A grid-based fault monitoring method to improve distribution network safety

Abstract

The present invention discloses a grid fault monitoring method that can improve the safety of distribution network. The method proposes to collect data for monitoring distribution network grid faults; then the monitoring data collected in the distribution network grid is trained, including convolution operation, function activation, pooling operation and full connection, etc., to build a distribution network grid node fault detection model and a distribution network grid equipment fault detection model based on convolutional neural network, and finally update the trained distribution network grid node fault detection model and distribution network grid equipment fault detection model to the edge computing terminal of the distribution network grid, and perform online calculation and analysis on the faults of grid nodes and equipment at the grid edge computing terminal, and perform corresponding disposal once a fault occurs. The present invention can more effectively capture and understand the complex characteristics in the power grid, efficiently process huge monitoring data, monitor and process the faults of distribution network grid nodes and equipment, and improve the safety of the distribution network grid.

Description

Technical Field

[0001] The present invention relates to the field of power safety technology, and in particular to a grid fault monitoring method capable of improving the safety of a distribution network. Background Art

[0002] Grid gridding refers to dividing a complex distribution network into several geographically and electrically independent power supply networks based on geography, grid status, load distribution, and upstream power supply conditions, and using the network as the smallest unit for distribution network planning, project management, and user access. Through grid-based power supply, we can clearly understand the equipment, resources, and investment conditions within the grid, assigning business management responsibilities such as operation, planning, and marketing to specific people, thereby accurately locating distribution network problems and improving the effectiveness of distribution network construction, transformation, and operation management.

[0003] At present, the distribution network grid fault detection still has the following shortcomings: First, the feature extraction capability is limited. The ability of traditional monitoring methods to extract key features in complex power grid environments is limited, and it is difficult to adapt to the diversity and complexity of power grid conditions. Second, the data processing efficiency is low. The large amount of data generated by distribution network monitoring needs to be processed efficiently. Traditional methods have bottlenecks in large-scale data processing and real-time performance, and it is difficult to meet the demand for rapid response to power grid conditions. Third, the accuracy of abnormal fault detection is insufficient. Some traditional methods may be insufficient in the accuracy of abnormal power grid fault detection, especially for rapidly occurring and complex fault conditions. The monitoring system may show low sensitivity and accuracy.

[0004] To this end, we designed a grid-based fault monitoring method that can improve the safety of distribution networks to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art in the low ability to extract key features in the power grid environment, low data processing efficiency, and insufficient accuracy of abnormal fault detection in the distribution network grid fault detection, and propose a grid fault monitoring method that can improve the safety of the distribution network. This method can not only improve the feature extraction capability and data processing efficiency, but also enhance the accuracy of abnormal fault detection. By implementing grid monitoring, the status of each node and device of the power grid can be monitored in real time, and potential faults or abnormal conditions can be detected and located in time. It not only helps to improve the stability and reliability of the power grid and reduce the possibility of power outages, but also accelerates fault location and repair, and improves the response speed of the power system, thereby ensuring the continuity and quality of power supply.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A grid-based fault monitoring method for improving the safety of a distribution network comprises the following steps:

[0008] Step 1, collecting monitoring data of distribution network grid faults, the monitoring data includes collecting current data, voltage data, power data, frequency data and voltage phase angle data of distribution network grid nodes, and current data, voltage data, power data and temperature data of distribution network grid equipment;

[0009] Step 2: Using the monitoring data of distribution network grid faults as training samples, a distribution network grid node fault detection model and a distribution network grid equipment fault detection model based on convolutional neural network are constructed;

[0010] Step 3: Update the distribution network grid node fault detection model and the distribution network grid device fault detection model to the edge computing terminal of the distribution network grid, perform online calculation and analysis on the faults of the distribution network grid nodes and devices at the grid edge computing terminal, and perform fault monitoring on the distribution network grid nodes and devices.

