Power distribution network fault identification method based on whale optimization algorithm LSSVM

Abstract

The invention relates to a power distribution network fault identification method based on a whale optimization algorithm LSSVM, and belongs to the technical field of power system fault diagnosis, the method uses an improved TS-GAN model to solve the problem of insufficient fault samples, and provides sufficient data support for subsequent diagnosis; the fault starting moment is efficiently determined by using a break variable monitoring algorithm, and complex mathematical transformation and model establishment do not need to be carried out; according to the DBO-VMD algorithm, the optimization efficiency and precision of key parameters are improved, complex fault signals can be effectively decomposed, IMF with abundant features can be separated out, envelope demodulation analysis is carried out by selecting a maximum kurtosis IMF component, and features such as envelope information are extracted; according to the WOA-LSSVM algorithm, kernel parameters and penalty factors of the LSSVM are optimized, local extreme values are avoided, various short-circuit faults of the power distribution network are recognized, more accurate and more comprehensive decision support is provided for operation and maintenance of the power distribution network, and finally the operation efficiency and reliability of the power distribution network are improved.

Description

Technical Field

[0001] This invention belongs to the field of power system fault diagnosis technology, and relates to a method for identifying distribution network faults based on the whale optimization algorithm LSSVM. Background Technology

[0002] Fault diagnosis technology for power distribution networks in complex mountainous areas at high altitudes is an important research direction in the field of power grid operation and maintenance. Mountainous power distribution networks face complex environments, often encountering severe natural conditions such as low temperatures, strong winds, snow, moisture erosion, and lightning strikes. These factors accelerate equipment aging, leading to frequent faults that are difficult to locate quickly. Furthermore, the complex terrain and insufficient communication network coverage in mountainous areas further increase the difficulty of fault diagnosis. Traditional fault diagnosis methods based on a single data source mainly rely on electrical signal acquisition and analysis. However, these methods have poor applicability to the diverse fault types encountered in complex environments and cannot meet the requirements for high real-time performance and high accuracy.

[0003] As a crucial link in the power system that directly interacts with users, the distribution network's operational status directly impacts power quality and safety. It also serves as a hub for transmission and distribution coordination and load regulation. A continuous and stable power supply is a fundamental prerequisite for all economic activities in modern society. Therefore, achieving efficient fault detection, location, isolation, and self-healing has become a key operational and maintenance aspect for ensuring the safe operation of the distribution network. In actual power grid operation, it is difficult to capture samples of progressive faults caused by ordinary faults, resulting in a small sample size and a lack of relevant data. Furthermore, accurate fault identification is crucial for ensuring the safe and stable operation of the power grid. Currently used signal decomposition methods suffer from low accuracy and classification model inaccuracies. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a method for identifying distribution network faults based on the whale optimization algorithm LSSVM.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for fault identification in distribution networks based on the whale optimization algorithm LSSVM includes the following steps: S1: Fault recording data is generated based on the improved TS-GAN network model; S2: Identify the time of fault occurrence based on mutation monitoring algorithm; S3: Decomposing fault signals based on the DOB-VMD network model; S4: Identify distribution network faults based on the WOA-LSSVM network model.

[0006] Furthermore, the improved TS-GAN network model described in step S1 incorporates the global attention mechanism GLFM into the traditional TS-GAN structure model, combines residual connections, and fuses the information captured by each layer; it introduces conditional WGAN to optimize and adjust the initial synthetic image generated by standard WGAN; it captures the conditional dependency relationship between the current electrical quantity state and historical data, and optimizes the generation model to make the approximate conditional distribution approximate the real conditional distribution.

[0007] Furthermore, the calculation steps of the Global Attention Mechanism (GLFM) are as follows: First, calculate the query matrix. and The key matrix To obtain the similarity, then divide by and through softmax The function normalizes the fraction matrix to obtain the fraction matrix. :

[0008] Then, the fraction matrix AND-value matrix Multiply by this to obtain the attention matrix. :

[0009] Input to the attention matrix The output expression of this layer is obtained after linear transformation as follows. i =1, 2, ..., s ;

[0010] in These are learnable network weight parameters; The computation process of a convolutional self-attention layer is as follows: First, the input from the previous layer is... X The whole is divided into multiple local blocks Then, one-dimensional convolution operations are used to learn each local block. of Q , K , V ,in:

[0011] In the formula Represents the convolution kernel ( p , q Weights at positions; k Indicates the kernel size; Indicates the size of the input / output channels; Indicates the feature location on the local block ( i ,j The characteristic tensor of ); calculate Q and K The dot product, then divided by the key vector K Square root of dimension ,use softmax The function is normalized and compared with the value matrix. V Multiplying them together yields the attention matrix between the slices and the slices. :

[0012] Convolutional self-attention layers use the input sequence h Convolution operations are performed on groups of distinct convolutional kernels, and their attention matrices are computed. ,Will Z The final output of this layer is obtained after linear transformation. Z : .

