power system restoration speed, and screen out the optimal switching strategy;

[0164] In this embodiment, a power system simulation platform (such as MATLAB / Simulink or DIgSILENT PowerFactory) is used to simulate all candidate switching strategies, and the execution efficiency of each strategy and the power system restoration speed are evaluated. By setting the same fault conditions, the response capabilities of each backup power supply automatic switching device strategy under power supply faults are simulated. For example, if candidate strategy A can successfully switch and restore the system power within 1 second, while candidate strategy B takes 3 seconds, strategy A will be selected as the preferred option. The simulation results will also evaluate the impact of the strategy on the power system stability, such as avoiding system instability or voltage overload and other phenomena. Finally, the optimal switching strategy is screened out by comparing the restoration speed, system stability and restoration quality.

[0165] Based on the optimal switching strategy, an instruction set is compiled to obtain the optimal switching instruction set of the backup power supply automatic switching device, and it is uploaded to the instruction management platform of the backup power supply automatic switching device to implement instruction replacement.

[0166] In this embodiment, corresponding instruction sets will be generated according to the selected optimal switching strategy. These instructions include specific operation steps, device numbers, power supply path numbers, and various control parameters. For example, the generated instruction is: "Within 5 seconds after the fault occurs, switch the power supply path of device 3 to the standby path 2 and keep the current not exceeding 10A." The instruction set takes into account factors such as the execution ability of the backup power supply automatic switching device, time limit, device capacity, etc., to ensure the efficiency and safety of execution. To ensure that the instructions can be correctly parsed and executed by the backup power supply automatic switching device, the instruction set is formatted into a machine-readable code or data format. This is a standardized instruction file, which may be encoded using XML, JSON or other protocol formats. The compiled instruction set will be verified in the system to ensure that all instructions can be executed smoothly as expected. This step will also check the logical consistency of the instruction set, for example, to ensure that there are no two instructions conflicting with each other, or there are no problems in the execution order. If there are problems, optimization will be carried out. For example, each instruction will be represented as a data block containing information such as device ID, operation type, parameter value, etc. After the instruction set compilation is completed, it will be uploaded to the instruction management platform of the backup power supply automatic switching device, and the platform will verify and send the instructions to the backup power supply automatic switching device for implementation to replace the original control instructions. Through this process, the device can quickly respond to faults according to the optimal switching strategy, ensuring the stable restoration of the power system.

[0167] Optionally, the fault evolution prediction described in step S5 is specifically:

[0168] Extract real-time operation characteristics based on the real-time operation data of the power transformation and distribution room equipment. Set the mother wavelet to the 4th order to extract the equipment operation stress characteristics, set the load fluctuation threshold to 0.05 pu to screen the equipment load characteristics, and set the equipment operation stress characteristics to 10 minutes to calculate the real-time operation temperature rise characteristics, so as to obtain the equipment operation stress characteristic data, equipment load characteristic data and real-time operation temperature rise characteristic data;

[0169] In this embodiment, the power transformation and distribution room equipment obtains various key operation data through real-time monitoring sensors, including current, voltage, temperature, etc. To extract the operation stress characteristics of the equipment, the mother wavelet is set to the 4th order, and the frequency and fluctuation characteristics of the equipment operation are analyzed through wavelet transform. As the basic function, the mother wavelet can capture the high-frequency and low-frequency characteristics of the equipment at different time scales, especially suitable for the extraction of signal mutation or oscillation characteristics. The equipment load characteristics are screened by setting the load fluctuation threshold to 0.05 pu (per unit value), which means that when the load fluctuation amplitude exceeds the threshold, the load characteristics are considered to have changed. The real-time operation temperature rise characteristics of the equipment are obtained by averaging the temperature data with a period of 10 minutes. Considering the heat dissipation time of the equipment, the real-time operation temperature rise characteristics are obtained. These characteristic data will be used to evaluate the operation state of the equipment, generate equipment operation stress characteristic data, load characteristic data and temperature rise characteristic data, and help subsequent analysis and fault prediction.

