Power System Fault Location Discrimination Method, Device, Terminal Equipment and Storage Medium

Abstract

The present invention belongs to the technical field of power system fault detection, and discloses a method, device, terminal device and storage medium for discriminating the fault location of a power system. The method includes: constructing a system simulation model of the power system to be detected, and obtaining a fault data set; performing data preprocessing on the fault data set to obtain a power training data set and a power test data set; constructing an initial convolutional neural network model, and training the initial convolutional neural network model according to the power training data set to obtain a convolutional neural network model to be tested; inputting the power test data set into the convolutional neural network model to be tested for testing, and when the test result meets the preset test conditions, using the convolutional neural network model to be tested as a fault test model; discriminating the fault location of the power system to be detected according to the fault test model. The present invention uses the convolutional neural network model to quickly judge the fault location in the power system through the above method.

Description

Technical Field

[0001] The present invention relates to the technical field of power system fault detection, and particularly to a method, device, terminal device and storage medium for discriminating the fault location in a power system. Background Art

[0002] The safe and stable operation of the power grid is of great significance to the national economy. Nowadays, the scale and coverage of the distribution network are constantly increasing, and its operation complexity is also increasing. Since the distribution network is approaching its maximum load during operation, the frequency and risk of faults are also rising day by day. Circuit faults in the distribution network seriously affect the stability and safety of system operation, causing huge losses to social production safety and further affecting the daily life and work of the people. If the fault location can be quickly determined when a fault occurs, and the faulty line segment is isolated for troubleshooting, losses can be stopped in time and the rest of the normal part of the system can be prevented from being affected, which can significantly shorten the fault recovery time, reduce economic losses and improve the reliability of the power system. Therefore, a fast, reliable and accurate fault location determination method is very important.

[0003] The existing fault location determination methods are roughly divided into two categories: The first category is the traditional fault location determination method, which can be mainly divided into the impedance method, the injection method and the traveling wave method. It mainly considers physical parameters such as voltage, node impedance, and traveling wave signals in the power grid, but the calculation is relatively cumbersome. In recent years, distribution lines have become more and more complex and have more and more branches, resulting in lower efficiency in fault location determination and positioning. The second category is the automatic fault location determination method, mainly including matrix algorithms, bat algorithms, genetic algorithms, particle swarm algorithms, neural network algorithms, etc. These algorithms take into account different structural characteristics and fault conditions of the distribution network, and have certain improvements in fault tolerance and accuracy. However, there are still limitations, and it is easy to fall into the local optimum situation. At the same time, how to construct appropriate objective functions and switching functions is the bottleneck of the solution model.

[0004] The above content is only used to assist in understanding the technical solution of the present invention, and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main object of the present invention is to provide a method, device, terminal device and storage medium for discriminating the fault location in a power system, aiming to solve the technical problem of low efficiency in discriminating the fault location in the power system in the prior art.

[0006] To achieve the above object, the present invention provides a method for discriminating the fault location in a power system, and the method includes the following steps:

[0007] Construct a system simulation model of the power system to be detected, and obtain the fault data set of the power system to be detected;

[0008] Preprocess the fault data set, and obtain a power training data set and a power test data set according to the processing results;

[0009] Construct an initial convolutional neural network model, and train the initial convolutional neural network model according to the power training data set to obtain a convolutional neural network model to be tested;

[0010] Input the power test data set into the convolutional neural network model to be tested for testing. When the test results meet the preset test conditions, use the convolutional neural network model to be tested as the fault test model corresponding to the power system to be detected;

[0011] Discriminate the fault location of the power system to be detected according to the fault test model.

[0012] Optionally, the steps of constructing a system simulation model of the power system to be detected and obtaining the fault data set of the power system to be detected specifically include:

[0013] Obtain the system type of the power system to be detected, and establish a system simulation model according to the system type;

[0014] Obtain a preset fault information set, and perform fault simulation on the system simulation model according to the preset fault information set;

[0015] Detect the fault data generated during the fault simulation to obtain a fault data set.

[0016] Optionally, the steps of preprocessing the fault data set and obtaining a power training data set and a power test data set according to the processing results specifically include:

[0017] Obtain abnormal data and normal data in the fault data set according to the preset numerical boundary, and perform numerical correction on the abnormal data to obtain corrected data;

[0018] Perform data normalization processing on the corrected data and the normal data to obtain the preprocessing result of the fault data set;

[0019] Perform set partitioning on the preprocessing result of the fault data set according to the preset set ratio to obtain a power training data set and a power test data set.

