Method, device and equipment for determining maintenance time of power distribution network

Abstract

The embodiment of the invention provides a method, a device and equipment for determining maintenance time of a power distribution network. The method comprises the following steps: acquiring real-time working condition data and short-time meteorological data of a target power distribution network, and updating a basic heterogeneous graph of the target power distribution network through the short-time meteorological data to obtain a dynamic heterogeneous graph set matched with the short-time meteorological data in an aging manner, and inputting the real-time working condition data and the dynamic heterogeneous graph set into a pre-trained fault risk time-varying prediction model to obtain a fault risk prediction time-varying rate corresponding to the target power distribution network, and determining target maintenance time of the target power distribution network based on the fault risk prediction time-varying rate and a preset maintenance time range. The method is used for improving the fault prediction precision, so that the target maintenance time determined based on fault prediction can be matched with the risk prevention and control target in extreme weather.

Description

Technical Field

[0001] This application relates to the field of power system operation and maintenance technology, and in particular to a method, apparatus and equipment for determining the maintenance time of a distribution network. Background Technology

[0002] As a crucial component of the power system, the stable operation of the distribution network directly impacts the reliability and security of power supply. With the increasing frequency of extreme weather events such as typhoons, thunderstorms, and torrential rains, the failure rate of distribution network equipment has risen significantly, necessitating the reduction of failure risks through scientifically scheduled maintenance.

[0003] In existing technologies, recurrent neural networks are used to learn the temporal dependencies of equipment maintenance history data, such as the number of minor repairs and operating time. At the same time, convolutional neural networks are used to extract geographical environmental features, such as altitude and humidity. Finally, the recurrent neural network and the convolutional neural network jointly predict the time of the next failure, and thus predict the next maintenance interval.

[0004] Existing technologies suffer from poor fault prediction accuracy under extreme weather conditions, which makes it difficult for maintenance time determined based on fault prediction to match the risk prevention and control objectives of the distribution network under extreme weather conditions. Summary of the Invention

[0005] This application provides a method, apparatus, and equipment for determining the maintenance time of a power distribution network, in order to improve the accuracy of fault prediction and enable the target maintenance time determined based on fault prediction to match the risk prevention and control target under extreme weather conditions.

[0006] In a first aspect, embodiments of this application provide a method for determining the maintenance time of a power distribution network, including:

[0007] Acquire real-time operating data and short-term weather data of the target power distribution network.

[0008] By updating the basic heterogeneous map of the target distribution network using short-term meteorological data, a dynamic heterogeneous map set matching the timeliness of the short-term meteorological data is obtained.

[0009] Real-time operating data and dynamic heterogeneous graph sets are input into a pre-trained time-varying fault risk prediction model to obtain the time-varying rate of fault risk prediction corresponding to the target distribution network.

[0010] Based on the time-varying rate of fault risk prediction and the preset maintenance time range, the target maintenance time of the target distribution network is determined.

[0011] In one possible implementation, before acquiring real-time operating data and short-term meteorological data of the target distribution network in conjunction with the first aspect, the method further includes:

[0012] Obtain the distribution network topology data and meteorological comprehensive data of the target distribution network.

[0013] By parsing the distribution network topology data, a set of device nodes and a set of electrical connections are generated. The set of device nodes contains the feature tuples of each device node in the distribution network topology data. The set of electrical connections contains the electrical connection relationships between each device.

[0014] By analyzing comprehensive meteorological data, a set of virtual meteorological nodes is generated; the set of virtual meteorological nodes contains feature tuples of virtual meteorological nodes corresponding to each meteorological station in the comprehensive meteorological data.

[0015] Based on the feature tuples of each device node and its corresponding device node, the electrical connection edges corresponding to each electrical connection relationship, the preset weights of the electrical connection edges, and the feature tuples of each meteorological virtual node and its corresponding meteorological virtual node, a basic heterogeneous graph of the target distribution network is constructed.

[0016] In one possible implementation, in conjunction with the first aspect, the basic heterogeneity map of the target distribution network is updated using short-term meteorological data to obtain a dynamic heterogeneity map set that matches the timeliness of the short-term meteorological data, including:

[0017] By analyzing meteorological change information under different time slices in short-term meteorological data, multiple sets of time-series correlated meteorological data corresponding to the meteorological virtual nodes are obtained.

[0018] Based on each set of time-series correlated meteorological data, the basic heterogeneous map is updated to obtain the dynamic heterogeneous map corresponding to each set of time-series correlated meteorological data.

[0019] Based on each dynamic heterogeneous graph, a set of dynamic heterogeneous graphs matching the timeliness of short-term meteorological data is obtained.

[0020] In one possible implementation, in conjunction with the first aspect, the basic heterogeneous graph is updated according to each set of time-series correlated meteorological data to obtain a dynamic heterogeneous graph corresponding to each set of time-series correlated meteorological data, including:

[0021] Based on time-series correlated meteorological data, dynamic distance thresholds that are related to both meteorological virtual nodes and device nodes are obtained.

[0022] Based on the meteorological station coordinates in the feature tuple of the meteorological virtual node and the device coordinates in the feature tuple of the device node, the coordinate distance between the meteorological station corresponding to the meteorological virtual node and the device corresponding to the device node is obtained.

[0023] When the coordinate distance is less than the dynamic distance threshold, a meteorological influence edge is generated between the meteorological virtual node and the device node. The meteorological influence edge weight is obtained based on the coordinate distance, time-series associated meteorological data, device type in the device node feature tuple, and preset parameters.

[0024] The basic heterogeneous graph is updated by updating each meteorological influence edge and its weight, resulting in a dynamic heterogeneous graph corresponding to the time-series correlated meteorological data.

[0025] In one possible implementation, in conjunction with the first aspect, the training process of the time-varying prediction model for fault risk includes:

[0026] Obtain the meta-task support set corresponding to the historical fault sample data of the target distribution network; the meta-task support set contains multiple sample multidimensional data and the fault label corresponding to each sample multidimensional data.

[0027] By inputting the multidimensional sample data into the time-varying prediction model for fault risk, adversarial sample data corresponding to the multidimensional sample data and fault labels are obtained.

[0028] Based on multidimensional sample data, fault labels, and a first preset loss function, the first parameter of the time-varying prediction model for fault risk is updated.

[0029] Based on the updated first parameter, the adversarial sample data is input into the time-varying prediction model of fault risk to obtain the time-varying rate of fault risk prediction corresponding to the adversarial sample data.

[0030] The generalization loss value between the time-varying rate of fault risk prediction and the fault label corresponding to the adversarial sample data is calculated using the second preset loss function.

[0031] The second parameter of the fault risk time-varying prediction model is updated based on the generalization loss value. Based on the updated second parameter, the process of inputting multidimensional sample data into the fault risk time-varying prediction model is repeated to obtain adversarial sample data corresponding to the multidimensional sample data and fault labels until both the first preset loss function and the second preset loss function converge to obtain the trained fault risk time-varying prediction model. The second parameter includes the first parameter.

[0032] In one possible implementation, in conjunction with the first aspect, the multidimensional sample data is input into the time-varying prediction model for fault risk to obtain adversarial example data corresponding to the multidimensional sample data and fault labels, including:

[0033] Feature fusion processing is performed on the sample dynamic heterogeneous graph in the multidimensional sample data to obtain a multidimensional spatiotemporal feature matrix.

[0034] Based on the multidimensional spatiotemporal feature matrix and preset physical constraints, the physical constraint loss value is obtained.

[0035] Adversarial example data is generated based on physical constraint loss values ​​and historical operating condition data and historical meteorological data from the multidimensional sample data.

[0036] In one possible implementation, in conjunction with the first aspect, the first parameters of the time-varying prediction model for fault risk are updated based on multidimensional sample data, fault labels, and a first preset loss function, including:

[0037] By inputting the multidimensional sample data into the time-varying prediction model for fault risk, the predicted time-varying rate of fault risk corresponding to the multidimensional sample data is obtained.

[0038] The comprehensive loss value between the time-varying rate of fault risk prediction and the fault label corresponding to the multidimensional data of the sample is calculated using the first preset loss function.

[0039] The first parameter of the time-varying prediction model for fault risk is updated based on the comprehensive loss value.

[0040] In one possible implementation, in conjunction with the first aspect, the target maintenance time for the target distribution network is determined based on the time-varying rate of fault risk prediction and a preset maintenance time range, including:

[0041] The preset maintenance time range is divided into multiple maintenance time windows; each maintenance time window consists of multiple maintenance time points.

[0042] Based on the time-varying rate of fault risk prediction, the risk cost corresponding to each maintenance time window is obtained.

[0043] Based on risk costs and preset sorting rules, multiple maintenance time windows are sorted.

[0044] The maintenance time window located at the preset sorting position is set as the target maintenance time.

