Power grid fault early warning method and device, electronic equipment and storage medium

Abstract

The invention discloses a power grid fault early warning method and device, electronic equipment and a storage medium. The method comprises the steps that target operation data of a target early warning area in a power grid are acquired, the target early warning area is an area to be subjected to early warning in the power grid, and the target operation data comprise power data and meteorological data; on the basis of the target operation data, graph network data is constructed, and the graph network data is used for representing the connection relation between the target early warning areas; predicting the graph network data based on a fault prediction model to obtain a fault risk value of the target early warning area, the fault risk value being used for representing a probability of fault occurrence of the target early warning area in a future time period; and based on the fault risk value, early warning information is determined, and the early warning information is used for representing the information of the fault of the power grid. The technical problem that the related technology cannot accurately warn the occurrence position of the power grid fault is solved.

Description

Technical Field

[0001] The present invention relates to the field of power systems and fault warning, and in particular to a power grid fault warning method, device, electronic equipment and storage medium. Background Art

[0002] As a key national infrastructure, the construction and stable operation of the power grid are of great significance to ensuring power supply and improving the reliability and stability of the power system. However, in actual operation, factors such as extreme weather events, equipment aging, overload operation, and human operational errors may occur, leading to power grid failures.

[0003] Due to the complexity and uncertainty of the power grid system, relevant technologies are unable to accurately warn of the specific location of power grid failures, thus affecting the safe and stable operation of the power grid.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present invention provide a method, device, electronic device and storage medium for early warning of power grid faults, so as to at least solve the technical problem that the related technologies cannot accurately warn the location where the power grid fault occurs.

[0006] According to one aspect of an embodiment of the present invention, there is provided a method for early warning of power grid faults, comprising: obtaining target operating data of a target early warning area in the power grid, wherein the target early warning area is an area to be warned in the power grid, and the target operating data comprises power data and meteorological data; constructing graph network data based on the target operating data, wherein the graph network data is used to characterize the connection relationship between the target early warning areas; predicting the graph network data based on a fault prediction model to obtain a fault risk value of the target early warning area, wherein the fault risk value is used to characterize the probability of a fault occurring in the target early warning area in a future time period; determining early warning information based on the fault risk value, wherein the early warning information is used to characterize information on a power grid fault.

[0007] Optionally, the target warning area includes an area obtained by dividing the power grid; obtaining target operation data of the target warning area in the power grid includes: collecting initial operation data of the target warning area at a preset time interval, wherein the initial operation data includes image data, power grid data, meteorological data and time domain waveform data; preprocessing the initial operation data to obtain target operation data.

[0008] Optionally, the initial operation data is preprocessed, including: interpolating and filling missing values ​​in the initial operation data; gray-scaling image data in the initial operation data; normalizing power grid data and meteorological data in the initial operation data; and converting time domain waveform data in the initial operation data into frequency domain waveform data.

[0009] Optionally, based on the target operation data, graph network data is constructed, including: based on the target warning area, determining the nodes and edges of the graph network data, wherein the nodes of the graph network data and the target warning area are in a one-to-one mapping relationship, and the edges of the graph network data are connection relationships between the target warning areas; based on the target operation data, determining the initial features of the nodes, wherein the initial features are used to characterize the target operation data of the target warning area at a preset time point; and constructing the graph network data based on the nodes, edges and initial features.

[0010] Optionally, the graph network data is predicted based on the fault prediction model to obtain the fault risk value of the target warning area, including: splitting the graph network data to obtain multiple sub-graphs, wherein different sub-graphs correspond to different warning areas, and the characteristics of the sub-nodes in each sub-graph are the same as the initial characteristics of the same nodes in the graph network data; predicting the multiple sub-graphs separately based on the fault prediction model to obtain the fault risk value of the target warning area.

[0011] Optionally, the graph network data is split to obtain multiple subgraphs, including: determining the similarity between two nodes connected by an edge in the graph network data; merging the two nodes to generate a subgraph in response to the similarity being greater than or equal to a similarity threshold; adjusting the similarity threshold based on the number of subgraphs that have been generated, and repeatedly performing the steps of determining the similarity between two nodes connected by an edge in the graph network data, merging the two nodes to generate a subgraph in response to the similarity being greater than or equal to the similarity threshold, until the number of subgraphs that have been generated reaches a preset number; determining that the subgraph that has been generated is multiple subgraphs.

[0012] Optionally, the fault prediction model includes multiple parallel modules, with multiple sub Figure 1 One-to-one correspondence; based on the fault prediction model, multiple sub-graphs are predicted respectively to obtain the fault risk value of the target warning area, including: using the target parallel module corresponding to each sub-graph to predict the sub-graph to obtain the fault risk value of the warning area corresponding to the sub-graph.

[0013] Optionally, the target parallel module includes a first hidden layer, a second hidden layer, a third hidden layer, an extraction layer and a first classifier; the target parallel module corresponding to each subgraph is used to predict the subgraph to obtain the fault risk value of the warning area corresponding to the subgraph, including: using the first hidden layer to extract features of the subgraph to obtain multiple first vectors, wherein the multiple first vectors correspond one-to-one to multiple time points of the features of the subnodes in the subgraph; using the second hidden layer to encode the first vector to obtain multiple second vectors, wherein the multiple second vectors correspond one-to-one to the features of the subnodes of the subgraph; using the third hidden layer to process the subgraph and the second vector to obtain a graph update matrix, wherein different row vectors of the graph update matrix are used to characterize the update features of different subnodes in the subgraph; using the extraction layer to extract the update features of each subnode from the graph update matrix; using the first classifier to determine the fault risk value of the target warning area based on the update features of different subnodes in the subgraph.