[0170] Based on the variable load fluctuation characteristics in the equipment load characteristic data, construct a load fluctuation viscous evolution model, perform Fourier transform to extract the load viscous component, calculate the viscous damping parameter in combination with the real-time operation temperature rise characteristic data, and set the calculation threshold of the viscous damping parameter to 0.02 to eliminate the invalid damping, so as to generate a load viscous characteristic matrix;

[0171] In this embodiment, after extracting the equipment load characteristics, a load fluctuation viscous evolution model is constructed according to the load fluctuation characteristics. Specifically, according to the fluctuation of historical load data, the frequency components of the load fluctuation are analyzed through Fourier transform, and the viscous component of the load is extracted therefrom. The load viscous component represents the internal friction effect generated by the load fluctuation on the equipment, and these viscous components are closely related to factors such as equipment operation stress and temperature rise characteristics. Combining the real-time operation temperature rise characteristic data, the system further calculates the viscous damping parameter, and the damping parameter reflects the influence of the load fluctuation on the thermal stress of the equipment. Set the calculation threshold of the viscous damping parameter to 0.02, and the damping below this threshold will be considered as invalid damping, and the invalid data will be eliminated to ensure the effectiveness and accuracy of the viscous characteristic matrix. Finally, a load viscous characteristic matrix is generated, which is used for subsequent fault evolution analysis and prediction.

[0172] Screen the historical equipment fault data based on the equipment operation mode library;

[0173] In this embodiment, historical equipment failure data related to the current load viscosity characteristics is screened through the equipment operation mode library. The equipment operation mode library contains historical failure modes, equipment performance data, and equipment operation data under various operating environments. By comparing the load characteristics of the current equipment with the historical failure data, potential failure modes can be identified. For example, if the current equipment load viscosity characteristic data has a high similarity with a certain historical equipment failure mode (such as overload failure or poor electrical contact), the system will extract the relevant failure data for further analysis. The historical equipment failure data will be used as a reference to help infer the possible failure types and occurrence times of the current equipment.

[0174] Dynamically evolve the load viscosity characteristic matrix, combine the equipment operation stress characteristic data and historical equipment failure data to solve the fault induction trend, set the viscosity fault evolution threshold to 0.1, and obtain the load viscosity fault evolution path;

[0175] In this embodiment, by dynamically evolving the load viscosity characteristic matrix, combining the real-time operation stress characteristic data of the equipment with the historical equipment failure data, the fault induction trend is inferred. To achieve dynamic evolution, time series analysis and sliding window techniques are adopted. First, the equipment load viscosity characteristic matrix is constructed as a time series data set, where each time step contains characteristics such as load fluctuation, temperature rise, and stress. By setting a sliding window (for example, the window length is 10 minutes), the change of load fluctuation can be tracked in each time step. Fourier transform is further used to analyze the periodic components in the load viscosity characteristic data, revealing the potential impact of long-term and short-term load changes on the equipment. By combining the real-time stress data of the equipment (such as temperature, pressure, etc.) and historical failure data, the system uses techniques such as regression analysis and fault tree analysis (FTA) to generate the fault induction trend. For example, through regression analysis, the relationship model between load viscosity and factors such as equipment thermal stress and load fluctuation can be calculated to further predict the time when the equipment may enter the fault mode. Set the viscosity fault evolution threshold to 0.1, indicating that when the evolution of the load viscosity characteristic exceeds this threshold, the equipment enters a critical stage of fault induction. At this stage, the risk of equipment operation will be alerted, and the trend path of fault induction will be generated, providing a basis for the next step of fault diagnosis and recovery.

