CN115733730A - Power grid fault detection method and device based on graph neural network - Google Patents

Abstract

The invention discloses a grid fault detection method and device based on a graph neural network. The method includes: constructing graph structure data based on a node system in a power system, and dividing the graph structure data into a training set and a test set; constructing an adjacency matrix Construct a graph neural network model based on the adjacency matrix and graph structure data, use the training set to train the graph neural network model to obtain a grid fault detection model, and use the test set to test the grid fault detection model to obtain a trained grid fault detection model; The data of the nodes to be detected in the power grid in the system is input into the trained grid fault detection model to detect the faults of the grid nodes, and the detection results of the grid node faults are obtained. The invention applies the proposed graph neural network to problems such as power grid fault location, power grid node correlation analysis, power grid fault key feature extraction, etc. After verification of simulated fault scenarios, it can greatly improve the positioning efficiency of fault nodes and improve positioning accuracy.

Description

A grid fault detection method and device based on graph neural network

technical field

The invention relates to the technical field of grid fault diagnosis, in particular to a grid fault detection method and device based on a graph neural network.

Background technique

With the continuous development of the power Internet of Things, power big data ushers in new opportunities. At the same time, with the vigorous development of smart distribution network construction and increasing investment, it has become an urgent task to evaluate the quality of smart distribution network construction. In recent years, data-driven modeling and artificial intelligence algorithms have developed rapidly, and distribution network evaluation methods based on sample learning have also become research hotspots. Regulatory institutions use artificial intelligence technology to simulate human thinking, and discover laws by learning massive power grid operating data and operating experience.

At present, the disposal and dispatching work after a power grid failure mainly depends on the subjective decision-making of the dispatcher. The dispatcher analyzes the status and parameter changes of the power grid after the fault in real time, finds out the cause of the fault, and formulates corresponding fault handling measures.

With the rapid development of the power system, the structure and operation mode of the power grid are becoming more and more complex, and the difficulty of handling after a fault is increasing. The traditional dispatching decision-making mechanism relying on manual experience is becoming more and more difficult to deal with the rapid fault analysis and fault handling of complex large power grids. In recent decades, domestic and foreign scholars have proposed a series of methods and ideas for fault diagnosis, mainly including Petri nets, artificial neural networks, genetic algorithms, rough set decision-making, expert systems, data mining and other intelligent methods.

The power system needs to use data processing to extract key features from unstructured fault data. Due to the electrical connection properties of the grid structure, the grid structure can be abstracted into a graph structure, and the graph neural network can be used to assist judgment. When a grid fault occurs, help The dispatcher quickly analyzes the cause of the accident, comprehensively grasps the key information of fault handling, and makes auxiliary decisions to improve the emergency response capability of the power grid. At present, there is still a lack of a means of using the graph structure to detect faults in the power grid.

Contents of the invention

The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

Therefore, the purpose of the present invention is to propose a grid fault detection method based on graph neural network. By constructing a grid feature extraction model and a knowledge graph, the scattered fault case knowledge in the grid field is effectively associated with complex fault data, and the application based on The neural network of graph convolution performs intelligent diagnosis of network faults and assists in solving fault problems in the field of network operation and maintenance. Applying the proposed graph neural network to power grid fault location, power grid node correlation analysis, power grid fault key feature extraction and other issues, after simulation fault scene verification, can greatly improve the positioning efficiency of fault nodes and improve positioning accuracy.

Another object of the present invention is to propose a grid fault detection device based on a graph neural network.

In order to achieve the above object, the present invention proposes a grid fault detection method based on a graph neural network on the one hand, including:

Constructing graph structure data based on the node system in the power system, and dividing the graph structure data into a training set and a test set;

Construct an adjacency matrix based on the graph structure data topology graph and the electrical connection relationship between the topology graph nodes;

Constructing a graph neural network model based on the adjacency matrix and the graph structure data, using the training set to train the graph neural network model to obtain a power grid fault detection model, and using the test set to test the power grid fault detection model to obtain a trained model Grid fault detection model;

Inputting the data of nodes to be detected in the grid in the power system into the trained grid fault detection model to detect grid node faults and obtain grid node fault detection results.

The grid fault detection method based on the graph neural network according to the embodiment of the present invention may also have the following additional technical features:

Further, in one embodiment of the present invention, the constructing the graph neural network model based on the adjacency matrix and the graph structure data includes: performing feature dimension processing on the graph structure data based on the graph neural network and the degree of the adjacency matrix, and outputting the first One-dimensional data; obtain second-dimensional data based on the first-dimensional data and a preset processing method, use the second-dimensional data and a preset function to obtain a final predicted value, and use the final predicted value and label to calculate a loss function to Build a neural network model.