[0014] Optionally, the method further includes: constructing training samples based on the topological structure of the power grid and the equipment information of the equipment in the power grid; and training the pre-trained model using the training samples to obtain a fault prediction model.

[0015] Optionally, training samples are constructed based on the topological structure of the power grid and the equipment information of the equipment in the power grid, including: constructing a three-dimensional model of the power grid based on the topological structure of the power grid and the equipment information of the equipment in the power grid; collecting operating data of the three-dimensional model to construct sample data; labeling the sample data based on the power grid state and state duration of the power grid to determine the sample label, wherein the sample label is used to characterize the probability of failure of the sample data in a future time period; and constructing training samples based on the sample data and the sample label.

[0016] Optionally, before using the training samples to train the pre-trained model to obtain the fault prediction model, the above method also includes: obtaining second training data, wherein the second training data includes: a second training sample, a second label and a third label, the second label is used to characterize the number of people in the target warning area, and the third label is used to characterize whether sparks occur in the target warning area; using the second training data to train the initial hidden layer of the initial prediction model to obtain the pre-trained model.

[0017] According to another aspect of an embodiment of the present invention, there is also provided an early warning device for power grid faults, comprising: an acquisition module, used to acquire target operation data of a target early warning area in the power grid, wherein the target early warning area is an area to be warned in the power grid, and the target operation data includes power data and meteorological data; a determination module, used to determine graph network data based on the target operation data, wherein the graph network data is used to characterize the connection relationship between the target early warning areas; a prediction module, used to predict the graph network data based on a fault prediction model to obtain a fault risk value of the target early warning area, wherein the fault risk value is used to characterize the probability of a fault occurring in the target early warning area in a future time period; an early warning module, used to determine early warning information based on the fault risk value, wherein the early warning information is used to characterize information on power grid faults.

[0018] According to another aspect of an embodiment of the present invention, there is further provided an electronic device, comprising: a memory storing an executable program; and a processor for running the program, wherein the method in each embodiment of the present invention is executed when the program is running.

[0019] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the methods in various embodiments of the present invention.

[0020] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0021] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0022] According to another aspect of the embodiments of the present invention, a computer program is further provided. When the computer program is executed by a processor, the methods in the embodiments of the present invention are implemented.

[0023] In an embodiment of the present invention, a method for early warning of power grid faults is provided, comprising: obtaining target operation data of a target early warning area in the power grid, wherein the target early warning area is an area to be warned in the power grid, and the target operation data includes power data and meteorological data; constructing graph network data based on the target operation data, wherein the graph network data is used to characterize the connection relationship between the target early warning areas; predicting the graph network data based on a fault prediction model to obtain a fault risk value of the target early warning area, wherein the fault risk value is used to characterize the probability of a fault occurring in the target early warning area in a future time period; determining early warning information based on the fault risk value, wherein the early warning information is used to characterize information about a fault occurring in the power grid. It is easy to notice that the present application considers the influence of different areas of the power grid, constructs graph network data based on power grid power data and meteorological data, and thereby uses a fault prediction model to predict the graph network data, and obtains the probability of a possible fault occurring in the warning area, thereby achieving the purpose of early warning of power grid faults, and realizing the technical effect of accurately locating the location where the fault risk occurs, thereby solving the technical problem that the related technology cannot accurately warn the location where the power grid fault occurs. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0025] Figure 1 is a flow chart of a power grid fault early warning method according to an embodiment of the present invention;

[0026] Figure 2 2 is a schematic diagram of a power grid fault early warning device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] According to an embodiment of the present invention, an embodiment of a method for early warning of power grid failure is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0030] Figure 1 A power grid fault early warning method according to an embodiment of the present invention is provided. Figure 1 As shown, the method comprises the following steps:

[0031] Step S102, obtaining target operation data of a target warning area in the power grid.

[0032] The target warning area is the area in the power grid to be warned, and the target operation data includes power data and meteorological data;

[0033] The above-mentioned power grid can be an electric network system composed of transmission lines, substation equipment, distribution equipment, etc., which is used to transmit the electric energy generated by the power plant to users and terminal devices. The power grid can be a transmission network, a distribution network, etc., which is not limited here and can be set as needed. The stable operation of the power grid is of great significance to ensure power supply and improve the reliability and stability of the power system. Therefore, it is particularly important to give early warning of power grid failures to avoid failures that affect the stable operation of the power grid.

[0034] The above-mentioned target warning area may be an area to be warned in the power grid. The area to be warned may be an area in the power grid that needs regular inspection or monitoring, an area that is prone to failure or an important area in the power grid. The target warning area may be at least one area obtained by dividing the power grid, and may be divided according to importance, geographical location or purpose. In order to achieve accurate warning, warnings may be issued for one or more areas in the power grid respectively.

[0035] The above-mentioned target operation data may include power data and meteorological data. The target operation data may be data of factors that affect the normal operation of the power grid, such as weather factors, equipment factors, human factors, etc. The target operation data may be used to analyze the laws that affect the occurrence of power failures, thereby providing early warnings to avoid failures or deploying maintenance as soon as possible.