[0045] Secondly, embodiments of this application provide a device for determining the maintenance time of a power distribution network, comprising:

[0046] The acquisition module is used to acquire real-time operating data and short-term meteorological data of the target power distribution network.

[0047] The graph construction module is used to update the basic heterogeneous graph of the target distribution network using short-term meteorological data, and obtain a dynamic heterogeneous graph set that matches the timeliness of the short-term meteorological data.

[0048] The model prediction module is used to input real-time operating condition data and dynamic heterogeneous graph set into a pre-trained fault risk time-varying prediction model to obtain the fault risk prediction time-varying rate corresponding to the target distribution network.

[0049] The determination module is used to determine the target maintenance time of the target distribution network based on the time variability of the fault risk prediction and the preset maintenance time range.

[0050] In one possible implementation, in conjunction with the second aspect, the acquisition module is further configured to:

[0051] Obtain the distribution network topology data and meteorological comprehensive data of the target distribution network.

[0052] Correspondingly, the graph building module is also used for:

[0053] By parsing the distribution network topology data, a set of equipment nodes and a set of electrical connections are generated. The equipment node set contains the feature tuples of each equipment node corresponding to each device in the distribution network topology data; the electrical connection set contains the electrical connection relationships between each device. By parsing comprehensive meteorological data, a set of meteorological virtual nodes is generated; the meteorological virtual node set contains the feature tuples of each meteorological station corresponding to each meteorological station in the comprehensive meteorological data. Based on each equipment node feature tuple and its corresponding equipment node, each electrical connection relationship and its corresponding electrical connection edge, the preset electrical connection edge weights, and each meteorological virtual node feature tuple and its corresponding meteorological virtual node, a basic heterogeneous graph of the target distribution network is constructed.

[0054] In one possible implementation, in conjunction with the second aspect, the graph construction module is specifically used for:

[0055] By analyzing meteorological change information under different time slices in short-term meteorological data, multiple sets of time-series correlated meteorological data corresponding to the meteorological virtual nodes are obtained.

[0056] Based on each set of time-series correlated meteorological data, the basic heterogeneous map is updated to obtain the dynamic heterogeneous map corresponding to each set of time-series correlated meteorological data.

[0057] Based on each dynamic heterogeneous graph, a set of dynamic heterogeneous graphs matching the timeliness of short-term meteorological data is obtained.

[0058] In one possible implementation, in conjunction with the second aspect, the graph construction module is specifically used for:

[0059] Based on time-series correlated meteorological data, dynamic distance thresholds that are related to both meteorological virtual nodes and device nodes are obtained.

[0060] Based on the meteorological station coordinates in the feature tuple of the meteorological virtual node and the device coordinates in the feature tuple of the device node, the coordinate distance between the meteorological station corresponding to the meteorological virtual node and the device corresponding to the device node is obtained.

[0061] When the coordinate distance is less than the dynamic distance threshold, a meteorological influence edge is generated between the meteorological virtual node and the device node. The meteorological influence edge weight is obtained based on the coordinate distance, time-series associated meteorological data, device type in the device node feature tuple, and preset parameters.

[0062] The basic heterogeneous graph is updated by updating each meteorological influence edge and its weight, resulting in a dynamic heterogeneous graph corresponding to the time-series correlated meteorological data.

[0063] In one possible implementation, in conjunction with the second aspect, the acquisition module is further configured to:

[0064] Obtain the meta-task support set corresponding to the historical fault sample data of the target distribution network; the meta-task support set contains multiple sample multidimensional data and the fault label corresponding to each sample multidimensional data.

[0065] Correspondingly, the model prediction module is also used for:

[0066] By inputting the multidimensional sample data into the time-varying prediction model for fault risk, adversarial sample data corresponding to the multidimensional sample data and fault labels are obtained.

[0067] Based on multidimensional sample data, fault labels, and a first preset loss function, the first parameter of the time-varying prediction model for fault risk is updated.

[0068] Based on the updated first parameter, the adversarial sample data is input into the time-varying prediction model of fault risk to obtain the time-varying rate of fault risk prediction corresponding to the adversarial sample data.

[0069] The generalization loss value between the time-varying rate of fault risk prediction and the fault label corresponding to the adversarial sample data is calculated using the second preset loss function.

[0070] The second parameter of the fault risk time-varying prediction model is updated based on the generalization loss value. Based on the updated second parameter, the process of inputting multidimensional sample data into the fault risk time-varying prediction model is repeated to obtain adversarial sample data corresponding to the multidimensional sample data and fault labels until both the first preset loss function and the second preset loss function converge to obtain the trained fault risk time-varying prediction model. The second parameter includes the first parameter.

[0071] In one possible implementation, in conjunction with the second aspect, the model prediction module is specifically used for:

[0072] Feature fusion processing is performed on the sample dynamic heterogeneous graph in the multidimensional sample data to obtain a multidimensional spatiotemporal feature matrix.

[0073] Based on the multidimensional spatiotemporal feature matrix and preset physical constraints, the physical constraint loss value is obtained.

[0074] Adversarial example data is generated based on physical constraint loss values ​​and historical operating condition data and historical meteorological data from the multidimensional sample data.

[0075] In one possible implementation, in conjunction with the second aspect, the model prediction module is specifically used for:

[0076] By inputting the multidimensional sample data into the time-varying prediction model for fault risk, the predicted time-varying rate of fault risk corresponding to the multidimensional sample data is obtained.

[0077] The comprehensive loss value between the time-varying rate of fault risk prediction and the fault label corresponding to the multidimensional data of the sample is calculated using the first preset loss function.

[0078] The first parameter of the time-varying prediction model for fault risk is updated based on the comprehensive loss value.

[0079] In one possible implementation, in conjunction with the second aspect, the determining module is specifically used for:

[0080] The preset maintenance time range is divided into multiple maintenance time windows; each maintenance time window consists of multiple maintenance time points.

[0081] Based on the time-varying rate of fault risk prediction, the risk cost corresponding to each maintenance time window is obtained.

[0082] Based on risk costs and preset sorting rules, multiple maintenance time windows are sorted.

[0083] The maintenance time window located at the preset sorting position is set as the target maintenance time.

[0084] Thirdly, embodiments of this application provide an electronic device, including: a processor, and a memory communicatively connected to the processor.

[0085] The memory stores the instructions that the computer executes.

[0086] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0087] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0088] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0089] This application provides a method, apparatus, and equipment for determining the maintenance time of a distribution network. It acquires real-time operating data and short-term meteorological data of the target distribution network, updates the basic heterogeneous graph of the target distribution network using the short-term meteorological data, and obtains a dynamic heterogeneous graph set that matches the timeliness of the short-term meteorological data. The real-time operating data and the dynamic heterogeneous graph set are input into a pre-trained fault risk time-varying prediction model to obtain the fault risk prediction time-varying rate corresponding to the target distribution network. Based on the fault risk prediction time-varying rate and a preset maintenance time range, the target maintenance time of the target distribution network is determined. By fusing the power grid topology with the spatiotemporal correlation of meteorology to obtain the dynamic heterogeneous graph set, the resource mismatch problem caused by insufficient meteorological modeling in traditional static models is avoided, thus improving the accuracy of fault prediction. This allows the target maintenance time determined based on fault prediction to match the risk prevention and control targets under extreme weather conditions. Attached Figure Description

[0090] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0091] Figure 1 A schematic diagram illustrating a method for determining the maintenance time of a power distribution network as provided in this application;

[0092] Figure 2 A flowchart illustrating a method for determining the maintenance time of a power distribution network provided in this application. Figure 1 ;

[0093] Figure 3 A flowchart illustrating a method for determining the maintenance time of a power distribution network provided in this application. Figure 2 ;

[0094] Figure 4 A flowchart illustrating a method for determining the maintenance time of a power distribution network provided in this application. Figure 3 ;

[0095] Figure 5 A flowchart illustrating a method for determining the maintenance time of a power distribution network provided in this application. Figure 4 ;

[0096] Figure 6 A flowchart illustrating a method for determining the maintenance time of a power distribution network provided in this application. Figure 5 ;

[0097] Figure 7 A specific example diagram illustrating a method for determining the maintenance time of a power distribution network provided in this application;

[0098] Figure 8 A schematic diagram of a device for determining the maintenance time of a power distribution network provided in this application;

[0099] Figure 9 A schematic diagram of the structure of the electronic device provided in this application.

[0100] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0101] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0102] First, the terms used in this application will be explained:

[0103] Dynamic heterogeneous graph: A graph structure that includes equipment nodes (such as transformers and switchgear) and meteorological virtual nodes (such as weather stations). Equipment nodes are physically connected via electrical connection edges, while meteorological virtual nodes are affected by meteorological influence edges, representing the time-varying impact of environmental factors on the equipment. For example, in a typhoon scenario, the dynamic heterogeneous graph includes a substation (equipment node) and a typhoon center weather station (meteorological virtual node). Electrical connection edges represent the connection between the transformer and the feeder, and meteorological influence edges represent the impact of typhoon wind speed on the substation equipment.