Further, in one embodiment of the present invention, the construction of the graph structure data based on the node system in the power system includes: based on the time when the grid fault occurs in the power system, filtering the fault data at the preset time to obtain the simulation data; based on the Obtain fault state data labels and non-fault state data labels from the above simulation data, and add them to the data set samples.

Further, in an embodiment of the present invention, the dividing the graph structure data into a training set and a test set includes: according to the timing relationship and preset ratio of the samples in the data set, taking every three time nodes as a Group, the first two time nodes are divided into training sets, and the latter time nodes are divided into test sets.

Further, in one embodiment of the present invention, the loss function adopts binary cross-entropy loss, and the expression is:

In the formula, N is the total number of samples in the data set, y _n is the probability of the real value, which is 0 or 1, and x _n is the probability of the predicted value.

In order to achieve the above object, another aspect of the present invention proposes a network fault detection device based on a graph neural network, including:

The data acquisition module is used to construct the graph structure data based on the node system in the power system, and divide the graph structure data into a training set and a test set;

A matrix building block for constructing an adjacency matrix based on the graph structure data topology graph and the electrical connection relationship between the nodes of the graph structure;

A model training module, configured to construct a graph neural network model based on the adjacency matrix and the graph structure data, use the training set to train the graph neural network model to obtain a grid fault detection model, and use the test set to test the grid fault The detection model obtains the trained grid fault detection model;

The fault detection module is used to input the data of nodes to be detected in the grid in the power system into the trained grid fault detection model to detect grid node faults and obtain grid node fault detection results.

The grid fault detection method and device based on the graph neural network in the embodiment of the present invention can apply the proposed graph neural network to grid fault location, grid node correlation analysis, grid fault key feature extraction and other issues, and is verified by simulated fault scenarios , which can greatly improve the location efficiency of the fault node and improve the location accuracy.

Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.

Description of drawings

The above and/or additional aspects and advantages of the present invention will become apparent and easy to understand from the following description of the embodiments in conjunction with the accompanying drawings, wherein:

Fig. 1 is a flow chart of a grid fault detection method based on a graph neural network according to an embodiment of the present invention;

Fig. 2 is a schematic diagram of IEEE10 machine 39 node topology according to an embodiment of the present invention;

FIG. 3 is a basic flowchart of a graph neural network according to an embodiment of the present invention;

Fig. 4 is the architectural diagram of graph neural network according to the embodiment of the present invention

Fig. 5 is a schematic structural diagram of a grid fault detection device based on a graph neural network according to an embodiment of the present invention.

Detailed ways

It should be noted that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and examples.

In order to enable those skilled in the art to better understand the solutions of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only It is an embodiment of a part of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

The graph neural network-based power grid fault detection method and device proposed according to the embodiments of the present invention are described below with reference to the accompanying drawings.

Fig. 1 is a flowchart of a grid fault detection method based on a graph neural network according to an embodiment of the present invention.

As shown in Figure 1, the method includes but is not limited to the following steps:

S1, construct the graph structure data based on the node system in the power system, and divide the graph structure data into a training set and a test set;

S2, constructing an adjacency matrix based on the graph structure data topology graph and the electrical connection relationship between nodes in the graph structure;

S3, build a graph neural network model based on the adjacency matrix and graph structure data, use the training set to train the graph neural network model to obtain a power grid fault detection model, and use the test set to test the power grid fault detection model to obtain a trained power grid fault detection model;

S4. Input the data of nodes to be detected in the grid in the power system into the trained grid fault detection model to detect grid node faults, and obtain grid node fault detection results.

The graph neural network-based power grid fault detection method of the embodiment of the present invention will be described in detail below in conjunction with the accompanying drawings.

The first choice is to construct graph structure data, and Fig. 2 is a schematic diagram of IEEE10 machine 39 node topology according to an embodiment of the present invention, as shown in Fig. 2:

The IEEE10 39-bus system is a well-known regional power transmission system network in the field of power systems, also known as the New England 39-bus system (NE39BS). This benchmark network is configured in New England, the United States, and consists of 39 bus Composition, including 10 generator buses and 19 load buses, is widely used in small signal stability research, dynamic stability analysis, power quality analysis and control and other fields. This research is based on a 39-node system, and the data mainly comes from the 39-node system simulation.