[0036] The above-mentioned power data can be used to characterize the load status of the target warning area. Power data can include but is not limited to voltage value, current value, frequency value, etc., which are not limited here and can be set as needed. Through power data such as voltage value and current value, the load condition of the power grid can be monitored. Therefore, potential faults can be discovered and resolved in time through power data to ensure the safe operation and reliable power supply of the power grid.

[0037] The above-mentioned meteorological data can be used to characterize the weather conditions in the target warning area. Meteorological data may include, but are not limited to, temperature, humidity, wind speed, rainfall, snowfall, or light intensity, etc., which are not limited here and can be set as needed. Meteorological data such as temperature and humidity will affect the performance of power equipment in the power grid. Therefore, meteorological data can be used to timely detect equipment failures in the power grid so that operation and maintenance personnel can perform equipment maintenance and take preventive measures.

[0038] In an optional embodiment, the stored target operation data can be obtained from a cloud storage platform, or the target operation data of the target warning area can be collected in real time based on sensors in the power grid, or the target operation data can be reported manually.

[0039] In another optional embodiment, the target operation data can be determined based on the initial operation data of the target warning area. The target operation data is obtained by preprocessing the initial operation data to avoid missing initial operation data or inconsistent formats, which is not conducive to analyzing the rules between operation data.

[0040] Step S104, constructing graph network data based on the target operation data.

[0041] Among them, graph network data is used to characterize the connection relationship between target warning areas.

[0042] The above-mentioned graph network data can be used to characterize the connection relationship between target warning areas. Graph network data is data in a graph network structure. Graph network data can be a collection of nodes and edges. Graph network data can represent the connection relationship and operating status of the target warning area in the power grid in the form of a graph, thereby simplifying the power grid data for the purpose of analyzing and predicting power grid failures. Graph network data can also include nodes, initial features of nodes, and edges between nodes, wherein the nodes represent the target warning areas, the initial features of the nodes can be a collection of target operating data of the target warning areas, and the edges represent the connection relationship between the target warning areas. By constructing graph network data in the form of a graph, the data of the time dimension and the space dimension can be combined to predict the fault risk value of the target warning area and improve the accuracy of the fault warning.

[0043] In an optional embodiment, the target operation data may be manually analyzed to construct the graph network data. Alternatively, the nodes and edges may be determined to construct the graph network structure, and the graph network structure data may be further determined based on the target operation data. Optionally, the boundary of the target warning area may be determined by the power equipment in the target warning area, and the power equipment may be further used as the boundary, the target warning area may be regarded as a node of the graph network structure, and the connection relationship of the target warning area may be regarded as an edge of the graph network structure, and the target operation data of the target warning area may be used to construct the graph network data.

[0044] In another optional embodiment, the nodes and edges of the graph network data can be determined through the target warning area, and further, the initial characteristics of the nodes can be determined based on the target operation data, so as to construct the graph network data based on the nodes, edges and initial characteristics.

[0045] Step S106: predict the graph network data based on the fault prediction model to obtain the fault risk value of the target warning area.

[0046] Among them, the fault risk value is used to characterize the probability of failure occurring in the target warning area in the future time period.

[0047] The above-mentioned fault prediction model can be a model for predicting potential faults in the power grid. The fault prediction model can be used to predict the fault risk value of the target warning area. The fault prediction model can be constructed based on a convolutional neural network model, a gated network, a long short-term memory network, or a graph attention neural network model, which is not limited here and can be determined as needed. The fault prediction model can be used to analyze data that affects power failure factors, such as power data, meteorological data, etc., to accurately assess the risk of power grid failures, so that measures can be taken in advance to reduce the risk of power operation. Through the fault prediction model, the fault risk value of the target warning area can be accurately predicted, and the speed and efficiency of power grid fault warning can be improved.

[0048] The above-mentioned fault risk value can be used to characterize the probability of a fault occurring in the target warning area in the future time period. The fault risk value can also be an indicator that characterizes the possibility of a fault in the target warning area. The value range of the indicator can be [0, 100], and the value is a positive integer. The value from low to high indicates that the possibility of the risk of a fault is increasing. Alternatively, the fault risk value can be compared with the preset risk value. In response to the fault risk value being higher than the preset risk value, the target warning area corresponding to the fault risk value is an area where a fault may occur, so as to trigger an emergency response mechanism and notify the operation and maintenance personnel to conduct timely maintenance to prevent potential power grid failures. The fault risk value can be obtained through graph network data analysis, that is, the risk value of a possible fault in the target warning area can be determined based on the target warning area and its target operation data, which can improve the accuracy of the fault risk value so that the operation and maintenance personnel can intuitively feel the possibility of a fault in the target warning area, so as to respond in time, avoid power paralysis, and ensure the stable operation of the power grid.

[0049] In an optional embodiment, a deep learning model, such as a graph convolutional network combined with a long short-term memory network, can be used to construct a fault prediction model. The graph network data is converted into a node feature matrix and an adjacency matrix as well as time series data, such as power data and meteorological data, and the graph data network is divided into a training set and a test set. Furthermore, the training set is preprocessed to screen features related to fault risk prediction. For example, data features can be screened out by using correlation analysis, principal component analysis, or autoencoders to reduce input dimensions and improve the training efficiency of the model. The model is trained using the training set, and the hyperparameters of the model are tuned by methods such as cross-validation or grid search to improve the prediction performance of the model. The test set is predicted based on the trained fault prediction model, and the fault risk value is output.