[0011] Furthermore, in step 1, the monitoring data of the distribution network grid fault is collected and expressed by the following formula:

[0012] (1)

[0013] (2)

[0014] In the formula, For collection Distribution network grid node data at all times, The distribution network grid nodes collected exist Current data at the moment, The distribution network grid nodes collected exist Voltage data at the moment, The distribution network grid nodes collected exist Power data at the moment, The distribution network grid nodes collected exist Frequency data at time, The distribution network grid nodes collected exist Voltage phase angle data at the moment; For collection All-time grid equipment data, Distribution network grid equipment for collection exist Current data at the moment, Distribution network grid equipment for collection exist Voltage data at the moment, Distribution network grid equipment for collection exist Power data at the moment, Distribution network grid equipment for collection exist Temperature data at the moment.

[0015] Furthermore, in step 2, the distribution network grid node data in the monitoring data of the distribution network grid fault is marked as a training sample, and the construction of a distribution network grid node fault detection model based on a convolutional neural network includes the following steps:

[0016] Step 2.11, input the training samples to the input layer of the convolutional neural network, perform convolution operation on the training samples, and the convolution operation slides the input training samples through the convolution kernel to extract the fault features in the distribution network grid nodes, as shown in the following formula:

[0017] (3)

[0018] In the formula, represents the convolution result, is the convolution kernel, , is the index of the matrix, Represents the matrix OK, Represents the matrix Column; m and n represent the row and column indices of the convolution kernel, respectively. For distribution network grid nodes The input data, Represents the element corresponding to the convolution kernel, Represents the convolution kernel OK, Represents the convolution kernel Columns;

[0019] Step 2.12, construct the activation function of the distribution network grid node fault detection model, introduce nonlinear transformation, and capture the abstract characteristics of the distribution network grid node fault, as shown in the following formula:

[0020] (4)

[0021] In the formula, represents the activation function of the distribution network grid node fault detection model, Represents the abstract features of captured distribution network grid node faults;

[0022] Step 2.13, perform a pooling operation by sampling the fault features of the distribution network grid nodes and using the maximum pooling operation, as shown in the following formula:

[0023] (5)

[0024] In the formula, is the function after pooling of the distribution network grid node fault detection model, is the activation function OK, is the activation function The columns of , m and n represent the row and column indices of the convolution kernel respectively;

[0025] Step 2.14, perform the convolutional neural network fully connected layer operation, the weight of the fully connected layer is , the bias of the fully connected layer is , the output of the fully connected layer of the convolutional neural network is shown as follows:

[0026] (6)

[0027] In the formula, Function represents the pooled function Flatten to vector;

[0028] Step 2.15, output the final distribution network grid node fault detection model, as shown in the following formula:

[0029] (7)

[0030] In the formula, is the distribution network grid node fault detection model, When the value is 1, it indicates that there is a node fault in the distribution network grid. When the value is 0, it indicates that there is no node fault in the distribution network grid.

[0031] Furthermore, in the training and construction process of the distribution network grid node fault detection model, the difference between the predicted output and the actual label of the distribution network grid node fault detection model is measured by minimizing the loss function, as shown in the following formula:

[0032] (8)

[0033] In the formula, represents the loss function of the distribution network grid node fault detection model, Represents the distribution network grid node fault detection model and the true value of the distribution network grid node fault detection model The difference between

[0034] Computing Gradients and , and update the weights of the fully connected layer , the bias of the fully connected layer After multiple iterations, the distribution network grid node fault detection model parameters are continuously adjusted to minimize the loss function and output the distribution network grid node fault detection model: .

[0035] Furthermore, in step 2, the monitoring data of the distribution network grid fault is used as a training sample to construct a distribution network grid equipment fault detection model based on a convolutional neural network, including the following steps:

[0036] Step 2.21, input the training samples to the input layer of the convolutional neural network, perform convolution operation on the training samples, and the convolution operation slides the input training samples through the convolution kernel to extract the fault features in the distribution network grid equipment, as shown in the following formula:

[0037] (9)

[0038] In the formula, represents the convolution result, is the convolution kernel, and is the index of the matrix, Represents the matrix OK, Represents the matrix List; and Represent the row and column indices of the convolution kernel respectively, For distribution grid equipment The input data, Represents the element corresponding to the convolution kernel, Represents the convolution kernel OK, Represents the convolution kernel Columns;

[0039] Step 2.22, construct the activation function of the distribution network grid equipment fault detection model, introduce nonlinear transformation, and capture the abstract characteristics of the distribution network grid equipment fault, as shown in the following formula:

[0040] (10)