[0176] Perform feature clustering according to the load viscosity fault evolution path, divide the fault evolution stage, and perform fault matching based on the historical equipment failure data. Set the fault matching similarity threshold to identify the fault type and generate the fault evolution prediction result;

[0177] In this embodiment, after obtaining the load viscosity fault evolution path, feature clustering will be performed to divide different fault evolution stages. By performing clustering analysis on the multi-dimensional data of the equipment load viscosity characteristics, the characteristic patterns of different fault development stages can be identified. For example, in the early stage, the load fluctuation shows slight fluctuations, while in the later stage, the load fluctuation shows severe fluctuations, indicating that the equipment is about to have a serious fault. Fault matching will also be performed based on historical equipment fault data, setting a fault matching similarity threshold to identify the similarity between the current load evolution and historical faults. If the similarity exceeds the set threshold, it is considered that the current equipment fault type has been confirmed, and a fault evolution prediction result is generated.

[0178] According to the optimal switching instruction set of the backup power supply automatic switching device, perform switching instruction adaptation and matching on the fault evolution prediction result to obtain the load viscosity fault switching instruction;

[0179] In this embodiment, through the optimal switching instruction set, switching instruction adaptation will be performed on the fault evolution prediction result. This step includes matching the predicted fault evolution stage with the instruction set of the backup power supply automatic switching device. The optimal switching instruction set of the backup power supply automatic switching device contains operation instructions for equipment restoration under different fault modes. For example, if the load fluctuation reaches a specific threshold, it will instruct the backup power supply automatic switching device to perform a switching operation on a specific piece of equipment to ensure the continuous and stable operation of the power system. The switching instruction will be dynamically adjusted according to the change of the fault type and fault evolution stage to achieve the best restoration of the power system.

[0180] Based on the fault evolution prediction result, construct a viscosity fault warning model, set the viscosity instability threshold to 0.08, and combine the equipment load characteristic data and the load viscosity fault switching instruction to adjust the dynamic warning strategy, and use the trusted computing framework in the equipment trusted twin verification chain to construct a global operation optimization evidence chain.

[0181] In this embodiment, a sticky fault warning model is constructed based on the fault evolution prediction result, and the dynamic warning strategy is adjusted in combination with the real-time device operation data. Specifically, first, a prediction model is trained by using machine learning algorithms (such as support vector machine SVM or decision tree) through historical data and real-time data of device load characteristics. This model can judge whether the device enters the high-risk area of fault evolution based on the real-time load characteristics. The core of the sticky fault warning model is to evaluate whether the device is about to become unstable according to characteristics such as load fluctuation, device operation stress, and temperature rise. The warning model plays a key role in this process. By setting the sticky instability threshold to 0.08, if the combination of the device's load fluctuation characteristics and thermal stress characteristics exceeds this threshold, the device is considered to be in a high-risk state and a warning is triggered. The dynamic warning strategy is used to adjust the response strategy in real time. When the operation characteristics of the device are close to instability, warning measures are taken in a timely manner, such as adjusting the load, optimizing the device operation load, or starting the backup power automatic switching device. The Bayesian network is used in combination with the credible twin verification chain of the device to optimize the warning strategy. Under the trusted computing framework, the system can not only generate fault predictions based on historical fault data and real-time data, but also automatically adjust the warning rules by using the device status information, so as to achieve personalized dynamic warning. When warning of device faults, the system can output accurate adjustment strategies according to the device load characteristic data and sticky fault switching instructions, such as modifying the load control parameters or adjusting the device operation mode, so as to minimize the system risk. In addition, based on the global operation optimization evidence chain in the device credible twin verification chain, credible support is provided for fault warning and adjustment strategies. This multi-level warning and response mechanism effectively improves the stability and response speed of the system, and helps the device to operate efficiently in the face of potential faults.

