A substation operation state monitoring method of a power system

Abstract

The application provides a substation operation state monitoring method of a power system, and relates to the technical field of substation monitoring, which comprises the following steps: collecting real-time state data and historical state data of a substation and corresponding fault labels, taking each device of the substation as a node, taking the causal relationship between device pairs as an edge, constructing a fault causal model, training the fault causal model by using the historical state data and the corresponding fault labels, inputting the real-time state data into the trained fault causal model, and obtaining real-time fault labels of the substation. The application constructs a causal relationship model, verifies the existence of a new fault mode by constructing a causal verification graph through a nonlinear Granger causal test method, adjusts the fault causal model, improves the flexibility of the model when facing new fault modes, and achieves the purpose of improving the accuracy of substation fault diagnosis.

Description

Technical Field

[0001] This application relates to the field of substation monitoring technology, and in particular to a method for monitoring the operating status of a substation in a power system. Background Technology

[0002] In modern power systems, substations serve as core nodes for power conversion, dispatch, and distribution, and their operational safety and stability are of paramount importance. With the increasing complexity of power grid structures and the growth of load demand, traditional substation control systems face numerous challenges, such as slow response times, insufficient data processing capabilities, and limited information sharing. Therefore, improving the intelligent management level of substations and achieving efficient and accurate status monitoring and fault diagnosis has become an important research direction.

[0003] Chinese invention patent application publication number CN119651923A provides a substation control system based on cloud control and local area network. This patent uses a dynamic multivariate autoregressive tree algorithm to construct a multidimensional causal relationship graph to obtain the nonlinear dependency relationship between multiple variables in order to accurately identify potential anomalies and fault modes. It also uses a sparse distributed optimization algorithm for global load scheduling and achieves fast local response control based on a Markov random field model.

[0004] Regarding the above technical solutions, the dynamic multivariate autoregressive tree algorithm is a data-driven statistical learning method. Therefore, the multidimensional causal relationship graph obtained based on the dynamic multivariate autoregressive tree algorithm is a causal hypothesis extracted from the data pattern, and does not represent the actual causal results between the data. When encountering new fault modes, it is impossible to correctly infer the causal relationship generated by the new fault modes, which affects the accuracy of fault diagnosis. Summary of the Invention

[0005] To improve the accuracy of substation fault diagnosis, this application provides a method for monitoring the operating status of substations in a power system.

[0006] Firstly, this application provides a method for monitoring the operating status of a substation in a power system, employing the following technical solution:

[0007] A method for monitoring the operating status of a substation in a power system includes the following steps:

[0008] Data acquisition: Collect real-time status data of the substation, including real-time equipment status data, real-time equipment location, and real-time equipment health data; collect historical status data of the substation and corresponding fault tags, including historical equipment status data, historical equipment location, and historical equipment health data.

[0009] Model Construction: Using each piece of equipment in the substation as a node and the causal relationships between equipment pairs as edges, a fault causal model is constructed.

[0010] Model training includes the first judgment, the first update, the first processing, the second processing, the third processing, and the first training.

[0011] First judgment: Based on historical state data, a causal verification graph between devices is constructed using the nonlinear Granger causality test method. The causal relationships in the causal verification graph are then compared with the edges between device pairs in the fault causality model to determine whether they are consistent.

[0012] If so, proceed with the first processing step;

[0013] If not, proceed with the first update step;

[0014] First update: Based on the causal verification graph, update the fault causal model and use the updated fault causal model as the new fault causal model;

[0015] First processing: Obtain the timestamp of historical status data. Based on the fault label, merge historical status data with the same fault label and consecutive timestamps according to the order of timestamps. Record the merged historical status data as the first data.

[0016] The second processing step involves calculating the Euclidean distance between device pairs based on historical device locations, and then aggregating the Euclidean distances between each device pair to obtain the first feature set.

[0017] Third processing: Obtain the time series of the first data and determine whether the device's historical health data has changed within the time series:

[0018] If not, proceed with the first training step;

[0019] If so, the time points of the device's historical health data changes are obtained, the first data is segmented based on the obtained time points to obtain the segmented first data, and the segmented first data is updated as the first data.

[0020] First training: The first feature set is concatenated with the first data, the concatenated data is updated with the first data, and the first data and the corresponding fault labels are used to train the new fault causal model to obtain the trained fault causal model.

[0021] Fault identification: Input real-time status data into the trained fault causal model to obtain real-time fault labels for the substation.

[0022] By adopting the above technical solutions and constructing a causal verification graph between devices using the nonlinear Granger causality test method, not only can linear causal relationships between devices be captured, but more complex nonlinear causal relationships can also be identified. When the causal relationships in the model are inconsistent with those in the causal verification graph, the fault causal model is dynamically adjusted, improving the model's flexibility in the face of new fault modes. Furthermore, merging historical state data based on timestamps to form corresponding time series helps the model capture the occurrence patterns and temporal dynamics of faults when facing new faults, thereby improving the accuracy of substation fault diagnosis. Moreover, segmenting historical state data based on equipment lifecycle data helps the model adapt to new fault modes occurring at different stages of the equipment's lifecycle, further improving the accuracy of model fault diagnosis.

[0023] Optionally, after performing the model building step and before performing the model training step, the following steps are also included:

[0024] First construction: Denote the nodes in the fault causal model as the first node, the fault causal model as the first causal layer, obtain the functional attributes of each device in the substation, classify the devices based on the functional attributes to obtain several subsystems, and construct the second causal layer with the subsystems as nodes and the causal relationship between subsystems as edges.

[0025] Second construction: The nodes of the second causal layer are denoted as the second node, and the substations are used as nodes and the causal relationships between substations are used as edges to construct the third causal layer;

[0026] Third construction: Record the nodes of the third causal layer as the third node, and update the first, second, and third causal layers to the fault causal model.

[0027] By adopting the above technical solution, hierarchical modeling is performed from equipment and subsystems to the entire substation, expanding the fault diagnosis problem of the entire substation from the single equipment level to a higher systemic level. This allows the model to capture fault modes at each level more meticulously. It helps the model to infer and analyze complex systemic faults or novel fault modes through hierarchical causal relationships, discovering potential novel fault modes across subsystems and equipment, and reducing the possibility of reduced diagnostic accuracy due to the failure to capture novel fault modes. Furthermore, setting up a hierarchical fault diagnosis model not only enhances the model's ability to identify novel local faults but also helps the model predict the chain reactions caused by equipment failures, achieving the technical effect of providing early warning before novel fault modes occur and improving the response speed of substation fault diagnosis.

[0028] Optionally, after performing the third construction step and before performing the model training step, the following steps are also included:

[0029] First transformation: Based on the first causal layer, each device is taken as a node of the first graph, the causal relationship between each pair of devices is transformed into an embedding vector, and the obtained embedding vector is used as the node feature of the corresponding first graph node to construct the first causal graph;

[0030] Second transformation: Based on the second causal layer, the subsystem is used as the node of the second graph. The embedding vector of the first causal graph is used to calculate the embedding vector between the subsystem pairs. The obtained embedding vector is used as the node feature of the second graph node to construct the second causal graph.

[0031] Third transformation: Based on the third causal layer, the substation is taken as the node of the third graph. The embedding vector of the second causal graph is used to calculate the embedding vector between substation pairs. The obtained embedding vector is used as the node feature of the third graph node to construct the third causal graph.

[0032] Fourth construction: Based on the first, second, and third causal graphs, a multi-causal relationship graph is constructed, and the constructed multi-causal relationship graph is used as a new fault causal model;

[0033] In the first update step, a new fault causal model is updated based on the causal verification graph.

