Power distribution network load transfer and recovery method, system and device and storage medium

Abstract

The invention discloses a power distribution network load transfer and recovery method, system and device, and a storage medium. The method comprises the steps of obtaining a conventional index and a historical fault index of a power distribution network; performing first calculation on non-fault duration and fault recovery time according to the historical fault indexes; performing second calculation on a self-healing coefficient of the system under a self-healing scheme according to the conventional index, the non-fault duration time and the fault recovery time; and comparing the self-healing coefficients to obtain a self-healing scheme for self-adaptive self-healing of the power distribution network fault, thereby providing powerful guarantee for efficient and stable operation of the power distribution network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution network monitoring, and in particular to a power distribution network load transfer and recovery method, system, device and storage medium. BACKGROUND

[0002] The current power distribution network fault transfer and recovery strategy faces significant technical bottlenecks, mainly in two dimensions of multi-source heterogeneous data processing and intelligent analysis method. First, the operation of the power distribution network involves SCADA systems, PMU devices, smart meters, weather monitoring and other multi-source heterogeneous data, and their data formats, sampling frequencies and communication protocols differ significantly. Traditional data fusion methods rely on manual rule configuration and are difficult to achieve dynamic correlation analysis across platforms and time scales, resulting in incomplete fault feature extraction and delayed situation awareness. For example, the existing topology verification method based on a single data source cannot effectively identify the implicit flow limit caused by the access of distributed power supply, and the manual threshold setting mechanism is difficult to adapt to the dynamic changes of complex working conditions.

[0003] Therefore, the present application provides a power distribution network load transfer and recovery method, system, device and storage medium to solve the technical problem that the prior art cannot analyze multi-source heterogeneous data of the power distribution network and generate a response strategy in a timely manner, and to achieve sufficient analysis of power distribution network operation data to make the determined power distribution network transfer and recovery strategy more meet the actual requirements. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a power distribution network load transfer and recovery method, system, device and storage medium to solve the problem that the prior art cannot analyze multi-source heterogeneous data of the power distribution network and generate a response strategy in a timely manner.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a power distribution network load transfer and recovery method, comprising:

[0008] Obtaining multi-source heterogeneous data, preprocessing the multi-source heterogeneous data to obtain first multi-source heterogeneous data;

[0009] Performing feature extraction on the first multi-source heterogeneous data to obtain multi-scale time series features, performing causal reasoning operation on the multi-scale time series features to obtain a plurality of key factors;

[0010] Selecting a fault prediction model, inputting the plurality of key factors into the fault prediction model to obtain a fault probability prediction result;

[0011] Obtaining a twin body parameter of a power distribution network, constructing a digital twin model according to the twin body parameter, and obtaining a power grid parameter;

[0012] According to the fault probability prediction result and the power grid parameter, an execution strategy of the power distribution network is obtained.

[0013] As a preferred scheme of the power distribution network load transfer and recovery method, the preprocessing of the multi-source heterogeneous data comprises:

[0014] An abnormal point detection is performed on the multi-source heterogeneous data by using a first abnormality detection algorithm, and a first repair method is used to repair the abnormal data points.

[0015] The repaired multi-source heterogeneous data is subjected to first feature standardization processing, and the standardized multi-source heterogeneous data is subjected to first feature alignment processing to obtain first multi-source heterogeneous data.

[0016] The beneficial effects of the preferred technical scheme are that through abnormality detection and repair, feature standardization and alignment processing, the quality and consistency of the multi-source heterogeneous data are effectively improved, and reliable data foundation is provided for subsequent analysis.

[0017] As a preferred scheme of the power distribution network load transfer and recovery method, the plurality of key factors comprises:

[0018] A first transformation analysis method is used to extract features from the first multi-source heterogeneous data to obtain multi-scale time sequence features.

[0019] A first causal reasoning algorithm is used to construct a causal directed acyclic graph on the multi-scale time sequence features to identify a plurality of key factors.

[0020] The beneficial effects of the preferred technical scheme are that through wavelet transform to extract multi-scale time sequence features and using a causal reasoning algorithm to identify key factors, key information in the data can be accurately captured, the data dimension is effectively reduced, the accuracy and efficiency of fault prediction are improved, and strong support is provided for subsequent strategy generation.

