Substation fault location analysis system based on BP neural network

Abstract

The application relates to the technical field of fault positioning analysis, and provides a transformer substation fault positioning analysis system based on a BP neural network. A data acquisition module acquires real-time information and historical fault information of a transformer substation, a topological feature extraction module extracts topological features of the transformer substation and parameterizes the topological features, then a feature analysis and fusion module fuses the topological features to form a comprehensive feature vector through a weighted average distribution weight method, a BP neural network module takes the comprehensive feature vector as input, constructs a multilayer neural network with a specific structure and an activation function, and outputs a fault positioning probability distribution, and a fault positioning module comprehensively determines a fault position based on the fault positioning probability distribution and a voting mechanism of a K nearest neighbor algorithm. Through extraction of topological features of the transformer substation and data fusion, the application inputs the topological features into a specific structure BP neural network to obtain a fault positioning probability distribution, and then realizes accurate fault positioning by combining a voting mechanism of a K nearest neighbor algorithm.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of fault location analysis, in particular to a transformer substation fault location analysis system based on a BP neural network. BACKGROUND

[0002] With the continuous expansion of the power system scale and the gradual improvement of the voltage level, the safe and stable operation of the transformer substation, as a key hub of power transmission and distribution, is crucial. Once a fault occurs in the transformer substation, quickly and accurately locating the fault position is of great significance for reducing power outage time, reducing economic losses and ensuring the reliable operation of the power system.

[0003] Traditional transformer substation fault location methods, such as logic reasoning based on protection and circuit breaker action information, gradually reveal limitations when facing increasingly complex transformer substation network structures and diversified fault types, such as poor fault tolerance and insufficient complex fault diagnosis capabilities.

[0004] With the development of artificial intelligence technology, the BP neural network is widely used in transformer substation fault location due to its powerful non-linear mapping ability and self-learning ability. It can establish a complex mapping relationship between fault features and fault positions through learning a large amount of historical fault data, thereby realizing intelligent diagnosis of fault positions. At present, there are many research results on transformer substation fault location based on BP neural networks, and they have been applied to some extent in some transformer substations. However, in the actual application process, the transformer substation fault location analysis system based on the BP neural network still faces many challenges.

[0005] 1. The electrical system of the transformer substation has a complex network topology, including the interconnection between busbars, circuit breakers, disconnectors, transmission lines and other devices. When a fault occurs, the fault signal will propagate and reflect in this complex network, which increases the difficulty of fault location.

[0006] 2. At the same time, the BP neural network needs to process a large amount of signals from different positions and devices, and accurately analyze the relationship between these signals and the fault position. However, the complex network topology makes the signal propagation path and fault characteristics become blurred, and the neural network may not be able to effectively extract accurate fault location information from these complex signals, leading to inaccurate fault location. SUMMARY

[0007] In order to solve the above technical problems, the application provides a substation fault positioning analysis system based on a BP neural network, which extracts the topological features of the substation and fuses the features of the substation data, inputs into a specific structure BP neural network to obtain a fault positioning probability distribution, and then combines the voting mechanism of the K nearest neighbor algorithm to accurately position the fault, so as to solve the problems in the prior art.

[0008] A substation fault positioning analysis system based on a BP neural network comprises:

[0009] A data acquisition module acquires three-phase voltage and three-phase current data by deploying voltage and current sensors in each area of the substation, acquires temperature and gas change rate data by deploying temperature and gas sensors, acquires historical fault positioning data of the substation and corresponding voltage and current data and temperature and gas data based on historical fault data of the substation, and the like.

[0010] A topological feature extraction module constructs a detailed topological graph of the substation based on the design blueprint and equipment connection information of the substation and extracts topological features for parameterization.

[0011] A feature analysis and fusion module is connected with the data acquisition module and the topological feature extraction module, extracts features from the collected voltage, current, temperature, pressure and gas data, and then fuses the topological feature data by a weighted average allocation weight method to form a comprehensive feature vector.

[0012] A BP neural network module is connected with the feature analysis and fusion module, takes the comprehensive feature vector of the feature analysis and fusion module as input, constructs a multi-layer neural network with a specific structure and an activation function, realizes accurate prediction of the fault position of the substation, adopts a weighted cross-entropy loss function and a momentum-based stochastic gradient descent algorithm in the training process, and ensures that the network will not be over-fitted when dealing with the problem of uneven sample distribution.

[0013] A fault positioning module is connected with the feature analysis and fusion module and the BP neural network module, positions and decides the fault based on the fault positioning probability distribution calculated by the BP neural network module and the voting mechanism of the K nearest neighbor algorithm, preliminarily positions the fault according to the fault position probability distribution output by the BP neural network, assists positioning by using the K nearest neighbor algorithm and performs voting, calculates a comprehensive score by combining the probability and the voting result, and determines the fault position according to the comprehensive score ranking.

