A method and system for power grid fault identification based on fusion neural network model

Abstract

The present invention discloses a method and system for identifying power grid faults based on a fusion neural network model, the method comprising: obtaining several power parameters representing the power grid operation status in the power grid line to be identified as characteristic values ​​and performing preprocessing; inputting the preprocessed characteristic values ​​into a pre-established fusion neural network model to obtain the fault probability of the power grid line to be identified. In view of the fact that power grid fault identification is easily affected by signal transmission noise, the present invention proposes a method and system for identifying power grid faults based on a fusion neural network model, utilizing the advantages of artificial intelligence, giving full play to the role of power grid operation big data, mining the correlation between operation data and fault characteristics, and performing comprehensive diagnosis of the power grid through multi-dimensional information, which can effectively reduce the influence of signal transmission noise or error on the accuracy of fault identification and improve the safety of power grid operation.

Description

Technical Field

[0001] The present invention belongs to the technical field of power grid operation and control, and in particular relates to a power grid fault identification method and system taking into account noise of measurement and signal transmission devices. Background Art

[0002] The power grid usually operates in a safe and stable state, but it is also constantly subjected to various types of disturbances. When a large disturbance occurs in the power grid, it is easy to cause a power grid failure. Improper fault handling will cause huge and irreversible economic and social losses, and even harm lives. Therefore, timely detection and handling of power grid failures is particularly important. Traditional fault alarm devices analyze the wiring method of the power grid and the connection relationship with the circuit breaker to determine the status and operation of the circuit breaker and analyze the fault.

[0003] When a power grid fails, a large amount of fault information will flow into the dispatching control center. Due to the influence of signal transmission noise, it is easy to misjudge the fault situation, resulting in untimely and inappropriate fault handling, causing more serious chain failures and causing huge losses. Summary of the invention

[0004] The purpose of the present invention is to provide a power grid fault identification method and system based on a fusion neural network model to solve the technical problem in the prior art that when a power grid fault occurs, a large amount of fault information will flow into the dispatching control center, and due to the influence of signal transmission noise, it is easy to cause misjudgment of the fault situation.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a power grid fault identification method based on a fusion neural network model, comprising:

[0007] Acquire several power parameters representing the operation status of the power grid in the power grid line to be identified as characteristic values ​​and perform preprocessing;

[0008] The preprocessed characteristic values ​​are input into the pre-obtained fusion neural network model to obtain the fault probability of the power grid line to be identified;

[0009] Among them, the fusion neural network model integrates a one-dimensional convolutional neural network and an LSTM neural network; the fusion neural network model is obtained by learning the fault operation data of the power grid.

[0010] A further improvement of the present invention is that in the step of obtaining the characteristic value of the power grid line to be identified for fault and performing preprocessing, the power grid line to be identified for fault is a power grid line with a topological structure.

[0011] A further improvement of the present invention is that the step of obtaining the characteristic value of the power grid line to be identified for fault and performing preprocessing specifically includes:

[0012] A tree search method is used to number nodes of the power grid lines to be identified for faults in the topological structure;

[0013] The characteristic values ​​include at least three of the voltage amplitude at the beginning of the line, the voltage amplitude at the end of the line, the injection power at the beginning of the line, the injection power at the end of the line, the active power of the line, and the reactive power of the line;

[0014] The beginning of the line is the end with the smallest node number, and the end of the line is the end with the largest node number.

[0015] A further improvement of the present invention is that in the step of obtaining the characteristic value of the power grid line to be identified and preprocessing, the line active power P ij The calculation method is:

[0016]

[0017] Line reactive power Q ij The calculation method is:

[0018]

[0019] Among them, U i is the voltage amplitude at the beginning of the line, U j is the voltage amplitude at the end of the line, B ij is the line susceptance, G ij is the line conductance, y ij0 is the admittance to ground, θ ij is the phase angle difference between the voltage at the first and the last nodes.

[0020] A further improvement of the present invention is that in the step of inputting the preprocessed eigenvalues ​​into a pre-established fusion neural network model to obtain the fault probability of the power grid line to be identified for fault, the preprocessing is specifically to convert the acquired eigenvalues ​​of the power grid line to be identified for fault into a one-dimensional vector format.