[0021] As a preferred scheme of the power distribution network load transfer and recovery method, the power grid parameter comprises:

[0022] A hierarchical step simulation mechanism is established.

[0023] Under normal operating conditions, a first simulation step is used for steady-state simulation to obtain the power grid parameter.

[0024] Under switch operating conditions, a second simulation step is used for fine simulation to obtain the power grid parameter.

[0025] The switching between the normal operating conditions and the switch operating conditions is controlled by a first trigger threshold.

[0026] As a preferred scheme of the power distribution network load transfer and recovery method, wherein: the feature extraction of the first multi-source heterogeneous data by using the first transformation analysis method comprises:

[0027] The coefficient square sum of each wavelet decomposition node is calculated as an energy value to obtain a frequency band energy feature;

[0028] The energy entropy of the high frequency band is calculated to obtain a high frequency band energy entropy feature;

[0029] The time domain statistics are calculated to obtain a time domain statistics feature;

[0030] The frequency band energy feature, the high frequency band energy entropy feature and the time domain statistics feature are integrated to obtain a feature set;

[0031] The feature set is screened, and a space-time feature matrix is obtained through dimension compression.

[0032] The beneficial effects of the preferred technical scheme are that through multi-dimensional feature extraction and screening, key information is effectively integrated, data dimensions are compressed, feature quality is improved, and the effectiveness and accuracy of model input are enhanced.

[0033] As a preferred scheme of the power distribution network load transfer and recovery method, wherein: the construction of the causal directed acyclic graph by using the first causal reasoning algorithm comprises:

[0034] An initial skeleton graph is established by using the Gaussian conditional independence test;

[0035] On the initial skeleton graph, collision nodes are determined by topological structure detection, edge orientation is performed by using the greedy equivalent search, and a directed acyclic graph is obtained;

[0036] According to the directed acyclic graph, the key factors with the greatest impact on the target variable are screened out through causal effect evaluation;

[0037] A dynamic updating mechanism of the Bayesian network is constructed, and when new data arrives, local structure learning is triggered to obtain a dynamically updated causal directed acyclic graph.

[0038] The beneficial effects of the preferred technical scheme are that key factors are effectively identified through causal reasoning, causal relationships are dynamically updated, fault prediction accuracy and adaptability are improved, and the scientificity and reliability of system decision-making are enhanced.

[0039] As a preferred scheme of the power distribution network load transfer and recovery method, wherein: the execution strategy of the power distribution network comprises:

[0040] When the fault probability prediction result is greater than the first threshold value, the corresponding fault occurrence time information is determined;

[0041] obtaining corresponding power grid parameters according to corresponding fault occurrence time information;

[0042] inputting the corresponding power grid parameters into a power distribution network execution strategy determination model to obtain a corresponding execution strategy.

[0043] In a second aspect, the present application provides a power distribution network load transfer and recovery system, comprising:

[0044] a preprocessing module configured to obtain multi-source heterogeneous data, preprocess the multi-source heterogeneous data, and obtain first multi-source heterogeneous data;

[0045] a feature extraction module configured to perform feature extraction on the first multi-source heterogeneous data, obtain multi-scale time series features, and perform causal reasoning on the multi-scale time series features to obtain a plurality of key factors;

[0046] a prediction module configured to select a fault prediction model, input the plurality of key factors into the fault prediction model, and obtain a fault probability prediction result;

[0047] a model construction module configured to obtain twin parameters of a power distribution network, construct a digital twin model according to the twin parameters, and obtain power grid parameters;

[0048] an execution strategy generation module configured to obtain an execution strategy of a power distribution network according to the fault probability prediction result and the power grid parameters.

[0049] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the steps of the power distribution network load transfer and recovery method when executing the computer program.

[0050] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the power distribution network load transfer and recovery method.