[0013] Furthermore, the generation of fault recording data based on the improved TS-GAN network model described in step S1 includes the following steps: In the data generation process, the generator first generates an image by combining a random noise vector with a text feature vector; then the generated image is continuously improved during the training process, gradually approaching the real image; the discriminator's function is to determine whether the input image comes from a real sample or is an image generated by the generator; The network structure includes: an input layer that receives real and generated images and combines them with the input text features; a convolutional layer that extracts image features through multiple convolutions and uses activation functions to avoid gradient vanishing; feature fusion that combines text features with temporal features; weighted fusion of the two through concatenation or attention mechanisms; a fully connected layer that outputs the final judgment result after a series of convolutional and pooling layers; and finally, an output layer that outputs a probability value indicating that the sample is a real sample. The training process of TS-GAN is adversarial, with the generator and discriminator being trained alternately in two stages. For a given electrical quantity sequence The model focuses on capturing the current state. Compared with historical data Conditional dependency According to probability model theory, the joint distribution is decomposed into: ;Model m ( φ The optimization objective of this generation is to approximate the conditional distribution. At any time Both approximate the true conditional distribution Fault generation model m ( φ )satisfy:

[0014] The implicit coding space in the TS-GAN framework is used to design and characterize the conditional distribution characteristics of fault sequences, providing the time-series information modeling capability for the generative model; the modules are connected by multi-level gated cyclic units, and the information transmission path is dynamically adjusted through an improved gating system; each module uses a multi-level GRU to process the sequence data. The improved TS-GAN network model includes: encoder e Establishing a temporal feature space To include space The mapping is used to encode the original sequence into a low-dimensional representation through a function;

[0015] State recursive computation: at each time step, the state inheritance history information is combined with the current input;

[0016] decoder r The inverse transformation from the spatial domain to the feature space is achieved through computation, and the reconstructed sequence is output.

[0017] The discriminator and generator computations in the TS-GAN model are both performed in the hidden coding space. The specific computation steps are as follows: The discriminator evaluates the input hidden-coded sequence and determines whether it comes from the real data distribution; the discriminator first uses a function... The hidden-coded sequence is scored, and the output score ranges from generation 0 to generation 1, representing the probability that the sequence belongs to a real fault sequence. The generator employs a two-layer cascaded architecture, and its operation mechanism is as follows: First, the encoding layer generates a random sequence... Convert to implicit coding Secondly, conditional distribution calculation. The first layer ensures the temporal coherence of the generated sequence by modeling state-dependent Markov dependencies; the second layer further approximates the true conditional distribution through recalculation. .

[0018] Furthermore, step S2, which involves identifying the time of fault occurrence based on the mutation monitoring algorithm, specifically includes the following steps: The mutation monitoring algorithm is shown in the following formula:

[0019]

[0020] in , Let i(n) and u(n) be the current and voltage abrupt changes per unit cycle, respectively, and i(n) and u(n) be the current and voltage values ​​at sampling point n. The number of times per wave sampling, , The setpoints for the mutation threshold criterion are all greater than 0; Multiple judgments are made using continuous multi-point sampling values. If the above two formulas are satisfied three times consecutively, the first time is considered the fault initiation time. The initiation time criterion is determined according to the following formula. , :

[0021]

[0022] in, , These are the threshold values ​​for current and voltage surges, both set to 0.5. , , These are the steady-state current amplitudes of phases A, B, and C, respectively. , , These are the steady-state voltage amplitudes of phases A, B, and C, respectively.

[0023] in, This indicates the sampling point corresponding to the start of the fault.