[0104] Equipment node: The representation of physical equipment (such as transformers and switchgear) in the distribution network in the diagram structure.

[0105] Virtual meteorological nodes: The representation of meteorological monitoring points (such as weather stations) in a graph structure.

[0106] Secondly, the application background of the embodiments of this application will be explained:

[0107] As a critical link in the power system, the stable operation of the distribution network directly affects the reliability and security of power supply. With the frequent occurrence of extreme weather events such as typhoons, thunderstorms, and torrential rains, the failure rate of distribution network equipment increases significantly, necessitating the reduction of failure risks through scientifically scheduled maintenance. Current technologies employ recurrent neural networks to learn the temporal dependencies of historical equipment maintenance data, such as the number of minor repairs and operating time; simultaneously, convolutional neural networks extract geographical environmental features, such as altitude and humidity; finally, the recurrent and convolutional neural networks jointly predict the time of the next failure, thereby predicting the next maintenance interval. Therefore, existing technologies suffer from poor failure prediction accuracy under extreme weather conditions, making it difficult for maintenance times determined based on failure prediction to align with the risk control objectives of the distribution network under extreme weather conditions.

[0108] To address the aforementioned issues, the inventors investigated whether it was possible to accurately predict the probability of distribution network faults by constructing a dynamic heterogeneous graph that integrates the spatiotemporal correlation between power grid topology and meteorological data, combined with physical constraint embedding and few-shot learning techniques, thereby determining the distribution network maintenance time based on the fault probability. The inventors proposed a method for determining the maintenance time of a distribution network. This method involves acquiring real-time operating data and short-term meteorological data of the target distribution network. The basic heterogeneous graph of the target distribution network is updated using the short-term meteorological data, resulting in a dynamic heterogeneous graph set that matches the timeliness of the short-term meteorological data. The real-time operating data and the dynamic heterogeneous graph set are then input into a pre-trained fault risk time-varying prediction model to obtain the fault risk prediction time-varying rate corresponding to the target distribution network. Based on the fault risk prediction time-varying rate and a preset maintenance time range, the target maintenance time of the target distribution network is determined. Ultimately, this method improves the accuracy of fault prediction, enabling the target maintenance time determined based on fault prediction to match risk prevention and control targets under extreme weather conditions.

[0109] Starting with existing technological problems, the inventors gradually explored solutions. Firstly, the insufficient ability of existing technologies to handle sudden faults stems from inadequate dynamic meteorological modeling. The inventors proposed constructing a dynamic heterogeneous graph containing equipment nodes and virtual meteorological nodes, dynamically adjusting meteorological weights based on real-time meteorological information to accurately characterize the time-varying impact of environmental factors. Secondly, considering the mismatch between physical laws and purely data-driven models, the inventors proposed adding a physical constraint loss function to the time-varying fault risk prediction model, forcing the prediction results to conform to the physical laws of the circuit. Thirdly, considering the scarcity of distribution network fault samples required for training the time-varying fault risk prediction model, the inventors proposed a gradient based on physical rule constraints, generating adversarial examples based on fault samples to improve the generalization ability of the time-varying fault risk prediction model under sparse samples.

[0110] Taking the intelligent maintenance decision-making system for power distribution networks in scenarios with frequent extreme weather events as an example, combined with Figure 1 This illustrates the specific application scenario of the method for determining the maintenance time of a power distribution network provided in this application. For example... Figure 1 As shown, the specific application scenarios of this application include a power grid data acquisition system 101, a meteorological information comprehensive analysis and processing system 102, and a distribution network intelligent maintenance decision-making system 103. The distribution network intelligent maintenance decision-making system 103 acquires real-time operating condition data of the target distribution network through the power grid data acquisition system 101 and acquires corresponding short-term meteorological data of the target distribution network through the meteorological information comprehensive analysis and processing system 102. The distribution network intelligent maintenance decision-making system 103 deploys the distribution network maintenance time determination device proposed in this application. Based on real-time operating condition data and short-term meteorological data, the device determines the target maintenance time, i.e., the optimal maintenance time, for the target distribution network.

[0111] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0112] Figure 2 A flowchart illustrating a method for determining the maintenance time of a power distribution network provided in this application. Figure 1 ,like Figure 2 As shown, the method includes:

[0113] S201. Obtain real-time operating data and short-term meteorological data of the target distribution network.

[0114] Short-term meteorological data includes real-time meteorological monitoring data and meteorological forecast data for the target future time period. For example, meteorological forecast data for the next week.

[0115] S202. Update the basic heterogeneous map of the target distribution network using short-term meteorological data to obtain a dynamic heterogeneous map set that matches the timeliness of the short-term meteorological data.

[0116] The basic heterogeneous graph includes device nodes and their corresponding device node feature tuples, meteorological virtual nodes and their corresponding meteorological virtual node feature tuples, and electrical connection edges.

[0117] In this step, by analyzing meteorological change information under different time slices in short-term meteorological data, multiple sets of time-series correlated meteorological data corresponding to the meteorological virtual nodes are obtained. Based on each set of time-series correlated meteorological data, the basic heterogeneous graph is updated to obtain a dynamic heterogeneous graph corresponding to each set of time-series correlated meteorological data. Based on each dynamic heterogeneous graph, a set of dynamic heterogeneous graphs matching the timeliness of the short-term meteorological data is obtained.

[0118] Specifically, based on time-series correlated meteorological data, dynamic distance thresholds related to both the meteorological virtual node and the device node are obtained. Based on the meteorological station coordinates in the feature tuple of the meteorological virtual node and the device coordinates in the feature tuple of the device node, the coordinate distance between the meteorological station corresponding to the meteorological virtual node and the device corresponding to the device node is obtained.

[0119] When the coordinate distance is less than the dynamic distance threshold, a meteorological influence edge is generated between the meteorological virtual node and the device node. The weight of the meteorological influence edge is then determined based on the coordinate distance, time-series associated meteorological data, the device type in the device node's feature tuple, and preset parameters. The basic heterogeneous graph is updated using each meteorological influence edge and its weight to obtain the dynamic heterogeneous graph corresponding to the time-series associated meteorological data.

[0120] S203. Input the real-time operating condition data and dynamic heterogeneous graph set into the pre-trained fault risk time-varying prediction model to obtain the fault risk prediction time-varying rate corresponding to the target distribution network.

[0121] Among them, the time-varying prediction model for fault risk is trained based on historical fault sample data of the target distribution network. This historical fault sample data includes multidimensional sample data and its corresponding fault labels. The multidimensional sample data includes historical operating condition data, historical meteorological data, historical distribution network topology data, and a sample dynamic heterogeneous diagram constructed based on historical distribution network topology data and historical meteorological data.

[0122] S204. Based on the time-varying rate of fault risk prediction and the preset maintenance time range, determine the target maintenance time of the target distribution network.

[0123] In this step, the preset maintenance time range is divided into multiple maintenance time windows. Based on the time-varying rate of fault risk prediction, the risk cost corresponding to each maintenance time window is obtained. Then, based on the risk cost and preset sorting rules, the multiple maintenance time windows are sorted, and the maintenance time window located at the preset sorting position is determined as the target maintenance time. Here, a maintenance time window consists of multiple maintenance time points; the risk cost includes a risk change rate penalty value and a risk integral.

[0124] Specifically, based on risk cost and preset sorting rules, multiple maintenance time windows are sorted, and the maintenance time window located at the preset sorting position is determined as the target maintenance time. This can be achieved by sorting the multiple maintenance time windows from low to high risk cost, and determining the maintenance time window at the top as the target maintenance time; that is, determining the maintenance time window with the lowest risk cost as the target maintenance time.

[0125] This application provides a method for determining the maintenance time of a distribution network. The method involves acquiring real-time operating data and short-term meteorological data of the target distribution network, updating the basic heterogeneous diagram of the target distribution network using the short-term meteorological data to obtain a dynamic heterogeneous diagram set that matches the timeliness of the short-term meteorological data, inputting the real-time operating data and the dynamic heterogeneous diagram set into a pre-trained fault risk time-varying prediction model to obtain the fault risk prediction time-varying rate corresponding to the target distribution network, and determining the target maintenance time of the target distribution network based on the fault risk prediction time-varying rate and a preset maintenance time range.

[0126] This application embodiment obtains a dynamic heterogeneous graph set by integrating the power grid topology and the spatiotemporal correlation of meteorology, avoiding the resource mismatch problem caused by insufficient meteorological modeling in traditional static models, and achieving the effect of improving the accuracy of fault prediction. This enables the target maintenance time determined based on fault prediction to match the risk prevention and control target under extreme weather conditions.