[0021] A further improvement of the present invention is that in the step of inputting the preprocessed eigenvalues ​​into a pre-established fusion neural network model to obtain the fault probability of the power grid line to be identified, the pre-established fusion neural network model includes a layer of one-dimensional convolutional neural network, a layer of pooling layer, two layers of LSTM neural network and three layers of fully connected layers connected in sequence.

[0022] A further improvement of the present invention is that in the step of inputting the preprocessed eigenvalues ​​into the pre-established fusion neural network model to obtain the fault probability of the power grid line to be identified, the number of convolution kernels of the one-dimensional convolutional neural network is equal to the number of eigenvalues ​​M; the size of each convolution kernel is a, and the value range of a is The step size is l, which is greater than or equal to the convolution kernel size a.

[0023] A further improvement of the present invention is that: in the step of inputting the preprocessed eigenvalues ​​into a pre-established fusion neural network model to obtain the fault probability of the power grid line to be identified, a three-layer fully connected layer is used as the output layer, the number of output units in the last output layer is the same as the number of fault types, and a sigmoid function is used as an activation function, and each output is classified into a (0, 1) interval to indicate the probability of occurrence of each type of fault.

[0024] In a second aspect, the present invention provides a power grid fault identification system based on a fusion neural network model, comprising:

[0025] A data acquisition module, used to obtain characteristic values ​​of the power grid line to be fault identified and perform preprocessing;

[0026] The fault identification module is used to input the preprocessed characteristic values ​​into the pre-established fusion neural network model to obtain the fault probability of the power grid line to be fault identified.

[0027] In the third aspect, the present invention provides a power grid fault identification system based on a fusion neural network model, which is applied to a power system, comprising: a processor and a memory coupled to the processor, the memory storing a computer program, and the computer program, when executed by the processor, implements the method steps of a power grid fault identification method based on a fusion neural network model.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] In view of the fact that power grid fault identification is easily affected by signal transmission noise, the present invention proposes a power grid fault identification method and system based on a fusion neural network model, which utilizes the advantages of artificial intelligence, gives full play to the role of power grid operation big data, mines the correlation between operation data and fault characteristics, and conducts comprehensive diagnosis of the power grid through multi-dimensional information, which can effectively reduce the influence of signal transmission noise or error on fault identification accuracy and improve the safety of power grid operation.

[0030] The fault identification model based on the fusion of one-dimensional convolutional neural network and LSTM neural network proposed in the present invention can effectively improve the accuracy of fault line identification. Moreover, since the present invention uses multi-dimensional operation data of power grid operation, it can effectively reduce the influence of signal transmission noise or error on the fault identification accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. 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 accompanying drawings:

[0032] Figure 1 The network topology diagram after numbering;

[0033] Figure 2 It is the flow chart of the fusion model fault identification method;

[0034] Figure 3 It is a flow chart of a fault identification method based on a fusion neural network model of the present invention;

[0035] Figure 4 This is a schematic diagram of a fault identification system based on a fusion neural network model of the present invention. DETAILED DESCRIPTION

[0036] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present invention can be combined with each other without conflict.

[0037] The following detailed description is an exemplary description, which is intended to provide further detailed description of the present invention. Unless otherwise specified, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art to which the present invention belongs. The terms used in the present invention are only for describing specific embodiments, and are not intended to limit the exemplary embodiments according to the present invention.

[0038] The fault identification method based on the fusion neural network model proposed in the present invention mainly integrates a one-dimensional convolutional neural network and an LSTM neural network to learn the fault operation data of the power grid, explore the correlation between the operation data and the fault characteristics, and perform comprehensive diagnosis of the power grid through multi-dimensional information, which can effectively reduce the impact of signal transmission noise or errors on fault identification.

[0039] Example 1

[0040] The present invention provides a fault identification method based on a fusion neural network model, comprising:

[0041] First, prepare the training data required for the fusion model, and number and sort the power grid lines for fault identification. Select relevant eigenvalues ​​as data training objects, normalize the eigenvalue data, and divide the processed data into training set and verification set to facilitate the training and accuracy verification of the fusion neural network after training.