[0024] Furthermore, step S3, which involves decomposing the fault signal based on the DOB-VMD network model, specifically includes the following steps: S31: Initialize the DBO parameters of the dung beetle optimization algorithm, and decompose the variational mode decomposition (VMD) into its decomposition numbers. K Penalty Factor α Set the location to the dung beetle; S32: Calculate the local minimum envelope entropy of the IMF component, compare and update the local minimum envelope entropy, global minimum, and dung beetle position information, repeat this process until the maximum number of iterations is met, and obtain the global minimum and optimal parameters; S33: Perform VMD decomposition on the fault signal using optimal parameters to obtain K IMF components; S34: Select the IMF component with the maximum kurtosis, perform envelope demodulation analysis to extract fault features, reconstruct the signal, and complete the fault signal preprocessing.

[0025] Furthermore, in the dung beetle optimization algorithm, the rolling dung beetle position update mechanism is expressed as follows:

[0026]

[0027] t This represents the current iteration number; For the first i Only one dung beetle individual in the first t The position at the next iteration; The natural coefficient is assigned to 1 or -1 by probability; k∈(0,2) is the deflection coefficient; b It is a constant belonging to (0, 1); Used to simulate changes in light intensity; This is the worst bit globally; The dung beetle's dance behavior can be represented by the following formula:

[0028] In the formula, For deflection angle; when θ When the value is 0, π / 2, or π, the dung beetle's position will not be updated; The method of selecting oviposition sites for dung beetles based on boundary-based oviposition site selection is defined as follows:

[0029]

[0030]

[0031] In the formula Indicates the lower boundary of the spawning area; Indicates the upper part of the spawning area; Indicates the current local optimal position; Lb Denotes the lower bound of the optimization problem; Ub This represents the upper bound of the optimization problem; Indicates the maximum number of iterations; The position of the reproductive body changes continuously during the iteration, as follows:

[0032] In the formula, This indicates the position of the i-th breeding ball in the t-th iteration; , These are independent random vectors of size 1×D; The optimal feeding range for dung beetles is as follows:

[0033]

[0034] Indicates the lower bound of the optimal foraging area; Indicates the upper limit of the optimal foraging area; This represents the globally optimal position; the dung beetle's position update formula is:

[0035] In the formula, These are random numbers that follow a normal distribution. Let be a random variable belonging to (0, 1); This represents the best food source, assuming... The vicinity is the best foraging ground; the location update formula for dung beetles is defined as follows:

[0036] In the formula, S Let g be a constant; g is an independent random vector following a normal distribution of 1×D. The submodal decomposition constructs a constrained variational model based on fault characteristics, which is then transformed into an unconstrained variational problem to obtain an unconstrained optimal solution, thereby achieving fault signal separation. Specifically, it includes the following steps: Step 1: Construct a constraint-conditional variational model, as shown in the following equation:

[0037] Step 2: Transform the constrained problem into an unconstrained problem by introducing a quadratic penalty factor. With Lagrange multiplication operators , is represented as:

[0038] Step 3: Solve the unconstrained model using the alternating direction multiplier method, continuously updating the solution during the process. , , The specific expression is as follows:

[0039] Its frequency domain expression is:

[0040] Similarly, we get:

[0041]

[0042] Calculated through the above steps, when the iterative loop satisfies... If the iteration stops, then stop. The formula for the DOB-VMD algorithm is:

[0043] in,

[0044] In the formula: a ( j () represents the probability of an event occurring; m The number of samples; a ( j ) is the Hilbert demodulated envelope sequence of VMD modal components.

[0045] Furthermore, step S4, which involves identifying distribution network faults based on the WOA-LSSVM network model, specifically includes the following steps: S41: Using the Whale Algorithm (WOA), calculate the individual fitness, take the minimum fitness as the optimal position, and update and iterate through prey encirclement and spiral search operations to output the optimal kernel parameters and penalty factor of the Least Squares Support Vector Machine (LSSVM). S42: Initialize LSSVM, calculate the total kurtosis of multiple intrinsic mode functions, and input the training and test sets; S43: Train LSSVM with optimal parameters to identify various short-circuit faults and output the results.