[0034] By adopting the above technical solution, causal relationships are transformed into embedded vectors. Using these embedded vectors as features of graph nodes helps capture complex nonlinear relationships between them. Furthermore, by propagating the embedded vectors layer by layer, the technical effect of gradually establishing complex causal chains between equipment, subsystems, and substations is achieved. This enhances the model's ability to identify novel fault modes, enabling it to infer potential causal relationships from existing data when faced with new and complex fault modes through a multi-layered embedded vector propagation mechanism. This helps improve the model's diagnostic accuracy for novel fault modes. In addition, integrating the causal graphs of each layer into a multi-causal graph helps to grasp the complex causal relationships between different levels from a global perspective, avoiding information fragmentation between levels, further enhancing the model's ability to identify complex fault modes and improving its diagnostic accuracy for novel fault modes.

[0035] Optionally, after performing the first judgment step and before performing the first update step, the method further includes:

[0036] Fourth processing: Add lag features to the historical state data to obtain the first dataset;

[0037] Fifth construction: Construct a causal forest, train the causal forest using historical state data to obtain the trained causal forest, input the first dataset and the causal verification graph into the trained causal forest to obtain the causal relationship prediction results between each device pair;

[0038] Second judgment: Determine whether the predicted causal relationship between each pair of devices is completely consistent with the edges between the node pairs in the first graph of the fault causal model.

[0039] If so, proceed with the first processing step;

[0040] If not, then based on the causal relationship prediction results between each device pair, a new causal verification graph is constructed, the new causal verification graph is updated to the causal verification graph, and the first update step is performed.

[0041] By adopting the above technical solution, lag features are added to historical state data, enabling causal forests to not only predict causal relationships based on the device state at a certain moment, but also to infer potential causal relationships between devices based on state changes before a certain moment. This provides causal forests with time series context information, improves the accuracy of causal relationships inferred between devices, and facilitates further fault mode judgment.

[0042] Optionally, after the step of performing the first judgment and before the step of performing the first processing, the method further includes:

[0043] The third judgment: Based on historical state data, a causal verification graph between each subsystem pair is constructed using the nonlinear Granger causality test method, denoted as the first verification graph. The judgment is made as to whether the causal relationship between each subsystem pair in the first verification graph is consistent with the edge between the second node pair in the fault causality model.

[0044] If so, proceed with the fourth judgment step;

[0045] If not, then use the causal verification graph and the first verification graph as the new causal verification graph, and perform the first update step;

[0046] Fourth Judgment: Based on historical state data, a causal verification graph between each substation pair is constructed using the nonlinear Granger causality test method, denoted as the second verification graph. The judgment is then made as to whether the causal relationships between each substation pair in the second verification graph are consistent with the edges between the third node pairs in the fault causality model.

[0047] If so, proceed with the first processing step;

[0048] If not, update the causal verification graph, the first verification graph, and the second verification graph to the causal verification graph, and perform the first update step.

[0049] By adopting the above technical solutions, the causal relationships between each layer are verified step by step, from equipment to subsystems and then to substations. The refined causal verification graph helps to more accurately identify new fault modes, reducing the occurrence of misdiagnosis and missed diagnosis. Furthermore, the verification of multi-level causal relationships enhances the model's adaptability to new fault modes, reduces the limitations caused by fixed causal graphs, improves the flexibility and scalability of model fault diagnosis, and simultaneously increases the speed of model fault detection and response.

[0050] Optionally, after performing the first update step and before performing the first processing step, the method further includes:

[0051] First data collection: Input the first dataset and the causal verification graph into the trained causal forest to obtain the confidence of the causal relationship between node pairs in the first graph;

[0052] First annotation: Based on the fault causal model before the update, annotate the added and deleted causal relationships in the updated fault causal model, and collect the annotated causal relationships as the updated causal relationships;

[0053] First screening: Obtain the confidence scores for updating causal relationships to obtain the confidence score dataset;

[0054] First calculation: Obtain the first graph node pair corresponding to the updated causal relationship, calculate the embedding vector of the first graph node pair corresponding to the updated causal relationship before and after the update, calculate the difference between the embedding vector before and after the update based on the calculated embedding vector, and use the calculated difference as the vector change amount of the updated causal relationship.

[0055] The second calculation involves weighting and summing the vector changes in the updated causal relationship with the corresponding confidence levels, and using the result as the degree of correlation in the updated causal relationship.

[0056] First criterion: Determine whether the correlation of the updated causal relationship is higher than the preset correlation threshold.

[0057] If so, proceed to the second judgment step;

[0058] If not, proceed to the third judgment step;

[0059] Second criterion: Determine whether the current updated causal relationship is an added causal relationship:

[0060] If so, then no action will be taken;

[0061] If not, then add the causal relationship of the first graph node pair corresponding to the current updated causal relationship;

[0062] Third criterion: Determine whether the current update causal relationship is a deletion causal relationship:

[0063] If so, then no action will be taken;

[0064] If not, then delete the causal relationship of the first graph node pair corresponding to the currently updated causal relationship.

[0065] By adopting the above technical solution, the changes in causal relationships before and after the model update are quantified based on the differences in embedding vectors. Combined with the confidence level of the causal relationships, the importance of the causal relationships in the model is calculated. This helps reduce noise and redundant information introduced by the model update, optimizes the modeling of causal relationships, improves the model's interpretability and the effectiveness of dynamic updates, and enhances the model's accuracy and robustness. Furthermore, the aforementioned dynamic adjustment and verification mechanism helps enhance the model's adaptability, enabling it to promptly detect new fault modes caused by faults or changes in the operating environment in substations, thus improving the accuracy and effectiveness of fault diagnosis in complex and dynamic environments.

[0066] Optionally, each device in the substation includes a device ID, and after the first processing step and before the second processing step, the following steps are also included:

[0067] Second labeling: Based on each device in the substation, label the collected historical status data with the corresponding device ID, and record the labeled data as the new first data;

[0068] First association: Obtain the edges between each pair of nodes, and convert the edges between each pair of nodes into causal relationship strength values, which are denoted as the first feature;

[0069] Location encoding: Based on the timestamps of historical state data, location encoding is performed on the timestamps of each node, and the encoded timestamps are recorded as time features;

[0070] Data update: The time feature, the first feature, and the new first data are concatenated, and the concatenated data is updated to the first data.

[0071] By employing the aforementioned technical solutions, causal relationships are quantified. Based on the quantified features, the model's ability to capture causal chains within substations is enhanced. Furthermore, adding location encoding to each node based on the timestamps of historical state data helps the model better understand the temporal sequence and time dependencies of time-series data, improving its ability to capture temporal trends and periodic changes, and enhancing its processing capabilities for time-series data. In addition, the concatenation of quantified causal relationships and temporal features enhances the expressive power of the feature space of historical state data, improving the model's ability to capture new fault modes and potential complex relationships when facing varied equipment fault patterns, thus improving the accuracy of substation fault diagnosis.

[0072] Optionally, after the second processing step and before the third processing step, the process further includes:

[0073] First acquisition: Acquire the environmental parameters of each device's location;

[0074] Difference calculation: Calculate the difference in environmental parameters between each pair of devices;

[0075] Third calculation: Based on the difference in environmental parameters and Euclidean distance between each pair of devices, the environmental spatial difference formula is used to calculate the environmental spatial difference between each pair of devices, and the calculated environmental spatial difference between each pair of devices is used as the new first feature set.