[0050] In another optional embodiment, the model can be trained based on historical fault data and historical operation data. After the model training is completed, the graph network data can be input into the fault prediction model, and the fault risk value of the target warning area can be obtained through the calculation of the model. For example, the fault risk value of the target warning area is 85, and according to the set warning threshold of 61-100, it is a high risk. Then the area will be marked as a high-risk warning area, triggering the emergency response mechanism, notifying relevant managers to conduct timely inspections and maintenance to prevent potential power grid failures.

[0051] Step S108: determining warning information based on the fault risk value.

[0052] Among them, the early warning information is used to characterize the information of power grid failure.

[0053] The above-mentioned warning information can be used to characterize information about power grid failure. The warning information may include a warning area and a fault risk value of the warning area, that is, the probability of a fault occurring in the warning area in a future time period. Alternatively, the warning information may include a warning area, a risk level of the fault in the area, and a control strategy. For example, the risk level of a possible fault in the target warning area may be determined based on the fault risk value. The risk level may be set according to user needs, and this application does not limit this.

[0054] For example, the risk level includes three levels, namely low risk level, medium risk level and high risk level. Furthermore, different control strategies are determined according to different risk levels to save resources and improve the efficiency and accuracy of fault risk response. The control strategy corresponding to the high risk level is to immediately inspect and handle the power grid warning area, the control strategy corresponding to the medium risk level is to immediately repair the relay protection device, and the control strategy corresponding to the low risk level is to continuously monitor the warning area.

[0055] In an embodiment of the present invention, a method for early warning of power grid faults is provided, comprising: obtaining target operation data of a target early warning area in the power grid, wherein the target early warning area is an area to be warned in the power grid, and the target operation data includes power data and meteorological data; constructing graph network data based on the target operation data, wherein the graph network data is used to characterize the connection relationship between the target early warning areas; predicting the graph network data based on a fault prediction model to obtain a fault risk value of the target early warning area, wherein the fault risk value is used to characterize the probability of a fault occurring in the target early warning area in a future time period; determining early warning information based on the fault risk value, wherein the early warning information is used to characterize information about a fault occurring in the power grid. It is easy to notice that the present application considers the influence of different areas of the power grid, constructs graph network data based on power grid power data and meteorological data, and thereby uses a fault prediction model to predict the graph network data, and obtains the probability of a possible fault occurring in the warning area, thereby achieving the purpose of early warning of power grid faults, and realizing the technical effect of accurately locating the location where the fault risk occurs, thereby solving the technical problem that the related technology cannot accurately warn the location where the power grid fault occurs.

[0056] Optionally, the target warning area includes an area obtained by dividing the power grid; obtaining target operation data of the target warning area in the power grid includes: collecting initial operation data of the target warning area at a preset time interval, wherein the initial operation data includes image data, power grid data, meteorological data and time domain waveform data; preprocessing the initial operation data to obtain target operation data.

[0057] The grid can be divided into several factors, such as the grid topology, geographical distribution, load characteristics, equipment type, and key nodes. Optionally, the grid can be divided based on historical fault data to determine the priority of fault prediction in the warning area. The grid can also be divided based on load characteristics to focus on and analyze high-load areas. By dividing the grid into target warning areas, each warning area can be analyzed independently to improve the accuracy of fault warning.

[0058] The above-mentioned initial operation data may be data of factors that affect the normal operation of the power grid, such as weather factors, equipment factors, human factors, etc. The initial operation data may include image data, power grid data, meteorological data, and time domain waveform data. In an optional embodiment, the stored initial operation data may be obtained from a cloud storage platform, or the initial operation data of the target warning area may be collected at time intervals based on sensors in the power grid, or the initial operation data may be reported manually.

[0059] The above-mentioned image data can be used to characterize the environmental state around the power grid. The environment around the power grid can be photographed by a camera in the power grid monitoring system to obtain image data. The image data can be static or dynamic. Through the image data, the physical state of the equipment in the target warning area, such as cracks, corrosion, etc., can be detected.

[0060] The above-mentioned power data can be used to characterize the load status of the target warning area. Through the power data, the load status of the power grid can be detected. The above-mentioned meteorological data can be used to characterize the weather status of the target warning area. Through the meteorological data, the environmental conditions of the location of the power grid can be detected.

[0061] The above-mentioned time domain waveform data can be used to characterize the state of the power line in the target warning area. The horizontal axis of the time domain waveform data represents the time point, and the vertical axis represents the reflection coefficient. The reflection coefficient represents the ratio between the impedance of the received reflected pulse signal and the impedance of the transmitted pulse signal. The time domain waveform data can be collected by transmitting the pulse signal.

[0062] By preprocessing the initial operation data of the target warning area, the target operation data can be obtained, so as to clean and convert the operation data of the target warning area and improve the efficiency of model use.

[0063] Optionally, the initial operation data is preprocessed, including: interpolating and filling missing values ​​in the initial operation data; gray-scaling image data in the initial operation data; normalizing power grid data and meteorological data in the initial operation data; and converting time domain waveform data in the initial operation data into frequency domain waveform data.