[0020] Optionally, the steps of constructing an initial convolutional neural network model and training the initial convolutional neural network model according to the power training data set to obtain a convolutional neural network model to be tested specifically include:

[0021] Construct a one-dimensional convolutional neural network model as the initial convolutional neural network model, and input the power training data set into the initial convolutional neural network model for training according to a preset optimization objective to obtain a convolutional neural network model to be tested.

[0022] Optionally, the preset optimization objective is to minimize the cross-entropy loss function.

[0023] Optionally, the step of determining the fault location of the power system to be detected according to the fault test model specifically includes:

[0024] Obtain the current operation data of the power system to be detected, and input the current operation data into the fault detection model;

[0025] When the output result indicates a fault, determine the current fault location of the power system to be detected according to the fault detection model and the system simulation model.

[0026] Optionally, after the step of constructing the initial convolutional neural network model and training the initial convolutional neural network model according to the power training data set to obtain a convolutional neural network model to be tested, the method further includes:

[0027] Input the power test data set into the convolutional neural network model to be tested for testing. When the test result does not meet the preset test conditions, perform optimization training on the convolutional neural network model to be tested.

[0028] In addition, to achieve the above object, the present invention also proposes a device for discriminating the fault location of a power system. The device includes:

[0029] A first model construction module, configured to construct a system simulation model of the power system to be detected and obtain a fault data set of the power system to be detected;

[0030] A data processing module, configured to perform data preprocessing on the fault data set and obtain a power training data set and a power test data set according to the processing result;

[0031] A second model construction module, configured to construct an initial convolutional neural network model and train the initial convolutional neural network model according to the power training data set to obtain a convolutional neural network model to be tested;

[0032] A model testing module, configured to input the power test data set into the convolutional neural network model to be tested for testing. When the test result meets the preset test conditions, use the convolutional neural network model to be tested as the fault test model corresponding to the power system to be detected;

[0033] A fault detection module, configured to determine the fault location of the power system to be detected according to the fault test model.

[0034] In addition, to achieve the above object, the present invention further provides a terminal device, which includes a memory, a processor, and a power system fault location discrimination program stored on the memory and executable on the processor. The power system fault location discrimination program is configured to implement the steps of the power system fault location discrimination method as described above.

[0035] In addition, to achieve the above object, the present invention further provides a storage medium, on which a power system fault location discrimination program is stored. When the power system fault location discrimination program is executed by a processor, it implements the steps of the power system fault location discrimination method as described above.

[0036] The present invention constructs a system simulation model of the power system to be detected, and obtains a fault data set of the power system to be detected; performs data preprocessing on the fault data set, and obtains a power training data set and a power test data set according to the processing results; constructs an initial convolutional neural network model, and trains the initial convolutional neural network model according to the power training data set to obtain a convolutional neural network model to be tested; inputs the power test data set into the convolutional neural network model to be tested for testing. When the test result meets the preset test conditions, the convolutional neural network model to be tested is used as the fault test model corresponding to the power system to be detected; determines the fault location of the power system to be detected according to the fault test model. Based on deep learning technology, the present invention can effectively determine the location of faults in the power system by establishing a model of the power system and training a detection model according to the power system data, and can greatly improve the efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on the structures shown in these drawings without creative efforts.

[0038] Figure 1 is a schematic structural diagram of a terminal device in the hardware operating environment related to the embodiment solution of the present invention;

[0039] Figure 2 is a schematic flowchart of the first embodiment of the power system fault location discrimination method of the present invention;

[0040] Figure 3Schematic diagram of the convolutional neural network for the method of discriminating the fault location in the power system of the present invention;

[0041] Figure 4 Flow schematic diagram of the second embodiment of the method for discriminating the fault location in the power system of the present invention;

[0042] Figure 5 Structural block diagram of the first embodiment of the device for discriminating the fault location in the power system of the present invention.

[0043] The realization, functional characteristics and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Detailed implementation manners

[0044] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0045] Refer to Figure 1 , Figure 1 Schematic diagram of the terminal device of the hardware operating environment involved in the embodiment solution of the present invention.

[0046] As Figure 1 shown, the terminal device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard). Optionally, the user interface 1003 may further include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (WI-FI) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM) or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0047] Those skilled in the art can understand that Figure 1 the structure shown in

[0048] does not constitute a limitation on the terminal device and may include more or fewer components than shown in the figure, or combine some components, or have different component arrangements. Figure 1As shown in the figure, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a power system fault location discrimination program.