[0014] An interaction module, which is in data connection with the fault location module, is configured to provide a visual interface and interaction function of the system, to display the substation topology and fault information in a three-dimensional graph, to display the characteristic space distribution, and to allow the operation and maintenance personnel to input expert judgment and adjust system parameters.

[0015] Preferably, the topology feature extraction module comprises a topology graph construction unit and a parameterization unit.

[0016] The topology graph construction unit constructs a topology graph according to the design blueprint and the device connection information of the substation, wherein the topology graph comprises various device nodes and lines connecting the devices; each edge is assigned with electrical parameters, including resistance, reactance, capacitance, admittance and length; and for any node, device characteristic parameters are added according to the device type, including bus, transformer, circuit breaker and disconnector.

[0017] The parameterization unit calculates the topology features of the nodes according to the topology graph constructed by the topology graph construction unit, including node degree, weighted degree centrality of the node, closeness centrality of the node and clustering coefficient of the node.

[0018] Preferably, the feature analysis and fusion module comprises an electrical feature analysis unit and a feature fusion unit.

[0019] The electrical feature analysis unit analyzes the collected three-phase voltage and three-phase current, and calculates the positive sequence, negative sequence and zero sequence components, as well as the total harmonic distortion rate, power factor and asymmetry degree of the voltage and current.

[0020] For temperature data and pressure data, the change rates of temperature and pressure are calculated, and for gas data, the change of gas composition is analyzed; the change rates of dissolved gas content including hydrogen, methane and ethane in oil-immersed transformers are calculated by using DGA technology, and whether there is an insulation fault is judged according to the change of different gas components.

[0021] The feature fusion unit fuses the data calculated by the topology feature extraction module and the electrical feature analysis unit into a comprehensive feature vector by using a weighted average method according to the importance of the features.

[0022] Preferably, the BP neural network module comprises a model construction unit and a model training unit.

[0023] The model construction unit is configured to build a BP neural network structure, the input layer receives the comprehensive feature vector of the feature analysis and fusion module and is normalized, three hidden layers in a diamond structure are set, different activation functions are used in each layer to mine the features, the number of neurons in the output layer corresponds to the number of historical fault positions, and the output is converted into a fault position probability distribution through linear transformation and softmax function, thereby providing a basis for fault location.

[0024] The model training unit is used to train the constructed BP neural network. To address the problem of imbalanced sample distribution, a weighted cross-entropy loss function is adopted, and the stochastic gradient descent algorithm with momentum is used to update the parameters to accelerate convergence. After the dataset is divided, the training set is trained using mini-batch gradient descent. After each round of training, the performance is evaluated and the accuracy is calculated using the validation set. An early stopping mechanism is introduced to prevent overfitting, and the learning rate is adjusted according to the performance of the validation set to improve the network's generalization ability and localization accuracy.

[0025] Preferably, the specific construction process of the model building unit is as follows:

[0026] Input layer: Receives the combined feature vector from the feature analysis and fusion module. As input;

[0027] set up The dimension is d, and the input layer has d neurons. First, for Preprocessing is performed to normalize the range, normalizing each element of the feature vector to the [0,1] interval;

[0028] Hidden layers are constructed, with three hidden layers: the first layer has n1 neurons, the second layer has n2 neurons, and the third layer has n3 neurons, where n1... <n2>n3, forming a network structure similar to a rhombus shape;

[0029] For the first layer of hidden layers, the input is x0=F, and the output h1 of the first layer of hidden layers is:

[0030] h1=f1(W1x0+b1)

[0031] where W1 is a weight matrix of d x n1, b1 is a bias vector of length n1, f1 is the activation function of the first layer of hidden layers, and an ELU activation function is used, and its formula is:

[0032]

[0033] For the second layer of hidden layers, the input is x1=h1, and the output h2 of the second layer of hidden layers is:

[0034] h2=f2(W2x1+b2)

[0035] where W2 is a weight matrix of n1 x n2, b2 is a bias vector of length n2, f2 is the activation function of the second layer of hidden layers, and a smooth nonlinear function of hyperbolic tangent function is used, and its formula is:

[0036]

[0037] For the third layer of hidden layers, the input is x2=h2, and the output h2 of the third layer of hidden layers is:

[0038] h3=f3(W3x2+b3)

[0039] where W3 is a weight matrix of n2 x n3, b3 is a bias vector of length n3, f3 is the activation function of the third layer of hidden layers, and a Swish function is used, and its formula is:

[0040] f3(x)=xσ(x)

[0041] where σ(x) is a Sigmoid function, i.e.:

[0042]

[0043] Output layer, the number of neurons k of the output layer corresponds to the number of fault positions of the transformer substation in the historical data, the input of the output layer is x3=h3, and the original output z of the output layer is obtained through linear transformation:

[0044] z=W4x3+b4

[0045] where W4 is an n3 x k weight matrix, b4 is a bias vector of length k, and softmax function is used to convert the raw output z into a probability distribution, and the final output is:

[0046]

[0047] where i and j represent different fault locations, z i is the raw output value corresponding to the i-th fault location, j is the raw output value corresponding to the j-th fault location, and P i represents the probability of the i-th fault location.

[0048] Preferably, the model training unit considers the uneven distribution of samples for different fault locations and uses a weighted cross-entropy loss function to train the model.

[0049] For the prediction result P i of the fault location, the loss function is:

[0050]

[0051] where m is the number of samples, y ri is the true label of the r-th sample at the i-th fault location, P ri is the predicted probability of the r-th sample at the i-th fault location, and ω i is the weight of the i-th fault location, which is set by the reciprocal of the number of samples at this fault location to balance the learning weights of different location samples.

[0052] The optimization algorithm uses a stochastic gradient descent algorithm with momentum to update the network parameters. For the update of the weight matrix W and the bias term b, first calculate the gradients g W and g b , and the momentum update formula is:

[0053] u W = μu W - ηg W

[0054] u b = μu b - ηg b

[0055] where u W and u b are momentum terms, μ is the momentum coefficient, which is 0.9, and η is the learning rate used to control the update speed of momentum update according to specific requirements.

[0056] Then update the parameters: W' = W + u W and b' = b + u b , the momentum term is used to accelerate the convergence, so that the network can find the optimal solution faster during training.

[0057] Preferably, when the model is trained using the weighted cross-entropy loss function, the data set needs to be divided into a training set, a validation set and a test set. For the training set, use mini-batch gradient descent to divide the training set into multiple batches;

[0058] For each training batch, perform forward propagation to calculate the output probability, calculate the loss according to the loss function, and then calculate the gradient g W and g b by back propagation, and update the network parameters using the stochastic gradient descent algorithm with momentum;

[0059] After each round of training is completed, the network performance is evaluated using the validation set, and the accuracy ACC is calculated:

[0060]

[0061] Where m a is the number of validation set samples, and argmax(P ri ) is 1 when the predicted fault location is consistent with the true location, otherwise it is 0.

[0062] Early stopping mechanism is introduced. If the performance of the validation set does not improve for multiple consecutive rounds, training is stopped in advance to prevent overfitting. At the same time, the learning rate is adjusted according to the performance of the validation set. When the performance no longer improves, the learning rate is reduced to find a better parameter space.

[0063] Preferably, the fault location module is based on the probability distribution output by the BP neural network module for fault location;

[0064] The feature vector after real-time acquisition and processing is input into the trained model to obtain the probability distribution of the fault location (P1, P2, P i ,...,P j ), and the possibility of the fault location is preliminarily determined according to the probability size.

[0065] K-Nearest Neighbor algorithm is used for auxiliary positioning;

[0066] The comprehensive feature vector H hist of the historical fault sample is stored in the feature space. For a new test sample feature vector H test , calculate its Euclidean distance o with the historical sample:

[0067]

[0068] Select c nearest neighbors. If the fault location j is in the selected c nearest neighbors, vote for each possible fault location. The voting score is:

[0069]

[0070] wherein, β j represents the actual fault location class of the jth sample in the selected c nearest neighbor samples, and δ is an indicator function, which is 1 when the condition β j = i is satisfied, and 0 otherwise;

[0071] Combining the probability and the voting result, the comprehensive score S

[0072] S i = ω p P i + ω V V i

[0073] wherein, ω p is the weight of the probability, ω V is the weight of the voting score, which can be adjusted according to the system performance and experience, and finally the fault location is ranked according to the comprehensive score S i .

[0074] Compared with the prior art, the present application has the following beneficial effects:

[0075] 1. The present application constructs a detailed substation topology graph through a topology feature extraction module, calculates node topology features, and fuses them with electrical features into a comprehensive feature vector by a feature analysis and fusion module. A fault location module determines the fault location based on the probability distribution output by the BP neural network combined with the voting mechanism. The present application realizes effective processing of the complex network topology and multi-source signals of the substation, accurately extracts fault information, and solves the problem of inaccurate positioning caused by complex topology in traditional and existing methods.