[0051] Compared with the prior art, the application has the beneficial effects that: through the collection of data in the aspects of equipment, weather and operation, the operation state of the power distribution network is comprehensively described, the data standardization and alignment processing solve the problems of inconsistent data formats and different time scales of multiple source data, provide high-quality input for subsequent analysis, and improve the data availability; the processed data is analyzed by using a machine learning model, the fault probability of the power distribution network is accurately predicted, the deficiencies of traditional methods in fault feature extraction and situation awareness are made up, and the accuracy and timeliness of fault prediction are improved; a digital twin model is established for the power distribution network, real-time power grid operation information is obtained, the digital twin model can simulate the real-time operation state of the power distribution network, and provide accurate basis for formulating transfer and recovery strategies, when the fault probability exceeds a preset value, an execution strategy is generated based on the real-time operation information of the digital twin model, the pertinence and effectiveness of the strategy are ensured, the fault can be quickly responded to, the power outage time and economic loss are reduced, and the reliability and economy of the power distribution network are improved. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and all other drawings obtained by those skilled in the art without creative labor should be within the protection scope of the present application.

[0053] Figure 1 The overall flow logic diagram of a power distribution network load transfer and recovery method provided by an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the protection scope of the present application.

[0055] Embodiment 1, refer to Figure 1 For an embodiment of the present application, a power distribution network load transfer and recovery method is provided, which comprises:

[0056] S100: acquiring multiple source heterogeneous data, pre-processing the multiple source heterogeneous data to obtain first multiple source heterogeneous data;

[0057] S200: performing feature extraction on the first multiple source heterogeneous data to obtain multi-scale time series features, performing causal reasoning operation on the multi-scale time series features to obtain multiple key factors;

[0058] S300: Select a failure prediction model, input multiple key factors into the failure prediction model, and obtain a failure probability prediction result;

[0059] Specifically, the prediction of the power distribution network failure probability adopts a CNN neural network (convolutional neural network), including three layers of an input layer, a processing layer and a fusion layer. The processing layer includes an LSTM time sequence processing module and a causal reasoning module.

[0060] The input layer receives a 128-dimensional feature vector (including 32-dimensional device state, 24-dimensional environmental factors, 48-dimensional load characteristics and 24-dimensional topological features).

[0061] The model training process also includes an online parameter updating mechanism.

[0062] The historical operation data of the power distribution network are selected as sample data, the operation data of the power distribution network are taken as input, and the failure of the power distribution network is taken as output to train the CNN model, so as to obtain the failure prediction model.

[0063] S400: Obtain the twin parameters of the power distribution network, construct a digital twin model according to the twin parameters, and obtain the grid parameters.

[0064] S500: Obtain the execution strategy of the power distribution network according to the failure probability prediction result and the grid parameters.

[0065] It should be noted that through the collection, preprocessing, feature extraction, causal reasoning of multi-source heterogeneous data and the construction of the digital twin model, the comprehensive perception and accurate prediction of the operation state of the power distribution network are realized. The application can effectively improve the accuracy of failure prediction, quickly generate the optimal transfer and recovery strategy, significantly improve the operation reliability, economy and safety of the power distribution network, reduce the operation and maintenance cost, and enhance the intelligent level and the ability to adapt to complex working conditions of the system.

[0066] In the embodiment of the application, the above step S100 includes the following sub-steps A1-A2.

[0067] In A1, the first anomaly detection algorithm is used to detect abnormal points in the multi-source heterogeneous data, and the first repair method is used to repair the abnormal data points.

[0068] In A2, the repaired multi-source heterogeneous data is subjected to first feature standardization processing, and the multi-source heterogeneous data subjected to the standardization processing is subjected to first feature alignment processing, to obtain first multi-source heterogeneous data.

[0069] In an optional embodiment, the first anomaly detection algorithm can be a DBSCAN algorithm. For each data point, the neighborhood is calculated, and the core point and the noise point are marked. The noise point is the abnormal point. From the core point, the cluster is expanded, and all points belonging to the same cluster are marked as normal points.

[0070] In an optional embodiment, the first anomaly detection algorithm can be a random forest, the multi-source heterogeneous data is input into the random forest model, and the data points are detected by using multiple decision trees of the random forest. Each decision tree classifies the data points according to its own training result, and the detection results of all decision trees are fused by a majority voting mechanism to finally determine which data points are abnormal points.

[0071] In the embodiment of the application, the first anomaly detection algorithm includes an isolation forest algorithm.