[0049] In Figure 1 In the terminal device shown in the figure, the network interface 1004 is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 in the terminal device of the present invention may be arranged in the terminal device. The terminal device calls the power system fault location discrimination program stored in the memory 1005 through the processor 1001 and executes the power system fault location discrimination method provided by the embodiments of the present invention.

[0050] An embodiment of the present invention provides a power system fault location discrimination method. Referring to Figure 2 , Figure 2 is a schematic flowchart of the first embodiment of a power system fault location discrimination method of the present invention.

[0051] In this embodiment, the power system fault location discrimination method includes the following steps:

[0052] Step S10: Construct a system simulation model of the power system to be detected, and obtain a fault data set of the power system to be detected.

[0053] It should be noted that the system simulation model can be constructed by using the visualization simulation tool Simulink, and system data of the power system to be detected is obtained, including but not limited to: branch impedance, load of each node, tie switch parameters, etc., to construct the power system to be detected. The fault data set includes various parameters of the power system to be detected during a fault. The various parameters may be historical fault parameters of the power system to be detected or fault parameters obtained by performing fault simulation through the system simulation model.

[0054] Further, step S10 specifically includes: obtaining the system type of the power system to be detected, and establishing a system simulation model according to the system type; obtaining a preset fault information set, and performing fault simulation on the system simulation model according to the preset fault information set; detecting the fault data generated during the fault simulation to obtain a fault data set.

[0055] This embodiment is described by taking the system type of the power system to be detected as the IEEE standard 33-node system as an example (in specific implementation, this method can also be used in power systems of other system types, and this embodiment does not limit the actual application scenarios of the present invention). In specific implementation, for example: use MATLAB / Simulink to build an IEEE 33-node distribution network model, and configure components for the standard 33-node distribution network through modules such as Simscape, Sources, and Sinks in Simulink.

[0056] Further, to obtain a fault data set, based on the above example: add a monitoring point module between preset nodes in the simulation model of the IEEE 33-node system. The above nodes can be between nodes 8 and 9; set four random value model parameters including short-circuit type, short-circuit occurrence location, short-circuit occurrence time (initial phase), and short-circuit resistance value, and perform a simulation of a short-circuit fault; further, to obtain more fault data, an Asynchronous Machine module can also be added, and the motor starting from a standstill is used to cause a voltage sag to simulate a load impact fault; by collecting circuit data at the monitoring point, including voltage, current, and power data of each node, as well as data of the branch where the fault is located, etc., the above data are all added to the fault data set.

[0057] Step S20: Perform data preprocessing on the fault data set, and obtain a power training data set and a power test data set according to the processing results;

[0058] It is easy to understand that in order to perform effective model training, some abnormal data in the fault data need to be deleted, so data preprocessing is performed on the fault data set. Process data abnormal points: During the simulation of the circuit, it is possible that the recorded circuit data is abnormal due to environmental and operation problems. These data points are unreasonable, and if not processed, it will lead to deviations in subsequent results.

[0059] Step S20 specifically includes: obtaining abnormal data and normal data in the fault data set according to a preset numerical boundary, and performing numerical correction on the abnormal data to obtain corrected data; performing data normalization processing on the corrected data and the normal data to obtain a preprocessing result of the fault data set; performing set division on the preprocessing result of the fault data set according to a preset set ratio to obtain a power training data set and a power test data set.

[0060] In specific implementation, the preset numerical boundary can be implemented through box plot analysis, that is, the values greater than or less than the upper and lower bounds set by the box plot are abnormal values, and then the identified abnormal values are corrected with the average value of their upper and lower neighboring points to obtain corrected data.

[0061] Furthermore, the data is normalized. Since the types of input parameters are different and the numerical differences are large, the main solution to this type of problem is to normalize the input parameter data. The most commonly used method for normalizing data is min-max normalization, also known as deviation normalization, which is a linear transformation of the original data that maps the result values to between [0,1]. The formula is as follows:

[0062]

[0063] where X max is the maximum value of the single-column feature data, and X min is the minimum value of the single-column feature data.

[0064] Furthermore, the preset ratio can be 7:3. Specifically, in implementation, it can be 700 sets of power training data and 300 sets of power test data.