[0076] The calculation model for the environmental spatial difference formula is as follows:

[0077] ;

[0078] in, The difference in environmental space between device i and device j. As a regulating factor, The difference in environmental parameters between device i and device j. Let be the Euclidean distance between device i and device j.

[0079] By adopting the above technical solution, the environmental parameter differences and Euclidean distance between each device are combined to construct environmental spatial difference features. This helps the model consider the influence of the environment and spatial location of the device on its behavior and state, enhances the model's ability to learn spatial correlation and environmental factors, improves the model's ability to capture the potential impact of environmental factors on equipment failure, helps the model automatically adapt to different operating environments and working conditions during training, improves the accuracy and stability of the model's prediction ability for various types of equipment under different environments and spaces, and thus improves the comprehensiveness and accuracy of the model in diagnosing multiple failure modes.

[0080] Optionally, the historical status data also includes historical device usage duration, historical load data, and historical resource data, and after the second processing step and before the third processing step, it also includes:

[0081] First, we collect historical load data and maximum load values ​​for each device, and calculate the total duration for which historical load data exceeds the maximum load value, the difference between historical load data exceeding the maximum load value and the maximum load value, and the corresponding duration.

[0082] Second statistics: Obtain historical resource data and minimum resource requirement value for each device, and calculate the total duration for which historical resource data for each device is lower than the minimum resource requirement value, the difference between historical resource data and minimum resource requirement value, and the corresponding duration.

[0083] Load calculation: The load impact formula is used to calculate the impact factor of load fluctuation on the equipment status of each device, which is denoted as the first impact.

[0084] Resource calculation: The resource impact formula is used to calculate the impact factor of resource fluctuations on the equipment status of each device, which is denoted as the second impact;

[0085] Fourth calculation: Based on the equipment's historical health data, the first influence and the second influence, the actual health data of the equipment is calculated using the equipment health formula, and the calculated actual health data is used as the equipment's new historical health data;

[0086] The calculation model for the load impact formula is as follows:

[0087] ;

[0088] in, Let 'a' be the number of historical load data points that exceeded the maximum load value, and 'a' be the 'a'th historical load data point that exceeded the maximum load value. This represents the difference between the historical load data point that exceeds the maximum load value and the maximum load value for the a-th instance. This represents the maximum load value of the device. The duration of the historical load data that exceeds the maximum load value for the *a*th time. This refers to the historical usage duration of the equipment.

[0089] The calculation model for the resource impact formula is as follows:

[0090] ;

[0091] in, Let be the number of historical resource data points that are below the minimum resource requirement value, and let b be the b-th historical resource data point that is below the minimum resource requirement value. The difference between the b-th historical resource data point that is lower than the minimum resource requirement and the minimum resource requirement. This represents the minimum resource requirements for the equipment. The duration of the b-th historical resource data point that is below the minimum resource requirement value;

[0092] The calculation model for the equipment health formula is as follows:

[0093] ;

[0094] in, Assign weights to the impact of load fluctuations on equipment status. The weighting of the impact of resource fluctuations on equipment status. This refers to the device's historical health data.

[0095] By adopting the above technical solution, the dynamic characteristics of load fluctuations and resource fluctuations are introduced, the impact of load fluctuations and resource fluctuations on equipment health is quantified, and the model's ability to handle various load and resource fluctuations is improved.

[0096] The adaptability of equipment under the same conditions helps to enhance the model's fault prediction capability under unstable operating conditions. In addition, based on the dynamic characteristics of the introduced load fluctuations and resource fluctuations, and considering the situation where equipment is under abnormal load and resource conditions for a long time, the model's ability to identify potential failure modes of equipment under abnormal operating conditions for a long time is enhanced, the model's sensitivity to changes in equipment state is improved, and the fault prediction accuracy of substations is enhanced.

[0097] Optionally, after performing the third processing step and before performing the first training step, the following steps are also included:

[0098] Second acquisition: Acquire the time points of historical health data changes of the device, and record them as the first time point;

[0099] The third acquisition: acquire the device's historical health data at the time point before the first time point, and record it as the first device data; acquire the device's historical health data at the time point after the first time point, and record it as the second device data;

[0100] Rate calculation: Calculate the time difference between the time point before the first time point and the time point after the first time point. Based on the calculated time difference, the first device data, and the second device data, calculate the rate of change of the device's historical health data, which is denoted as the first rate of change.

[0101] Fifth judgment: Determine whether the first rate of change is less than a preset rate threshold.

[0102] If so, the first data segmented based on the first time point will be merged, and the merged data will be updated to the first data.

[0103] If not, no action will be taken.

[0104] By adopting the above technical solution and introducing dynamic features that reflect the rate of change in equipment health, it helps to improve the model's ability to capture long-term trend changes in equipment, reduce training errors caused by noisy data or irrelevant changes, thereby enhancing the model's ability to predict equipment status and further improving the model's accuracy in identifying substation faults.

[0105] In summary, this application includes at least one of the following beneficial technical effects:

[0106] 1. A nonlinear Granger causality test is used to construct a causal verification graph between devices. This not only captures linear causal relationships between devices, but also identifies more complex nonlinear causal relationships. When the causal relationships in the model are inconsistent with those in the causal verification graph, the fault causal model is dynamically adjusted, which improves the flexibility of the model when facing new fault modes.

[0107] 2. Transforming causal relationships into embedded vectors and using these vectors as features of graph nodes helps capture complex nonlinear relationships between them. Furthermore, by propagating the embedded vectors layer by layer, the model achieves the technical effect of gradually establishing complex causal chains between equipment, subsystems, and substations. This enhances the model's ability to identify novel fault modes, enabling it to infer potential causal relationships from existing data through a multi-layer embedded vector propagation mechanism when facing new and complex fault modes. This helps improve the model's diagnostic accuracy for novel fault modes.

[0108] 3. Based on the difference in embedding vectors before and after the update, the changes in causal relationships before and after the model update are quantified. Combined with the confidence of causal relationships, the importance of causal relationships in the model is calculated. This helps to reduce noise and redundant information introduced by model updates, optimize the modeling of causal relationships, improve the explanatory power and effectiveness of dynamic updates of the model, and improve the accuracy and robustness of the model.

[0109] 4. By introducing the dynamic characteristics of load fluctuations and resource fluctuations, the impact of load fluctuations and resource fluctuations on equipment health is quantified, which improves the model's adaptability to equipment under different conditions, helps to enhance the model's fault prediction capability under unstable operating conditions, improves the model's sensitivity to changes in equipment status, and enhances the fault prediction accuracy of substations. Attached Figure Description

[0110] Figure 1 This is a flowchart of Embodiment 1 of this application;

[0111] Figure 2 This is a flowchart of the S3 model training in Embodiment 1 of this application;

[0112] Figure 3 This is a flowchart of the first reconstruction (S21) in Embodiment 2 of this application;

[0113] Figure 4 This is a flowchart of the second reconstruction in S22 of Embodiment 4 of this application;

[0114] Figure 5 This is a flowchart of S321 update verification in Embodiment 4 of this application. Detailed Implementation

[0115] The following combination Figures 1 to 5 This application will be described in further detail.

[0116] Example 1: This example discloses a method for monitoring the operating status of a substation in a power system, such as... Figure 1 As shown, the method includes: collecting real-time and historical status data of the substation and corresponding fault labels; constructing a fault causal model using each device in the substation as a node and the causal relationship between device pairs as an edge; training the fault causal model using historical status data and corresponding fault labels to obtain the trained fault causal model; and inputting real-time status data into the trained fault causal model to obtain real-time fault labels for the substation. This embodiment includes the following steps:

[0117] S1 Data Acquisition: Acquires real-time status data of the substation; acquires historical status data of the substation and corresponding fault tags. Historical status data includes historical equipment status data, historical equipment location, and historical health data of the equipment.