[0076] 2. The present application constructs a multi-layer neural network with a specific diamond structure and different activation function hidden layers through a BP neural network module, converts the input comprehensive feature vector into a fault location probability distribution, adopts a weighted cross-entropy loss function to balance the learning weights of different position samples during the training process, uses a momentum-based stochastic gradient descent algorithm to update the network parameters to accelerate convergence, introduces an early stopping mechanism to prevent overfitting, adjusts the learning rate according to the performance of the validation set, effectively solves the problem of uneven sample distribution, improves the network generalization ability, more accurately learns the complex mapping relationship between fault features and fault locations, and improves the fault location accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0077] Fig. 1 is a system workflow schematic diagram of the present application;

[0078] Fig. 2 is a schematic diagram of the BP neural network model construction process of the present application;

[0079] Fig. 3 is a schematic diagram of the specific fault location process of the present application. DETAILED DESCRIPTION

[0080] The embodiments of the present application will be further described in detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the present application, but cannot be used to limit the scope of the present application.

[0081] The present application provides a substation fault location analysis system based on BP neural network, comprising:

[0082] A data acquisition module, which acquires three-phase voltage and three-phase current data by deploying voltage and current sensors in each area of the substation, and acquires temperature and gas change rate data in the substation by deploying temperature and gas sensors, and acquires historical fault location data of the substation based on historical fault data of the substation, as well as corresponding voltage and current data and temperature and gas data;

[0083] A topological feature extraction module, which constructs a detailed topological graph of the substation based on the design blueprint and equipment connection information of the substation and extracts topological features for parameterization;

[0084] A feature analysis and fusion module, which is connected with the data acquisition module and the topological feature extraction module, first extracts features from the collected voltage, current, temperature, pressure and gas data, and then fuses them with the topological feature data to form a comprehensive feature vector by a weighted average allocation weight method

[0085] A BP neural network module, which is connected with the feature analysis and fusion module, takes the comprehensive feature vector of the feature analysis and fusion module as input, and realizes accurate prediction of the fault location of the substation by constructing a multi-layer neural network with a specific structure and activation function, and in the training process, adopts a weighted cross-entropy loss function and a stochastic gradient descent algorithm with momentum to ensure that the network will not overfit when dealing with uneven sample distribution problems;

[0086] A fault location module, which is connected with the feature analysis and fusion module and the BP neural network module, performs fault location and decision based on the fault location probability distribution calculated by the BP neural network module and the voting mechanism of the K-nearest neighbor algorithm, preliminarily locates the fault position according to the fault position probability distribution output by the BP neural network, assists in positioning and voting by using the K-nearest neighbor algorithm, calculates the comprehensive score by combining the probability and the voting result, and determines the fault position according to the comprehensive score ranking;

[0087] The interactive module is connected to the fault location module and provides a visual interface and interactive functions for the system. It displays the substation topology and fault information in three-dimensional graphics, shows the spatial distribution of features, and allows operation and maintenance personnel to input expert judgments and adjust system parameters.

[0088] Example 1:

[0089] like Figs. 1-3 As shown in this embodiment, a large substation in a certain region has numerous devices and a complex network topology, including multiple busbars, multiple transformers, a large number of circuit breakers and disconnectors. Traditional fault location methods are insufficient in diagnosing complex faults. To improve the accuracy and efficiency of fault location, it was decided to use a substation fault location analysis system based on a BP neural network to construct a new fault location model based on the substation's historical fault information.

[0090] The data acquisition module deploys voltage and current sensors in various areas of the substation to collect three-phase voltage and current data in real time. It also deploys temperature, pressure, and gas sensors to collect temperature, pressure, and gas change rate data from equipment such as transformers. For example, gas sensors are installed on oil-immersed transformers to monitor changes in the content of dissolved gases such as hydrogen, methane, and ethane.

[0091] Collect historical fault data of substations, including historical fault location data and corresponding electrical and non-electrical signal data, to provide samples for model training.

[0092] The topology construction unit constructs a topology diagram based on the substation design blueprint and equipment connection information, which includes various equipment nodes and lines connecting these devices; it assigns electrical parameters to each edge, including resistance, reactance, capacitance, admittance, and length; at the same time, for any node, it adds equipment characteristic parameters according to the equipment type, including busbar, transformer, circuit breaker, and disconnector.

[0093] The parameterization unit calculates the topological characteristics of the nodes based on the topological graph constructed by the topological graph construction unit, including node degree, node weighted degree centrality, node proximity centrality, and node clustering coefficient.