[0065] Step S30: Construct an initial convolutional neural network model, and train the initial convolutional neural network model according to the power training data set to obtain a convolutional neural network model to be tested;

[0066] Step S30 specifically includes: constructing a one-dimensional convolutional neural network model as the initial convolutional neural network model, and inputting the power training data set into the initial convolutional neural network model for training according to a preset optimization objective to obtain a convolutional neural network model to be tested.

[0067] It should be noted that a one-dimensional CNN convolutional neural network model is built, the input layer, hidden layer, and output layer of the neural network are determined, the number of neurons in the input layer corresponds to the input parameter features, and the network weights are initialized. The one-dimensional CNN convolutional neural network designed by this method has a total of 9 layers, including 1 input layer, 4 convolutional layers, 2 pooling layers, 1 dropout layer, and 1 fully connected layer output layer. The output layer is the location of the fault predicted by the neural network;

[0068] It should be noted that referring to Figure 3 , Figure 3 is a schematic diagram of the convolutional neural network of the power system fault location discrimination method of the present invention. The preset optimization objective is to minimize the cross-entropy loss function. The training data set is input into the 1D CNN neural network model for training. With minimizing the cross-entropy loss function as the optimization objective, the Adam optimizer is used to update and adjust the model parameters to reduce the prediction error, and the test set is used for testing to obtain an ideal prediction model. The ideal network model architecture and network parameters are saved, and the cross-entropy loss function loss and accuracy accuracy are used as evaluation indicators.

[0069]

[0070] Among them, M represents the number of categories, and y ic represents the sign function (0 or 1). If the true category of sample i is equal to c, then y ic takes 1, otherwise takes 0, and p ic represents the predicted probability that sample i belongs to category c.

[0071]

[0072] Among them, TP represents the number of correctly predicted positive examples, FP represents the number of incorrectly predicted negative examples, TN represents the number of correctly predicted negative examples, and FN represents the number of incorrectly predicted positive examples.

[0073] It should be noted that the specific network structure parameters are set according to actual needs and will not be explained one by one in this embodiment. The meaning of each layer of the convolutional neural network model to be tested is as follows: (1) Input layer: A data collection interval of 0.00001 seconds, only collecting data during the fault occurrence period, with a total of 13,000 sampling points. After data preprocessing, each group of data contains the voltage values Ua, Ub, Uc of three phases and the current values Ia, Ib, Ic of three phases. In the neural network, the data is flattened into a 13,000×6 vector and then input into the neural network.

[0074] (2) The first 1D CNN layer: The first layer defines a filter with a convolutional kernel size of 100 and a stride of 1. In order to learn more data features, 100 filters are defined for multi-dimensional extraction, and 100 different characteristics can be trained in the first layer of the network.

[0075] (3) The second 1D CNN layer: The output result of the first CNN will be input into the second CNN layer. 50 different filters are defined for training on this network layer, and the other parameters are the same as those of the first 1D CNN layer.

[0076] (4) Max pooling layer: In order to reduce the complexity of the output and prevent overfitting of the data, a pooling layer is used after the CNN layer. The pooling layer uses a 3×1 window and a stride of 1.

[0077] (5) The third 1D CNN layer: In order to learn higher-level features, a one-dimensional convolutional network is continued to be used for feature extraction here. The convolutional kernel size is 50×1, the stride is 1, the number of convolutional kernels is 160, and the activation function uses Relu.

[0078] (6) The fourth 1D CNN layer: The parameters are the same as those of the third 1D CNN layer, and feature extraction is continued.

[0079] (7) Average pooling layer: Add one more pooling layer to further avoid overfitting. Average pooling takes the average of two weights in the neural network. The size of the output matrix is 1×160. Each feature detector only has one weight left in this layer of the neural network.

[0080] (8) Dropout layer: Here, a ratio of 0.5 is selected, and the other parameters are the same as those of the first Dropout layer.

[0081] (9) Fully connected layer with Softmax activation: Since there are 2 classes to be predicted (i.e., "upstream" and "downstream"), the last layer reduces the vector of length 160 to a vector of length 2. Softmax is used as the activation function. It forces the sum of the 2 output values of the neural network to be 1, and the output values will represent the probabilities of each of the 2 classes occurring. The class with the higher probability is the predicted classification result.