[0118] Real-time status data includes real-time device status data, real-time device location, real-time device health data, real-time device usage time, real-time load data, and real-time resource data. Real-time device status data includes real-time device switch status data, real-time device operating mode, real-time device electrical parameter data, and real-time device communication status. Real-time electrical parameter data includes real-time device current, real-time device voltage, real-time device frequency, and real-time device power. Real-time device location includes real-time device geographic location and real-time device configuration information, where the real-time configuration information is the device's location within the substation system and its connection relationships with other devices. Each device is assigned a device ID. The real-time load data includes, but is not limited to, real-time calculated load, real-time working load, and real-time energy load. The real-time resource data includes, but is not limited to, real-time temperature, real-time power capacity, real-time power network traffic, and real-time power consumption. The real-time device health data represents the device's current remaining lifespan.

[0119] Historical status data includes historical equipment status data, historical equipment location, historical equipment health data, historical equipment usage duration, historical load data, and historical resource data. Historical equipment status data includes historical equipment switch status data, historical equipment operating modes, historical equipment electrical parameter data, and historical equipment communication status. Historical electrical parameter data includes historical equipment current, historical equipment voltage, historical equipment frequency, and historical equipment power. Historical equipment location includes historical equipment geographical location and historical equipment configuration information. The historical load data includes, but is not limited to, historical computing load, historical working load, and historical energy load. The historical resource data includes, but is not limited to, historical temperature, historical power capacity, historical power network traffic, and historical electricity consumption. The historical health data of the equipment represents its historical remaining lifespan.

[0120] The fault labels include, but are not limited to, short circuit faults, overload faults, overvoltage faults, low voltage faults, grounding faults, current imbalance, power outage faults, equipment mechanical faults, equipment over-temperature faults, low oil level faults, vibration faults, communication loss, circuit breaker faults, power distribution line faults, frequency fluctuation faults, low power factor faults, and harmonic faults.

[0121] In this embodiment, the collected data needs to be preprocessed. The data preprocessing steps include data cleaning and data normalization.

[0122] Data cleaning: Performing operations such as noise reduction, handling missing values, and removing outliers on the collected data.

[0123] Data normalization: Normalize the collected data.

[0124] S2 Model Construction: Using each device in the substation as a node and the causal relationship between device pairs as an edge, a fault causal model is constructed.

[0125] In this embodiment, the basic model of the fault causal model can be a Bayesian network model. Using each device in the substation as a node and the substation's status data as node features, causal relationships between devices are established through a physical model and current domain expert experience. The causal relationships between device pairs are then used as edges to construct the fault causal model. This fault causal model has the function of outputting real-time fault labels based on the input real-time substation status data.

[0126] Model training in S3 includes S31 first judgment, S32 first update, S33 first processing, S34 feature concatenation, S35 second processing, S36 first compensation, S37 second compensation, S38 third processing, S39 first verification, and S310 first training. Figure 2 As shown.

[0127] S31 First Judgment: Based on historical state data, a causal verification graph between each device is constructed using the nonlinear Granger causality test method. The causal relationship between each pair of devices in the causal verification graph is matched one-to-one with the edges between each pair of devices in the fault causality model. The causal relationship in the causal verification graph is then judged to be consistent with the edges between each pair of devices in the fault causality model.

[0128] If so, then execute S33, the first process.

[0129] If not, then execute the first update of S32.

[0130] Nonlinear Granger causality tests include the Hiemstra-Jones test, the Diks-Panchenko test, and neural network-based tests. In this embodiment, a neural network-based test is used to construct a causal verification graph between the devices. The specific steps are as follows:

[0131] Using a multilayer perceptron as the base model, experimental and control models were constructed. All device pair combinations were obtained, with one pair selected as device A and device B. Historical state data for both device A and device B were acquired. The historical state data of device B was labeled with experimental tags for the historical state data of device A. The labeled historical state data of device A was input into the experimental model, and the experimental fault prediction result for device B was output. The error of the experimental model was calculated based on the experimental fault prediction result of device B and its true tag, and this calculated error was recorded as the experimental error. The historical state data of device B was input into the control model, and the control fault prediction result for device B was output. The error of the control model was calculated based on the control fault prediction result of device B and its true tag, and this calculated error was recorded as the control error. It was determined whether the experimental error of device B was less than the control error of device B: if so, a causal relationship was determined between device A and device B; otherwise, it was not. The above operations were repeated to obtain the causal relationship test results for all device pair combinations. A causal verification graph was constructed based on the causal relationship test results for all device pair combinations.

[0132] In the above steps, within the experimental model, the gradient between the input data of device A and the output data of device B is calculated using the backpropagation algorithm. This calculated gradient is used as the causal strength value between device A and device B. The causal strength values ​​for all device pair combinations are then calculated based on these steps. When constructing the causal verification graph, the corresponding causal strength values ​​are labeled for the causal relationships of all device pair combinations.

[0133] S32 First Update: Based on the causal relationships in the causal verification graph, the edges in the fault causal model are added and deleted to make the edges between each pair of devices in the processed fault causal model consistent with the causal relationships in the causal verification graph, and the processed fault causal model is updated to the fault causal model.

[0134] S33 First Processing: Obtain the timestamps of historical status data. Based on the fault tag, merge historical status data with the same fault tag and consecutive timestamps according to the order of timestamps. Record the merged historical status data as the first data. Example: A set of historical status data tables is shown in Table 1.

[0135] Serial Number time Historical status data Fault Label 1 2-11 10:00 Historical Status Data A Short circuit fault 2 2-11 11:00 Historical state data B normal 3 2-11 12:00 Historical state data C normal 4 2-11 13:00 Historical state data D Low voltage fault 5 2-11 14:00 Historical state data E Low voltage fault

[0136] Among them, the fault labels of the historical status data with serial number 2 and serial number 3 are consistent, and the timestamps are continuous. The fault labels of the historical status data with serial number 4 and serial number 5 are consistent, and the timestamps are continuous. Therefore, the historical status data with serial number 2 and serial number 3 are merged, and the historical status data with serial number 4 and serial number 5 are merged. All the merged historical status data is recorded as the first data. The data in the above historical status data table are merged to obtain the first data table, as shown in Table 2.

[0137] Serial Number time Historical status data Fault Label 1 2-11 10:00 Historical Status Data A Short circuit fault 2 2-11 11:00-12:00 Historical state data B Historical state data C normal 3 2-11 13:00-14:00 Historical state data D Historical state data E Low voltage fault

[0138] S34 Feature Concatenation: Includes S341 Second Annotation, S342 First Association, S343 Position Encoding, and S344 Data Update.

[0139] S341 Second annotation: Based on each device in the substation, obtain the device ID of each device, associate the historical status data of the same device from different sources with the device ID, and annotate the historical status data of the device with the corresponding device ID, and record the annotated data as the new first data.

[0140] S342 First Association: Obtain the edges between each pair of nodes. Based on the causal verification graph, obtain the causal relationship strength value corresponding to the edge between each pair of nodes. For each device, construct the causal relationship strength value matrix between the device and all other devices. Record the constructed matrix as the first matrix of the device. Collect the first matrices of all devices to obtain the first feature.

[0141] S343 Location Encoding: Based on the timestamps of historical status data, extract the device ID and the corresponding timestamps of historical status data for each device, perform location encoding on the timestamps of historical status data for each device ID, store the encoded data in the time feature matrix corresponding to each device, and collect the time feature matrices of all devices to obtain the time features.