[0182] Optionally, this specification also provides a system for evaluating the operation state of a substation and distribution room equipment, which is used to execute the method for evaluating the operation state of a substation and distribution room equipment as described above. The system for evaluating the operation state of a substation and distribution room equipment includes:

[0183] A data acquisition module, which is used to obtain multi-dimensional sensing data of the substation and distribution room and construct a dynamic spatio-temporal adjacency matrix; identify the spatio-temporal abnormal characteristics of the equipment according to the dynamic spatio-temporal adjacency matrix, and obtain the equipment spatio-temporal feature tensor;

[0184] An equipment operation trend deduction module, which is used to deduce the equipment operation trend based on the equipment spatio-temporal feature tensor and obtain the equipment operation trend prediction data;

[0185] A DC panel insulation performance analysis module, which is used to perform DC panel insulation degradation federated twin modeling according to the equipment operation trend prediction data and perform dynamic characteristic diffeomorphic mapping to obtain the equipment credible twin verification chain;

[0186] An instruction compilation module, which is used to reconstruct the thermal stress field of the busbar connection points of the device trusted twin verification chain, analyze the tap-changing strategy, and obtain the inrush current prevention control strategy; compile the optimal switching instruction set of the backup power supply automatic switching device based on the inrush current prevention control strategy;

[0187] A fault tracing module, which is used to obtain the real-time operation data of the substation equipment, and perform fault evolution prediction on the real-time operation data of the substation equipment according to the optimal switching instruction set of the backup power supply automatic switching device, and generate a global operation optimization evidence chain; perform fault tracing analysis according to the global operation evidence chain, and screen high-reliability fault warning instructions to obtain the maintenance strategy of the substation equipment.

[0188] Therefore, from any perspective, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application document are intended to be included in the present invention.

[0189] The above description is only the specific implementation manners of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for evaluating the operating status of power transformation and distribution room equipment, characterized in that, Including the following steps: Step S1: Obtain the multi-dimensional sensing data of the variable substation and construct a dynamic spatio-temporal adjacency matrix; identify the spatio-temporal anomaly features of the equipment according to the dynamic spatio-temporal adjacency matrix to obtain the equipment spatio-temporal feature tensor; Step S2: Deduce the equipment operation trend based on the equipment spatio-temporal feature tensor to obtain the equipment operation trend prediction data; Step S3: Execute the DC panel insulation deterioration federated twin modeling according to the equipment operation trend prediction data and perform dynamic characteristic diffeomorphic mapping to obtain the equipment credible twin verification chain; Step S4: Reconstruct the thermal stress field of the busbar connection points for the equipment credible twin verification chain, analyze the tap-changing strategy to obtain the inrush current control strategy; compile the optimal switching instruction set of the backup power supply automatic switching device based on the inrush current control strategy; Step S5: Obtain the real-time operation data of the variable substation equipment, and perform fault evolution prediction on the real-time operation data of the variable substation equipment according to the optimal switching instruction set of the backup power supply automatic switching device to generate a global operation optimization evidence chain; Perform fault traceability analysis according to the global operation evidence chain and screen high-reliability fault warning instructions to obtain the maintenance strategy of the variable substation equipment.

2. The evaluation method for the operating status of the power transformation and distribution room equipment according to claim 1, wherein Step S1 is specifically as follows: Step S11: Obtain the multi-dimensional sensing data of the variable substation and perform data preprocessing on the multi-dimensional sensing data of the variable substation to obtain the multi-dimensional sensing data to be analyzed; Step S12: Divide the multi-dimensional sensing data to be analyzed to obtain the equipment sensing data and the environmental sensing data, and perform equipment operation environment perception according to the equipment sensing data and the environmental sensing data to obtain the equipment operation environment data; Step S13: Extract the equipment correlation features and interaction frequency features based on the equipment sensing data, and construct a preliminary spatio-temporal adjacency matrix according to the equipment correlation features and the interaction frequency features; Step S14: Perform spatio-temporal interaction feature mapping and adaptive aggregation on the preliminary spatio-temporal adjacency matrix according to the equipment operation environment data to obtain a dynamic spatio-temporal adjacency matrix; Step S15: Perform equipment spatio-temporal anomaly identification according to the dynamic spatio-temporal adjacency matrix, and mark the equipment anomaly spatio-temporal anomaly features to obtain the equipment spatio-temporal feature tensor.