[0041] In the formula, represents the activation function of the distribution network grid equipment fault detection model, An abstract feature representing the captured distribution network grid equipment fault;

[0042] Step 2.23, perform a pooling operation by sampling the fault characteristics of the distribution network grid equipment and using the maximum pooling operation, as shown in the following formula:

[0043] (11)

[0044] In the formula, is the function after pooling of the distribution network equipment fault detection model, is the activation function OK, is the activation function Columns, and Respectively represent the row and column indices of the convolution kernel;

[0045] Step 2.24, perform the convolutional neural network fully connected layer operation, the weight of the fully connected layer is , the bias of the fully connected layer is , the output of the fully connected layer of the convolutional neural network is shown as follows:

[0046] (12)

[0047] In the formula, Function represents the pooled function Flatten to vector;

[0048] Step 2.25, output the final distribution network grid equipment fault detection model, as shown in the following formula:

[0049] (13)

[0050] In the formula, It is a distribution network equipment fault detection model. When the value is 1, it indicates that there is a device fault in the distribution network grid. When the value is 0, it indicates that there is no device fault in the distribution network grid.

[0051] Furthermore, in step 3, the grid edge computing terminal performs online calculation and analysis on the faults of the distribution network grid nodes and devices, and performs fault monitoring on the distribution network grid nodes and devices, as shown in the following formula:

[0052] ;

[0053] In the formula, Indicates the command issued by the edge computing terminal of the distribution network grid, including the nodes in the distribution network grid Fault Isolation Instructions and Equipment Fault isolation instructions; when The value is 1, indicating that the node in the distribution network grid is monitored. A failure occurs, the node Issue a fault isolation command; when The value is 1, indicating that the device in the distribution network grid is monitored. Failure occurs, the device Issue fault isolation instructions.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention proposes a grid fault monitoring method that can improve the safety of the distribution network, proposes to collect data for monitoring the distribution network grid fault, train the monitoring data collected in the distribution network grid, including convolution operations, function activation, pooling operations and full connection processes, construct a distribution network grid node fault detection model and a distribution network grid equipment fault detection model based on a convolutional neural network, and finally update the trained distribution network grid node fault detection model and the distribution network grid equipment fault detection model to the edge computing terminal of the distribution network grid, and perform online calculation and analysis on the faults of the grid nodes and equipment at the grid edge computing terminal, and perform corresponding disposal once a fault occurs. The present invention can more effectively capture and understand the complex characteristics in the power grid, has high data processing efficiency, and can efficiently process huge monitoring data; accurately judge, monitor and process the faults of the distribution network grid nodes and equipment, improve the sensitivity and accuracy of monitoring, reduce the transmission of sensitive information, and improve the safety of the distribution network grid. By implementing grid monitoring, the present invention can monitor the status of each node and equipment of the power grid in real time, and detect and locate potential faults or abnormal conditions in time. It not only helps to improve the stability and reliability of the power grid and reduce the possibility of power outages, but also accelerates fault location and repair, and improves the response speed of the power system, thereby ensuring the continuity and quality of power supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A schematic diagram of a flow chart of a grid fault monitoring method for improving the safety of a distribution network proposed by the present invention;

[0056] Figure 2 The present invention is a schematic diagram of constructing a training distribution network grid node fault detection model in a grid fault monitoring method that can improve the safety of distribution networks. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present invention will be described clearly and completely 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, rather than all the embodiments.

[0058] Reference Figure 1 The present invention proposes a grid fault monitoring method for improving the safety of distribution network, comprising the following steps:

[0059] Step 1: Collect monitoring data of distribution network grid faults. The monitoring data includes collecting current data, voltage data, power data, frequency data and voltage phase angle data of distribution network grid nodes, and current data, voltage data, power data and temperature data of distribution network grid equipment.

[0060] The distribution grid node current is a key parameter for assessing the state of the grid. By monitoring the current at key nodes within the grid, current anomalies or imbalances can be detected, indicating possible faults. Voltage is another key monitoring parameter. Abnormal voltage levels may indicate problems in the grid, such as excessive or low voltage. Power data provides insights into the flow and consumption of electrical energy, helping to detect abnormal load faults. The frequency of the grid is an important reference indicator. Abnormal frequency may indicate grid imbalance or other problems. The voltage phase angle plays an important role in detecting grid stability, faults, asymmetry problems, etc. The current, voltage, power, temperature and other data of the equipment are mainly used to monitor the operating status of the equipment within the distribution grid.