[0094] The electrical characteristic analysis unit analyzes the collected three-phase voltage and three-phase current, and calculates their positive sequence, negative sequence, zero sequence components, as well as the total harmonic distortion rate, power factor, and asymmetry of voltage and current.

[0095] For temperature data and pressure data, the rate of change of temperature and pressure is calculated, and for gas data, the change of gas composition is analyzed, for example, the rate of change of dissolved gas content including hydrogen, methane, ethane in oil-immersed transformer by DGA technology, whether there is an insulation fault is judged according to the change of different gas components;

[0096] The feature fusion unit fuses the data analyzed and calculated by the topological feature and electrical feature analysis unit into a comprehensive feature vector by using weighted average method according to the importance of the features

[0097] The BP neural network module includes a model construction unit and a model training unit;

[0098] The model construction unit is used to build the structure of the BP neural network, the input layer receives the comprehensive feature vector of the feature analysis and fusion module and is normalized, three hidden layers in a diamond structure are set, different activation functions are used in each layer to mine the features, the number of neurons in the output layer corresponds to the number of historical fault positions, and the output is converted into a probability distribution of fault positions through linear transformation and softmax function, thereby providing a basis for fault location;

[0099] The specific construction process of the model construction unit is as follows:

[0100] The input layer receives the comprehensive feature vector from the feature analysis and fusion module as input;

[0101] Let the dimension of d, the input layer has d neurons, first, the is preprocessed for range normalization, and each element of the feature vector is normalized to the interval [0, 1];

[0102] Hidden layer, three hidden layers are constructed, the first layer has n1 neurons, the second layer has n2 neurons, and the third layer has n3 neurons, and n1 <n2>n3, forming a network structure similar to a rhombus shape;

[0103] For the first layer of hidden layers, the input is x0=F, and the output h1 of the first layer of hidden layers is:

[0104] h1=f1(W1x0+b1)

[0105] where W1 is a d x n1 weight matrix, b1 is a bias vector of length n1, f1 is the activation function of the first layer of hidden layers, and the ELU activation function is used, and its formula is:

[0106]

[0107] For the second layer of hidden layers, the input is x1=h1, and the output h2 of the second layer of hidden layers is:

[0108] h2=f2(W2x1+b2)

[0109] where W2 is an n1 x n2 weight matrix, b2 is a bias vector of length n2, f2 is the activation function of the second layer of hidden layers, and the smooth nonlinear function hyperbolic tangent function is used, and its formula is:

[0110]

[0111] For the third layer of hidden layers, the input is x2=h2, and the output h2 of the third layer of hidden layers is:

[0112] h3=f3(W3x2+b3)

[0113] where W3 is an n2 x n3 weight matrix, b3 is a bias vector of length n3, f3 is the activation function of the third layer of hidden layers, and the Swish function is used, and its formula is:

[0114] f3(x)=xσ(x)

[0115] where σ(x) is the Sigmoid function, i.e.:

[0116]

[0117] Output layer, the number of neurons k of the output layer corresponds to the number of fault positions of the substation in the historical data, and the input of the output layer is x3=h3. The original output z of the output layer is obtained by linear transformation:

[0118] z=W4x3+b4

[0119] where W4 is an n3 x k weight matrix, b4 is a bias vector of length k, and softmax function is used to convert the original output z into a probability distribution, and the final output is:

[0120]

[0121] where i and j represent different fault locations, z i is the original output value corresponding to the i-th fault location, j is the original output value corresponding to the j-th fault location, P i represents the probability of the i-th fault location.

[0122] The model training unit takes into account the uneven distribution of samples in different fault locations, and uses a weighted cross-entropy loss function to train the model.

[0123] For the prediction result P i of the fault location, the loss function is:

[0124]

[0125] where m is the number of samples, y ri is the true label of the r-th sample in the i-th fault location, P ri is the probability predicted by the r-th sample in the i-th fault location, and ω i is the weight of the i-th fault location, which is set by the reciprocal of the number of samples in this fault location to balance the learning weights of samples in different locations.

[0126] Optimization algorithm, using stochastic gradient descent algorithm with momentum to update network parameters, for the update of weight matrix W and bias term b, first calculate the gradient g W and g b , the momentum update formula is:

[0127] u W = μu W - ηg W

[0128] u b = μu b - ηg b

[0129] where u W and u b are momentum terms, μ is the momentum coefficient, which is 0.9, and η is the learning rate used to control the update speed of momentum update according to specific requirements.