[0042] Then, the fusion model structure is designed to adapt to the characteristic data of the power grid.

[0043] Finally, the training set data is input into the model that integrates the one-dimensional convolutional neural network and the LSTM neural network. After the model is trained, the parameters are saved, and the network performance is verified using the validation set data. The model finally outputs the fault line number.

[0044] The specific technical solutions provided by the present invention are as follows:

[0045] S1. Preparation of fusion model training data

[0046] The grid operation characteristic data is the basis for fault identification using the fusion model. Therefore, the network data is first processed to make it more suitable for the neural network fusion model used for fault identification.

[0047] S11. Use a tree search method to number power grid lines with a certain topological structure. The specific steps are as follows: select the branch connected to node 1, number it, select the branch connected to node 2 to continue numbering, and if it is repeated with the previous branch, remove the branch. Repeat the above operation until the branch corresponding to the last node is numbered.

[0048] S12, select line characteristic parameters. When a power grid fails, the network current will be redistributed, and the various network operating parameters will also change. It can be said that the operating parameters of the network correspond one-to-one to the operating state of the network. There are many network parameters. Selecting appropriate network operation characteristic parameters can effectively reduce hardware requirements and save training time. The characteristic values ​​of the fault operation data of the power grid selected by the present invention include: the voltage amplitude at the beginning of the line (corresponding to the end with a smaller node number), the voltage amplitude at the end of the line (corresponding to the end with a larger node number), the injected power at the beginning of the line, the injected power at the end of the line, the active power of the line and the reactive power of the line.

[0049] Among them, the line active power P ij The calculation method is:

[0050]

[0051] Line reactive power Q ij The calculation method is:

[0052]

[0053] Among them, U i is the voltage amplitude at the beginning of the line, U j is the voltage amplitude at the end of the line, B ij is the line susceptance, G ij is the line conductance, yij0 is the admittance to ground, θ ij is the phase angle difference between the voltage at the first and the last nodes.

[0054] S13, fault data set generation and recording. The number of real power system faults is relatively small, and it is difficult to meet the amount of data required for fusion model training. Therefore, it is necessary to use simulation software to simulate the power grid fault situation, which can not only meet the number requirements of training samples, but also cover different fault types. The specific process is: set a branch to fail, randomly change the generator output and load size in the network, record the eigenvalues ​​of each branch, and sort them according to the line number. In the present invention, each simulation will form an N×M matrix. Here N is the number of branches, and M is the number of selected eigenvalues. Change the fault location, generator output and load size to generate multiple samples. Perform k simulations to form k N×M matrix data samples.

[0055] S14, data normalization. Due to the different data types, their size distribution is also relatively scattered. In order to improve the training efficiency and effect of the fusion model, the data samples are normalized to the maximum and minimum. The formula is as follows:

[0056]

[0057] in, x is the data to be processed, x max is the maximum value in the data set, x min is the minimum value in the data set.

[0058] S15, training set and validation set data division. The processed data is divided into two parts according to the ratio, one part is the training set data, which is used to train the network, and the other part is the validation set data, which is used to verify the performance of the trained neural network.

[0059] S2. Design of Fusion Neural Network Model

[0060] The fusion forms of various neural networks can usually be divided into series type and parallel type. The present invention adopts a series fusion method. Its structure includes a one-dimensional convolutional neural network, a pooling layer, two LSTM neural networks, and finally three fully connected layers as the output layer. The specific situation is as follows:

[0061] S21. Since the data processed by the one-dimensional convolutional neural network is in one-dimensional vector format, it is necessary to first convert the N×M dimensional data matrix into (N×M)×1 dimensional vector data as the network input. Then the processed data is input into the one-dimensional convolutional network.

[0062] S22. Set t convolution kernels to extract data features. The size of each convolution kernel is a, and the value range of a is It can retain more data features, with a step size of l, which is greater than or equal to the convolution kernel size.

[0063] S23. After convolution, the output data is pooled, and the maximum pooling operation is selected.

[0064] S24, then input the data into a two-layer LSTM neural network. The LSTM neural network has a memory function and can learn the topological connection relationship of the branch to improve the effect of fault identification.