[0061] The specific monitoring data of distribution network grid faults collected is expressed by the following formula:

[0062] (1)

[0063] (2)

[0064] In the formula, For collection Distribution network grid node data at all times, The distribution network grid nodes collected exist Current data at the moment, The distribution network grid nodes collected exist Voltage data at the moment, The distribution network grid nodes collected exist Power data at the moment, The distribution network grid nodes collected exist Frequency data at time, The distribution network grid nodes collected exist Voltage phase angle data at the moment; For collection All-time grid equipment data, Distribution network grid equipment for collection exist Current data at the moment, Distribution network grid equipment for collection exist Voltage data at the moment, Distribution network grid equipment for collection exist Power data at the moment, Distribution network grid equipment for collection exist Temperature data at the moment.

[0065] Step 2: Use the monitoring data of distribution network grid faults as training samples to build a distribution network grid node fault detection model and a distribution network grid equipment fault detection model based on convolutional neural network.

[0066] The distribution network grid node data in the monitoring data of distribution network grid faults are marked as training samples, and then the sample data are trained using a convolutional neural network, including convolution operations, function activation, pooling operations, back propagation and other processes. A distribution network grid node fault detection model and a distribution network grid equipment fault detection model based on convolutional neural networks are constructed to detect node faults in the distribution network and distribution network grid equipment faults.

[0067] This paper focuses on building a distribution network grid node fault detection model based on convolutional neural network. Figure 2 As shown in the figure, the technology inputs the collected data into the input layer of the convolutional neural network, and then performs convolution operations, pooling operations and full connection operations in the hidden layer. According to the fault samples of the nodes in the distribution network grid, the fault features of the sampled nodes in the distribution network grid are extracted, and finally the distribution network grid node fault detection model is output through the output layer, thereby realizing the accurate identification of node faults in the distribution network grid.

[0068] Specifically, in the present invention, constructing and training a distribution network grid node fault detection model based on a convolutional neural network includes the following steps:

[0069] Step 2.11, input the training samples to the input layer of the convolutional neural network, perform convolution operation on the training samples, and the convolution operation slides the input training samples through the convolution kernel to extract the fault features in the distribution network grid nodes, as shown in the following formula:

[0070] (3)

[0071] In the formula, represents the convolution result, is the convolution kernel, , is the index of the matrix, Represents the matrix OK, Represents the matrix Column; m and n represent the row and column indices of the convolution kernel, respectively. For distribution network grid nodes The input data, Represents the element corresponding to the convolution kernel, Represents the convolution kernel OK, Represents the convolution kernel Columns.

[0072] Step 2.12, construct the activation function of the distribution network grid node fault detection model, introduce nonlinear transformation, so that the convolutional neural network can learn and represent more complex distribution network grid fault characteristics, so that the convolutional neural network can more effectively capture the abstract characteristics of distribution network grid node faults, as shown in the following formula:

[0073] (4)

[0074] In the formula, represents the activation function of the distribution network grid node fault detection model, Represents the abstract features of captured distribution network grid node faults.

[0075] Step 2.13, perform a pooling operation, by sampling the distribution network grid node fault features, reduce the spatial dimension of the distribution network grid node fault data, while retaining important fault feature information, reduce the complexity of fault identification calculations, reduce the number of model parameters, and improve the model's robustness to translation invariance. The present invention adopts a maximum pooling operation, as shown in the following formula:

[0076] (5)

[0077] In the formula, is the function after pooling of the distribution network grid node fault detection model, is the activation function OK, is the activation function The columns of , m and n represent the row and column indices of the convolution kernel respectively.

[0078] Step 2.14, perform the convolutional neural network fully connected layer operation, the weight of the fully connected layer is , the bias of the fully connected layer is , the output of the fully connected layer of the convolutional neural network is shown as follows:

[0079] (6)

[0080] In the formula, Function represents the pooled function Flatten to a vector.