[0072] Specifically, the multi-source heterogeneous data is input into the isolation forest model, the random partition mechanism of the isolation forest is used to perform isolation operation on each data point, and the anomaly score is calculated. According to a preset confidence threshold (such as 0.95), the data points with an anomaly score higher than the threshold are determined as abnormal points, and the first repair method is used to repair the detected abnormal data points to obtain clean data.

[0073] In an optional embodiment, the first repair method can be a repair method based on mean or median, the mean or median of normal data points is calculated, and if a data point is detected as an abnormal point or a missing point, the mean or median of the feature is used to replace the value of the point.

[0074] In an optional embodiment, the first repair method can be a repair method based on K-nearest neighbors, the distance between each abnormal point or missing point and other normal data points is calculated, the K nearest data points are selected as neighbors, and the repair is performed by weighted average value.

[0075] In the embodiment of the application, the first repair method includes cubic spline repair, the relationship between data points is fitted by constructing a cubic polynomial function, and the abnormal points or missing points are repaired.

[0076] In an optional embodiment, the first feature standardization processing can be robust standardization, the median, the first quartile and the third quartile of each feature are calculated for current, voltage and weather data, and the standardized value of each data point is obtained according to the median and the difference between the third quartile and the first quartile.

[0077] In an optional embodiment, the first feature standardization processing can be maximum absolute value standardization, the maximum absolute value of each feature is calculated for current, voltage and weather data, and the standardized value is obtained by comparing each data point with the maximum absolute value.

[0078] In the embodiment of the application, the first feature standardization processing includes Z-score standardization for current and voltage, Min-Max normalization for weather parameters, and One-hot encoding for operation events.

[0079] In an optional embodiment, the first feature alignment processing includes event-driven time alignment, for switch operation logs, aligning related current, voltage and weather data based on switch action time, for fault alarm signals, extracting time series data before and after the fault occurs, and ensuring the time consistency of different data sources;

[0080] In an optional embodiment, the first feature alignment processing can be data interpolation-based spatial alignment, for temperature and humidity, wind speed and other data collected by weather sensors, aligning the data to a unified GIS grid through spatial interpolation, and for device status data distributed in different geographical locations, aligning to a unified grid through spatial interpolation for spatial analysis;

[0081] In the embodiment of the application, the first feature alignment processing includes designing a space-time alignment engine, using GPS second pulse synchronization time stamp (error <1ms) to establish a GIS coordinate grid (resolution 50m x 50m) to realize spatial matching, and developing a sliding time window (length 10s, step 1s) for dynamic alignment.

[0082] Specifically, the distributed data collection network includes a plurality of distributed nodes, each of which includes a multi-modal sensor, an intelligent electric meter terminal, a weather sensor node, and a SCADA system node.

[0083] The intelligent electric meter terminal (sampling frequency >=4kHz) is used to obtain electrical quantity time series data, the weather sensor node (including temperature and humidity, wind speed, and atmospheric pressure sensor group) is used to collect micro-weather parameters, and the SCADA system is used to synchronously access switch operation logs; the LoRa-WAN protocol is used to realize wide-area heterogeneous data transmission, a data cache queue (capacity >=1TB) is configured to cope with collection delay, a device-weather-operation three-dimensional data mapping table is established, and the data in the three-dimensional data mapping table is multi-source heterogeneous data.

[0084] Based on the isolated forest algorithm, abnormal data points in the multi-source heterogeneous data are detected, and the abnormal data points are repaired by using cubic spline interpolation; since the multi-source heterogeneous data collected by the multi-modal sensor may not be dimensionally uniform, the device-weather-operation three-dimensional data needs to be standardized accordingly to facilitate subsequent processing; at the same time, since the device-weather-operation three-dimensional data are all time-related data, in order to uniformly analyze the three-dimensional data, the above data needs to be time-aligned.

[0085] It should be noted that through abnormality detection and repair, standardization processing and time alignment, the quality and consistency of the multi-source heterogeneous data are effectively improved, providing a reliable data basis for subsequent analysis.

[0086] In the embodiment of the present application, the step S200 comprises the following sub-steps B1-B7.

[0087] In B1, the first multi-source heterogeneous data is subjected to feature extraction by using a first transform analysis method to obtain multi-scale time sequence features.