[0142] S344 Data Update: The obtained time feature, the first feature, and the new first data are added and concatenated, and the concatenated data is updated to the first data.

[0143] S35 Second Processing: Based on the historical device locations, the Euclidean distance between each pair of devices is calculated using the Euclidean distance formula. The Euclidean distances between each pair of devices are then aggregated to obtain the first feature set.

[0144] In this embodiment, the location of each device in the substation is fixed by default, so the historical device locations in each historical status data are the same.

[0145] S36 First Compensation: Includes S361 First Acquisition, S362 Difference Calculation and S363 Third Calculation.

[0146] S361 First Acquisition: Acquire the environmental parameters of the location of each device, including temperature and humidity.

[0147] S362 Difference Calculation: Based on the temperature and humidity of each device's location, calculate the temperature difference and humidity difference between each pair of devices.

[0148] S363 Third Calculation: Based on the temperature difference and humidity difference between each device pair and the Euclidean distance, the environmental spatial difference formula is used to calculate the environmental spatial difference between each device pair, and the calculated environmental spatial difference between each device pair is used as a new first feature set. In this embodiment, the environmental spatial difference between each device pair includes temperature spatial difference and humidity spatial difference.

[0149] The calculation model for the environmental spatial difference formula is as follows:

[0150] ;

[0151] in, For equipment With equipment Environmental spatial differences between them As a regulating factor, For equipment With equipment The difference in environmental parameters between them For equipment With equipment The Euclidean distance between them. Adjustment factor. The value of is obtained through cross-validation calculation. In this embodiment, =0.5.

[0152] For example, if the temperature difference between device A and device B is set to -2℃, the humidity difference between device A and device B is set to 5%, and the Euclidean distance between device A and device B is 14.14, then the spatial temperature difference between device A and device B is... The difference in humidity between device A and device B is: The difference in the ambient space between device A and device B is [ ]. , ].

[0153] In this embodiment, the humidity is expressed as a percentage. When performing calculations, the humidity expressed as a percentage is converted to a decimal form to participate in the calculation of the environmental space difference formula.

[0154] S37 Second Compensation: Includes S371 First Statistics, S372 Second Statistics, S373 Load Calculation, S374 Resource Calculation and S375 Fourth Calculation.

[0155] S371 First Statistics: Obtain historical load data and maximum load value for each device, and calculate the total duration for which historical load data exceeds the maximum load value, the difference between historical load data exceeding the maximum load value and the maximum load value, and the corresponding duration.

[0156] S372 Second Statistics: Obtain historical resource data and minimum resource requirement value for each device, and calculate the total duration for which historical resource data for each device is lower than the minimum resource requirement value, the difference between historical resource data and the minimum resource requirement value, and the corresponding duration.

[0157] S373 Load Calculation: The load impact formula is used to calculate the impact factor of load fluctuation on the equipment status of each device, which is denoted as the first impact.

[0158] S374 Resource Calculation: The resource impact formula is used to calculate the impact factor of resource fluctuations on the equipment status of each device, which is denoted as the second impact.

[0159] S375 Fourth Calculation: Based on the equipment's historical health data, first influence, and second influence, the equipment health formula is used to calculate the equipment's actual health data, and the calculated actual health data is used as the equipment's new historical health data.

[0160] S375 Fourth Calculation: Based on the equipment's historical health data, first influence, and second influence, the equipment health formula is used to calculate the equipment's actual health data, and the calculated actual health data is used as the equipment's new historical health data.

[0161] The calculation model for the load impact formula is as follows:

[0162] .

[0163] in, Let 'a' be the number of historical load data points that exceeded the maximum load value, and 'a' be the 'a'th historical load data point that exceeded the maximum load value. This represents the difference between the historical load data point that exceeds the maximum load value and the maximum load value for the a-th instance. This represents the maximum load value of the device. The duration of the historical load data that exceeds the maximum load value for the *a*th time. This refers to the historical usage time of the device.

[0164] The calculation model for the resource impact formula is as follows:

[0165] .

[0166] in, Let be the number of historical resource data points that are below the minimum resource requirement value, and let b be the b-th historical resource data point that is below the minimum resource requirement value. The difference between the b-th historical resource data point that is lower than the minimum resource requirement and the minimum resource requirement. This represents the minimum resource requirements for the equipment. The duration of the b-th historical resource data point below the minimum resource requirement value.

[0167] The calculation model for the equipment health formula is as follows:

[0168] .

[0169] in, Assign weights to the impact of load fluctuations on equipment status. The weighting of the impact of resource fluctuations on equipment status. This refers to the device's historical health data. Specifically, regression analysis was performed using historical status data to obtain... and In this embodiment, the value of is... =0.6, .

[0170] Example: Set the historical health data of device A to 2000 hours, and the historical usage duration of device A to 5 hours (this is just an example, listing only a small portion of the historical status data; the first sampling time is assumed to be the initial usage time of the device, so the historical usage duration of device A is the interval between the first and last sampling times). The maximum load value of device A is 100L, and the minimum resource requirement of device A is 40R. Set the sampling time and interval of the historical load data to be consistent with the sampling time and interval of the historical status data. See Table 3 for the historical load data and historical resource data of device A.

[0171] time Historical load data (L) Historical resource data (R) 2-11 10:00 80 50 2-11 11:00 120 40 2-11 12:00 110 35 2-11 13:00 130 45 2-11 14:00 90 60 2-11 15:00 140 30

[0172] According to Table 3 above, the number of historical load data points exceeding the maximum load value for device A is 4. Therefore, the first impact of device A is... .

[0173] If the number of historical resource data points for device A that are below the minimum resource requirement is 2, then the first impact of device A is: .

[0174] Therefore, the actual health data of device A is =1999.85 Therefore, the new historical health data for device A is 1999.85h.

[0175] S38 Third Processing: Obtain each time series in the first data and determine whether the device's historical health data has changed within the time series.

[0176] If not, then perform the first training step of S310.

[0177] If so, the time point of the historical health data change of the device is obtained, the first data is segmented based on the obtained time point to obtain the segmented first data, the segmented first data is updated as the first data, and then S39 first verification is executed.

[0178] S39 First Verification: Includes S391 Second Acquisition, S392 Third Acquisition, S393 Rate Calculation, and S394 Fifth Judgment.

[0179] S391 Second Acquisition: Acquire the time points of historical health data changes of the device, denoted as the first time point.

[0180] S392 Third Acquisition: Acquire the device's historical health data from the time point preceding the first time point, denoted as the first device data. Acquire the device's historical health data from the time point following the first time point, denoted as the second device data.

[0181] S393 Rate Calculation: Calculate the time difference between the time point before the first time point and the time point after the first time point. Based on the calculated time difference, the first device data, and the second device data, calculate the rate of change of the device's historical health data, which is denoted as the first rate of change.

[0182] S394 Fifth Judgment: Determine whether the first rate of change is less than the preset rate threshold.

[0183] If so, the first data segmented based on the first time point will be merged, the merged data will be updated to the first data, and the first training of S310 will be executed.

[0184] If not, then perform the first training step of S310.

[0185] S310 First Training: The first feature set and the first data are added and concatenated. The concatenated data is updated to the first data. The first data and the corresponding fault labels are used to train the new fault causal model to obtain the trained fault causal model.

[0186] S4 Fault Identification: Input real-time status data into the trained fault causal model to obtain real-time fault labels for the substation.