[0081] Step 2.15, output the final distribution network grid node fault detection model, as shown in the following formula:

[0082] (7)

[0083] In the formula, is the distribution network grid node fault detection model, When the value is 1, it indicates that there is a node failure in the distribution network. When the value is 0, it means that there is no node failure in the distribution network grid.

[0084] In the construction and training process of the distribution network grid node fault detection model, the model is guided to adjust in the direction of more accurately predicting the target by minimizing the loss function; the difference between the predicted output of the distribution network grid node fault detection model and the actual label is measured as shown in the following formula:

[0085] (8)

[0086] In the formula, represents the loss function of the distribution network grid node fault detection model, Represents the distribution network grid node fault detection model and the true value of the distribution network grid node fault detection model The difference between .

[0087] Computing Gradients and , and update the weights of the fully connected layer , the bias of the fully connected layer After multiple iterations, the distribution network grid node fault detection model parameters are continuously adjusted to minimize the loss function and output the distribution network grid node fault detection model: .

[0088] After the above process, the distribution network grid node fault detection model based on convolutional neural network is finally output , perform fault detection on distribution network grid nodes.

[0089] A distribution network grid equipment fault detection model based on convolutional neural network is constructed. This technology extracts fault features of equipment in the distribution network through fault samples of key equipment in the distribution network learned by convolutional neural network, thereby realizing accurate identification of key equipment faults in the distribution network. The process of constructing a distribution network grid equipment fault detection model based on convolutional neural network is similar to that of constructing a distribution network grid node fault detection model.

[0090] The present invention generally introduces the training and construction steps of the distribution network grid equipment fault detection model based on convolutional neural network as follows:

[0091] Step 2.21, input the training samples to the input layer of the convolutional neural network, perform convolution operation on the training samples, and the convolution operation slides the input training samples through the convolution kernel to extract the fault features in the distribution network grid equipment, as shown in the following formula:

[0092] (9)

[0093] In the formula, represents the convolution result, is the convolution kernel, and is the index of the matrix, Represents the matrix OK, Represents the matrix List; and Represent the row and column indices of the convolution kernel respectively, For distribution grid equipment The input data, Represents the element corresponding to the convolution kernel, Represents the convolution kernel OK, Represents the convolution kernel Columns.

[0094] Step 2.22, construct the activation function of the distribution network grid equipment fault detection model, introduce nonlinear transformation, and capture the abstract characteristics of the distribution network grid equipment fault, as shown in the following formula:

[0095] (10)

[0096] In the formula, represents the activation function of the distribution network grid equipment fault detection model, An abstract feature representing the captured distribution network grid equipment fault;

[0097] Step 2.23, perform a pooling operation by sampling the fault characteristics of the distribution network grid equipment and using the maximum pooling operation, as shown in the following formula:

[0098] (11)

[0099] In the formula, is the function after pooling of the distribution network equipment fault detection model, is the activation function OK, is the activation function Columns, and Respectively represent the row and column indices of the convolution kernel;

[0100] Step 2.24, perform the convolutional neural network fully connected layer operation, the weight of the fully connected layer is , the bias of the fully connected layer is , the output of the fully connected layer of the convolutional neural network is shown as follows:

[0101] (12)

[0102] In the formula, Function represents the pooled function Flatten to vector;

[0103] Step 2.25, output the final distribution network grid equipment fault detection model, as shown in the following formula:

[0104] (13)

[0105] In the formula, It is a distribution network equipment fault detection model. When the value is 1, it indicates that there is a device failure in the distribution network. A value of 0 indicates that there is no equipment failure in the distribution network grid.

[0106] In the training and construction process of the distribution network grid equipment fault detection model, we can also refer to formula (8) to minimize the loss function so that the model can be guided to adjust in the direction of more accurately predicting the target. After the above process, the distribution network grid equipment fault detection model based on the convolutional neural network is finally output: , perform fault detection on distribution network grid equipment.

[0107] Step 3, propose a distribution network grid fault monitoring method based on edge computing and convolutional neural network. In the distribution network grid fault online monitoring stage, the trained distribution network grid node fault detection model and distribution network grid equipment fault detection model are updated to the edge computing terminal of the distribution network grid. The grid edge computing terminal performs online calculation and analysis on the faults of the distribution network grid nodes and equipment, and performs fault monitoring on the distribution network grid nodes and equipment. Once a fault occurs, the corresponding treatment is carried out.