[0088] In B2, a causal directed acyclic graph is constructed by using a first causal inference algorithm on the multi-scale time sequence features to identify a plurality of key factors.

[0089] In B3, the coefficient square sum of each wavelet decomposition node is calculated as an energy value to obtain a frequency band energy feature.

[0090] In B4, the energy entropy of the high frequency band is calculated to obtain an energy entropy feature of the high frequency band.

[0091] In B5, a time domain statistic is calculated to obtain a time domain statistic feature.

[0092] In B6, the frequency band energy feature, the high frequency band energy entropy feature and the time domain statistic feature are integrated to obtain a feature set.

[0093] In B7, the feature set is screened and a space-time feature matrix is obtained through dimension compression.

[0094] In an optional embodiment, the first transform analysis method can be Fourier transform. For time sequence data of current, voltage and meteorological parameters, the time sequence data is converted into frequency domain signals by Fourier transform, the frequency spectrum of the signals is calculated, and the energy, amplitude and phase information of the main frequency components are extracted.

[0095] In an optional embodiment, the first transform analysis method can be principal component analysis. The multi-source heterogeneous data is subjected to standardization processing, the covariance matrix of the data is calculated to reflect the correlation between the features, the eigenvalue decomposition is performed on the covariance matrix to obtain the eigenvalue and the eigenvector, the eigenvectors corresponding to the first k largest eigenvalues are selected as the principal components, the projection matrix is constructed, and the original data is projected into the principal component space to obtain the feature representation after dimension reduction.

[0096] In the embodiment of the present application, the first transform analysis method comprises wavelet transform.

[0097] Specifically, Daubechies9 wavelet basis is used for 6-layer wavelet packet decomposition, the frequency band energy feature vector is constructed by calculating the coefficient square sum of each node as the energy value, the energy entropy feature of the 1-5kHz high frequency band is extracted, the time domain statistics (including mean, variance, skewness and kurtosis) are calculated synchronously, the feature filter is designed, the key features are retained by the mutual information method (threshold > 0.3), and finally the space-time feature matrix after dimension compression is generated.

[0098] It should be noted that the multi-scale time sequence features are extracted by wavelet transform, combined with energy values, entropy features and time domain statistics, and an efficient and accurate space-time feature matrix is generated through feature screening and dimension compression, so that the accuracy of fault prediction and analysis is significantly improved.

[0099] In the embodiment of the present application, the step S200 after the completion of the steps B1-B7 further comprises the following steps B8-B10.

[0100] In B8: an initial skeleton graph is established by using Gaussian conditional independence test;

[0101] In B9: collision nodes are determined on the initial skeleton graph by topological structure detection, and edge orientation is performed by using greedy equivalent search to obtain a directed acyclic graph;

[0102] In B10: according to the directed acyclic graph, the key factor with the greatest impact on the target variable is screened out through causal effect evaluation;

[0103] In B11: a dynamic updating mechanism of the Bayesian network is constructed, and local structure learning is triggered when new data arrives to obtain a dynamically updated causal directed acyclic graph.

[0104] In an optional embodiment, the first causal reasoning algorithm can be a GES algorithm, which starts from an empty graph or a complete graph, and finds a causal graph structure that can maximize a score function by gradually adding, deleting or reversing edges, uses Bayesian information criterion (BIC) or Bayesian Dirichlet equivalent uniform (BDeu) or other score functions to evaluate the pros and cons of the causal graph, and finds the optimal causal graph structure through multiple iterations to identify the key factor;

[0105] In an optional embodiment, the first causal reasoning algorithm can be a ridge regression, which takes the features in the multi-source heterogeneous data as independent variables and the target variable as the dependent variable, constructs a ridge regression model, performs regularization processing on the model by selecting a suitable regularization parameter, and calculates the weight coefficient of each feature after the model training is completed. The greater the absolute value of the weight coefficient, the greater the impact of the feature on the target variable, and according to the size of the weight coefficient, the feature with a larger weight absolute value is selected as the key factor;

[0106] In the embodiment of the present application, the first causal reasoning algorithm can be a PC algorithm to construct a causal directed acyclic graph;

[0107] Specifically, the initial skeleton graph is established by using Gaussian conditional independence test (significance level alpha = 0.01); the collision nodes are determined by V-shaped structure detection, and the edge orientation is performed by using a greedy equivalent search; a causal effect evaluator is designed, the average causal effect is calculated, and the top 5 key factors are screened as key factors; and a dynamic updating mechanism of the Bayesian network is constructed, and local structure learning (updating period <= 1h) is triggered when new data arrives.