[0064] In an optional embodiment, the image data and time domain waveform data can be filled with the image data and time domain waveform data of the time points adjacent to the missing time points. The power data and meteorological data can be filled with the data of two time points adjacent to the missing time points. By filling the data, the integrity and quality of the data can be improved, thereby improving the accuracy and reliability of the prediction. The time domain waveform data can be converted by fast Fourier transform, and the horizontal axis of the frequency domain waveform data represents the frequency value, and the vertical axis represents the power spectrum density corresponding to the frequency value.

[0065] Optionally, based on the target operation data, graph network data is constructed, including: based on the target warning area, determining the nodes and edges of the graph network data, wherein the nodes of the graph network data and the target warning area are in a one-to-one mapping relationship, and the edges of the graph network data are connection relationships between the target warning areas; based on the target operation data, determining the initial features of the nodes, wherein the initial features are used to characterize the target operation data of the target warning area at a preset time point; and constructing the graph network data based on the nodes, edges and initial features.

[0066] The above initial features can be used to characterize the target operation data of the target warning area at a preset time point. The initial features of the node can include information such as power data and meteorological data of the warning area at different time points. The initial features of the node can reflect the dynamic changes of the warning area of ​​the power grid and the influence of environmental factors.

[0067] The warning area can be regarded as a node of a graph network. Then, the edges between nodes can be constructed based on the connection relationship between the warning areas. The initial feature of the node is the operation data of the warning area at a preset time point. Based on this, the graph network data is constructed through the nodes, edges and initial features of the nodes.

[0068] Optionally, the graph network data is predicted based on the fault prediction model to obtain the fault risk value of the target warning area, including: splitting the graph network data to obtain multiple sub-graphs, wherein different sub-graphs correspond to different warning areas, and the characteristics of the sub-nodes in each sub-graph are the same as the initial characteristics of the same nodes in the graph network data; predicting the multiple sub-graphs separately based on the fault prediction model to obtain the fault risk value of the target warning area.

[0069] In an optional embodiment, the above-mentioned graph network data can be split into multiple subgraphs, and multiple subgraphs constitute a graph network. It can be divided according to the weight or type of the edge, or according to the weight or characteristic attribute of the node, or it can be divided using a community discovery algorithm. Furthermore, the fault prediction model is used to predict the multiple subgraphs separately to obtain the fault risk value of the warning area corresponding to the subgraph, so as to accurately order the location where the fault occurs.

[0070] Optionally, the graph network data is split to obtain multiple subgraphs, including: determining the similarity between two nodes connected by an edge in the graph network data; merging the two nodes to generate a subgraph in response to the similarity being greater than or equal to a similarity threshold; adjusting the similarity threshold based on the number of subgraphs that have been generated, and repeatedly performing the steps of determining the similarity between two nodes connected by an edge in the graph network data, merging the two nodes to generate a subgraph in response to the similarity being greater than or equal to the similarity threshold, until the number of subgraphs that have been generated reaches a preset number; determining that the subgraph that has been generated is multiple subgraphs.

[0071] The similarity between adjacent nodes connected by edges can be calculated by cosine similarity, and the similarity is compared with the similarity threshold. When the similarity is greater than or equal to the similarity threshold, it indicates that the adjacent nodes are highly correlated, and the two nodes are merged into one subgraph. When the number of subgraphs obtained by the similarity threshold is less than the target number of subgraphs, the similarity threshold is reduced; if the number of subgraphs is greater than the preset number, the similarity threshold is increased, and this process continues until the number of generated subgraphs reaches the preset number.

[0072] By splitting the graph network data into multiple subgraphs, we can consider the correlation between warning areas, focus on the local characteristics of each warning area, reduce the amount of data for model analysis, and improve processing efficiency.

[0073] Optionally, the fault prediction model includes multiple parallel modules, with multiple sub Figure 1 One-to-one correspondence; based on the fault prediction model, multiple sub-graphs are predicted respectively to obtain the fault risk value of the target warning area, including: using the target parallel module corresponding to each sub-graph to predict the sub-graph to obtain the fault risk value of the warning area corresponding to the sub-graph.

[0074] The above fault prediction model may include multiple parallel modules, and the multiple parallel modules and multiple submodules Figure 1 One-to-one correspondence. Multiple parallel modules can run in parallel on multiple processors, reducing the computational load of a single parallel module, thereby reducing the prediction time and improving the real-time nature of the early warning.

[0075] The above-mentioned target parallel modules have corresponding subgraphs. The target parallel modules in the fault prediction model can predict the corresponding subgraphs, thereby improving the relevance of the model processing data and reducing the complexity of the model calculation.

[0076] Optionally, the target parallel module includes a first hidden layer, a second hidden layer, a third hidden layer, an extraction layer and a first classifier; the target parallel module corresponding to each subgraph is used to predict the subgraph to obtain the fault risk value of the warning area corresponding to the subgraph, including: using the first hidden layer to extract features of the subgraph to obtain multiple first vectors, wherein the multiple first vectors correspond one-to-one to multiple time points of the features of the subnodes in the subgraph; using the second hidden layer to encode the first vector to obtain multiple second vectors, wherein the multiple second vectors correspond one-to-one to the features of the subnodes of the subgraph; using the third hidden layer to process the subgraph and the second vector to obtain a graph update matrix, wherein different row vectors of the graph update matrix are used to characterize the update features of different subnodes in the subgraph; using the extraction layer to extract the update features of each subnode from the graph update matrix; using the first classifier to determine the fault risk value of the target warning area based on the update features of different subnodes in the subgraph.