[0108] The grid edge computing terminal performs online calculation and analysis on the faults of the distribution network grid nodes and devices, and performs fault monitoring on the distribution network grid nodes and devices, as shown in the following formula:

[0109]

[0110] In the formula, Indicates the command issued by the edge computing terminal of the distribution network grid, including the nodes in the distribution network grid Fault Isolation Instructions and Equipment Fault isolation instructions; when The value is 1, indicating that the node in the distribution network grid is monitored. A failure occurs, the node Issue a fault isolation command; when The value is 1, indicating that the device in the distribution network grid is monitored. Failure occurs, the device Issue fault isolation instructions.

[0111] The introduction of convolutional neural networks and edge computing technology in the present invention can improve the feature extraction capability, data processing efficiency, and enhance the accuracy of abnormal fault detection. It is mainly manifested in the following aspects: 1) Solve the problem of complex feature extraction. The traditional method has limited ability to extract features of power grids in complex environments, while the convolutional neural network can more effectively capture and understand the complex features in the power grid by learning the features in the data. 2) It can effectively cope with the challenge of large-scale data processing. The large-scale data transmission and processing problems generated by distribution network monitoring are a challenge for traditional systems, and the parallel computing capability of convolutional neural networks helps to efficiently process huge monitoring data. 3) Improve the real-time performance of fault monitoring. Traditional monitoring systems may face limitations in real-time performance, while edge computing technology can push calculations to a location closer to the data source, reduce transmission delays, and achieve faster real-time response. 4) Reduce privacy and security risks. Data transmission in traditional monitoring systems may involve privacy and security risks, while edge computing helps to process and store data locally, reducing the transmission of sensitive information and improving privacy and security. Therefore, the present invention effectively overcomes the shortcomings of traditional monitoring methods and improves the monitoring capability of distribution network grid faults through convolutional neural networks and edge computing technology.