[0108] It should be noted that the key factors are accurately identified by causal reasoning, the causal graph is dynamically updated, and the fault prediction accuracy and adaptability are improved.

[0109] In the embodiment of the application, the above step S400 includes the following sub-steps C1-C4.

[0110] In C1, a hierarchical step simulation mechanism is established.

[0111] In C2, under normal operating conditions, steady-state simulation is performed using a first simulation step to obtain power grid parameters.

[0112] In C3, under switching operating conditions, fine simulation is performed using a second simulation step to obtain power grid parameters.

[0113] In C4, the switching between normal operating conditions and switching operating conditions is controlled by a first trigger threshold.

[0114] Specifically, the digital twin model is a twin body cooperative construction of the physical layer and the model layer for the distribution network, the physical layer uses an RTDS real-time simulator to construct an electromagnetic transient simulation environment, sets a 50 microsecond simulation step to realize microsecond-level electromagnetic transient process simulation, and configures an FPGA acceleration card to establish a switching transient processing channel, and the FPGA acceleration card realizes nanosecond-level switching state response through hardware logic programming.

[0115] The model layer includes a device-level modeling unit and a network-level modeling unit, wherein the device-level modeling unit uses a multi-physical field coupling modeling method to construct fine models of transformers and circuit breakers, and the network-level modeling unit establishes a topology-adaptive power flow calculation model based on an improved forward-backward substitution algorithm.

[0116] The power grid parameters include voltage qualification rate, load recovery rate and other parameters for characterizing the operating state of the distribution network.

[0117] The hierarchical step simulation mechanism is established, under normal operating conditions, the first simulation step is to perform steady-state simulation using a 1 second macro step, and when a switching action event is detected, it is automatically switched to a second simulation step of 10 milliseconds micro step to implement transient process fine simulation, and the switching process is controlled by a preset trigger threshold, the first trigger threshold is that the switching current rate exceeds 200A / us.

[0118] It should be noted that through the hierarchical simulation mechanism, the normal and switching conditions of the power distribution network are accurately simulated, the power grid parameters are obtained in real time, and the operation state perception and fault response capability are improved.

[0119] In the embodiment of the application, the step S500 includes the following sub-steps D1-D3.

[0120] In D1, when the fault probability prediction result is greater than the first threshold value, the corresponding fault occurrence time information is determined.

[0121] In D2, the corresponding power grid parameters are obtained according to the corresponding fault occurrence time information.

[0122] In D3, the corresponding power grid parameters are input into the power distribution network execution strategy determination model to obtain the corresponding execution strategy.

[0123] Specifically, when the predicted fault probability result is greater than the first threshold value, it indicates that a fault event occurs in the power distribution network, and a monitoring event needs to be triggered to facilitate subsequent determination of the execution strategy; the first threshold value can be 80%; the fault occurrence time information refers to the start time of the fault;

[0124] According to the corresponding fault occurrence time information, the corresponding power grid parameters are obtained, which refer to the operation parameter information of the power distribution network simulated by the digital twin model;

[0125] The power distribution network execution strategy determination model is obtained by training a CNN model. In the training sample, historical operation data of an existing power distribution network is selected, and in each sample data, the operation parameter information corresponding to an existing power distribution network, as well as the constraints of economy, reliability and safety, and the corresponding optimal power distribution network execution strategy are included.

[0126] In the training process of the model, the operation parameter information and the constraints are used as inputs, and the optimal power distribution network execution strategy is used as output to train the power distribution network strategy determination model.

[0127] Specifically, after confirming the execution strategy, an execution state board is established to display the operation of the power distribution network in real time, and the displayed content includes the switch action success rate, the load recovery progress, the real-time network loss change curve, etc.

[0128] It should be noted that through the fault probability threshold value judgment, the fault occurrence time and the corresponding power grid parameters are accurately determined, the well-trained CNN model is used to quickly generate the optimal execution strategy, and the fault response speed and processing efficiency are significantly improved.