[0082] Step S40: Input the power test data set into the convolutional neural network model to be tested. When the test result meets the preset test conditions, regard the convolutional neural network model to be tested as the fault test model corresponding to the power system to be detected;

[0083] It should be noted that after the neural network model is trained, the effects of the training set and the validation set are evaluated respectively. Further illustration based on the above embodiments, for example: the cross-entropy loss function loss of the training set is 0.0821, and the cross-entropy loss function loss of the validation set is 0.0026. Both values are very low and close to 0. And the accuracy rates both reach 1. The results of these two data sets show that the constructed one-dimensional convolutional neural network model has a very high prediction accuracy and quite excellent predictability, and it can be used as a fault test model. During the training process of this embodiment, the hyperparameters can be set as: batch size batch_size = 10, number of epochs epochs = 20, learning rate learning_rate = 0.00001.

[0084] Step S50: Determine the fault location of the power system to be detected according to the fault test model.

[0085] Step S50 specifically includes: Obtain the current operation data of the power system to be detected, and input the current operation data into the fault detection model; when the output result indicates a fault, determine the current fault location of the power system to be detected according to the fault detection model and the system simulation model.

[0086] In specific implementation, when performing fault detection, a pre-trained fault detection model is loaded, and the current operating data is input into the fault detection model to obtain an output result. When the output result indicates a fault, the type of device and the parameter that caused the fault are analyzed based on the output result, and the fault location can be quickly determined in combination with the system simulation model.

[0087] Through the above method, the present invention, based on deep learning technology, can effectively determine the location of faults in the power system by establishing a model for the power system and training a detection model according to the power system data, and can greatly improve the efficiency and accuracy.

[0088] Reference Figure 4 , Figure 4 is a schematic flowchart of the second embodiment of a method for discriminating the fault location of a power system according to the present invention. Based on the above first embodiment, after step S30 of the method for discriminating the fault location of the power system in this embodiment, the following steps are further included:

[0089] Step S60: Input the power test data set into the to-be-tested convolutional neural network model for testing. When the test result does not meet the preset test conditions, optimize and train the to-be-tested convolutional neural network model.

[0090] Based on the first embodiment, if neither the cross-entropy loss function loss of the training set nor the cross-entropy loss function loss of the validation set is close to 0, and the accuracy rate does not reach 1. Then the to-be-tested convolutional neural network model should be continuously trained to optimize the model and improve the test accuracy rate.

[0091] Refer to Figure 5 , Figure 5 is a structural block diagram of the first embodiment of a device for discriminating the fault location of a power system according to the present invention.

[0092] As Figure 5 shown, the device includes: a first model construction module 10, configured to construct a system simulation model of the to-be-detected power system and obtain a fault data set of the to-be-detected power system;

[0093] It should be noted that the system simulation model can be constructed through a visualization simulation tool Simulink to obtain the system data of the to-be-detected power system, including but not limited to: branch impedance, load of each node, tie switch parameters, etc., to construct the to-be-detected power system. The fault data set includes various parameters of the to-be-detected power system during a fault, and the various parameters can be historical fault parameters of the to-be-detected power system or fault parameters obtained through fault simulation of the system simulation model.

[0094] Further, the first model construction module 10 is specifically configured to obtain the system type of the power system to be detected, and establish a system simulation model according to the system type; obtain a preset fault information set, and perform fault simulation on the system simulation model according to the preset fault information set; detect the fault data generated during the fault simulation process to obtain a fault data set.

[0095] In this embodiment, the system type of the power system to be detected is taken as an example of the IEEE standard 33-node system for illustration (in specific implementation, this method can also be used in power systems of other system types, and this embodiment does not limit the actual application scenarios of the present invention). In specific implementation, for example: use MATLAB / Simulink to build an IEEE 33-node distribution network model, and use modules such as Simscape, Sources, and Sinks in Simulink to configure components according to the standard 33-node distribution network.

[0096] Further, to obtain a fault data set, based on the above example: add a monitoring point module between preset nodes in the simulation model of the IEEE 33-node system, and the above nodes can be between nodes 8 and 9; set four random value model parameters of short-circuit type, short-circuit occurrence location, short-circuit occurrence time (initial phase), and short-circuit resistance value, and perform simulation of a short-circuit fault; further, to obtain more fault data, an Asynchronous Machine asynchronous motor module can also be added, and the motor is started from a standstill to cause a voltage sag to simulate a load impact fault; by collecting the circuit data at the monitoring point, including the voltage, current, and power data of each node, and the data of the branch where the fault is located, etc., the above data are all added to the fault data set.