[0077] The above-mentioned first hidden layer can be constructed based on a convolutional neural network model. The first hidden layer can be used to extract features from the subgraph to obtain multiple first vectors. Optionally, the first hidden layer may include a first hidden unit, a first logic unit, and a splicing unit. The first hidden unit can be used to extract features from the initial features of the nodes in the subgraph, and feature extraction can be performed on the image data at each time point in the initial features to obtain an image feature map. Furthermore, the feature map is represented by a vector through the first logic unit, and the vector representation can be spliced ​​with the target running data through the splicing unit to obtain multiple first vectors, so that each time point of the initial features of each node corresponds to a first vector.

[0078] The above-mentioned second hidden layer can be constructed based on the gated network unit. The first vector can be encoded by the second hidden layer to obtain multiple second vectors. Optionally, the second hidden layer may include a second hidden unit and a second logic unit. Multiple first vectors can be encoded by the second hidden unit to obtain multiple second vectors, so as to convert the initial features of the node into a vector representation. Furthermore, the second logic unit can be a fully connected layer or an autoencoder, which encodes or pools the first vector so that the initial features of each node correspond to a second vector.

[0079] The third hidden layer mentioned above can be constructed based on the graph attention neural network model. The third hidden layer can be used to process the subgraph and the second vector to obtain a graph update matrix, and different row vectors of the graph update matrix are used to characterize the update features of different subnodes in the subgraph. Optionally, the attention scores between the nodes in the subgraph can be calculated, and the importance of the nodes can be reflected by the attention scores, and the attention scores can be normalized. Furthermore, based on the subgraph, normalized attention scores, weight parameters, bias parameters and activation functions, the update features of each node are calculated. Then, the update features of the nodes can be stacked to obtain a graph update matrix. Through the graph update matrix, different weights can be assigned to the nodes to improve the accuracy of fault warning.

[0080] The extraction layer can be used to extract the update features of each child node from the graph update matrix, so that the classifier can perform classification calculations based on the update features.

[0081] The first classifier mentioned above can be constructed based on one of the following: logistic regression, decision tree, random forest, support vector machine, which is not limited here and can be set as needed. The first classifier can be used to determine the fault risk value of the target warning area based on the updated features of different sub-nodes in the sub-graph.

[0082] Optionally, the method further includes: constructing training samples based on the topological structure of the power grid and the equipment information of the equipment in the power grid; and training the pre-trained model using the training samples to obtain a fault prediction model.

[0083] The above topology is the connection relationship between the devices in the power grid. Based on the easily accessible topology and device information, training samples are constructed to train the pre-trained model and adjust the model parameters and generalization ability.

[0084] The above-mentioned pre-trained model can be used for power grid fault prediction. The pre-trained model can be constructed based on a convolutional neural network model, a gated network, a long short-term memory network, or a graph attention neural network model, which is not limited here and can be determined as needed.

[0085] Optionally, training samples are constructed based on the topological structure of the power grid and the equipment information of the equipment in the power grid, including: constructing a three-dimensional model of the power grid based on the topological structure of the power grid and the equipment information of the equipment in the power grid; collecting operating data of the three-dimensional model to construct sample data; labeling the sample data based on the power grid state and state duration of the power grid to determine the sample label, wherein the sample label is used to characterize the probability of failure of the sample data in a future time period; and constructing training samples based on the sample data and the sample label.

[0086] The above-mentioned three-dimensional model of the power grid can be simulation data of the power grid, which can be obtained by simulating the topological structure of the power grid and the equipment information of the equipment in the power grid through a simulation platform, so as to construct training samples based on the three-dimensional model.

[0087] The above-mentioned power grid state may include one of the following: normal state, single-phase grounding fault, two-phase short circuit, three-phase short circuit and phase-to-phase short circuit grounding, which is not limited here and can be determined as needed.

[0088] The sample data can be labeled by the grid state and the corresponding state duration, and the probability of fault occurrence under the grid state can be labeled to form a sample label for the sample data. Repeat the process until a preset number of training samples are obtained. Optionally, the labeling can be performed by relevant experts to improve the accuracy and reliability of grid fault warning.

[0089] Optionally, before using the training samples to train the pre-trained model to obtain the fault prediction model, the above method also includes: obtaining second training data, wherein the second training data includes: a second training sample, a second label and a third label, the second label is used to characterize the number of people in the target warning area, and the third label is used to characterize whether sparks occur in the target warning area; using the second training data to train the initial hidden layer of the initial prediction model to obtain the pre-trained model.

[0090] The above initial prediction model can be used for power grid fault prediction. The initial prediction model can be one of the following: a convolutional neural network model, a gated network, a long short-term memory network, or a graph attention neural network model, which is not limited here and can be determined as needed.

[0091] The above-mentioned initial hidden layer can extract features from the training samples. Before training the pre-trained model, the initial hidden layer of the initial prediction model can be trained. The second training sample can be labeled with the second label and the third label, and optionally, can be labeled by relevant experts, thereby improving the accuracy and reliability of the power grid fault warning.

[0092] According to an embodiment of the present invention, an embodiment of a power grid fault early warning device is provided. It should be noted that the device can be used to execute the above-mentioned power grid fault early warning method.