[0129] The above is a schematic scheme of the power distribution network load transfer and restoration method of the embodiment. It should be noted that the technical scheme of the power distribution network load transfer and restoration system and the technical scheme of the power distribution network load transfer and restoration method described above belong to the same concept. The technical scheme of the power distribution network load transfer and restoration system in the embodiment is not described in detail, and the description of the technical scheme of the power distribution network load transfer and restoration method described above can be referred to.

[0130] The power distribution network load transfer and restoration system in the embodiment comprises:

[0131] The preprocessing module is configured to obtain multi-source heterogeneous data, preprocess the multi-source heterogeneous data, and obtain first multi-source heterogeneous data.

[0132] The feature extraction module is configured to perform feature extraction on the first multi-source heterogeneous data, obtain multi-scale time sequence features, and perform causal reasoning operation on the multi-scale time sequence features to obtain a plurality of key factors.

[0133] The prediction module is configured to select a fault prediction model, input the plurality of key factors into the fault prediction model, and obtain a fault probability prediction result.

[0134] The model construction module is configured to obtain twin parameters of the power distribution network, construct a digital twin model according to the twin parameters, and obtain power grid parameters.

[0135] The execution strategy generation module is configured to obtain an execution strategy of the power distribution network according to the fault probability prediction result and the power grid parameters.

[0136] The embodiment also provides a computer device suitable for power distribution network load transfer and restoration, which comprises:

[0137] The memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions to implement the power distribution network load transfer and restoration method according to the above embodiment.

[0138] The embodiment also provides a storage medium having a computer program stored thereon, and the program is executed by a processor to implement the power distribution network load transfer and restoration method according to the above embodiment.

[0139] The storage medium according to the embodiment and the power distribution network load transfer and restoration method according to the above embodiment belong to the same inventive concept. The technical details not described in detail in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0140] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware, and of course can also be realized by hardware. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH memory, a hard disk, or an optical disc, etc., and includes a number of instructions to make a computing device (which can be a personal computer, a server, or a network device, etc.) execute the methods of various embodiments of the present application.

[0141] Embodiment 2, which is different from the first embodiment, provides a verification test of the power distribution network load transfer and recovery method, to verify the technical effects adopted in the method.

[0142] Taking a 10KV power distribution network in a certain city as an example, the fault scenario is that line L1 causes a short-circuit fault due to lightning strike;

[0143] Collecting multi-source heterogeneous data such as humidity (90%), voltage fluctuation rate (0.5%), and protection action times (3 times) along line L1;

[0144] Using the causal reasoning algorithm to analyze the collected multi-source heterogeneous data, extracting key factors, and inputting the extracted key factors into the fault prediction model, the model predicts the fault probability P fault = 0.78, since the fault probability is greater than 80%, an early warning signal is triggered, and line L1 may fail;

[0145] The digital twin model simulates the running state of the power distribution network in real time, and displays that the capacity of tie switch K2 is sufficient (S switch = 1200A), and the prepared path power flow carrying capacity is P capacity = 800kW, indicating that the tie switch K2 has sufficient capacity to transfer the load of line L1.

[0146] Inputting the power grid parameters provided by the digital twin model into the power distribution network execution strategy determination model, an optimal execution strategy is generated: transferring the load of L1 to L2 through K2.

[0147] This strategy ensures that when a fault occurs, the affected load can be quickly transferred to other normally operating lines, reducing the power outage time;

[0148] The load transfer operation is completed in 18 seconds, ensuring that the 15 affected users have no perceptible power outage. The real-time monitoring system displays the switch action success rate, load recovery progress, and real-time network loss change curve, ensuring the accuracy and reliability of the operation.

[0149] It should be noted that the above examples are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A method for load transfer and restoration in a power distribution network, characterized in that, include: Acquire multi-source heterogeneous data, preprocess the multi-source heterogeneous data to obtain first multi-source heterogeneous data; Feature extraction is performed on the first multi-source heterogeneous data to obtain multi-scale time-series features. Causal inference is then performed on the multi-scale time-series features to obtain multiple key factors. Select a fault prediction model, input the multiple key factors into the fault prediction model, and obtain the fault probability prediction result; Obtain the twin parameters of the distribution network, construct a digital twin model based on the twin parameters, and obtain the network parameters; Based on the fault probability prediction results and the power grid parameters, the execution strategy of the distribution network is obtained.