[0065] S25. Finally, the three-layer fully connected layer is used as the output layer. The number of output units in the final output layer is the same as the number of fault types. The sigmoid function is used as the activation function, and each output is classified into the (0, 1) interval to indicate the probability of each type of fault. The type of fault with the highest probability indicates which type of fault has occurred in the power grid, and finally the line number with the highest fault probability is output. The calculation formula is as follows:

[0066]

[0067] Among them, S(x) represents the sigmoid function operation, x The variable that is input to the final output layer for sigmoid calculation.

[0068] S3. Training and evaluation of neural network models

[0069] According to the data that has been divided into training set and validation set, the training set is input into the built fusion neural network according to step S2 to train the fusion neural network model. The validation set data is input into the neural network to verify the accuracy. The mean square error RMSE is used as the evaluation index. The smaller the RMSE, the smaller the error and the better the performance of the fusion neural network. The calculation formula is as follows:

[0070]

[0071] Among them, n is the number of samples in the validation set, is the actual value of the i-th validation set data, y i The output value obtained by inputting the fusion neural network into the i-th validation set data.

[0072] When the mean square error RMSE reaches the set value, the neural network model training ends.

[0073] S4, obtaining characteristic values ​​of the power grid line to be identified for fault and performing preprocessing; inputting the preprocessed characteristic values ​​into the fusion neural network model evaluated in step S3 to obtain the fault probability of the power grid line to be identified for fault.

[0074] Example 2

[0075] The following is an example of the fault identification process using a 9-node system:

[0076] S1. Preparation of fusion model training data

[0077] The grid operation characteristic data is the basis for fault identification using the fusion model. Therefore, the network data is first processed to make it more suitable for the neural network fusion model used for fault identification.

[0078] S11. Use tree search to number the topological power grid lines. The specific steps are as follows: select the branch connected to node 1, number it, select the branch connected to node 2 to continue numbering, and if it is repeated with the previous branch, remove the branch. Repeat the above operation until the branch corresponding to the last node is numbered. The final line numbering result is as shown in the attached figure. Figure 1 shown.

[0079] S12, select line characteristic parameters. When a power grid fails, the network flow will be redistributed and the network operation parameters will also change.

[0080] S13, fault data set generation and recording. The number of real power system faults is relatively small, and it is difficult to meet the amount of data required for fusion model training. Therefore, the present invention simulates the power grid fault situation with the help of simulation software, which can not only meet the number requirements of training samples, but also cover different fault types. The specific process is: set a branch to fail, randomly change the generator output and load size in the network, record the eigenvalues ​​of each branch, and sort them according to the line number. In the present invention, each simulation will form a 9×6 matrix. Here 9 is the number of branches, and 6 is the number of selected eigenvalues. Change the fault location, generator output and load size to generate multiple samples. Perform 50,000 simulations to form 50,000 9×6 matrix data samples.

[0081] S14, data normalization. Due to the different data types, their size distribution is also relatively scattered. In order to improve the training efficiency and effect of the fusion model, the data samples are normalized to the maximum and minimum. The formula is as follows:

[0082]

[0083] in, x is the data to be processed, x max is the maximum value in the data set, x min is the minimum value in the data set.

[0084] S15, training set and validation set data division. The processed data is divided into 80% training set data for network training and 20% validation set data for validation of the trained neural network performance.

[0085] S2. Design of Fusion Neural Network Model

[0086] The fusion form of the neural network model in this embodiment adopts a serial fusion mode. Its structure includes a one-dimensional convolutional neural network, a pooling layer, two LSTM neural networks, and finally three fully connected layers as the output layer. The specific situation is as follows:

[0087] S21. Data conversion: Since the data processed by the one-dimensional convolutional neural network is in one-dimensional vector format, it is necessary to first convert the 9×6-dimensional data matrix into (9×6)×1-dimensional vector data as the network input. Then the processed data is input into the one-dimensional convolutional network.

[0088] S22. Set 6 convolution kernels in total to extract data features. The size of each convolution kernel is 2 and the step size is 2.

[0089] S23. After convolution, the output data is pooled, and the maximum pooling operation is selected.