[0093] according to Figure 2 is a schematic diagram of the structure of a power grid fault early warning device according to an embodiment of the present invention. Figure 2As shown, the device includes: an acquisition module 22, which is used to obtain target operation data of a target warning area in a power grid, wherein the target warning area is an area to be warned in the power grid, and the target operation data includes power data and meteorological data; a determination module 24, which is used to determine graph network data based on the target operation data, wherein the graph network data is used to characterize the connection relationship between the target warning areas; a prediction module 26, which is used to predict the graph network data based on a fault prediction model to obtain a fault risk value of the target warning area, wherein the fault risk value is used to characterize the probability of a fault occurring in the target warning area in a future time period; an early warning module 28, which is used to determine early warning information based on the fault risk value, wherein the early warning information is used to characterize information on a fault in the power grid.

[0094] Optionally, the acquisition module 22 is also used to collect initial operation data of the target warning area at preset time intervals, wherein the initial operation data includes image data, power grid data, meteorological data and time domain waveform data; and pre-process the initial operation data to obtain target operation data.

[0095] Optionally, the above-mentioned acquisition module 22 is also used to interpolate and fill missing values ​​in the initial operation data; grayscale the image data in the initial operation data; normalize the power grid data and meteorological data in the initial operation data; and convert the time domain waveform data in the initial operation data into frequency domain waveform data.

[0096] Optionally, the above-mentioned determination module 24 is also used to determine the nodes and edges of the graph network data based on the target warning area, wherein the nodes of the graph network data and the target warning area are in a one-to-one mapping relationship, and the edges of the graph network data are the connection relationships between the target warning areas; based on the target operation data, determine the initial characteristics of the nodes, wherein the initial characteristics are used to characterize the target operation data of the target warning area at a preset time point; and construct the graph network data based on the nodes, edges and initial characteristics.

[0097] Optionally, the above-mentioned prediction module 26 is also used to split the graph network data to obtain multiple sub-graphs, wherein different sub-graphs correspond to different warning areas, and the characteristics of the sub-nodes in each sub-graph are the same as the initial characteristics of the same node in the graph network data; based on the fault prediction model, the multiple sub-graphs are predicted separately to obtain the fault risk value of the target warning area.

[0098] Optionally, the above-mentioned prediction module 26 is also used to determine the similarity between two nodes connected by an edge in the graph network data; in response to the similarity being greater than or equal to a similarity threshold, merge the two nodes to generate a subgraph; based on the number of subgraphs that have been generated, adjust the similarity threshold, and repeat the steps of determining the similarity between two nodes connected by an edge in the graph network data, and in response to the similarity being greater than or equal to the similarity threshold, merge the two nodes to generate a subgraph, until the number of subgraphs that have been generated reaches a preset number; determine that the subgraph that has been generated is multiple subgraphs.

[0099] Optionally, the above fault prediction model includes multiple parallel modules and multiple submodules. Figure 1 The prediction module 26 is also used to use the target parallel module corresponding to each subgraph to predict the subgraph and obtain the fault risk value of the warning area corresponding to the subgraph.

[0100] Optionally, the target parallel module includes a first hidden layer, a second hidden layer, a third hidden layer, an extraction layer and a first classifier; the prediction module 26 is further used to use the first hidden layer to extract features from the subgraph to obtain multiple first vectors, wherein the multiple first vectors correspond one-to-one to multiple time points of the features of the subnodes in the subgraph; use the second hidden layer to encode the first vector to obtain multiple second vectors, wherein the multiple second vectors correspond one-to-one to the features of the subnodes of the subgraph; use the third hidden layer to process the subgraph and the second vector to obtain a graph update matrix, wherein different row vectors of the graph update matrix are used to characterize the update features of different subnodes in the subgraph; use the extraction layer to extract the update features of each subnode from the graph update matrix; use the first classifier to determine the fault risk value of the target warning area based on the update features of different subnodes in the subgraph.

[0101] Optionally, the above device also includes a training module.

[0102] Among them, the training module is used to construct training samples based on the topological structure of the power grid and the equipment information of the equipment in the power grid; the pre-training model is trained using the training samples to obtain a fault prediction model.

[0103] Optionally, the above-mentioned training module is also used to construct a three-dimensional model of the power grid based on the topological structure of the power grid and the equipment information of the equipment in the power grid; collect the operating data of the three-dimensional model to construct sample data; label the sample data based on the power grid status and status duration of the power grid, and determine the sample label, wherein the sample label is used to characterize the probability of failure of the sample data in a future time period; and construct training samples based on the sample data and sample labels.

[0104] Optionally, the above device also includes a pre-training module.

[0105] Among them, the above-mentioned pre-training module is used to obtain second training data, wherein the second training data includes: a second training sample, a second label and a third label, the second label is used to characterize the number of people in the target warning area, and the third label is used to characterize whether sparks occur in the target warning area; the initial hidden layer of the initial prediction model is trained using the second training data to obtain a pre-training model.

[0106] An embodiment of the present application further provides an electronic device, comprising: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of the present invention when running.

[0107] An embodiment of the present application further provides a computer-readable storage medium, which includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the methods in various embodiments of the present invention.

[0108] An embodiment of the present application further provides a computer program product, including a computer program, which implements the methods in various embodiments of the present invention when executed by a processor.