[0127] Figure 3 A flowchart illustrating a method for determining the maintenance time of a power distribution network provided in this application. Figure 2 ,like Figure 3 As shown, in this embodiment... Figure 2 Based on the embodiments, a method for determining the maintenance time of a power distribution network is described in detail. Before S201, the method further includes:

[0128] S301. Obtain the distribution network topology data and meteorological comprehensive data of the target distribution network.

[0129] S302. By parsing the power distribution network topology data, generate a set of equipment nodes and a set of electrical connection relationships.

[0130] The device node set contains the device node feature tuples corresponding to each device in the power distribution network topology data; the electrical connection relationship set contains the electrical connection relationships between each device.

[0131] In one possible implementation, regular expressions are used to match key fields in the distribution network topology data and extract identifiers corresponding to equipment types. For example, 0 represents critical power supply equipment, 1 represents switch control equipment, and 2 represents ordinary terminal equipment. The longitude and latitude of the WGS84 geographic coordinates of each device in the distribution network topology data are then converted to UTM projected coordinates. Simultaneously, the equipment parameters in the distribution network topology data are vectorized, that is, the equipment parameters are encoded into 128-dimensional vectors. These 128-dimensional vectors contain electrical parameters such as rated current, insulation class, and impedance, as well as status parameters such as service life and cumulative number of faults. Finally, the parsed results constitute a set of equipment nodes. Its expression is:

[0132]

[0133] in, Indicates device The device node feature tuple has a total of One device; Indicates device UTM coordinates; Indicates device The type of equipment, 0 represents critical power supply equipment, 1 represents switch control equipment, and 2 represents ordinary terminal equipment; This represents the device parameter vector of device i.

[0134] Subsequently, by parsing the electrical adjacency relationships between devices in the distribution network topology data, the electrical connection edge set is obtained. Its expression is:

[0135]

[0136] in, and Representing the equipment and equipment The device node feature tuple, Indicates from node To the node The directed electrical connection edge; The values ​​are 0 and 1. A value of 1 indicates the device and equipment There is a physical electrical connection between them, that is, there is an electrical connection edge. A value of 0 indicates the device and equipment There is no electrical connection between them.

[0137] S303. Generate a set of virtual meteorological nodes by analyzing comprehensive meteorological data.

[0138] The meteorological virtual node set contains feature tuples of meteorological virtual nodes corresponding to each meteorological station in the comprehensive meteorological data.

[0139] In one possible implementation, a set of virtual meteorological nodes is established based on integrated meteorological data, and its expression is:

[0140]

[0141] in, The feature tuples representing the meteorological virtual nodes have a total of A feature tuple of a virtual meteorological node; and They represent weather stations. latitude and longitude It represents a meteorological feature vector, which includes at least one of temperature, humidity, wind speed, atmospheric pressure, rainfall intensity, and light intensity.

[0142] S304. Based on the set of equipment nodes, the set of electrical connection relationships, and the set of meteorological virtual nodes, construct the basic heterogeneous diagram of the target distribution network.

[0143] In this step, a basic heterogeneous graph of the target distribution network is constructed based on the feature tuples of each device node and its corresponding device node, the electrical connection edges corresponding to each electrical connection relationship, the preset weights of the electrical connection edges, and the feature tuples of each meteorological virtual node and its corresponding meteorological virtual node. The preset weights of the electrical connection edges can be constant values ​​such as 1.

[0144] This application provides a method for determining the maintenance time of a distribution network. It acquires distribution network topology data and comprehensive meteorological data of the target distribution network. By parsing the distribution network topology data, it generates a set of equipment nodes and a set of electrical connection relationships. By parsing the comprehensive meteorological data, it generates a set of meteorological virtual nodes. Based on the set of equipment nodes, the set of electrical connection relationships, and the set of meteorological virtual nodes, it constructs a basic heterogeneous diagram of the target distribution network. Therefore, by constructing the basic heterogeneous diagram of the target distribution network in advance, it provides a data foundation for obtaining a dynamic heterogeneous diagram based on short-term meteorological data. This avoids the repeated construction of the basic structure and data redundancy during the dynamic update process based on short-term meteorological data, thus ensuring the stability and efficiency of the dynamic heterogeneous diagram generation.

[0145] Figure 4 A flowchart illustrating a method for determining the maintenance time of a power distribution network provided in this application. Figure 3 ,like Figure 4 As shown, this embodiment provides a detailed description of S202 based on any of the above embodiments. The method includes:

[0146] S401. Based on time-series correlated meteorological data, obtain dynamic distance thresholds that are related to both meteorological virtual nodes and equipment nodes.

[0147] S402. Based on the coordinates of the weather station and the equipment, obtain the coordinate distance between the weather station and the equipment.

[0148] In this step, since meteorological impacts have a spatial attenuation effect, such as the impact of heavy rain gradually weakening from the center to the outside, it is necessary to quantify the geometric spatial distance between the equipment and the meteorological source and calculate the coordinate distance between any equipment and any meteorological station in order to quantify the spatial attenuation effect and provide an existence criterion for the meteorological impact edge.

[0149] Specifically, based on the meteorological station coordinates in the feature tuple of the meteorological virtual node and the device coordinates in the feature tuple of the device node, the coordinate distance between the meteorological station corresponding to the meteorological virtual node and the device corresponding to the device node is obtained.

[0150] In one possible implementation, coordinate distance The calculation formula is:

[0151]

[0152] in, This represents the Earth's radius, typically taken as 6371 km. Indicates device and weather station The difference in latitude between them; Indicates device and weather station The longitude difference between them will determine the device's UTM coordinates. Convert to geographic coordinates Then the corresponding difference can be calculated, that is and The weather station coordinates are kept in WGS84 format. and Representing the equipment and weather station The geographical latitude.

[0153] S403. When the coordinate distance is less than the dynamic distance threshold, generate a meteorological influence edge between the meteorological virtual node and the device node, and determine the weight of the meteorological influence edge.

[0154] In this step, since the scope of meteorological influence changes dynamically with the environment, it is necessary to adjust the connection relationship between the meteorological virtual node and the device node in real time. Based on the coordinate distance and time-series associated meteorological data between any device node and any meteorological virtual node, a meteorological influence edge set containing multiple meteorological influence edges is obtained, and the meteorological influence edge weight of each meteorological influence edge is determined.

[0155] Specifically, the meteorological influence edge weights are obtained based on coordinate distance, time-series correlated meteorological data, equipment type in the equipment node feature tuple, and preset parameters. The preset parameters are sensitivity factors shared by both equipment type and meteorological type, including temperature sensitivity factor, humidity sensitivity factor, wind speed sensitivity factor, atmospheric pressure sensitivity factor, rainfall intensity sensitivity factor, and light intensity sensitivity factor, representing the equipment's sensitivity to a certain meteorological type.

[0156] In one possible implementation, the meteorological influence boundary set is calculated. The expression is:

[0157]

[0158] in, Indicates from the meteorological virtual node To device node The directional meteorological influence, Indicates connection with meteorological virtual nodes and device nodes The dynamic distance threshold related to the real-time meteorological information of the corresponding area.

[0159] Meteorological influence edge weight The calculation formula is:

[0160]

[0161] in, This represents the learnable spatial decay coefficient, with an initial value of 0.5; Representation of meteorological feature vectors Each element in the vector corresponds to a device type sensitivity vector, which includes at least one of the following: temperature sensitivity factor, humidity sensitivity factor, wind speed sensitivity factor, atmospheric pressure sensitivity factor, rainfall intensity sensitivity factor, and light intensity sensitivity factor. These vectors correspond one-to-one with the previous meteorological feature vectors. This represents the Hadamard product.

[0162] S404. Update the basic heterogeneous graph by each meteorological influence edge and each meteorological influence edge weight to obtain the dynamic heterogeneous graph corresponding to the time-series correlated meteorological data.

[0163] In one possible implementation, a multimodal meteorological data fusion mechanism is introduced to fuse short-term meteorological data (such as temperature and humidity) with remote sensing data (such as satellite cloud images and radar echoes) at multiple scales. The weights of meteorological influence edges are determined by integrating the multimodal meteorological data, thereby further optimizing the dynamic heterogeneous map. Specifically, a cross-modal attention mechanism is used to dynamically adjust the influence of different modalities of meteorological data on the weights of meteorological influence edges. For example, in scenarios where typhoons and thunderstorms coexist, satellite cloud images can provide large-scale meteorological structure information, while radar echoes can capture localized severe convection characteristics. By dynamically fusing the features of these two modalities through the attention mechanism, the ability of the dynamic heterogeneous map to represent complex meteorological conditions is improved.

[0164] By inputting the optimized dynamic heterogeneous graph based on the aforementioned techniques into a pre-trained time-varying failure risk prediction model, the model can more accurately capture the time-varying impact of complex meteorological factors such as typhoons and thunderstorms on equipment failure rates, thereby optimizing the prediction of time-varying risk rates. Simultaneously, the cross-modal attention mechanism can dynamically suppress interference from redundant meteorological information, enhancing the sensitivity of the dynamic heterogeneous graph to key meteorological features.