[0097] The data processing module 20 is configured to perform data preprocessing on the fault data set, and obtain a power training data set and a power test data set according to the processing result;

[0098] It is easy to understand that in order to perform effective model training, some abnormal data in the fault data need to be deleted, so data preprocessing is performed on the fault data set. Process data abnormal points: During the simulation of the circuit, it is possible that the recorded circuit data is abnormal due to environmental and operation problems, and these data points are unreasonable. If not processed, it will lead to deviations in subsequent results.

[0099] The data processing module 20 is specifically configured to obtain abnormal data and normal data in the fault data set according to preset numerical boundaries, and perform numerical correction on the abnormal data to obtain corrected data; perform data normalization processing on the corrected data and the normal data to obtain a preprocessing result of the fault data set; perform set partitioning on the preprocessing result of the fault data set according to a preset set ratio to obtain a power training data set and a power test data set.

[0100] In specific implementation, the preset numerical boundary can be analyzed through a box plot, that is, the values greater than or less than the upper and lower bounds set by the box plot are abnormal values, and then the identified abnormal values are corrected with the average value of their upper and lower neighbor points to obtain corrected data.

[0101] Furthermore, for data normalization, since the types of input parameters are different and the numerical differences are large, the main solution to this problem is to perform normalization processing on the input parameter data. The most commonly used method for data normalization is min-max normalization, also known as deviation normalization, which is a linear transformation of the original data to map the result values to the range [0,1]. The formula is:

[0102]

[0103] where X max is the maximum value of a single-column feature data, and X min is the minimum value of a single-column feature data.

[0104] Furthermore, the preset ratio can be 7:3. In specific implementation, it can be 700 groups for the power training data set and 300 groups for the power test data set.

[0105] The second model construction module 30 is used to construct an initial convolutional neural network model, and train the initial convolutional neural network model according to the power training data set to obtain a convolutional neural network model to be tested;

[0106] The second model construction module 30 is specifically configured to construct a one-dimensional convolutional neural network model as the initial convolutional neural network model, input the power training data set into the initial convolutional neural network model and train it according to a preset optimization objective to obtain a convolutional neural network model to be tested.

[0107] It should be noted that a one-dimensional CNN convolutional neural network model is built, the input layer, hidden layer and output layer of the neural network are determined, the number of neurons in the input layer corresponds to the input parameter features, and the network weights are initialized. The one-dimensional CNN convolutional neural network designed by this method has a total of 9 layers, including 1 input layer, 4 convolutional layers, 2 pooling layers, 1 dropout layer, and 1 fully connected layer output layer. The output layer is the orientation of the fault predicted by the neural network.

[0108] It should be noted that referring to Figure 3 , Figure 3 is a schematic diagram of the convolutional neural network of the power system fault orientation discrimination method of the present invention, and the preset optimization goal is to minimize the cross-entropy loss function. The training data set is input into the 1D CNN neural network model for training. With the minimization of the cross-entropy loss function as the optimization goal, the Adam optimizer is used to update and adjust the model parameters to reduce the prediction error, and the test set is used for testing to obtain an ideal prediction model. The ideal network model architecture and network parameters are saved, and the cross-entropy loss function loss and accuracy accuracy are used as evaluation indicators.

[0109]

[0110] Among them, M represents the number of categories, and y ic represents the sign function (0 or 1). If the true category of sample i is equal to c, then y ic takes 1, otherwise takes 0, and p ic represents the predicted probability that sample i belongs to category c.

[0111]

[0112] Among them, TP represents the number of correctly predicted positive examples, FP represents the number of incorrectly predicted negative examples, TN represents the number of correctly predicted negative examples, and FN represents the number of incorrectly predicted positive examples.

[0113] It should be noted that the specific network structure parameters are set according to actual needs and will not be explained one by one in this embodiment. The meaning of each layer of the convolutional neural network model to be tested is as follows: (1) Input layer: One acquisition interval is 0.00001 seconds, and only the data during the fault occurrence period is acquired. There are a total of 13,000 sampling points. After the data is preprocessed, each group of data contains the voltage values Ua, Ub, Uc of three phases and the current values Ia, Ib, Ic of three phases. In the neural network, the data is flattened into a 13,000×6 vector and then input into the neural network.

[0114] (2) The first 1D CNN layer: The first layer defines filters with a convolutional kernel size of 100 and a stride of 1. To learn more data features, 100 filters are defined for multi-dimensional extraction, and 100 different characteristics can be trained in the first layer of the network.