3. The evaluation method for the operating state of the power transformation and distribution room equipment according to claim 2, wherein, Step S15 is specifically as follows: Step S151: Perform auto-encoding of the correlation features of the equipment operation state - environmental change based on the dynamic spatio-temporal adjacency matrix, extract the non-linear correlation features, and set the dimension reduction target dimension to 16 for feature dimension reduction to generate the equipment operation state - environmental change feature encoding tensor; Step S152: Perform local outlier factor detection based on the equipment operation state - environmental change feature encoding tensor, set the number of neighborhood samples to 20 to calculate the local anomaly factor score, set the anomaly determination threshold to 1.5 to screen the anomaly feature points to obtain the spatio-temporal anomaly detection data; Step S153: Use the spatio-temporal anomaly detection data, combine with the preset equipment operation mode library for pattern matching and deviation metric analysis, calculate the equipment state drift deviation, and set the deviation threshold to 0.2 according to the distribution of the drift deviation to extract the equipment abnormal state features to obtain the equipment spatio-temporal anomaly feature data; Step S154: Based on the device spatio-temporal anomaly feature data, perform adaptive spatio-temporal feature mapping in combination with the dynamic spatio-temporal adjacency matrix, execute backpropagation feature mapping weight adjustment, and set the feature embedding optimization threshold to 0.01 to optimize the feature embedding representation, obtaining the device abnormal spatio-temporal feature tensor and the device normal spatio-temporal feature tensor; Step S155: Merge the device abnormal spatio-temporal feature tensor and the device normal spatio-temporal feature tensor for spatio-temporal feature tensor merging, construct a spatio-temporal feature expression, and obtain the device spatio-temporal feature tensor.

4. The evaluation method for the operating state of the power transformation and distribution room equipment according to claim 1, characterized in that Step S2 is specifically as follows: Step S21: Based on the device spatio-temporal feature tensor, perform variational autoencoder latent variable decomposition, extract the potential factors of the relay protection device operation state, and execute information bottleneck constraints to generate the device latent variable feature vector; Step S22: Perform structured Bayesian network learning on the device latent variable feature vector, construct a device causal association graph model, and execute MAP device causal path weight optimization to obtain the device causal association weight matrix; Step S23: Based on the device causal association weight matrix, execute distributed policy gradient optimization and perform proximal policy reinforcement learning iteration to obtain the optimal causal path decision parameters; Step S24: Obtain historical device operation data, construct a device operation state transition model in combination with the optimal causal path decision parameters; perform causal inference on the device operation state evolution path according to the device operation state transition model, thereby generating a device causal decision graph; Step S25: Perform device state prediction based on the device causal decision graph and execute particle filter prediction error correction to obtain device operation trend prediction data.

5. The evaluation method for the operation status of the power transformation and distribution room equipment according to claim 1, wherein Step S3 is specifically as follows: Step S31: Based on the device causal decision graph and the device operation trend prediction data, identify the DC screen insulation degradation variable, and perform multi-scale wavelet decomposition to extract the degradation feature spectrum, generating the DC screen insulation degradation feature space; Step S32: Perform multi-site device distributed modeling based on the DC screen insulation degradation feature space, and execute cross-site time series feature aggregation in combination with the preset adaptive gated recurrent unit to generate the DC screen insulation degradation federated twin model; Step S33: According to the DC screen insulation degradation federated twin model, perform diffeomorphic manifold embedding mapping and execute manifold Laplacian regularization optimization to obtain the dynamic characteristic diffeomorphic mapping result; Step S34: Use the dynamic characteristic diffeomorphic mapping result to perform local geometric consistency constraints on the device operation state transition model and execute variational Bayesian inference to obtain the device credible twin verification chain.