[0165] This application provides a method for determining the maintenance time of a power distribution network. It obtains a dynamic distance threshold related to both meteorological virtual nodes and equipment nodes based on time-series correlated meteorological data. The coordinate distance between the meteorological station and the equipment is then determined based on the coordinates of the meteorological station and the equipment. When the coordinate distance is less than the dynamic distance threshold, a meteorological influence edge is generated between the meteorological virtual node and the equipment node, and the weight of this edge is determined. Finally, the basic heterogeneous graph is updated using each meteorological influence edge and its weight to obtain a dynamic heterogeneous graph corresponding to the time-series correlated meteorological data.

[0166] This application's embodiments address the problem of insufficient dynamic meteorological impact modeling in existing technologies by updating a basic heterogeneous graph based on time-series correlated meteorological data to obtain a dynamic heterogeneous graph. The dynamic heterogeneous graph integrates the power grid topology with the spatiotemporal correlation of meteorology. The heterogeneous structure of equipment nodes and meteorological virtual nodes allows the dynamic heterogeneous graph to simultaneously represent the influence of physical connections and environmental factors. By dynamically suppressing interference from distant equipment based on time-series correlated meteorological data and the coordinate distance between equipment and meteorological stations, priority is given to the failure risk of equipment in severe meteorological areas (such as strong wind areas). Overall, the above technical means enable the dynamic heterogeneous graph to accurately represent the time-varying impact of environmental factors on equipment, thereby improving the accuracy of fault prediction under extreme weather conditions.

[0167] Figure 5 A flowchart illustrating a method for determining the maintenance time of a power distribution network provided in this application. Figure 4 ,like Figure 5 As shown, this embodiment, based on any of the above embodiments, provides a detailed description of a method for determining the maintenance time of a power distribution network. The training process of the time-varying fault risk prediction model in this method includes:

[0168] S501. Obtain the meta-task support set corresponding to the historical fault sample data of the target distribution network. The meta-task support set contains multiple sample multidimensional data and the fault label corresponding to each sample multidimensional data.

[0169] S502. Input the multidimensional sample data into the time-varying prediction model for fault risk to obtain adversarial sample data corresponding to the multidimensional sample data and fault labels.

[0170] Among them, adversarial sample data is enhanced sample data generated by physically perturbing the multidimensional sample data based on sample multidimensional data. The adversarial sample data strictly satisfies the current balance constraint.

[0171] In this step, feature fusion processing is performed on the sample dynamic heterogeneous graph in the multidimensional sample data to obtain a multidimensional spatiotemporal feature matrix. Then, based on the multidimensional spatiotemporal feature matrix and preset physical constraints, a physical constraint loss value is obtained. Finally, adversarial sample data is generated based on the physical constraint loss value and historical operating condition data and historical meteorological data in the multidimensional sample data.

[0172] Specifically, a physical constraint loss function is constructed using a differentiable physics engine, embedding Kirchhoff's current law and Ohm's law into the training process of the time-varying prediction model for fault risk. First, based on the topology and electrical parameters of devices in the sample dynamic heterogeneous graph, a topological correlation matrix and an electrical admittance matrix are generated. Then, combined with voltage time-series data from the multidimensional sample data, the voltage dynamic rate of change is calculated using the five-point difference method. Finally, a dual physical constraint loss function is constructed, containing the current continuity residual term corresponding to Kirchhoff's current law (the algebraic sum of currents flowing into nodes approaches 0) and the voltage-current relationship residual term corresponding to Ohm's law. The physical constraint loss value is obtained by calculating the sum of squared residuals.

[0173] Based on historical operating condition data and historical meteorological data in the multidimensional sample data, small perturbations are added to generate initial adversarial samples. Then, the physical constraint loss function mentioned above is used to verify the physical rationality of the initial adversarial samples, and finally adversarial sample data that conforms to physical laws is obtained.

[0174] This step enhances the model's generalization ability by generating adversarial examples guided by physics, improving the model's robustness in small sample scenarios and avoiding overfitting issues.

[0175] In one possible implementation, adversarial example data is generated based on the physical constraint loss value and historical operating condition data and historical meteorological data from the multidimensional sample data. The corresponding expression is:

[0176]

[0177] in, This represents an adversarial example projected based on the physical constraint loss value; As a projection operator for the physical feasible region, after perturbation along the gradient direction for a short-circuit fault sample, the projection operator ensures that the node currents after perturbation still satisfy the condition that the sum of the currents of the input nodes is equal to the sum of the currents of the output nodes. , where is the amplitude of the Gaussian distribution perturbation, controlling for sample diversity; It is a symbolic function; This represents the physical loss gradient of the physical constraint loss value with respect to the multidimensional data X of the sample.

[0178] S503. Based on the multidimensional sample data, fault labels, and the first preset loss function, update the first parameter of the time-varying prediction model for fault risk.

[0179] The first parameter includes a meta-learning parameter, which is optimized to generate adversarial examples specific to a particular fault mode.

[0180] The first preset loss function is the physical constraint loss function, which is used to determine whether the adversarial examples generated based on multidimensional sample data conform to physical laws.

[0181] In this step, the multidimensional sample data is input into the time-varying prediction model for fault risk to obtain the time-varying rate of fault risk prediction corresponding to the multidimensional sample data. Then, the comprehensive loss value between the time-varying rate of fault risk prediction and the fault label is calculated using a first preset loss function. Finally, the first parameter of the time-varying prediction model for fault risk is updated based on the comprehensive loss value.

[0182] S504. Based on the updated first parameter, input the adversarial sample data into the time-varying prediction model of fault risk to obtain the time-varying rate of fault risk prediction corresponding to the adversarial sample data.

[0183] S505. Calculate the generalization loss value between the time-varying rate of the fault risk prediction and the fault label corresponding to the adversarial sample data using the second preset loss function.

[0184] The second preset loss function is the task prediction loss function, which is used to determine whether the entire model has converged.

[0185] S506. Determine whether both the first preset loss function and the second preset loss function have converged. If yes, proceed to S508; otherwise, proceed to S507.

[0186] In this step, if both the first and second preset loss functions converge, a trained time-varying failure risk prediction model is obtained. If either the first or second preset loss function fails to converge, the second parameter of the time-varying failure risk prediction model is updated based on the generalization loss value.

[0187] S507. Update the second parameter of the time-varying prediction model of failure risk according to the generalization loss value, and execute S501~S506 based on the updated second parameter.

[0188] The second parameter includes the first parameter, the gated spatiotemporal convolutional network parameters (type weight matrix and meteorological projection matrix), and ResNet. 18 Residual network parameters, probability prediction weight matrix, and residual ODE network parameters.

[0189] In this step, if the first preset loss function fails to converge or the second preset loss function fails to converge, the second parameter of the fault risk time-varying prediction model is updated according to the generalization loss value. Based on the updated second parameter, the process of inputting the multidimensional sample data into the fault risk time-varying prediction model is executed again to obtain the adversarial sample data corresponding to the multidimensional sample data and fault labels.

[0190] S508. Obtain the trained time-varying prediction model for fault risk.

[0191] In this step, if both the first and second preset loss functions converge, a well-trained time-varying prediction model for fault risk is obtained.

[0192] This application provides a method for determining the maintenance time of a distribution network. It acquires multiple sample multidimensional data corresponding to historical fault sample data of the target distribution network, as well as fault labels corresponding to each sample multidimensional data. The sample multidimensional data is then input into a time-varying fault risk prediction model to obtain adversarial sample data corresponding to the sample multidimensional data and fault labels. Based on the sample multidimensional data, fault labels, and a first preset loss function, the first parameters of the time-varying fault risk prediction model are updated. Then, based on the updated first parameters, the adversarial sample data is input into the time-varying fault risk prediction model to obtain the time-varying fault risk prediction rate corresponding to the adversarial sample data.

[0193] Subsequently, the generalization loss value between the time-varying rate of fault risk prediction and the fault label corresponding to the adversarial sample data is calculated through the second preset loss function. The second parameter of the fault risk time-varying prediction model is updated according to the generalization loss value. Based on the updated second parameter, the process of inputting multidimensional sample data into the fault risk time-varying prediction model is repeated to obtain adversarial sample data corresponding to the multidimensional sample data and fault label until both the first preset loss function and the second preset loss function converge, and the trained fault risk time-varying prediction model is obtained.

[0194] This application's embodiments address the problem of scarce fault samples in distribution networks by generating adversarial example data based on multidimensional sample data, thereby improving the model's generalization ability under sparse samples. By performing an inner loop (updating the first parameter) on the model based on multidimensional sample data from the meta-task support set, and then performing an outer loop (updating the second parameter) based on the adversarial example data, the model can quickly generalize to new fault modes within a few samples, thus improving model training efficiency.