[0115] (3) The second 1D CNN layer: The output result of the first CNN will be input into the second CNN layer. 50 different filters are defined for training on this network layer, and the remaining parameters are the same as those of the first 1D CNN layer.

[0116] (4) The max pooling layer: To reduce the complexity of the output and prevent overfitting of the data, a pooling layer is used after the CNN layer. The pooling layer uses a 3×1 window with a stride of 1.

[0117] (5) The third 1D CNN layer: To learn higher-level features, a one-dimensional convolutional network is continued here for feature extraction. The convolutional kernel size is 50×1, the stride is 1, the number of convolutional kernels is 160, and the activation function uses Relu.

[0118] (6) The fourth 1D CNN layer: The parameters are the same as those of the third 1D CNN layer, and feature extraction continues.

[0119] (7) The average pooling layer: An additional pooling layer is added to further avoid overfitting. The average pooling takes the average of two weights in the neural network. The size of the output matrix is 1×160. Each feature detector only has one weight left in this layer of the neural network.

[0120] (8) The Dropout layer: A ratio of 0.5 is selected here, and the remaining parameters are the same as those of the first Dropout layer.

[0121] (9) The fully connected layer with Softmax activation: Since there are 2 classes to be predicted (i.e., "upstream" and "downstream"), the last layer will reduce the vector of length 160 to a vector of length 2. Softmax is used as the activation function. It forces the sum of the 2 output values of the neural network to be one, and the output values will represent the probabilities of each of the 2 classes occurring. The class with the higher probability is the predicted classification result.

[0122] The model testing module 40 is used to input the power test data set into the convolutional neural network model to be tested for testing. When the test result meets the preset test conditions, the convolutional neural network model to be tested is used as the fault test model corresponding to the power system to be detected;

[0123] It should be noted that after the neural network model is trained, the effects of the training set and the validation set are evaluated respectively. Further illustration is based on the above embodiments. For example, the cross-entropy loss function loss of the training set is 0.0821, and the cross-entropy loss function loss of the validation set is 0.0026. Both values are very low and close to 0. And the accuracy rates both reach 1. The results of these two data sets show that the constructed one-dimensional convolutional neural network model has a very high prediction accuracy and quite excellent predictability, and it can be used as a fault test model. During the training process of this embodiment, the hyperparameters can be set as follows: batch size batch_size = 10, number of epochs epochs = 20, learning rate learning_rate = 0.00001.

[0124] The fault detection module 50 is used to determine the fault location of the power system to be detected according to the fault test model.

[0125] The fault detection module 50 is specifically configured to obtain the current operation data of the power system to be detected and input the current operation data into the fault detection model; when the output result indicates a fault, determine the current fault location of the power system to be detected according to the fault detection model and the system simulation model.

[0126] In specific implementation, when performing fault detection, load the pre-trained fault detection model, input the current operation data into the fault detection model, and obtain the output result. When the output result indicates a fault, analyze what kind of device and what kind of parameters cause the fault according to the output result, and combine the system simulation model to quickly determine the fault location.

[0127] Through the above device, the present invention, based on deep learning technology, can effectively determine the location of faults in the power system by establishing a model for the power system and training a detection model according to the power system data, and can greatly improve the efficiency and accuracy.

[0128] In addition, an embodiment of the present invention also proposes a storage medium, on which a power system fault location discrimination program is stored, and the power system fault location discrimination program is executed by a processor to perform the steps of the power system fault location discrimination method as described above.

[0129] Since this storage medium adopts all the technical solutions of the above all embodiments, it at least has all the beneficial effects brought by the technical solutions of the above embodiments, and will not be elaborated here one by one.

[0130] It should be understood that the above is only for illustration and does not constitute any limitation to the technical solution of the present invention. In specific applications, those skilled in the art can set according to needs, and the present invention does not make any restrictions on this.

[0131] It should be noted that the workflow described above is only illustrative and does not limit the protection scope of the present invention. In actual applications, those skilled in the art can select some or all of them according to actual needs to achieve the purpose of the solution of this embodiment, and no limitation is made here.

[0132] In addition, for the technical details not described in detail in this embodiment, reference can be made to the power system fault location discrimination method provided in any embodiment of the present invention, and details are not described here again.

[0133] In addition, it should be noted that in this article, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or system including that element.