[0090] S24, then input the data into a two-layer LSTM neural network. The LSTM neural network has a memory function and can learn the topological connection relationship of the branch to improve the effect of fault identification.

[0091] S25. Finally, the three-layer fully connected layer is used as the output layer. The number of output units in the final output layer is the same as the number of fault types. In this example, there are 9 lines that may fail, and the number of output units in the output layer is 9. The output unit uses the sigmoid function as the activation function, and classifies each output into the (0, 1) interval, indicating the probability of each type of fault. The type of fault with the highest probability indicates which type of fault has occurred in the power grid.

[0092]

[0093] S3. Training and evaluation of neural network models

[0094] According to the data that has been divided into training set and validation set, the training set is input into the built fusion neural network according to step S2 to train the fusion neural network model. The validation set data is input into the neural network to verify the accuracy. The mean square error RMSE is used as the evaluation index. The smaller the RMSE, the smaller the error and the better the performance of the fusion neural network. The calculation formula is as follows:

[0095]

[0096] S4, obtaining characteristic values ​​of the power grid line to be identified for fault and performing preprocessing; inputting the preprocessed characteristic values ​​into the fusion neural network model evaluated in step S3 to obtain the fault probability of the power grid line to be identified for fault.

[0097] Example 3

[0098] See also Figure 3 As shown, a 9-node system is used as an example to illustrate the fault identification process. A fault identification method based on a fusion neural network model includes:

[0099] 1) Obtain the characteristic values ​​of the power grid line to be identified and perform preprocessing:

[0100] The network operation parameters correspond to the network operation status one by one. The voltage amplitude at the beginning of the line (corresponding to the end with a smaller node number), the voltage amplitude at the end of the line (corresponding to the end with a larger node number), the injected power at the beginning of the line, the injected power at the end of the line, the active power of the line, and the reactive power of the line are selected as the characteristic values.

[0101] Among them, the line active power P ij The calculation method is:

[0102]

[0103] Line reactive power Q ij The calculation method is:

[0104]

[0105] Among them, U i is the voltage amplitude at the beginning of the line, U j is the voltage amplitude at the end of the line, B ij is the line susceptance, G ij is the line conductance, y ij0 is the admittance to ground, θ ij is the phase angle difference between the voltage at the first and the last nodes.

[0106] Normalization of eigenvalue data: Perform maximum and minimum normalization on the data samples; the formula is as follows:

[0107]

[0108] in, x is the data to be processed, x max is the maximum value in the data set, x min is the minimum value in the data set.

[0109] 2) Input the preprocessed eigenvalues ​​into the pre-established fusion neural network model to obtain the fault probability of the power grid line to be identified.

[0110] The steps of pre-establishing the fusion neural network model include:

[0111] S1. Preparation of fusion model training data

[0112] The grid operation characteristic data is the basis for fault identification using the fusion model. Therefore, the network data is first processed to make it more suitable for the neural network fusion model used for fault identification.

[0113] S11. Use tree search to number the topological power grid lines. The specific steps are as follows: select the branch connected to node 1, number it, select the branch connected to node 2 to continue numbering, and if it is repeated with the previous branch, remove the branch. Repeat the above operation until the branch corresponding to the last node is numbered. The final line numbering result is as shown in the attached figure. Figure 1 shown.

[0114] S12, select line characteristic parameters. When a power grid fails, the network flow will be redistributed and the network operation parameters will also change.

[0115] S13, fault data set generation and recording. The number of real power system faults is relatively small, and it is difficult to meet the amount of data required for fusion model training. Therefore, the present invention simulates the power grid fault situation with the help of simulation software, which can not only meet the number requirements of training samples, but also cover different fault types. The specific process is: set a branch to fail, randomly change the generator output and load size in the network, record the eigenvalues ​​of each branch, and sort them according to the line number. In the present invention, each simulation will form a 9×6 matrix. Here 9 is the number of branches, and 6 is the number of selected eigenvalues. Change the fault location, generator output and load size to generate multiple samples. Perform 50,000 simulations to form 50,000 9×6 matrix data samples.