[0195] Figure 6 A flowchart illustrating a method for determining the maintenance time of a power distribution network provided in this application. Figure 5 ,like Figure 6 As shown, this embodiment provides a detailed description of S203 based on any of the above embodiments. The method includes:

[0196] S601. By extracting the device type labels of the device nodes in the dynamic heterogeneous graph, the type weight matrix is ​​obtained.

[0197] In this step, since different device types play different roles in fault propagation (e.g., switching equipment affects topology connectivity), it is necessary to differentiate the feature extraction weights for different devices. This can be achieved by extracting the device type labels from the device nodes in the dynamic heterogeneous graph, thus obtaining a type weight matrix.

[0198] In one possible implementation, the type weight matrix Its expression is:

[0199]

[0200] in, This represents the basic weight matrix, which is shared by all devices. The type offset matrix represents the type of equipment. Based on the different equipment types, an independent weight increment is set for each type of equipment. This allows for adaptive adjustment of the feature extraction intensity for switching equipment and key equipment by superimposing the type offset matrix.

[0201] In one possible implementation, due to short-term changes in electricity demand in the target distribution network, the equipment type of certain devices may change accordingly. The equipment type label of the device nodes in the dynamic heterogeneous diagram can be dynamically updated based on relevant parameters in the real-time operating data. The corresponding expression is:

[0202]

[0203] in, Represents device node Device type label; Indicates equipment capacity parameters, This represents the device type identifier parameter; This is an empty set. The expression categorizes device nodes into three types: a value of 0 represents critical power supply equipment, such as substations; a value of 1 represents switch control equipment, such as sectionalizing switches; and a value of 2 represents ordinary terminal equipment, such as meter boxes. This provides a classification basis for subsequent type weight allocation and enhances the ability to extract features of critical equipment.

[0204] S602. Based on the dynamic heterogeneous graph and type weight matrix, obtain the node characteristics of device state evolution.

[0205] In this step, dynamic heterogeneous graphs and type weight matrices are processed through multi-hop neighborhood aggregation to integrate topological connectivity and dynamic meteorological influences, thereby obtaining node features that characterize the evolution of equipment state.

[0206] In one possible implementation, the node characteristics of device state evolution are obtained by processing the dynamic heterogeneous graph and type weight matrix through multi-hop neighborhood aggregation, and the corresponding expression is:

[0207]

[0208] in, and Each represents any device node in the dynamic heterogeneous graph. In the Layer and first The feature vector of the layer; Represents the ReLU activation function; Represents device node In the K-hop field, K is generally 3, thus covering devices directly connected to within three hops; Represents device node To device node The edge feature vectors; It is a non-zero minimum value to prevent the denominator from being 0; This represents a vector concatenation operation; and Representing device nodes and device nodes The number of neighbors at one time.

[0209] S603. Based on the meteorological feature vectors in the dynamic heterogeneous graph, obtain the gating matrix.

[0210] In this step, meteorological feature vectors are first extracted from the dynamic heterogeneous graph. These vectors are then mapped to a high-dimensional feature space consistent with the feature dimensions of the device nodes, completing the dimensional alignment and spatial adaptation of cross-modal features. The mapped meteorological feature vectors are then subjected to nonlinear enhancement and range constraint processing to obtain range-constrained and nonlinearly enhanced meteorological feature vectors. These range-constrained and nonlinearly enhanced meteorological feature vectors are then diagonalized to obtain a gating matrix. This gating matrix is ​​used to amplify or suppress feature channels related to the current weather conditions during temporal convolution.

[0211] In one possible implementation, the gating matrix is ​​obtained based on the meteorological feature vectors of the meteorological virtual nodes in the dynamic heterogeneous graph set, and the corresponding expression is as follows:

[0212]

[0213] in, Let represent the gate matrix at time t; The meteorological projection matrix maps meteorological features to a high-dimensional feature space consistent with the feature dimensions of the equipment nodes. Dimensions of meteorological characteristics; Dimensions of device characteristics; express Meteorological feature vector at time, The function is used to compress the gate value to the interval [-1, 1].

[0214] S604. Based on the gating matrix and the characteristics of the device state evolution nodes, a multidimensional spatiotemporal feature matrix is ​​obtained.

[0215] In this step, based on the characteristics of the equipment state evolution nodes, dilated convolution and temporal weighting operations are performed to capture the long-term cumulative effect of equipment state evolution in the target period. Then, the convolution results are controlled channel by channel through the gating matrix corresponding to each time slice to enhance the nonlinear correlation between meteorological factors and equipment state evolution, and obtain the spatiotemporal feature vector corresponding to each time slice. Finally, the spatiotemporal features of each time slice are extracted to form a multidimensional spatiotemporal feature matrix for the target period.

[0216] The target period is the time period corresponding to the short-term meteorological data.

[0217] In one possible implementation, a multidimensional spatiotemporal feature matrix is ​​obtained based on the gating matrix and the device state evolution node characteristics, and the corresponding expression is:

[0218]

[0219] in, This represents the spatiotemporal feature vector of node v at time t. Represents the temporal convolution kernel, learning different time lags. The weight, This represents the expanded convolution operator with an expansion rate of d, for example, To exponentially expand the receptive field, The time-related features are sampled at intervals, where T represents the size of the time window. This represents the Hadamard product, used here to implement weather gating. Indicates that at time The device state evolution node features of time node v are used to extract the spatiotemporal features of each time slice, thereby forming a multidimensional spatiotemporal feature matrix.

[0220] In S601~S604, it can be regarded as processing real-time operating data and dynamic heterogeneous graph sets through the gated spatiotemporal convolutional network in the pre-trained fault risk time-varying prediction model to obtain a multi-dimensional spatiotemporal feature matrix.

[0221] S605. Based on the multidimensional spatiotemporal feature matrix, the risk feature matrix and the failure probability matrix are obtained.

[0222] In this step, the first residual network in the pre-trained time-varying fault risk prediction model is used to extract features from the multidimensional spatiotemporal feature matrix to obtain the risk feature matrix. This matrix can be a continuous high-dimensional feature matrix that can encode details of continuous states such as gradual changes in insulation performance. The first residual network can be a ResNet. 18 Residual network.

[0223] Then, by using the probability prediction weight matrix in the pre-trained time-varying fault risk prediction model, the risk feature matrix is ​​weighted to obtain the fault probability at any time within the target time period.

[0224] In one possible implementation, a risk feature matrix is ​​obtained by extracting features from the multidimensional spatiotemporal feature matrix using a residual network, and the corresponding expression is:

[0225]

[0226] The failure probability is obtained by weighting the risk feature matrix using the probability prediction weight matrix. The corresponding expression is:

[0227]

[0228] in, Represents the equipment risk characteristic matrix; express The probability of failure at any given moment; This is the probability prediction weight matrix; It is a residual network.

[0229] S606. Based on the risk feature matrix and the fault probability matrix, the time-varying rate of fault risk prediction is obtained.

[0230] In this step, the second residual network in the pre-trained fault risk time-varying prediction model is used to model the continuous change of equipment risk status over time based on the risk feature matrix and the fault probability matrix, thus obtaining the fault risk prediction time-varying rate. The second residual network can be a residual ODE network.

[0231] In one possible implementation, a second residual network is used to model the continuous change of equipment risk status over time based on the risk feature matrix and the fault probability matrix, resulting in the time-varying rate of fault risk prediction, expressed as follows:

[0232]

[0233] in, Represents the equipment risk state vector; This indicates the time-varying rate of fault risk prediction; This represents the residual ODE network.

[0234] This application provides a method for determining the maintenance time of a power distribution network. It extracts equipment type labels from equipment nodes in a dynamic heterogeneous graph to obtain a type weight matrix. Then, based on the dynamic heterogeneous graph and the type weight matrix, it obtains the equipment state evolution node features. Subsequently, based on the meteorological feature vectors in the dynamic heterogeneous graph, it obtains a gating matrix, and based on the gating matrix and the equipment state evolution node features, it obtains a multi-dimensional spatiotemporal feature matrix. Finally, based on the multi-dimensional spatiotemporal feature matrix, it obtains a risk feature matrix and a fault probability matrix, thereby obtaining the time-varying rate of fault risk prediction.

[0235] This application embodiment obtains a multi-dimensional spatiotemporal feature matrix for calculating the time variability of fault risk prediction based on meteorological characteristics and equipment state evolution characteristics. This breaks through the bottleneck of coupled modeling of equipment heterogeneity and meteorological time variability, and achieves the effect of improving the accuracy and precision of fault risk prediction.

[0236] Based on any of the above embodiments, the following, in conjunction with Figure 7 This paper provides a detailed explanation of a method for determining the maintenance time of a power distribution network through specific examples.

[0237] S701. Construct the basic heterogeneous diagram of the target distribution network.

[0238] In this step, we first obtain the distribution network topology data and meteorological comprehensive data of the target distribution network. The distribution network topology data is in JSON format, and the meteorological comprehensive data is in CSV format.