[0046] Furthermore, the Whale Algorithm (WOA) includes the following steps: Surrounding the prey: The position of each whale is determined based on the positional relationships between them, using the following mathematical model:

[0047]

[0048] t This represents the current iteration number; This represents the whale's position in the current iteration. This indicates the location of a randomly selected individual whale. Indicates the distance between other whales and the target location; and The calculation formula is:

[0049]

[0050] In the formula, and All are random numbers between [0, 1]; let This represents the maximum number of algorithm iterations.a The expression is:

[0051] Enclosure contraction: The formula for updating the whale's enclosure position is as follows:

[0052]

[0053] In the formula, It is the best one in the current population. This represents the distance between other whales and the optimal individual; Spiral hunting: Approaching and successfully capturing prey using a spiral-shaped upward trajectory. The value range is set to ≤1, the corresponding mathematical model is as follows:

[0054]

[0055] In the formula, k It is a constant; w It is a random number between [-1, 1]. w When the value is 1, the whale is furthest from its food; w When the value is -1, the whale is closest to its food; Indicates the distance between the whale and its food; The formula for updating the whale's position is expressed as follows:

[0056] In the formula, rand The result is a random number, and its value ranges from [0, 1]. The objective function for the least squares support vector machine (LSSVM) optimization is summarized as follows:

[0057] In the formula: Indicates the error value; γ Represents the regularization parameter; By introducing the Lagrange function to solve, we obtain:

[0058] The least squares regression prediction model is: .

[0059] The beneficial effects of this invention are as follows: By generating samples using an improved TS-GAN model, the invention enhances the realism and accuracy of small-sample fault waveform data generation, solving the problem of insufficient small-sample fault samples and providing sufficient data support for subsequent diagnosis; by simulating the division of labor and cooperation of dung beetles, it optimizes the key parameters of variational mode decomposition, solving the problem of balancing exploration and development in traditional algorithms, and improving the efficiency and accuracy of parameter optimization; by simulating the hunting mechanism of whales, it optimizes the kernel parameters and penalty factors of the least squares support vector machine, possessing good global optimization capabilities, avoiding local extrema, improving the parameter optimization effect, and increasing the accuracy of identifying various short-circuit faults in the distribution network. This provides more accurate and comprehensive decision support for the operation and maintenance of the distribution network, ultimately improving the operational efficiency and reliability of the distribution network.

[0060] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0061] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a flowchart of the power distribution network fault identification method based on the whale optimization algorithm LSSVM as described in this invention; Figure 2 A flowchart for generating a model computation process based on small sample data of TS-GAN; Figure 3 Optimize the flowchart for DBO-VMD; Figure 4 Optimization flowchart for WOA-LSSVM. Detailed Implementation

[0062] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0063] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0064] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0065] Example 1: like Figure 1 As shown, this invention provides a method for identifying distribution network faults based on the whale optimization algorithm LSSVM. The method includes the following steps: Step 1: Fault recording data generation method based on improved TS-GAN; The basic architecture of TS-GAN consists of two WGANs (Wasserstein Generative Adversarial Networks): a standard WGAN and a conditional WGAN. WGAN uses the Wasserstein distance to measure the similarity between two probability distributions, which is used to simulate the real distribution of synthetic data. This two-layer TS-GAN architecture can efficiently generate time series data.

[0066] An improvement to the traditional TS-GAN architecture is proposed. First, a Global Attention Mechanism (GLFM) is incorporated. The features of the sequence extracted by the multi-head attention layer are further learned to learn more complex nonlinear mappings through a feedforward layer with a multilayer perceptron, enabling the model to learn richer representations. Finally, the local correlation features of the sequence are extracted through the attention layer, and the information captured by each layer is fused through residual connections between the multi-head self-attention layer, the feedforward layer, and the convolutional attention layer.

[0067] In the first layer, standard WGAN uses random vectors to generate synthetic spectrograms. This process produces preliminary synthetic data, which, while similar to real data, is not yet accurate enough. The second layer introduces conditional WGAN to transform the synthetic image generated in the first layer into a more accurate synthetic time series. Conditional WGAN plays a crucial role in this process, further optimizing and adjusting the generated data based on specific conditions to make it closer to the actual time series data.

[0068] This two-stage approach not only generates high-quality synthetic time-series data but also significantly improves the accuracy and reliability of subsequent model identification. This method enables researchers to better simulate and analyze power grid faults, thereby enhancing the safety and stability of the power grid.

[0069] The embedded attention module enables the model to dynamically select features, autonomously focusing on key signal changes during the fault occurrence period. This lightweight architecture offers advantages over traditional deep networks: reduced parameter count and shorter average diagnostic response time. Relying solely on the eigenvalues ​​of random matrices for data analysis, without considering eigenvectors, may lead to the neglect of crucial signal information.