[0187] In this embodiment, a nonlinear Granger causality test is used to construct a causal verification graph between devices. This not only captures linear causal relationships between devices but also identifies more complex nonlinear causal relationships. When the causal relationships in the model are inconsistent with those in the causal verification graph, the fault causal model is dynamically adjusted, improving the model's flexibility in the face of new fault modes. Furthermore, the introduction of dynamic characteristics of load and resource fluctuations quantifies their impact on device health, enhancing the model's adaptability to different conditions. This helps strengthen the model's fault prediction capability under unstable operating conditions, improves its sensitivity to changes in device state, and enhances the fault prediction accuracy of the substation.

[0188] Example 2: The difference from Example 1 is that:

[0189] like Figure 3 As shown, after performing S2 to build the model and before performing S3 to train the model, there is also S21 first reconstruction, which includes S211 first construction, S212 second construction and S213 third construction.

[0190] S211 First Construction: The nodes in the fault causal model are designated as the first node, and the fault causal model itself is designated as the first causal layer. The functional attributes of each device in the substation are obtained, and the devices are classified based on these attributes, resulting in several subsystems. In this embodiment, the devices are classified based on their functional attributes to obtain the following subsystems: power transmission subsystem (e.g., busbars, cables, overhead lines), switchgear subsystem (e.g., circuit breakers, disconnectors), protection subsystem (e.g., relay protection devices, surge arresters), transformer subsystem (e.g., transformers, reactors), and measurement and monitoring subsystem (e.g., current transformers, voltage transformers). Based on the causal relationships between the devices in the first causal layer, the causal strength between each device is obtained. In this embodiment, since the first causal layer sets the causal relationships between devices based on the physical model and the experience of current domain experts, the causal strength between each device is also known. A device causal matrix is ​​constructed based on the causal strength between each device. Based on the constructed device causal matrix and the devices contained in each subsystem, the causal strength of the devices contained in each subsystem is collected as the causal strength between subsystem pairs. The causal relationship and causal strength between each subsystem pair are obtained. The second causal layer is constructed with subsystems as nodes and the causal relationship between subsystem pairs as edges.

[0191] S212 Second Construction: The nodes of the second causal layer are denoted as the second node. Taking the substation as the node, the causal relationship between substation pairs is set based on the experience of experts in the current domain. The causal relationship between substation pairs is used as the edge to construct the third causal layer.

[0192] S213 Third Construction: Record the nodes of the third causal layer as the third node, and update the first, second, and third causal layers to the fault causal model.

[0193] In this embodiment, the causal relationships are modeled hierarchically from equipment and subsystems to the entire substation. This expands the fault diagnosis problem of the entire substation from the level of a single equipment to a higher level of systemic issues. This allows the model to capture fault modes at each level more meticulously. It helps the model to infer and analyze complex systemic faults or new fault modes through hierarchical causal relationships, discover potential new fault modes across subsystems and equipment, and reduce the situation where the diagnostic accuracy is reduced due to the failure to capture new fault modes.

[0194] Example 3: The difference from Example 2 is that:

[0195] After executing the first judgment in S31 and before executing the first processing in S33, there is also the model judgment in S311, which includes the third judgment in S3111 and the fourth judgment in S3112.

[0196] S3111 Third Judgment: Based on historical state data, a causal verification graph between each subsystem pair is constructed using the nonlinear Granger causality test method, denoted as the first verification graph. It is then determined whether the causal relationship between each subsystem pair in the first verification graph is consistent with the edge between the second node pair in the fault causality model.

[0197] If so, then execute the fourth judgment in S3112.

[0198] If not, then use the causal verification graph and the first verification graph as the new causal verification graph, and perform the first update in S32.

[0199] S3112 Fourth Judgment: Based on historical state data, a causal verification graph between each substation pair is constructed using the nonlinear Granger causality test method, denoted as the second verification graph. The judgment is made as to whether the causal relationships between each substation pair in the second verification graph are consistent with the edges between the third node pairs in the fault causality model.

[0200] If so, then execute S33, the first process.

[0201] If not, update the causal verification graph, the first verification graph, and the second verification graph to the causal verification graph, and execute S32 First Update.

[0202] In this embodiment, the causal relationships between each layer are verified progressively, from equipment to subsystems and then to the substation. A detailed causal verification graph helps to more accurately identify new fault modes, reducing misdiagnosis and missed diagnosis. Furthermore, the verification of multi-level causal relationships enhances the model's adaptability to new fault modes, reduces the limitations of fixed causal graphs, improves the flexibility and scalability of the model's fault diagnosis, and simultaneously increases the speed of model fault detection and response.

[0203] Example 4: The difference from Example 2 is that:

[0204] like Figure 4 As shown, after performing the third construction in S213 and before performing the model training in S3, there is also a second reconstruction in S22. The second reconstruction in S22 includes the first transformation in S221, the second transformation in S222, the third transformation in S223, and the fourth construction in S224.

[0205] S221 First Transformation: Based on the first causal layer, each device is taken as a node of the first graph, the causal relationship between each pair of devices is transformed into an embedding vector, and the obtained embedding vector is taken as the node feature of the corresponding first graph node to construct the first causal graph.

[0206] In this embodiment, the Node2Vec algorithm is used to calculate the embedding vector. The specific steps are as follows: Based on the first causal layer, the parameters of the Node2Vec algorithm are defined, including vector dimension, random walk length, number of random walks, return parameter, and input / output parameter. According to the defined parameters, starting from the first graph node of the first causal layer, the next graph node is selected for random walk based on the neighboring graph nodes and the corresponding causal relationship strength values ​​of the first graph node, until the preset random walk length is reached, resulting in a random walk sequence. The above steps are performed on each graph node in the new medical knowledge graph to generate multiple random walk sequences. The generated random walk sequences are used as sample data to train the Skip-Gram model, resulting in the trained Skip-Gram model. A random first graph node is input into the trained Skip-Gram model to obtain the embedding vector of that first graph node.

[0207] S222 Second Transformation: Based on the third causal layer, the substation is used as the node of the third graph. The embedding vector of the second causal graph and the Node2Vec algorithm are used to calculate the embedding vector between substation pairs. The obtained embedding vector is used as the node feature of the third graph node to construct the third causal graph.

[0208] S223 Third Transformation: Based on the third causal layer, the substation is used as the node of the third graph. The embedding vector of the second causal graph and the Node2Vec algorithm are used to calculate the embedding vector between substation pairs. The obtained embedding vector is used as the node feature of the third graph node to construct the third causal graph.

[0209] S224 Fourth Construction: Based on the first causal graph, the second causal graph, and the third causal graph, a multi-causal relationship graph is constructed, and the constructed multi-causal relationship graph is used as a new fault causal model.

[0210] In the first update of S32, a new fault causal model is updated based on the causal verification graph.

[0211] After executing the first judgment in S31 and before executing the first update in S32, there is also the second verification in S312. The second verification in S312 includes the fourth processing in S3121, the fifth construction in S3122, and the second judgment in S3123.

[0212] S3121 Fourth Processing: Add a lag value as a lag feature to the historical state data to obtain the first dataset.

[0213] S3122 Fifth Construction: Construct a causal forest, train the causal forest using historical state data to obtain the trained causal forest, input the first dataset and the causal verification graph into the trained causal forest to obtain the causal relationship prediction results between each device pair.

[0214] S3123 Second Judgment: Determine whether the causal relationship prediction results between each pair of devices are completely consistent with the edges between the nodes in the first graph of the fault causal model.

[0215] If so, then execute S33, the first process.

[0216] If not, then based on the causal relationship prediction results between each pair of devices, construct a new causal verification graph, update the new causal verification graph to the causal verification graph, and execute S32 first update.