[0239] The distribution network topology data is then parsed into a set of device nodes. Each device node set contains a feature tuple corresponding to each device in the distribution network topology data. This feature tuple includes UTM coordinates, device type, and device parameter vector. Device types include critical power supply equipment, switch control equipment, and general terminal equipment. In the basic heterogeneous graph, 0 represents critical power supply equipment, 1 represents switch control equipment, and 2 represents general terminal equipment. The device parameter vector is 128-dimensional.

[0240] Based on the set of device nodes, an electrical connection edge set is generated according to the electrical connection relationships between the devices. In the electrical connection edge set, directed edges represent the current direction.

[0241] Meanwhile, the comprehensive meteorological data is parsed into a set of virtual meteorological nodes, which contains WGS84 coordinates and meteorological feature vectors, including temperature, humidity, wind speed, etc.

[0242] Based on the aforementioned set of equipment nodes, electrical connection edges, and meteorological virtual nodes, a basic heterogeneous graph of the target distribution network is constructed.

[0243] S702. Obtain real-time operating data and short-term meteorological data of the target distribution network.

[0244] S703. By analyzing the meteorological change information under different time slices in short-term meteorological data, multiple sets of time-series correlated meteorological data corresponding to the meteorological virtual nodes are obtained.

[0245] S704. Based on each set of time-series correlated meteorological data, update the basic heterogeneous map to obtain a set of dynamic heterogeneous maps that match the timeliness of short-term meteorological data.

[0246] In this step, the basic heterogeneous graph is updated according to each set of time-series correlated meteorological data to obtain a dynamic heterogeneous graph corresponding to each set of time-series correlated meteorological data. Based on each dynamic heterogeneous graph, a set of dynamic heterogeneous graphs matching the timeliness of short-term meteorological data is obtained.

[0247] Specifically, the update of a basic heterogeneous graph using a set of time-series correlated meteorological data is illustrated as an example. The main meteorological feature in this time-series correlated meteorological data is wind speed. First, based on the basic heterogeneous graph, the coordinate distance between the equipment corresponding to the device node and the meteorological station corresponding to the meteorological virtual node is calculated using the Haversine formula. Then, a meteorological influence edge set and its weights are generated based on the real-time wind speed and coordinate distance. Finally, a dynamic heterogeneous graph containing device nodes, meteorological virtual nodes, electrical connection edges, meteorological influence edges, and their weights are generated.

[0248] The Haversine formula is used to calculate the spherical distance between two points on the Earth's surface, and is used to quantify the coordinate distance between the equipment and the weather station.

[0249] S705. Input the real-time operating condition data and dynamic heterogeneous graph set into the pre-trained fault risk time-varying prediction model to obtain the fault risk prediction time-varying rate corresponding to the target distribution network.

[0250] S706. Divide the preset maintenance time range into multiple maintenance time windows.

[0251] S707. Based on the time-varying rate of fault risk prediction, the risk cost corresponding to each maintenance time window is obtained.

[0252] S708. The maintenance time window with the lowest risk cost shall be determined as the target maintenance time.

[0253] In this step, multiple maintenance time windows are sorted based on risk cost and preset sorting rules, and the maintenance time window located at the preset sorting position is determined as the target maintenance time. Specifically, multiple maintenance time windows can be sorted from low to high risk cost, and the maintenance time window at the top of the list is determined as the target maintenance time; that is, the maintenance time window with the lowest risk cost is determined as the target maintenance time.

[0254] It should be noted that, in Figure 7 The processing steps S701-S708 shown in the embodiments do not constitute a specific limitation on a method for determining the maintenance time of a distribution network. In other embodiments of this application, a method for determining the maintenance time of a distribution network may include a ratio... Figure 7 The embodiments may include more or fewer steps; for example, a method for determining the maintenance time of a power distribution network may include... Figure 7 Some steps in the embodiments, or, Figure 7 Some steps in the embodiments can be replaced by steps with the same function, or, Figure 7 Some steps in the embodiments can be broken down into multiple steps, etc.

[0255] Figure 8 A schematic diagram of a device for determining the maintenance time of a power distribution network provided in this application is shown below. Figure 8 As shown, the maintenance time determination device 80 for a power distribution network provided in this embodiment includes:

[0256] The acquisition module 801 is used to acquire real-time operating data and short-term meteorological data of the target power distribution network.

[0257] The graph construction module 802 is used to update the basic heterogeneous graph of the target distribution network using short-term meteorological data, so as to obtain a dynamic heterogeneous graph set that matches the timeliness of the short-term meteorological data.

[0258] The model prediction module 803 is used to input real-time operating condition data and dynamic heterogeneous graph set into a pre-trained fault risk time-varying prediction model to obtain the fault risk prediction time-varying rate corresponding to the target distribution network.

[0259] The determination module 804 is used to determine the target maintenance time of the target distribution network based on the time-varying rate of fault risk prediction and the preset maintenance time range.

[0260] In one possible implementation, the acquisition module 801 is further configured to:

[0261] Obtain the distribution network topology data and meteorological comprehensive data of the target distribution network.

[0262] Accordingly, the graph construction module 802 is also used for:

[0263] By parsing the distribution network topology data, a set of equipment nodes and a set of electrical connections are generated. The equipment node set contains the feature tuples of each equipment node corresponding to each device in the distribution network topology data; the electrical connection set contains the electrical connection relationships between each device. By parsing comprehensive meteorological data, a set of meteorological virtual nodes is generated; the meteorological virtual node set contains the feature tuples of each meteorological station corresponding to each meteorological station in the comprehensive meteorological data. Based on each equipment node feature tuple and its corresponding equipment node, each electrical connection relationship and its corresponding electrical connection edge, the preset electrical connection edge weights, and each meteorological virtual node feature tuple and its corresponding meteorological virtual node, a basic heterogeneous graph of the target distribution network is constructed.

[0264] In one possible implementation, the graph construction module 802 is specifically used for:

[0265] By analyzing meteorological change information under different time slices in short-term meteorological data, multiple sets of time-series correlated meteorological data corresponding to the meteorological virtual nodes are obtained.

[0266] Based on each set of time-series correlated meteorological data, the basic heterogeneous map is updated to obtain the dynamic heterogeneous map corresponding to each set of time-series correlated meteorological data.

[0267] Based on each dynamic heterogeneous graph, a set of dynamic heterogeneous graphs matching the timeliness of short-term meteorological data is obtained.

[0268] In one possible implementation, the graph construction module 802 is specifically used for:

[0269] Based on time-series correlated meteorological data, dynamic distance thresholds that are related to both meteorological virtual nodes and device nodes are obtained.

[0270] Based on the meteorological station coordinates in the feature tuple of the meteorological virtual node and the device coordinates in the feature tuple of the device node, the coordinate distance between the meteorological station corresponding to the meteorological virtual node and the device corresponding to the device node is obtained.

[0271] When the coordinate distance is less than the dynamic distance threshold, a meteorological influence edge is generated between the meteorological virtual node and the device node. The meteorological influence edge weight is obtained based on the coordinate distance, time-series associated meteorological data, device type in the device node feature tuple, and preset parameters.

[0272] The basic heterogeneous graph is updated by updating each meteorological influence edge and its weight, resulting in a dynamic heterogeneous graph corresponding to the time-series correlated meteorological data.

[0273] In one possible implementation, the acquisition module 801 is further configured to:

[0274] Obtain the meta-task support set corresponding to the historical fault sample data of the target distribution network; the meta-task support set contains multiple sample multidimensional data and the fault label corresponding to each sample multidimensional data.

[0275] Correspondingly, the model prediction module 803 is also used for:

[0276] By inputting the multidimensional sample data into the time-varying prediction model for fault risk, adversarial sample data corresponding to the multidimensional sample data and fault labels are obtained.

[0277] Based on multidimensional sample data, fault labels, and a first preset loss function, the first parameter of the time-varying prediction model for fault risk is updated.

[0278] Based on the updated first parameter, the adversarial sample data is input into the time-varying prediction model of fault risk to obtain the time-varying rate of fault risk prediction corresponding to the adversarial sample data.

[0279] The generalization loss value between the time-varying rate of fault risk prediction and the fault label corresponding to the adversarial sample data is calculated using the second preset loss function.

[0280] The second parameter of the fault risk time-varying prediction model is updated based on the generalization loss value. Based on the updated second parameter, the process of inputting multidimensional sample data into the fault risk time-varying prediction model is repeated to obtain adversarial sample data corresponding to the multidimensional sample data and fault labels until both the first preset loss function and the second preset loss function converge to obtain the trained fault risk time-varying prediction model. The second parameter includes the first parameter.

[0281] In one possible implementation, the model prediction module 803 is specifically used for:

[0282] Feature fusion processing is performed on the sample dynamic heterogeneous graph in the multidimensional sample data to obtain a multidimensional spatiotemporal feature matrix.