[0109] An embodiment of the present application further provides a computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0110] The embodiments of the present application further provide a computer program, which implements the methods in the above-mentioned embodiments of the present invention when executed by a processor.

[0111] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0112] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0113] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0114] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0115] If the integrated unit is implemented in the form of 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 the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0116] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for early warning of power grid failure, characterized in that: include: Acquire target operation data of a target warning area in the power grid, wherein the target warning area is an area to be warned in the power grid, and the target operation data includes power data and meteorological data; Based on the target operation data, constructing graph network data, wherein the graph network data is used to characterize the connection relationship between the target warning areas; Predicting the graph network data based on the fault prediction model to obtain a fault risk value of the target warning area, wherein the fault risk value is used to characterize the probability of a fault occurring in the target warning area in a future time period; Based on the fault risk value, early warning information is determined, wherein the early warning information is used to characterize information about a fault in the power grid.

2. The early warning method according to claim 1, characterized in that: The target early warning area includes an area obtained by dividing the power grid; The step of obtaining target operation data of a target early warning area in a power grid includes: Collecting initial operation data of the target warning area at preset time intervals, wherein the initial operation data includes image data, power grid data, meteorological data and time domain waveform data; The initial operation data is preprocessed to obtain the target operation data.

3. The early warning method according to claim 2, characterized in that: The preprocessing of the initial operation data comprises: Performing interpolation and filling processing on missing values ​​in the initial running data; Performing grayscale processing on the image data in the initial operation data; Normalizing the power grid data and the meteorological data in the initial operation data; The time domain waveform data in the initial operation data is converted into frequency domain waveform data.

4. The early warning method according to claim 1, characterized in that: The step of constructing graph network data based on the target operation data includes: Based on the target warning area, determine the nodes and edges of the graph network data, wherein the nodes of the graph network data and the target warning areas are in a one-to-one mapping relationship, and the edges of the graph network data are connection relationships between the target warning areas; Based on the target operation data, determining the initial characteristics of the node, wherein the initial characteristics are used to characterize the target operation data of the target warning area at a preset time point; The graph network data is constructed based on the nodes, the edges and the initial features.

5. The early warning method according to claim 1, characterized in that: The predicting the graph network data based on the fault prediction model to obtain the fault risk value of the target warning area includes: Splitting the graph network data to obtain a plurality of subgraphs, wherein different subgraphs correspond to different warning areas, and the features of the subnodes in each subgraph are the same as the initial features of the same node in the graph network data; The multiple sub-graphs are predicted respectively based on the fault prediction model to obtain the fault risk value of the target warning area.

6. The early warning method according to claim 5, characterized in that: The graph network data is split to obtain multiple subgraphs, including: Determining a similarity between two nodes connected by an edge in the graph network data; In response to the similarity being greater than or equal to a similarity threshold, merging the two nodes to generate a subgraph; Based on the number of subgraphs that have been generated, the similarity threshold is adjusted, and the steps of determining the similarity between two nodes connected by an edge in the graph network data, and merging the two nodes to generate a subgraph in response to the similarity being greater than or equal to the similarity threshold are repeatedly performed until the number of subgraphs that have been generated reaches a preset number; Determine that the sub-graphs that have been generated are the multiple sub-graphs.

7. The early warning method according to claim 5, characterized in that: The fault prediction model includes a plurality of parallel modules corresponding one to one to the plurality of subgraphs; Predicting the multiple sub-graphs respectively based on the fault prediction model to obtain the fault risk value of the target warning area includes: The target parallel module corresponding to each sub-graph is used to predict the sub-graph to obtain the fault risk value of the warning area corresponding to the sub-graph.

8. The early warning method according to claim 7, characterized in that: The target parallel module includes a first hidden layer, a second hidden layer, a third hidden layer, an extraction layer and a first classifier; the target parallel module corresponding to each subgraph is used to predict the subgraph to obtain the fault risk value of the warning area corresponding to the subgraph, including: Using the first hidden layer to extract features from the subgraph, a plurality of first vectors are obtained, wherein the plurality of first vectors correspond one-to-one to a plurality of time points of features of subnodes in the subgraph; Encoding the first vector using the second hidden layer to obtain a plurality of second vectors, wherein the plurality of second vectors correspond one-to-one to features of the sub-nodes of the sub-graph; Processing the subgraph and the second vector using the third hidden layer to obtain a graph update matrix, wherein different row vectors of the graph update matrix are used to represent update features of different subnodes in the subgraph; extracting an update feature of each child node from the graph update matrix using the extraction layer; The first classifier is used to determine the fault risk value of the target warning area based on the updated features of different sub-nodes in the sub-graph.

9. The early warning method according to any one of claims 1 to 8, characterized in that: Also includes: Constructing training samples based on the topological structure of the power grid and device information of devices in the power grid; The pre-training model is trained using the training samples to obtain the fault prediction model.

10. The early warning method according to claim 9, characterized in that: The constructing of training samples based on the topological structure of the power grid and the device information of the devices in the power grid includes: Constructing a three-dimensional model of the power grid based on the topological structure of the power grid and device information of devices in the power grid; Collecting the operating data of the three-dimensional model and constructing sample data; Based on the grid state and state duration of the grid, label the sample data to determine a sample label, wherein the sample label is used to characterize the probability of the sample data failing within the future time period; A training sample is constructed based on the sample data and the sample label.

Images