[0116] S14, data normalization. Due to the different data types, their size distribution is also relatively scattered. In order to improve the training efficiency and effect of the fusion model, the data samples are normalized to the maximum and minimum. The formula is as follows:

[0117]

[0118] Among them, x is the data to be processed, x max is the maximum value in the data set, x min is the minimum value in the data set.

[0119] S15, training set and validation set data division. The processed data is divided into 80% training set data for network training and 20% validation set data for validation of the trained neural network performance.

[0120] S2. Design of Fusion Neural Network Model

[0121] The fusion form of the neural network model in this embodiment adopts a serial fusion mode. Its structure includes a one-dimensional convolutional neural network, a pooling layer, two LSTM neural networks, and finally three fully connected layers as the output layer. The specific situation is as follows:

[0122] S21. Data conversion: Since the data processed by the one-dimensional convolutional neural network is in one-dimensional vector format, it is necessary to first convert the 9×6-dimensional data matrix into (9×6)×1-dimensional vector data as the network input. Then the processed data is input into the one-dimensional convolutional network.

[0123] S22. Set 6 convolution kernels in total to extract data features. The size of each convolution kernel is 2 and the step size is 2.

[0124] S23. After convolution, the output data is pooled, and the maximum pooling operation is selected.

[0125] S24, then input the data into a two-layer LSTM neural network. The LSTM neural network has a memory function and can learn the topological connection relationship of the branch to improve the effect of fault identification.

[0126] S25. Finally, the three-layer fully connected layer is used as the output layer. The number of output units in the final output layer is the same as the number of fault types. In this example, there are 9 lines that may fail, and the number of output units in the output layer is 9. The output unit uses the sigmoid function as the activation function, and classifies each output into the (0, 1) interval, indicating the probability of each type of fault. The type of fault with the highest probability indicates which type of fault has occurred in the power grid.

[0127]

[0128] S3. Training and evaluation of neural network models

[0129] According to the data that has been divided into training set and validation set, the training set is input into the built fusion neural network according to step S2 to train the fusion neural network model. The validation set data is input into the neural network to verify the accuracy. The mean square error RMSE is used as the evaluation index. The smaller the RMSE, the smaller the error and the better the performance of the fusion neural network. The calculation formula is as follows:

[0130]

[0131] Example 4

[0132] See also Figure 4 As shown, the present invention provides a power grid fault identification system based on a fusion neural network model, comprising:

[0133] A data acquisition module, used to obtain characteristic values ​​of the power grid line to be fault identified and perform preprocessing;

[0134] The fault identification module is used to input the preprocessed characteristic values ​​into the pre-established fusion neural network model to obtain the fault probability of the power grid line to be fault identified.

[0135] Example 5

[0136] The present invention provides a power grid fault identification system based on a fusion neural network model, which is applied to a power system, comprising: a processor and a memory coupled to the processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method steps described in one of embodiments 1-3 are implemented.

[0137] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0138] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0139] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A power grid fault identification method based on a fusion neural network model, characterized in that: include: Acquire several power parameters representing the operation status of the power grid in the power grid line to be identified as characteristic values ​​and perform preprocessing; The preprocessed characteristic values ​​are input into the pre-obtained fusion neural network model to obtain the fault probability of the power grid line to be identified; The fusion neural network model integrates a one-dimensional convolutional neural network and an LSTM neural network; the fusion neural network model is obtained by learning the fault operation data of the power grid; In the step of obtaining characteristic values ​​of the power grid line to be identified for fault and performing preprocessing, the power grid line to be identified for fault is a power grid line with a topological structure; The step of obtaining the characteristic value of the power grid line to be identified for fault and performing preprocessing specifically includes: A tree search method is used to number nodes of the power grid lines to be identified for faults in the topological structure; The characteristic values ​​include at least three of the voltage amplitude at the beginning of the line, the voltage amplitude at the end of the line, the injection power at the beginning of the line, the injection power at the end of the line, the active power of the line, and the reactive power of the line; The beginning of the line is the end with the smallest node number, and the end of the line is the end with the largest node number; In the step of inputting the preprocessed feature values ​​into a pre-established fusion neural network model to obtain the fault probability of the power grid line to be identified, the pre-established fusion neural network model includes a one-dimensional convolutional neural network layer, a pooling layer, two LSTM neural network layers and three fully connected layers connected in sequence; In the step of obtaining the characteristic value of the power grid line to be identified and preprocessing, the line active power P ij The calculation method is: Line reactive power Q ij The calculation method is: Among them, U i is the voltage amplitude at the beginning of the line, U j is the voltage amplitude at the end of the line, B ij is the line susceptance, G ij is the line conductance, y ij0 is the admittance to ground, θ ij is the phase angle difference of the voltage at the first and last nodes; In the step of inputting the preprocessed eigenvalues ​​into the pre-established fusion neural network model to obtain the fault probability of the power grid line to be identified, the number of convolution kernels of the one-dimensional convolutional neural network is equal to the number of eigenvalues ​​M; the size of each convolution kernel is a, and the value range of a is The step size is l, which is greater than or equal to the convolution kernel size a.