In the IEEE10 machine 39-node system, the node attributes include bus amplitude, phase angle, generator excitation voltage, power angle, active power, and reactive power. In order to introduce the characteristics of the time dimension, use the information of adjacent moments to assist in the prediction of the state of the intermediate moment, set the sliding window size of the time dimension to 5 moments, and calculate the bus amplitude and phase angle between the current moment and the two moments before and after. Mean and variance, also as properties of the node. Therefore, the graph structure data of the 39-node grid system can be expressed as

Data construction. The fault occurrence time is set to 0.1s, and the faults are removed at 0.70s, 0.72s, 0.74s, 0.76s, and 0.78s respectively to obtain simulation data. Set the label of the fault state data generated by removing the fault at 0.70s, 0.72s, 0.74s, 0.76s, and 0.78s to 1, a total of 320 samples. The non-fault state data label of the time node from 0.74s removal of the fault to the final time node is set to 0, a total of 927 samples. The data set has a total of 1247 samples. Divide the obtained data set into a training set and a test set. The specific division method is: according to the time series relationship, every three time nodes are grouped, the first two are set as the training set, and the latter is set as the test set. Among them, the training set has a total of 832 samples, and the test set has a total of 415 samples, with a ratio of about 2:1. The topology of all samples in this dataset is the same.

Further, build an adjacency matrix:

It is understandable that the topological graph abstracted from the IEEE 10-machine 39-node system is regarded as an undirected graph regardless of the power flow direction. According to the connection relationship between nodes, build an adjacency matrix

Further, the structure of the graph neural network, Figure 3 is the basic flowchart of the graph neural network, and Figure 4 is the architecture diagram of the graph neural network, as shown in Figure 3 and Figure 4,

The overall architecture of the graph neural network adopts a U-shaped structure. The input graph structure data is first increased in dimension and then reduced in dimension. The final output feature dimension is 39×1, and each graph neural network layer is nonlinearly activated by an activation function. transform.

Loss function settings. Apply a mask to the output of the network, shield the data of other dimensions except 15 nodes, and get the final predicted value through the Sigmoid function. Use the final predicted value and label to calculate the loss function. The loss function adopts the binary cross entropy loss (BCEloss), and the expression is:

In the formula, N is the total number of samples, y _n is the probability of the real value, which is 0 or 1, and x _n is the probability of the predicted value.

Furthermore, set the learning rate of the network and the weight of each part of the loss function. As an example, use the deep pytorch to train the above convolutional neural network until the loss converges to generate a training model.

After determining the network structure, input the training data to the network;

In the network training phase, the learning rate is set to 0.0001, and after every 50 epochs, the optimization method adopts the Adam method, in which the learning rate is set to 0.001, and the weight decay is set to 5e-4

Perform training until the network converges to generate a training model.

Using the test set to test the above training model to obtain a trained grid fault detection model; input the data of the nodes to be detected in the power system into the trained grid fault detection model for grid node fault detection, and obtain the grid node fault detection results.

According to the grid fault detection method based on the graph neural network in the embodiment of the present invention, a graph-based data structure is constructed based on the grid topology, and data features are extracted efficiently. Fault detection based on graph neural network is faster and more accurate than traditional methods.

In order to realize the above embodiment, as shown in FIG. 5 , a grid fault detection device 10 based on a graph neural network is also provided in this embodiment, and the device 10 includes: a data acquisition module 100, a matrix construction module 200, a model training module 300 and Fault detection module 400.

The data acquisition module 100 is used to construct the graph structure data based on the node system in the power system, and divide the graph structure data into a training set and a test set;

A matrix construction module 200, configured to construct an adjacency matrix based on the graph structure data topology graph and the electrical connection relationship between the topology graph nodes;

The model training module 300 is used to construct a graph neural network model based on the adjacency matrix and the graph structure data, use the training set to train the graph neural network model to obtain a grid fault detection model, and use the test set to test the grid fault detection model to obtain a trained grid Fault detection model;

The fault detection module 400 is used to input the data of the nodes to be detected in the power grid in the power system into the trained grid fault detection model to detect the faults of the grid nodes and obtain the detection results of the grid node faults.

Further, the above-mentioned model training module 300 is also used for:

Based on the graph neural network and adjacency matrix degree, the feature dimension processing of the graph structure data is performed, and the first dimension data is output;

The second-dimensional data is obtained based on the first-dimensional data and the preset processing method, the final predicted value is obtained by using the second-dimensional data and the preset function, and the loss function is calculated by using the final predicted value and labels to build a neural network model.

Further, the data acquisition module 100 is also used for:

Based on the time when the power grid fault occurs in the power system, the fault data at the preset time is filtered to obtain the simulation data;

Based on the simulation data, the fault state data labels and non-fault state data labels are obtained, and the data set samples are obtained.

Further, the data acquisition module 100 is also used for:

According to the timing relationship and preset ratio of the data set samples, every three time nodes are divided into a group, and the first two time nodes are divided into training sets, and the latter time nodes are divided into test sets.

Further, the loss function adopts the binary classification cross entropy loss, and the expression is:

In the formula, N is the total number of samples in the data set, y _n is the probability of the real value, which is 0 or 1, and x _n is the probability of the predicted value.