[0283] Based on the multidimensional spatiotemporal feature matrix and preset physical constraints, the physical constraint loss value is obtained.

[0284] Adversarial example data is generated based on physical constraint loss values ​​and historical operating condition data and historical meteorological data from the multidimensional sample data.

[0285] In one possible implementation, the model prediction module 803 is specifically used for:

[0286] By inputting the multidimensional sample data into the time-varying prediction model for fault risk, the predicted time-varying rate of fault risk corresponding to the multidimensional sample data is obtained.

[0287] The comprehensive loss value between the time-varying rate of fault risk prediction and the fault label corresponding to the multidimensional data of the sample is calculated using the first preset loss function.

[0288] The first parameter of the time-varying prediction model for fault risk is updated based on the comprehensive loss value.

[0289] In one possible implementation, the determining module 804 is specifically used for:

[0290] The preset maintenance time range is divided into multiple maintenance time windows; each maintenance time window consists of multiple maintenance time points.

[0291] Based on the time-varying rate of fault risk prediction, the risk cost corresponding to each maintenance time window is obtained.

[0292] Based on risk costs and preset sorting rules, multiple maintenance time windows are sorted.

[0293] The maintenance time window located at the preset sorting position is set as the target maintenance time.

[0294] This embodiment provides a device for determining the maintenance time of a power distribution network, which can execute the method provided in the above-described method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0295] Figure 9 A schematic diagram of the structure of the electronic device provided in this application. Figure 9 As shown, the electronic device 90 provided in this embodiment includes at least one processor 901 and a memory 902. Optionally, the device 90 further includes a communication component 903. The processor 901, memory 902, and communication component 903 are connected via a bus 904.

[0296] In a specific implementation, at least one processor 901 executes computer execution instructions stored in memory 902, causing at least one processor 901 to perform the above-described method.

[0297] The specific implementation process of processor 901 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0298] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0299] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0300] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0301] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0302] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0303] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0304] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0305] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0306] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0307] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0308] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0309] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0310] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for determining the maintenance time of a power distribution network, characterized in that, include: Acquire real-time operating data and short-term weather data of the target power distribution network; The basic heterogeneous map of the target distribution network is updated using the short-term meteorological data to obtain a dynamic heterogeneous map set that matches the timeliness of the short-term meteorological data. The real-time operating data and the dynamic heterogeneous graph set are input into a pre-trained fault risk time-varying prediction model to obtain the fault risk prediction time-varying rate corresponding to the target distribution network. Based on the time-varying rate of the fault risk prediction and the preset maintenance time range, the target maintenance time of the target distribution network is determined.

2. The method according to claim 1, characterized in that, Before acquiring the real-time operating data and short-term meteorological data of the target distribution network, the method further includes: Obtain the distribution network topology data and meteorological comprehensive data of the target distribution network; By parsing the power distribution network topology data, a set of device nodes and a set of electrical connection relationships are generated; the set of device nodes contains the feature tuples of each device node corresponding to each device in the power distribution network topology data; the set of electrical connection relationships contains the electrical connection relationships between each device. By parsing the comprehensive meteorological data, a set of virtual meteorological nodes is generated; the set of virtual meteorological nodes contains feature tuples of virtual meteorological nodes corresponding to each meteorological station in the comprehensive meteorological data. Based on the feature tuple of each device node and its corresponding device node, the electrical connection edge corresponding to each electrical connection relationship, the preset electrical connection edge weight, and the feature tuple of each meteorological virtual node and its corresponding meteorological virtual node, the basic heterogeneous graph of the target distribution network is constructed.

3. The method according to claim 2, characterized in that, The step of updating the basic heterogeneous map of the target distribution network using the short-term meteorological data to obtain a dynamic heterogeneous map set that matches the timeliness of the short-term meteorological data includes: By analyzing the meteorological change information under different time slices in the short-term meteorological data, multiple sets of time-series correlated meteorological data corresponding to the meteorological virtual node are obtained; Based on each set of time-series correlated meteorological data, the basic heterogeneous graph is updated to obtain the dynamic heterogeneous graph corresponding to each set of time-series correlated meteorological data. Based on each of the aforementioned dynamic heterogeneous graphs, a set of dynamic heterogeneous graphs matching the timeliness of the short-term meteorological data is obtained.

4. The method according to claim 3, characterized in that, The step of updating the basic heterogeneous graph based on each set of time-series correlated meteorological data to obtain a dynamic heterogeneous graph corresponding to each set of time-series correlated meteorological data includes: Based on the time-series correlated meteorological data, a dynamic distance threshold related to both the meteorological virtual node and the device node is obtained; Based on the meteorological station coordinates in the feature tuple of the meteorological virtual node and the device coordinates in the feature tuple of the device node, the coordinate distance between the meteorological station corresponding to the meteorological virtual node and the device corresponding to the device node is obtained; When the coordinate distance is less than the dynamic distance threshold, a meteorological influence edge is generated between the meteorological virtual node and the device node, and the meteorological influence edge weight is obtained based on the coordinate distance, the time-series associated meteorological data, the device type in the feature tuple of the device node, and preset parameters. The basic heterogeneous graph is updated by updating each meteorological influence edge and each meteorological influence edge weight to obtain the dynamic heterogeneous graph corresponding to the time-series associated meteorological data.

5. The method according to claim 1, characterized in that, The training process of the time-varying prediction model for fault risk includes: Obtain the meta-task support set corresponding to the historical fault sample data of the target distribution network; the meta-task support set contains multiple sample multidimensional data and a fault label corresponding to each sample multidimensional data. The multidimensional sample data is input into the time-varying prediction model for fault risk to obtain adversarial sample data corresponding to the multidimensional sample data and the fault label. Based on the sample multidimensional data, the fault label, and the first preset loss function, update the first parameter of the fault risk time-varying prediction model; Based on the updated first parameter, the adversarial sample data is input into the fault risk time-varying prediction model to obtain the fault risk prediction time-varying rate corresponding to the adversarial sample data; The generalization loss value between the time-varying rate of the fault risk prediction corresponding to the adversarial sample data and the fault label is calculated using a second preset loss function. The second parameter of the fault risk time-varying prediction model is updated according to the generalization loss value. Based on the updated second parameter, the process of inputting the sample multidimensional data into the fault risk time-varying prediction model to obtain adversarial sample data corresponding to the sample multidimensional data and the fault label is repeated until the first preset loss function and the second preset loss function converge to obtain the trained fault risk time-varying prediction model. The second parameter includes the first parameter.

6. The method according to claim 5, characterized in that, The step of inputting the multidimensional sample data into the time-varying prediction model for fault risk to obtain adversarial sample data corresponding to the multidimensional sample data and the fault label includes: The sample dynamic heterogeneous graph in the multidimensional sample data is subjected to feature fusion processing to obtain a multidimensional spatiotemporal feature matrix. Based on the multidimensional spatiotemporal feature matrix and the preset physical constraints, the physical constraint loss value is obtained; The adversarial sample data is generated based on the physical constraint loss value and the historical operating condition data and historical meteorological data in the sample multidimensional data.

7. The method according to claim 5, characterized in that, The step of updating the first parameter of the time-varying prediction model of fault risk based on the sample multidimensional data, the fault label, and the first preset loss function includes: The sample multidimensional data is input into the fault risk time-varying prediction model to obtain the fault risk prediction time-varying rate corresponding to the sample multidimensional data. The comprehensive loss value between the time-varying rate of the fault risk prediction corresponding to the multidimensional data of the sample and the fault label is calculated by the first preset loss function. The first parameter of the time-varying prediction model for fault risk is updated based on the comprehensive loss value.

8. The method according to claim 1, characterized in that, The step of determining the target maintenance time for the target distribution network based on the predicted fault risk time variability and the preset maintenance time range includes: The preset maintenance time range is divided into multiple maintenance time windows; each maintenance time window consists of multiple maintenance time points. Based on the time-varying rate of the fault risk prediction, the risk cost corresponding to each maintenance time window is obtained; Based on the aforementioned risk costs and preset sorting rules, the multiple maintenance time windows are sorted. The maintenance time window located at the preset sorting position is determined as the target maintenance time.

9. A device for determining the maintenance time of a power distribution network, characterized in that, include: The acquisition module is used to acquire real-time operating data and short-term meteorological data of the target power distribution network; The graph construction module is used to update the basic heterogeneous graph of the target distribution network using the short-term meteorological data, and obtain a dynamic heterogeneous graph set that matches the timeliness of the short-term meteorological data. The model prediction module is used to input the real-time operating data and the dynamic heterogeneous graph set into a pre-trained fault risk time-varying prediction model to obtain the fault risk prediction time-varying rate corresponding to the target distribution network. The determination module is used to determine the target maintenance time of the target distribution network based on the time-varying rate of the fault risk prediction and the preset maintenance time range.

10. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1 to 8.

Images