6. The evaluation method for the operating state of the power transformation and distribution room equipment according to claim 5, characterized in that, Step S31 is specifically as follows: Step S311: Extract the DC screen operation trend from the device operation trend prediction data, and set the voltage fluctuation threshold to 2%, the current fluctuation threshold to 3%, and the abnormal determination threshold to 5% to evaluate the DC screen operation performance, obtaining the DC screen operation performance trend data; Step S312: Screen the DC panel insulation performance degradation factors for the DC panel operating performance trend data according to the equipment causal decision map, set the causal decision factor screening threshold to 0.1, and perform contribution evaluation, select the top 5 causal decision factors in contribution ranking to construct the DC panel insulation state variable set; Step S313: Based on the dynamic spatiotemporal adjacency matrix, a correlation analysis is performed on the set of DC screen insulation state variables, the Pearson correlation coefficient threshold is set to 0.6, and variables with a value less than the Pearson correlation coefficient threshold are eliminated, thereby extracting a subset of DC screen insulation degradation variables; Step S314: perform empirical mode analysis on the DC panel insulation degradation variable subset, extract the intrinsic mode function, calculate the kurtosis and energy ratio, set the degradation feature threshold to 3, the energy proportion threshold to 5%, screen out the main degradation feature components, and generate the insulation degradation feature variable matrix; Step S315: Set the decomposition level to 4, perform wavelet transform on the insulation degradation characteristic variable matrix for multi-scale time-frequency decomposition, extract the high-frequency abnormal component energy ratio greater than the high-frequency energy ratio threshold of 10%, and thus construct the DC screen insulation degradation feature space.

7. The method for evaluating the operation status of the power transformation and distribution room equipment according to claim 1, characterized in that, The reconstruction of the thermal stress field of the busbar connection point described in step S4 is specifically as follows: Based on the trusted twin verification chain of the device, the trusted operation status data and trusted environment parameter data of the device are obtained, and the busbar connection point status characteristics and on-load tap changer status characteristics in the trusted operation status data of the device are extracted; Construct a thermal stress field model of the busbar connection point based on the state characteristics of the busbar connection point and the credible environmental parameter data; Perform time-space coordinate transformation on the thermal stress field model of the busbar connection point, and set the initial scenario of the thermal stress field of the busbar connection point in combination with the trusted operation status data of the equipment; The thermal pressure field of the initial scenario of the thermal stress field of the busbar connection point is calculated through the thermodynamic model of the busbar connection point, and it is dynamically updated according to the trusted twin verification chain of the equipment to obtain the stress field distribution diagram of the busbar connection point; Identify the thermal stress concentration area based on the stress field distribution diagram of the busbar connection point and obtain the busbar thermal stress concentration area data; A non-commutative decision model is constructed based on the busbar thermal stress concentration area data and the on-load tap changer status characteristics. The initial scenario of the thermal stress field at the busbar connection point is input into the non-commutative decision model for non-variable decision analysis. The equipment operation strategy is optimized according to the results of the non-variable decision analysis to generate an anti-surge control strategy.

8. The evaluation method for the operation status of the power transformation and distribution room equipment according to claim 1, wherein The optimal switching instruction set of the compiled standby automatic switching device described in step S4 is specifically: Identify power failure modes based on the trusted operating status data of the device and establish a device operating status-failure mode mapping model; Perform fault source location analysis based on the equipment operation status-fault mode mapping model, identify faulty equipment and power supply paths, and generate preliminary fault source location data; Acquire historical switching data of the standby automatic switching device, and perform switching mode recognition on the historical switching data of the standby automatic switching device to obtain switching mode data of the standby automatic switching device; Combined with the preliminary fault source location data and the power supply fault mode, the switching strategy of the standby automatic switching device is matched with the switching mode data of the standby automatic switching device to obtain the candidate switching strategy of the standby automatic switching device; Perform simulation tests on the candidate switching strategies of the backup power supply automatic switching device, evaluate the execution efficiency of the candidate switching strategies of the backup power supply automatic switching device and the power system restoration speed, and screen out the optimal switching strategy; Based on the optimal switching strategy, compile the instruction set to obtain the optimal switching instruction set of the backup power supply automatic switching device, and upload it to the instruction management platform of the backup power supply automatic switching device to implement instruction replacement.