According to the graph neural network-based power grid fault detection device of the embodiment of the present invention, a graph-based data structure is constructed based on the power grid topology, and data feature extraction is performed efficiently. Fault detection based on graph neural network is faster and more accurate than traditional methods.

In addition, the terms "first" and "second" are used for descriptive purposes only, and cannot be interpreted as indicating or implying relative importance or implicitly specifying the quantity of indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include at least one of these features. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

In the description of this specification, descriptions referring to the terms "one embodiment", "some embodiments", "example", "specific examples", or "some examples" mean that specific features described in connection with the embodiment or example , structure, material or characteristic is included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the described specific features, structures, materials or characteristics may be combined in any suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine different embodiments or examples and features of different embodiments or examples described in this specification without conflicting with each other.

Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention, those skilled in the art can make the above-mentioned The embodiments are subject to changes, modifications, substitutions and variations.

Claims (10)

1. A grid fault detection method based on graph neural network, is characterized in that, comprises the following steps: Constructing graph structure data based on the node system in the power system, and dividing the graph structure data into a training set and a test set; Construct an adjacency matrix based on the graph structure data topology graph and the electrical connection relationship between the topology graph nodes; Constructing a graph neural network model based on the adjacency matrix and the graph structure data, using the training set to train the graph neural network model to obtain a power grid fault detection model, and using the test set to test the power grid fault detection model to obtain a trained model Grid fault detection model; Inputting the data of nodes to be detected in the grid in the power system into the trained grid fault detection model to detect grid node faults and obtain grid node fault detection results. 2. The method according to claim 1, wherein said building a graph neural network model based on adjacency matrix and said graph structure data comprises: Based on the graph neural network and adjacency matrix degree, the feature dimension processing of the graph structure data is performed, and the first dimension data is output; Obtain second-dimensional data based on the first-dimensional data and a preset processing method, use the second-dimensional data and preset functions to obtain a final predicted value, and use the final predicted value and labels to calculate a loss function to build a neural network Model. 3. The method according to claim 2, wherein the graph structure data based on the node system in the construction of the power system comprises: Based on the time when the power grid fault occurs in the power system, the fault data at the preset time is filtered to obtain the simulation data; Based on the simulation data, the fault state data label and the non-fault state data label are obtained, and the data set samples are obtained. 4. The method according to claim 3, wherein said graph structure data is divided into a training set and a test set, comprising: According to the timing relationship and preset ratio of the data set samples, every three time nodes are used as a group, and the first two time nodes are divided into training sets, and the latter time nodes are divided into test sets. 5. method according to claim 4, is characterized in that, described loss function adopts binary classification cross-entropy loss, and expression is:

In the formula, N is the total number of samples in the data set, y _n is the probability of the real value, which is 0 or 1, and x _n is the probability of the predicted value. 6. A power grid fault detection device based on graph neural network, characterized in that, comprising: The data acquisition module is used to construct the graph structure data based on the node system in the power system, and divide the graph structure data into a training set and a test set; A matrix building block for constructing an adjacency matrix based on the graph structure data topology graph and the electrical connection relationship between the nodes of the graph structure; A model training module, configured to construct a graph neural network model based on the adjacency matrix and the graph structure data, use the training set to train the graph neural network model to obtain a grid fault detection model, and use the test set to test the grid fault The detection model obtains the trained grid fault detection model; The fault detection module is used to input the data of nodes to be detected in the grid in the power system into the trained grid fault detection model to detect grid node faults and obtain grid node fault detection results. 7. The device according to claim 6, wherein the model training module is also used for: Based on the graph neural network and adjacency matrix degree, the feature dimension processing of the graph structure data is performed, and the first dimension data is output; Obtain second-dimensional data based on the first-dimensional data and a preset processing method, use the second-dimensional data and preset functions to obtain a final predicted value, and use the final predicted value and labels to calculate a loss function to build a neural network Model. 8. The device according to claim 7, wherein the data acquisition module is also used for: Based on the time when the power grid fault occurs in the power system, the fault data at the preset time is filtered to obtain the simulation data; Based on the simulation data, the fault state data label and the non-fault state data label are obtained, and the data set samples are obtained. 9. The device according to claim 8, wherein the data acquisition module is also used for: According to the timing relationship and preset ratio of the data set samples, every three time nodes are used as a group, and the first two time nodes are divided into training sets, and the latter time nodes are divided into test sets. 10. The device according to claim 9, wherein the loss function adopts binary cross-entropy loss, and the expression is:

In the formula, N is the total number of samples in the data set, y _n is the probability of the real value, which is 0 or 1, and x _n is the probability of the predicted value.

Images