[0130] Then update the parameters: W' = W + u W and b' = b + u b , the momentum term is used to accelerate convergence, so that the network can find the optimal solution faster during training;

[0131] When training the model using the weighted cross-entropy loss function, the data set also needs to be divided into a training set, a validation set and a test set. For the training set, use mini-batch gradient descent to divide the training set into multiple batches;

[0132] For each training batch, perform forward propagation to calculate the output probability, calculate the loss according to the loss function, and then calculate the gradient g W and g b using the stochastic gradient descent algorithm with momentum to update the network parameters;

[0133] After each round of training is completed, the network performance is evaluated using the validation set, and the accuracy ACC is calculated:

[0134]

[0135] where m a is the number of validation set samples, and argmax(P ri ) is 1 when the predicted fault location is consistent with the true location, otherwise it is 0;

[0136] The early stopping mechanism is introduced, if the performance of the validation set does not improve for multiple consecutive rounds, the training is stopped in advance to prevent overfitting, and the learning rate is adjusted according to the performance of the validation set, when the performance no longer improves, the learning rate is reduced to find a better parameter space, and finally a trained BP neural network model is obtained.

[0137] A multi-layer neural network with a specific diamond structure and different activation function hidden layers is constructed through the BP neural network module, which converts the input comprehensive feature vector into a fault location probability distribution. In the training process, the weighted cross-entropy loss function is used to balance the learning weights of different position samples, the stochastic gradient descent algorithm with momentum is used to update the network parameters to accelerate convergence, and the early stopping mechanism is introduced to prevent overfitting, and the learning rate is adjusted according to the performance of the validation set;

[0138] The problem of uneven sample distribution is effectively solved and the network generalization ability is improved, which more accurately learns the complex mapping relationship between fault features and fault location, and improves the fault location accuracy.

[0139] Embodiment two:

[0140] As Figs. 1-3 As shown, in the normal operation process of the substation in Embodiment One, the monitoring system finds abnormal fluctuations in the operating parameters of some equipment in the substation on a certain day, and preliminarily judges that a fault may have occurred. At this time, the BP neural network-based substation fault location analysis system previously established needs to be used to quickly and accurately locate the fault position, so that timely maintenance measures can be taken to reduce power outage time and economic losses.

[0141] The data acquisition module is immediately started to collect voltage, current, temperature, pressure and gas data of each area of the substation in real time, while the topology feature extraction module provides the topology feature information of the current substation, and the feature analysis and fusion module performs feature extraction and fusion on the real-time collected data to form a comprehensive feature vector.

[0142] The real-time processed comprehensive feature vector is input into the trained BP neural network module, and the probability distribution of the fault position is output through forward propagation calculation of the network. For example, the probability of a fault occurring in bus A is 80%, the probability of a fault occurring in transformer B is 15%, and the probability of a fault occurring in other positions is relatively low.

[0143] The fault location module locates the fault based on the probability distribution output by the BP neural network module, and stores the comprehensive feature vectors of historical fault samples in the feature space in advance. For the current test sample feature vector, the Euclidean distance between it and the historical samples is calculated.

[0144] The real-time collected and processed feature vector is input into the trained model to obtain the probability distribution (P1, P2, P i ,...,P j ) of the fault position, and the possibility of the fault position is preliminarily determined according to the probability size.

[0145] K-Nearest Neighbor algorithm is used for auxiliary positioning;

[0146] The comprehensive feature vectors H hist of historical fault samples are stored in the feature space, and for a new test sample feature vector H test , the Euclidean distance o between it and the historical samples is calculated.

[0147]

[0148] Select c nearest neighbors, if the fault position j is in the selected c nearest neighbors, vote for each possible fault position, and the voting score is:

[0149]

[0150] Where, β j This represents the actual fault location category of the j-th sample among the selected c nearest neighbor samples, where δ is the indicator function, and δ is the value of the sample that meets the condition β. j =1 if the condition is met, otherwise 0;

[0151] Combining probability and voting results, calculate the overall score for fault location i:

[0152] S i =ω p P i +ω V V i

[0153] Where, ω p It is a weight related to probability, ω V This refers to the weighting of voting scores, which can be adjusted based on system performance and experience, ultimately based on the overall score S. i The fault locations are sorted, and the location with the highest score is selected as the fault location.

[0154] For example, if the probability weight is set to 0.6 and the voting score weight is set to 0.4, bus A has a higher probability score and a higher voting score, and its overall score is the highest among all possible fault locations.

[0155] Based on the overall scores, bus A with the highest score was selected as the fault location. The maintenance personnel quickly went to bus A for inspection and repair, and finally successfully identified and repaired the fault, restoring the substation to normal operation.

[0156] By utilizing the probability distribution of the BP neural network output, the fault location module combines a voting mechanism to determine the fault location. This achieves effective processing of complex substation network topologies and multi-source signals, accurately extracting fault information and solving the problem of inaccurate location due to topological complexity in traditional and existing methods.