[0134] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0135] Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as a read-only memory (ROM) / RAM, magnetic disk, optical disk), and includes several instructions for causing a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present invention.

[0136] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied to other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A method for discriminating the fault location of a power system, characterized in that, the method includes: Obtain the system type of the power system to be detected, and establish a system simulation model according to the system type; Obtain a preset fault information set, and perform fault simulation on the system simulation model according to the preset fault information set, where the preset fault information set includes the following types of fault information: short-circuit type, short-circuit occurrence location, short-circuit occurrence time, short-circuit resistance value, load impact fault obtained based on motor static start, voltage, current, power data of preset nodes, and branch data where the fault is located; Detect the fault data generated during the fault simulation to obtain a fault data set; Obtain abnormal data and normal data in the fault data set according to preset numerical boundaries, and perform numerical correction on the abnormal data to obtain corrected data; Perform data normalization processing on the corrected data and the normal data to obtain a preprocessing result of the fault data set; Perform set partitioning on the preprocessing result of the fault data set according to a preset set ratio to obtain a power training data set and a power test data set; Construct a one-dimensional convolutional neural network model as an initial convolutional neural network model, input the power training data set into the initial convolutional neural network model and train it according to a preset optimization objective to obtain a convolutional neural network model to be tested, where the preset optimization objective is to minimize the cross-entropy loss function, and the Adam optimizer is used to update and adjust the model parameters during the training process; Input the power test data set into the convolutional neural network model to be tested for testing. When the test result meets the preset test conditions, use the convolutional neural network model to be tested as the fault test model corresponding to the power system to be detected; Discriminate the fault location of the power system to be detected according to the fault test model.

2. The method for discriminating the fault location of a power system according to claim 1, characterized in that, the step of discriminating the fault location of the power system to be detected according to the fault test model specifically includes: Obtain the current operating data of the power system to be detected, and input the current operating data into the fault test model; When the output result indicates a fault, determine the current fault location of the power system to be detected according to the fault test model and the system simulation model.

3. The method for discriminating the fault location of a power system according to claim 2, characterized in that, after the steps of constructing an initial convolutional neural network model and training the initial convolutional neural network model according to the power training data set to obtain a convolutional neural network model to be tested, it further includes: Input the power test data set into the convolutional neural network model to be tested for testing. When the test result does not meet the preset test conditions, perform optimization training on the convolutional neural network model to be tested.

4. A device for discriminating the fault location of a power system, characterized in that, the device includes: The first model construction module is used to obtain the system type of the power system to be detected and establish a system simulation model according to the system type; Obtain a preset fault information set, and perform fault simulation on the system simulation model according to the preset fault information set. The preset fault information set includes the following types of fault information: short-circuit type, short-circuit occurrence location, short-circuit occurrence time, short-circuit resistance value, load impact fault obtained based on motor static start, voltage, current, power data of preset nodes, and branch data where the fault is located; Detect the fault data generated during the fault simulation process to obtain a fault data set; The data processing module is used to obtain abnormal data and normal data in the fault data set according to preset numerical boundaries, and perform numerical correction on the abnormal data to obtain corrected data; Perform data normalization processing on the corrected data and the normal data to obtain a preprocessing result of the fault data set; Perform set division on the preprocessing result of the fault data set according to a preset set ratio to obtain a power training data set and a power test data set; The second model construction module is used to construct a one-dimensional convolutional neural network model as an initial convolutional neural network model, and input the power training data set into the initial convolutional neural network model for training according to a preset optimization target to obtain a convolutional neural network model to be tested. The preset optimization target is to minimize the cross-entropy loss function, and the Adam optimizer is used to update and adjust the model parameters during the training process; The model testing module is used to input the power test data set into the convolutional neural network model to be tested for testing. When the test result meets the preset test conditions, the convolutional neural network model to be tested is used as the fault test model corresponding to the power system to be detected; The fault detection module is used to determine the fault location of the power system to be detected according to the fault test model.

5. A terminal device, Characterized in that, The terminal device includes: a memory, a processor, and a power system fault location discrimination program stored on the memory and executable on the processor. The power system fault location discrimination program is configured to implement the steps of the power system fault location discrimination method according to any one of claims 1 to 3.

6. A storage medium, Characterized in that, A power system fault location discrimination program is stored on the storage medium. When the power system fault location discrimination program is executed by a processor, it implements the steps of the power system fault location discrimination method according to any one of claims 1 to 3.

Images