2. A method for identifying power grid faults based on a fusion neural network model according to claim 1, characterized in that: In the step of inputting the preprocessed eigenvalues ​​into a pre-established fusion neural network model to obtain the fault probability of the power grid line to be identified for fault, the preprocessing is specifically to convert the acquired eigenvalues ​​of the power grid line to be identified for fault into a one-dimensional vector format.

3. A method for identifying power grid faults based on a fusion neural network model according to claim 1, characterized in that: In the step of inputting the preprocessed eigenvalues ​​into a pre-established fusion neural network model to obtain the fault probability of the power grid line to be identified, three fully connected layers are used as the output layer, the number of output units in the last output layer is the same as the number of fault types, and the sigmoid function is used as the activation function, and each output is classified into the (0, 1) interval, indicating the probability of occurrence of each type of fault.

4. A power grid fault identification system based on a fusion neural network model, characterized in that: include: A data acquisition module, used to obtain characteristic values ​​of the power grid line to be fault identified and perform preprocessing; A fault identification module is used to input the preprocessed characteristic values ​​into a pre-established fusion neural network model to obtain the fault probability of the power grid line to be fault identified; The fusion neural network model integrates a one-dimensional convolutional neural network and an LSTM neural network; the fusion neural network model is obtained by learning the fault operation data of the power grid; In the step of obtaining characteristic values ​​of the power grid line to be identified for fault and performing preprocessing, the power grid line to be identified for fault is a power grid line with a topological structure; The step of obtaining the characteristic value of the power grid line to be identified for fault and performing preprocessing specifically includes: A tree search method is used to number nodes of the power grid lines to be identified for faults in the topological structure; The characteristic values ​​include at least three of the voltage amplitude at the beginning of the line, the voltage amplitude at the end of the line, the injection power at the beginning of the line, the injection power at the end of the line, the active power of the line, and the reactive power of the line; The beginning of the line is the end with the smallest node number, and the end of the line is the end with the largest node number; In the step of inputting the preprocessed feature values ​​into a pre-established fusion neural network model to obtain the fault probability of the power grid line to be identified, the pre-established fusion neural network model includes a one-dimensional convolutional neural network layer, a pooling layer, two LSTM neural network layers and three fully connected layers connected in sequence; In the step of obtaining the characteristic value of the power grid line to be identified and preprocessing, the line active power P ij The calculation method is: Line reactive power Q ij The calculation method is: Among them, U i is the voltage amplitude at the beginning of the line, U j is the voltage amplitude at the end of the line, B ij is the line susceptance, G ij is the line conductance, y ij0 is the admittance to ground, θ ij is the phase angle difference of the voltage at the first and last nodes; In the step of inputting the preprocessed eigenvalues ​​into the pre-established fusion neural network model to obtain the fault probability of the power grid line to be identified, the number of convolution kernels of the one-dimensional convolutional neural network is equal to the number of eigenvalues ​​M; the size of each convolution kernel is a, and the value range of a is The step size is l, which is greater than or equal to the convolution kernel size a.

5. A power grid fault identification system based on a fusion neural network model, applied to a power system, characterized in that: include: A processor and a memory coupled to the processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method steps according to any one of claims 1 to 3 are implemented.

Images