[0217] like Figure 5 As shown, after executing the first update in S32 and before executing the first processing in S33, there is also an update verification in S321. The update verification in S321 includes the first acquisition in S3211, the first annotation in S3212, the first filtering in S3213, the first calculation in S3214, the second calculation in S3215, the first discrimination in S3216, the second discrimination in S3217, and the third discrimination in S3218.

[0218] S3211 First Collection: Input the first dataset and the causal verification graph into the trained causal forest to obtain the confidence of the causal relationship between the node pairs in the first graph.

[0219] In this embodiment, since the first dataset is obtained from historical state data, the confidence level of the causal relationship between the first graph node pairs obtained here actually refers to the confidence level of the causal relationship between the first graph node pairs before the model update.

[0220] S3212 First annotation: Based on the fault causal model before the update, annotate the added and deleted causal relationships in the updated fault causal model, and collect the annotated causal relationships as the updated causal relationships.

[0221] S3213 First Filter: Obtain the confidence score of the updated causal relationship to obtain the confidence score dataset.

[0222] S3214 First Calculation: Obtain the first graph node pair corresponding to the updated causal relationship, calculate the embedding vector of the first graph node pair corresponding to the updated causal relationship before and after the update, calculate the difference between the embedding vectors before and after the update based on the calculated embedding vectors, and use the calculated difference as the vector change amount of the updated causal relationship.

[0223] In this embodiment, the confidence scores obtained for updated causal relationships actually refer to the confidence scores of added and deleted causal relationships. Since causal relationships may be deleted after the model update, the confidence score of deleted causal relationships after the update is the same as the confidence score of the causal relationship before deletion, and the embedding vector of deleted causal relationships after the update is 0. For added causal relationships, a corresponding confidence score can be set for the added causal relationships using knowledge in the art or methods based on prior assumptions. The embedding vector of the added causal relationship before the model update is 0, and the vector change of the updated causal relationship is positive.

[0224] S3215 Second Calculation: The vector change of the updated causal relationship and the corresponding confidence level are weighted and summed, and the result is used as the correlation degree of the updated causal relationship.

[0225] In this embodiment, the weights of the vector change in updating the causal relationship and the confidence level in updating the causal relationship are both set based on expert knowledge or domain experience. Specifically, the weights of both the vector change in updating the causal relationship and the confidence level in updating the causal relationship are 0.5.

[0226] S3216 First criterion: Determine whether the correlation of the updated causal relationship is higher than the preset correlation threshold.

[0227] If so, then execute the second judgment in S3217.

[0228] If not, then execute the third judgment in S3218.

[0229] S3217 Second criterion: Determine whether the current updated causal relationship is an added causal relationship.

[0230] If so, then execute S33, the first process.

[0231] If not, then add the causal relationship of the first graph node pair corresponding to the current updated causal relationship.

[0232] S3218 Third criterion: Determine whether the current update causal relationship is a deletion causal relationship:

[0233] If so, then execute S33, the first process.

[0234] If not, then delete the causal relationship of the first graph node pair corresponding to the currently updated causal relationship.

[0235] In this embodiment, causal relationships are transformed into embedding vectors, which are then used as features of graph nodes. This helps capture complex nonlinear relationships between graph nodes. Furthermore, by propagating embedding vectors layer by layer, the technical effect of gradually establishing complex causal chains between equipment, subsystems, and substations is achieved. This enhances the model's ability to identify novel fault modes, enabling it to infer potential causal relationships from existing data through a multi-layer embedding vector propagation mechanism when facing new and complex fault modes. This helps improve the model's diagnostic accuracy for novel fault modes. Based on the differences in embedding vectors before and after the update, the changes in causal relationships before and after the model update are quantified. Combined with the confidence level of the causal relationships, the importance of the causal relationships in the model is calculated. This helps reduce noise and redundant information introduced by the model update, optimizes the modeling of causal relationships, improves the model's interpretability and the effectiveness of dynamic updates, and enhances the model's accuracy and robustness.

[0236] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method of substation operating state monitoring of an electric power system, characterized by, Comprise: Data collection: collect real-time state data of the substation, the real-time state data including real-time device state data, real-time device position and real-time health data of the device; collect historical state data of the substation and corresponding fault labels, the historical state data including historical device state data, historical device position and historical health data of the device; Model construction: taking each device of the substation as a node and taking the causal relationship between the device pairs as an edge, a fault causal model is constructed; Model training: including first judgment, first update, first processing, second processing, third processing and first training; First judgment: based on the historical state data, a causal verification graph between the devices is constructed by using a nonlinear Granger causality test method, and whether the causal relationship in the causal verification graph is consistent with the edge between the device pairs in the fault causal model is judged: If yes, the step of first processing is executed; If no, the step of first update is executed; First update: based on the causal verification graph, the fault causal model is updated, and the updated fault causal model is taken as a new fault causal model; First processing: the time stamp of the historical state data is obtained, based on the fault label, the historical state data with the same fault label and continuous time stamp are merged in the order of the time stamp, and the merged historical state data is recorded as first data; Second processing: based on the historical device position, the Euclidean distance between the device pairs is calculated, and the Euclidean distance between each device pair is collected to obtain a first feature set; Third processing: the time sequence of the first data is obtained, and whether the historical health data of the device changes in the time sequence is judged: If no, the step of first training is executed; If yes, the time point of the change of the historical health data of the device is obtained, the first data is segmented based on the obtained time point to obtain segmented first data, and the segmented first data is updated as the first data; First training: the first feature set and the first data are spliced, the spliced data is updated as the first data, the first data and the corresponding fault label are used to train the new fault causal model to obtain a trained fault causal model; Fault discrimination: the real-time state data is input into the trained fault causal model to obtain a real-time fault label of the substation.

2. The substation operating state monitoring method of a power system according to Claim 1, wherein, After executing the step of model construction and before executing the step of model training, it further comprises: First construction: the nodes in the fault causal model are recorded as first nodes, the fault causal model is recorded as a first causal layer, the functional attributes of each device in the substation are obtained, the devices are classified based on the functional attributes to obtain a plurality of subsystems, and a second causal layer is constructed taking the subsystems as nodes and taking the causal relationship between the subsystem pairs as edges; Second construction: the nodes of the second causal layer are recorded as second nodes, the substation is taken as a node, and a third causal layer is constructed taking the causal relationship between the substation pairs as edges; Third construction: the nodes of the third causal layer are recorded as third nodes, and the first causal layer, the second causal layer and the third causal layer are updated as the fault causal model.

3. The substation operating state monitoring method of a power system according to Claim 2, characterized by, After executing the step of third construction and before executing the step of model training, it further comprises: The first conversion: based on the first causal layer, each device is taken as a first graph node, the causal relationship between each device pair is converted into an embedding vector, and the obtained embedding vector is taken as the node feature of the corresponding first graph node, and a first causal graph is constructed; The second conversion: based on the second causal layer, the subsystem is taken as a second graph node, the embedding vector of the first causal graph is used to calculate the embedding vector between the subsystem pairs, and the obtained embedding vector is taken as the node feature of the second graph node, and a second causal graph is constructed; The third conversion: based on the third causal layer, the transformer substation is taken as a third graph node, the embedding vector of the second causal graph is used to calculate the embedding vector between the transformer substation pairs, and the obtained embedding vector is taken as the node feature of the third graph node, and a third causal graph is constructed; The fourth construction: based on the first causal graph, the second causal graph and the third causal graph, a multi-causal relationship graph is constructed, and the constructed multi-causal relationship graph is taken as a new fault causal model; In the step of the first update, the new fault causal model is updated based on the causal verification graph.