[0157] The embodiments of the present invention are given for the purposes of illustration and description. Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A substation fault location analysis system based on a BP neural network, characterized in that, include: The data acquisition module deploys voltage and current sensors in various areas of the substation to collect three-phase voltage and three-phase current data, and deploys temperature and gas sensors to collect temperature and gas change rate data within the substation. Simultaneously, it collects historical fault location data of the substation based on historical fault data, as well as corresponding voltage, current, temperature, and gas data. The topology feature extraction module constructs a detailed topology diagram of the substation based on the substation's design blueprint and equipment connection information, and extracts and parameterizes the topology features. The feature analysis and fusion module, which is connected to the data acquisition module and the topology feature extraction module, first extracts features from the acquired voltage, current, temperature, pressure, and gas data. Then, it fuses these features with the topology feature data using a weighted average method to form a comprehensive feature vector. ; The BP neural network module is connected to the feature analysis and fusion module. Based on the comprehensive feature vector of the feature analysis and fusion module as input, it constructs a multi-layer neural network with a specific structure and activation function to achieve accurate prediction of substation fault location. At the same time, during the training process, a weighted cross-entropy loss function and a stochastic gradient descent algorithm with momentum are used to ensure that the network does not overfit when dealing with the problem of uneven sample distribution. The fault location module is connected to the feature analysis and fusion module and the BP neural network module. It performs fault location and decision-making based on the fault location probability distribution calculated by the BP neural network module and the voting mechanism of the K-nearest neighbor algorithm. It performs preliminary location based on the fault location probability distribution output by the BP neural network, uses the K-nearest neighbor algorithm to assist in location and voting, calculates a comprehensive score by combining the probability and voting results, and determines the fault location based on the comprehensive score. The fault location module performs fault location based on the probability distribution output by the BP neural network module; The feature vectors acquired and processed in real time are input into the trained model to obtain the probability distribution of the fault location. Based on the probability magnitude, the likelihood of the fault location is initially determined and ranked. Use the K-nearest neighbor algorithm for assisted localization; The comprehensive feature vector of historical fault samples Stored in the feature space, for a new test sample feature vector Calculate its Euclidean distance o with historical samples: Select c nearest neighbors. If the fault location j is among the c nearest neighbors, vote on each possible fault location. The voting score is as follows: in, This indicates the actual fault location category of the j-th sample among the selected c nearest neighbor samples. It is an indicator function, when it satisfies The value is 1 if the condition is met, otherwise it is 0. Combining probability and voting results, calculate the overall score for fault location i: in, It's about the weights of probability. It concerns the weighting of voting scores, which can be adjusted based on system performance and experience, ultimately based on a comprehensive score. The fault locations are sorted, and the location with the highest score is selected as the fault location. The interactive module is connected to the fault location module and provides a visual interface and interactive functions for the system. It displays the substation topology and fault information in three-dimensional graphics, shows the spatial distribution of features, and allows operation and maintenance personnel to input expert judgments and adjust system parameters.

2. The substation fault location analysis system based on BP neural network as described in claim 1, characterized in that: The topology feature extraction module includes a topology graph construction unit and a parameterization unit; The topology construction unit constructs a topology diagram based on the substation design blueprint and equipment connection information, which includes various equipment nodes and lines connecting these devices; it assigns electrical parameters to each edge, including resistance, reactance, capacitance, admittance, and length; at the same time, for any node, it adds equipment characteristic parameters according to the equipment type, including busbar, transformer, circuit breaker, and disconnector. The parameterization unit calculates the topological features of the nodes based on the topological graph constructed by the topological graph construction unit, including node degree, node weighted degree centrality, node proximity centrality, and node clustering coefficient.

3. The substation fault location analysis system based on BP neural network as described in claim 1, characterized in that: The feature analysis and fusion module includes an electrical feature analysis unit and a feature fusion unit; The electrical characteristic analysis unit analyzes the collected three-phase voltage and three-phase current, and calculates their positive sequence, negative sequence, zero sequence components, as well as the total harmonic distortion rate, power factor, and asymmetry of voltage and current. For temperature and pressure data, calculate the rate of change of temperature and pressure. For gas data, analyze the changes in gas composition. Use DGA technology to detect the content of dissolved gases in oil-immersed transformers, including hydrogen, methane, and ethane, and calculate the rate of change of the above gases. Based on the changes in different gas compositions, determine whether there is an insulation fault. The feature fusion unit uses a weighted average to fuse the data analyzed and calculated by the topology feature and electrical feature analysis units into a comprehensive feature vector, assigning weights according to the importance of the features. .