2. The method for load transfer and restoration in a distribution network as described in claim 1, characterized in that, Preprocessing of the multi-source heterogeneous data includes: The first anomaly detection algorithm is used to detect anomalies in multi-source heterogeneous data, and the first repair method is used to repair the anomaly data points. The repaired multi-source heterogeneous data is subjected to first feature standardization processing, and the standardized multi-source heterogeneous data is subjected to first feature alignment processing to obtain the first multi-source heterogeneous data.

3. The method for load transfer and restoration in a distribution network as described in claim 2, characterized in that, Several key factors include: The first transformation analysis method is used to extract features from the first multi-source heterogeneous data to obtain multi-scale time-series features; A causal directed acyclic graph was constructed using the first causal inference algorithm for multi-scale time-series features, and several key factors were identified.

4. The method for load transfer and restoration in a distribution network as described in claim 3, characterized in that, Power grid parameters include: Establish a hierarchical step-by-step simulation mechanism; Under normal operating conditions, steady-state simulation is performed using the first simulation step size to obtain the power grid parameters; Under switching conditions, a fine simulation is performed using the second simulation step size to obtain the power grid parameters; The switching between normal operating conditions and on / off operating conditions is controlled by the first trigger threshold.

5. The method for load transfer and restoration in a distribution network as described in claim 3, characterized in that, Feature extraction of the first multi-source heterogeneous data using the first transformation analysis method includes: The sum of squared coefficients of each wavelet decomposition node is calculated as the energy value to obtain the frequency band energy characteristics; The energy entropy of the high-frequency band is calculated to obtain the energy entropy characteristics of the high-frequency band; The time-domain statistics are calculated to obtain the characteristics of the time-domain statistics; By integrating the frequency band energy characteristics, high-frequency band energy entropy characteristics, and time-domain statistical characteristics, a feature set is obtained; The feature set is filtered, and the spatiotemporal feature matrix is ​​obtained through dimensionality compression.

6. A method for load transfer and restoration in a distribution network as described in claim 3 or 5, characterized in that, Constructing a causal directed acyclic graph using the first causal reasoning algorithm includes: An initial skeleton diagram was constructed using the Gaussian conditional independence test; On the initial skeleton graph, collision nodes are identified through topological structure detection, and edge orientation is performed using greedy equivalence search to obtain a directed acyclic graph; Based on the directed acyclic graph, the key factors with the greatest impact on the target variable are screened out through causal effect assessment; A dynamic update mechanism for Bayesian networks is constructed, which triggers local structure learning when new data arrives, resulting in a dynamically updated causal directed acyclic graph.

7. The method for load transfer and restoration in a distribution network as described in claim 1, characterized in that, The implementation strategies for the distribution network include: When the fault probability prediction result is greater than the first threshold, the corresponding fault occurrence time information is determined; The corresponding power grid parameters are obtained based on the fault occurrence time information; The corresponding power grid parameters are input into the distribution network execution strategy determination model to obtain the corresponding execution strategy.

8. A distribution network load transfer and restoration system, employing the distribution network load transfer and restoration method as described in any one of claims 1-7, characterized in that, include: The preprocessing module is used to acquire multi-source heterogeneous data, preprocess the multi-source heterogeneous data, and obtain the first multi-source heterogeneous data. The feature extraction module is used to extract features from the first multi-source heterogeneous data to obtain multi-scale time-series features, and to perform causal inference operations on the multi-scale time-series features to obtain multiple key factors. The prediction module is used to select a fault prediction model, input the multiple key factors into the fault prediction model, and obtain the fault probability prediction result. The model building module is used to obtain the twin parameters of the distribution network, construct a digital twin model based on the twin parameters, and obtain the power grid parameters. The execution strategy generation module is used to obtain the execution strategy of the distribution network based on the fault probability prediction results and the power grid parameters.

9. A computer device, characterized in that, include: A memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a distribution network load transfer and restoration method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a processor, implements the steps of a power distribution network load transfer and restoration method according to any one of claims 1 to 7.

Images