9. The evaluation method for the operating state of the power transformation and distribution room equipment according to claim 1, wherein The specific fault evolution prediction described in step S5 is as follows: Extract real-time operation characteristics according to the real-time operation data of the substation equipment. Set the mother wavelet to the 4th order to extract the equipment operation stress characteristics, set the load fluctuation threshold to 0.05 pu to screen the equipment load characteristics, and set the equipment operation stress characteristics to 10 min to calculate the real-time operation temperature rise characteristics, so as to obtain the equipment operation stress characteristic data, equipment load characteristic data and real-time operation temperature rise characteristic data; Based on the variable load fluctuation characteristics in the equipment load characteristic data, construct a load fluctuation viscous evolution model, perform Fourier transform to extract the load viscous component, calculate the viscous damping parameter in combination with the real-time operation temperature rise characteristic data, and set the viscous damping parameter calculation threshold to 0.02 to eliminate the invalid damping, so as to generate the load viscous characteristic matrix; Screen historical equipment fault data based on the equipment operation mode library; Perform dynamic evolution on the load viscous characteristic matrix, solve the fault induction trend in combination with the equipment operation stress characteristic data and historical equipment fault data, set the viscous fault evolution threshold to 0.1, and obtain the load viscous fault evolution path; Perform feature clustering according to the load viscous fault evolution path, divide the fault evolution stage, and perform fault matching based on the historical equipment fault data. Set the fault matching similarity threshold to identify the fault type and generate the fault evolution prediction result; Perform switching instruction adaptation matching on the fault evolution prediction result according to the optimal switching instruction set of the backup power supply automatic switching device to obtain the load viscous fault switching instruction; Based on the fault evolution prediction result, construct a viscous fault warning model, set the viscous instability threshold to 0.08, and adjust the dynamic warning strategy in combination with the equipment load characteristic data and the load viscous fault switching instruction. Use the trusted computing framework in the equipment trusted twin verification chain to construct a global operation optimization evidence chain.

10. A power transformation and distribution room equipment operation status evaluation system, characterized in that, For implementing the substation equipment operation status evaluation method as described in claim 1, the substation equipment operation status evaluation system includes: A data acquisition module, configured to acquire multi-dimensional sensing data of the substation and construct a dynamic spatio-temporal adjacency matrix; identify the spatio-temporal abnormal characteristics of the equipment according to the dynamic spatio-temporal adjacency matrix to obtain the equipment spatio-temporal characteristic tensor; An equipment operation trend deduction module, configured to perform equipment operation trend deduction based on the equipment spatio-temporal characteristic tensor to obtain equipment operation trend prediction data; A DC screen insulation performance analysis module, configured to perform DC screen insulation degradation federated twin modeling according to the equipment operation trend prediction data and perform dynamic characteristic diffeomorphic mapping to obtain the equipment trusted twin verification chain; An instruction compilation module, configured to reconstruct the thermal stress field of the bus connection point for the equipment trusted twin verification chain, analyze the tap-changing strategy, and obtain the inrush current control strategy; compile the optimal switching instruction set of the backup power supply automatic switching device based on the inrush current control strategy; A fault tracing module, which is used to obtain the real-time operation data of the power transformation and distribution room equipment, and perform fault evolution prediction on the real-time operation data of the power transformation and distribution room equipment according to the optimal switching instruction set of the backup power supply automatic switching device to generate a global operation optimization evidence chain; perform fault tracing analysis according to the global operation evidence chain, and screen high-reliability fault warning instructions to obtain the maintenance strategy of the power transformation and distribution room equipment.

Images