4. The substation fault location analysis system based on BP neural network as described in claim 1, characterized in that: The BP neural network module includes a model building unit and a model training unit; The model building unit is used to build a BP neural network structure. The input layer receives the comprehensive feature vector from the feature analysis and fusion module and normalizes it. Three hidden layers with a diamond structure are set. Each layer uses a different activation function to mine features. The number of neurons in the output layer corresponds to the number of historical fault locations. The output is converted into a fault location probability distribution through linear transformation and softmax function to provide a basis for fault localization. The model training unit is used to train the constructed BP neural network. To address the problem of imbalanced sample distribution, a weighted cross-entropy loss function is adopted, and the stochastic gradient descent algorithm with momentum is used to update the parameters to accelerate convergence. After the dataset is divided, the training set is trained using mini-batch gradient descent. After each round of training, the performance is evaluated and the accuracy is calculated using the validation set. An early stopping mechanism is introduced to prevent overfitting, and the learning rate is adjusted according to the performance of the validation set to improve the network's generalization ability and localization accuracy.

5. The substation fault location analysis system based on BP neural network as described in claim 4, characterized in that: The specific construction process of the model building unit is as follows: Input layer: Receives the combined feature vector from the feature analysis and fusion module. As input; set up The dimension is d, and the input layer has d neurons. First, for Preprocessing is performed to normalize the range, normalizing each element of the feature vector to the [0,1] interval; Hidden layers, construct three hidden layers, the first layer has There are 1 neuron in the second layer. There are 1 neuron, and the third layer has 1 neuron. one neuron, and This forms a network structure resembling a rhombus shape. For the first hidden layer, the input is The output of its first hidden layer for: in, It is The weight matrix, It is a length of The bias vector, The activation function for the first hidden layer is the ELU activation function, and its formula is: For the second hidden layer, the input is The output of its second hidden layer for: in, It is The weight matrix, It is a length of The bias vector, The activation function for the second hidden layer is the smooth, non-linear hyperbolic tangent function, with the following formula: For the third hidden layer, the input is The output of its third hidden layer for: in, It is The weight matrix, It is a length of The bias vector, The activation function for the third hidden layer is the Swish function, whose formula is: in, It is the Sigmoid function, that is: The output layer has a number of neurons, k, that corresponds to the number of fault locations in the substation from the historical data. The input to the output layer is... The original output z of the output layer is obtained through linear transformation: in, It is The weight matrix, It is a length of The bias vector is used to transform the original output z into a probability distribution. The softmax function is applied, and the final output is: Where i and j represent different fault locations, The original output value corresponding to the i-th fault location, The original output value corresponding to the j-th fault location, This represents the probability of the i-th fault location.

6. The substation fault location analysis system based on BP neural network as described in claim 5, characterized in that: The model training unit takes into account the uneven distribution of samples at different fault locations and uses a weighted cross-entropy loss function to train the model. Prediction results for fault location The loss function is: Where m is the number of samples. It is the true label of the r-th sample at the i-th fault location. It is the probability predicted for the r-th sample at the i-th fault location. It is the weight of the i-th fault location, which is set by the reciprocal of the number of samples at that fault location to balance the learning weights of samples at different locations; The optimized algorithm employs stochastic gradient descent with momentum to update the network parameters. For updating the weight matrix W and the bias term b, the gradient is first calculated. and The momentum update formula is: in and It is the momentum term. It is the momentum coefficient, taken as 0.

9. The learning rate is used to control the speed of momentum updates according to specific needs; Then update the parameters: and The momentum term is used to accelerate convergence, enabling the network to find the optimal solution faster during training.

7. The substation fault location analysis system based on BP neural network as described in claim 6, characterized in that: When training the model using the weighted cross-entropy loss function, the dataset also needs to be divided into training set, validation set and test set. For the training set, mini-batch gradient descent is used to divide the training set into multiple batches. For each training batch, forward propagation is performed to calculate the output probability, the loss is calculated according to the loss function, and then the gradient is calculated through backpropagation. and And update the network parameters using the stochastic gradient descent algorithm with momentum; After each round of training, the network performance is evaluated using a validation set, and the accuracy is calculated. : in It is the number of samples in the validation set. A value of 1 indicates that the predicted fault location matches the actual location; otherwise, a value of 0 indicates the fault location. An early stopping mechanism is introduced. If the performance of the validation set does not improve in multiple consecutive rounds, training is stopped early to prevent overfitting. At the same time, the learning rate is adjusted according to the performance of the validation set. When the performance no longer improves, the learning rate is reduced to find a better parameter space.

Images