4. The substation operating state monitoring method of a power system according to Claim 3, characterized by, After the step of performing the first judgment, before the step of performing the first processing, further comprising: The fourth processing: adding a lag feature to the historical state data to obtain a first data set; The fifth construction: constructing a causal forest, training the causal forest by using the historical state data to obtain a trained causal forest, inputting the first data set and the causal verification graph into the trained causal forest to obtain a causal relationship prediction result between each device pair; The second judgment: judging whether the causal relationship prediction result between each device pair is completely consistent with the edge between the first graph node pairs in the fault causal model: If yes, the step of the first processing is performed; If no, a new causal verification graph is constructed based on the causal relationship prediction result between each device pair, the new causal verification graph is updated as the causal verification graph, and the step of the first update is performed.

5. The substation operating state monitoring method of a power system according to Claim 2, characterized by, After the step of performing the first judgment, before the step of performing the first processing, further comprising: The third judgment: based on the historical state data, a causal verification graph between each subsystem pair is constructed by using a nonlinear Granger causal test method, which is recorded as a first verification graph, and it is judged whether the causal relationship between each subsystem pair in the first verification graph is consistent with the edge between the second node pairs in the fault causal model: If yes, the step of the fourth judgment is performed; If no, the causal verification graph and the first verification graph are taken as a new causal verification graph, and the step of the first update is performed; The fourth judgment: based on the historical state data, a causal verification graph between each transformer substation pair is constructed by using a nonlinear Granger causal test method, which is recorded as a second verification graph, and it is judged whether the causal relationship between each transformer substation pair in the second verification graph is consistent with the edge between the third node pairs in the fault causal model: If yes, the step of the first processing is performed; If no, the causal verification graph, the first verification graph and the second verification graph are updated as the causal verification graph, and the step of the first update is performed.

6. The substation operating state monitoring method of a power system according to Claim 4, wherein After the step of performing the first update, before the step of performing the first processing, further comprising: The first collection: the first data set and the causal verification graph are input into the trained causal forest to obtain the confidence of the causal relationship between the first graph node pairs; The first collection: the first data set and the causal verification graph are input into the trained causal forest to obtain the confidence of the causal relationship between the first graph node pairs; The first labeling: based on the fault causal model before the update, label the added causal relationship and the deleted causal relationship in the updated fault causal model, and collect the labeled causal relationship as an updated causal relationship; The first screening: obtain the confidence of the updated causal relationship to obtain a confidence dataset; The first calculation: obtain a first graph node pair corresponding to the updated causal relationship, calculate the embedding vectors of the first graph node pair before and after the update corresponding to the updated causal relationship, based on the calculated embedding vectors, calculate the difference between the embedding vectors before and after the update, and take the calculated difference as a vector change of the updated causal relationship; The second calculation: weight and sum the vector change of the updated causal relationship and the corresponding confidence, and take the calculation result as a correlation degree of the updated causal relationship; The first discrimination: judge whether the correlation degree of the updated causal relationship is higher than a preset correlation degree threshold: If yes, execute the step of the second discrimination; If no, execute the step of the third discrimination; The second discrimination: judge whether the current updated causal relationship is an added causal relationship: If yes, do not process; If no, add a causal relationship of the first graph node pair corresponding to the current updated causal relationship; The third discrimination: judge whether the current updated causal relationship is a deleted causal relationship: If yes, do not process; If no, delete the causal relationship of the first graph node pair corresponding to the current updated causal relationship.

7. The substation operating state monitoring method of a power system according to Claim 1, wherein, Each device of the substation contains a device ID, after the step of the first processing and before the step of the second processing, further comprising: The second labeling: based on each device of the substation, label the corresponding device ID on the collected historical state data, and record the labeled data as new first data; The first association: obtain the edges between each node pair, convert the edges between each node pair into a causal relationship strength value, and record it as a first feature; Position coding: based on the timestamp of the historical state data, perform position coding on the timestamp of each node, and record the coded timestamp as a time feature; Data update: splice the time feature, the first feature and the new first data, and update the spliced data to the first data.

8. The substation operating state monitoring method of a power system according to Claim 7, characterized by, After the step of the second processing and before the step of the third processing, further comprising: The first acquisition: acquire the environmental parameters of the location where each device is located; Difference calculation: calculate the environmental parameter difference between each device pair; The third calculation: based on the environmental parameter difference and the Euclidean distance between each device pair, calculate the environmental space difference between each device pair by using an environmental space difference formula, and take the calculated environmental space difference between each device pair as a new first feature set; The calculation model of the environmental space difference formula is: ； wherein, is the environmental space difference between device i and device j, is the adjustment factor, is the environmental parameter difference between device i and device j, is the Euclidean distance between device i and device j.

9. The substation operating state monitoring method of a power system according to Claim 1, wherein, The historical state data further includes device historical use time length, historical load data and historical resource data, after the step of the second processing and before the step of the third processing, further comprising: The first statistics: obtain the historical load data of each device and the maximum load value of each device, and statistically obtain the total time length of the historical load data of each device exceeding the maximum load value, the difference between the historical load data exceeding the maximum load value and the maximum load value, and the corresponding duration; Second statistics: obtain the historical resource data of each device and the minimum resource requirement value of each device, and count the total duration of the historical resource data of each device being lower than the minimum resource requirement value, the difference between the historical resource data being lower than the minimum resource requirement value and the minimum resource requirement value, and the corresponding duration; Load calculation: calculate the influence factor of the load fluctuation of each device on the device state by using a load influence formula, denoted as a first influence; Resource calculation: calculate the influence factor of the resource fluctuation of each device on the device state by using a resource influence formula, denoted as a second influence; Fourth calculation: based on the historical health data of the device, the first influence and the second influence, calculate the actual health data of the device by using a device health formula, and take the calculated actual health data as the new historical health data of the device; The calculation model of the load influence formula is: ； wherein, is the number of historical load data exceeding the maximum load value in the historical load data, a is the a-th historical load data exceeding the maximum load value, is the difference between the a-th historical load data exceeding the maximum load value and the maximum load value, is the maximum load value of the device, is the duration of the a-th historical load data exceeding the maximum load value, is the historical usage duration of the device; The calculation model of the resource influence formula is: ； wherein, is the number of historical resource data below the minimum resource requirement value in the historical resource data, b is the bth historical resource data below the minimum resource requirement value, is the difference between the bth historical resource data below the minimum resource requirement value and the minimum resource requirement value, is the minimum resource requirement value of the device, is the duration of the bth historical resource data below the minimum resource requirement value; The calculation model of the device health formula is: ； wherein, is a weight of the impact of the load fluctuation on the device state, is a weight of the impact of the resource fluctuation on the device state, is historical health data of the device.

10. The substation operating state monitoring method of a power system according to Claim 1, wherein After the step of performing the third processing, before the step of performing the first training, further comprising: Second acquisition: acquire the time point of the change of the historical health data of the device, denoted as a first time point; Third acquisition: acquire the historical health data of the device at a time point before the first time point, denoted as first device data; and acquire the historical health data of the device at a time point after the first time point, denoted as second device data; Rate calculation: calculate the time difference between the time point before the first time point and the time point after the first time point, and calculate the change rate of the historical health data of the device based on the calculated time difference, the first device data and the second device data, denoted as a first change rate; Fifth judgment: judge whether the first change rate is less than a preset rate threshold: If yes, merge the first data segmented based on the first time point, and update the merged data as the first data; If no, do not process.

Images