[0112] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A grid fault monitoring method for improving the safety of distribution network, characterized in that: The following steps are involved: Step 1: Collect monitoring data of distribution network grid faults. The monitoring data includes collecting current data, voltage data, power data, frequency data and voltage phase angle data of distribution network grid nodes, current data, voltage data, power data and temperature data of distribution network grid equipment. The monitoring data of distribution network grid faults is expressed by the following formula: （1） （2） In the formula, For collection Distribution network grid node data at all times, The distribution network grid nodes collected exist Current data at the moment, The distribution network grid nodes collected exist Voltage data at the moment, The distribution network grid nodes collected exist Power data at the moment, The distribution network grid nodes collected exist Frequency data at time, The distribution network grid nodes collected exist Voltage phase angle data at the moment; For collection All-time grid equipment data, Distribution network grid equipment for collection exist Current data at the moment, Distribution network grid equipment for collection exist Voltage data at the moment, Distribution network grid equipment for collection exist Power data at the moment, Distribution network grid equipment for collection exist Temperature data at the moment; Step 2, using the monitoring data of distribution network grid faults as training samples, constructing and training distribution network grid node fault detection models and distribution network grid equipment fault detection models based on convolutional neural networks, the input layer of the convolutional neural network inputs the collected data, and then performs convolution operations, pooling operations and full connection operations in the hidden layer, extracting fault features in the sampled distribution network nodes according to the fault samples of the nodes in the distribution network grid, and finally outputting the distribution network grid node fault detection model through the output layer, wherein, Building and training a distribution network grid node fault detection model based on a convolutional neural network includes the following steps: Step 2.11, input the training samples to the input layer of the convolutional neural network, perform convolution operation on the training samples, and the convolution operation slides the input training samples through the convolution kernel to extract the fault features in the distribution network grid nodes, as shown in the following formula: （3） In the formula, represents the convolution result, is the convolution kernel, , is the index of the matrix, Represents the matrix OK, Represents the matrix Column; m and n represent the row and column indices of the convolution kernel, respectively. For distribution network grid nodes The input data, Represents the element corresponding to the convolution kernel, Represents the convolution kernel OK, Represents the convolution kernel Columns; Step 2.12, construct the activation function of the distribution network grid node fault detection model, introduce nonlinear transformation, and capture the abstract characteristics of the distribution network grid node fault, as shown in the following formula: （4） In the formula, represents the activation function of the distribution network grid node fault detection model; Represents the abstract features of captured distribution network grid node faults; Step 2.13, perform a pooling operation by sampling the fault features of the distribution network grid nodes and using the maximum pooling operation, as shown in the following formula: （5） In the formula, is the function after pooling of the distribution network grid node fault detection model, is the activation function OK, is the activation function The columns of , m and n represent the row and column indices of the convolution kernel respectively; Step 2.14, perform the convolutional neural network fully connected layer operation, the weight of the fully connected layer is , the bias of the fully connected layer is , the output of the fully connected layer of the convolutional neural network is shown as follows: （6） In the formula, Function represents the pooled function Flatten to vector; Step 2.15, output the final distribution network grid node fault detection model, as shown in the following formula: （7） In the formula, is the distribution network grid node fault detection model, When the value is 1, it indicates that there is a node fault in the distribution network grid, and when the value is 0, it indicates that there is no node fault in the distribution network grid; In the training and construction process of the distribution network grid node fault detection model, the difference between the predicted output and the actual label of the distribution network grid node fault detection model is measured by minimizing the loss function, as shown in the following formula: （8） In the formula, represents the loss function of the distribution network grid node fault detection model, Represents the distribution network grid node fault detection model and the true value of the distribution network grid node fault detection model The difference between Computing Gradients and , and update the weights of the fully connected layer , the bias of the fully connected layer After multiple iterations, the distribution network grid node fault detection model parameters are continuously adjusted to minimize the loss function and output the distribution network grid node fault detection model: ; Building and training a distribution network grid equipment fault detection model based on a convolutional neural network includes the following steps: Step 2.21, input the training samples to the input layer of the convolutional neural network, perform convolution operation on the training samples, and the convolution operation slides the input training samples through the convolution kernel to extract the fault features in the distribution network grid equipment, as shown in the following formula: （9） In the formula, represents the convolution result, is the convolution kernel, and is the index of the matrix, Represents the matrix OK, Represents the matrix List; and Represent the row and column indices of the convolution kernel respectively, For distribution grid equipment The input data, Represents the element corresponding to the convolution kernel, Represents the convolution kernel OK, Represents the convolution kernel Columns; Step 2.22, construct the activation function of the distribution network grid equipment fault detection model, introduce nonlinear transformation, and capture the abstract characteristics of the distribution network grid equipment fault, as shown in the following formula: （10） In the formula, represents the activation function of the distribution network grid equipment fault detection model, An abstract feature representing the captured distribution network grid equipment fault; Step 2.23, perform a pooling operation by sampling the fault characteristics of the distribution network grid equipment and using the maximum pooling operation, as shown in the following formula: （11） In the formula, is the function after pooling of the distribution network equipment fault detection model, is the activation function OK, is the activation function Columns, and Respectively represent the row and column indices of the convolution kernel; Step 2.24, perform the convolutional neural network fully connected layer operation, the weight of the fully connected layer is , the bias of the fully connected layer is , the output of the fully connected layer of the convolutional neural network is shown as follows: （12） In the formula, Function represents the pooled function Flatten to vector; Step 2.25, output the final distribution network grid equipment fault detection model, as shown in the following formula: （13） In the formula, It is a distribution network equipment fault detection model. When the value is 1, it indicates that there is a device fault in the distribution network grid; when the value is 0, it indicates that there is no device fault in the distribution network grid; Step 3: Update the distribution network grid node fault detection model and the distribution network grid device fault detection model to the edge computing terminal of the distribution network grid, and perform online calculation and analysis on the faults of the distribution network grid nodes and devices at the grid edge computing terminal. The method of fault monitoring of the distribution network grid nodes and devices is as follows: ; In the formula, Indicates the command issued by the edge computing terminal of the distribution network grid, including the nodes in the distribution network grid Fault Isolation Instructions and Equipment Fault isolation instructions; is the distribution network node fault detection model. The value is 1, indicating that the node in the distribution network grid is monitored. A failure occurs, the node Issue fault isolation instructions; is the distribution network equipment fault detection model. The value is 1, indicating that the device in the distribution network grid is monitored. Failure occurs, the device Issue fault isolation instructions.

Images