[0070] The multi-head attention calculation used in the GLFM module is shown below: First, calculate the query matrix. and The key matrix To obtain the similarity, then divide by and through softmax The function normalizes the fraction matrix to obtain the fraction matrix. : (1) Then, the fraction matrix AND-value matrix Multiply by this to obtain the attention matrix. : (2) Input to the attention matrix , ( i =1, 2, ..., s The output expression of this layer, obtained after linear transformation, is as follows: These are learnable network weight parameters.

[0071] (3) The computation flow of the convolutional self-attention layer is shown below, where conv , k Indicates the kernel size as k One-dimensional convolution. First, the input from the previous layer... X The whole is divided into multiple local blocks Then, one-dimensional convolution operations are used to learn each local block. of Q , K , V .in: (4) In the formula: Represents the convolution kernel (p , q Weights at positions; k Indicates the kernel size; Indicates the size of the input / output channels; Indicates the feature location on the local block ( i , j The characteristic tensor of ).

[0072] choose k =3 convolutional kernels are used to focus on short-term temporal correlations. These short-term correlations are more critical to the local details of the sequence.

[0073] calculate Q and K The dot product, then divided by the key vector K Square root of dimension ,use softmax The function is normalized and compared with the value matrix. V Multiplying them together yields the attention matrix between the slices and the slices. : (5) Convolutional self-attention layers use the input sequence Convolution operations are performed on groups of distinct convolutional kernels, and their attention matrices are computed. ,Will Z The final output of this layer is obtained after linear transformation. Z : (6) In the data generation process, the generator first generates an image using a random noise vector (following a normal distribution) and a text feature vector. The generated image then continuously improves during training, gradually approaching the real image. The discriminator's function is to determine whether the input image comes from a real sample or is generated by the generator. The network structure includes: an input layer that receives real and generated images and combines them with input text features; convolutional layers that extract image features through multiple convolutions and use Leaky or ReLU activation functions to avoid gradient vanishing; feature fusion that combines text features with temporal features; this fusion can be achieved through concatenation or an attention mechanism; and a fully connected layer that outputs the final judgment result after a series of convolutional and pooling layers. Finally, the output layer outputs a probability value indicating whether the sample is a real sample. The training process of TS-GAN is adversarial, with the generator and discriminator being trained alternately in two phases.

[0074] The core of TS-GAN lies in the accurate modeling of time-series conditional probabilities. For a given electrical quantity sequence... The model focuses on capturing the current state. Compared with historical data Conditional dependency According to probability model theory, the joint distribution can be decomposed into: .Model m ( φ The optimization objective of this generation is to approximate the conditional distribution. At any time Both approximate the true conditional distribution. Fault generation model m ( φ )satisfy: (7) To ensure the model's ability to learn the dynamic characteristics of fault sequences, this invention adopts the implicit coding space design from the TS-GAN framework. This coding space can effectively characterize the conditional distribution characteristics of fault sequences, providing the necessary temporal information modeling capability for the generative model. Multi-level gated recurrent units (GRUs) are used to connect the modules, serving as a specialized architecture for recurrent neural networks, specifically designed for time-series data modeling. Through an improved gating system (reset gate and update gate), it dynamically regulates the information transmission path, effectively alleviating the long-term dependency learning difficulty problem of traditional RNNs, and demonstrating excellent temporal modeling capabilities in sequence analysis tasks.

[0075] like Figure 2 As shown, Represents the set of loss functions; Indicates component loss; These represent the discrimination outputs for the generated sequence and the real sequence, respectively. Indicates the reconstructed sequence; Indicates auxiliary discrimination parameters; These represent the implicit codes for generated and real data, respectively. This design allows the model to focus on the temporal dynamics of fault sequence data, rather than paying too much attention to static features. Furthermore, the model can better simulate and predict the relationships between sample points in power grid faults. Each module uses a multi-level GRU to process the sequence data, thereby enhancing the model's ability to understand and process time series data.

[0076] The core components of the fault sample generation model include: (1) Encoder e Establishing a temporal feature space To include space The mapping is used to encode the original sequence into a low-dimensional representation through a function; (8) (2) State recursive calculation: the state inheritance history information at each time step is combined with the current input; (9) (3) Decoder r The inverse transformation from the spatial domain to the feature space is achieved through computation, and the reconstructed sequence is output. This bidirectional mapping mechanism fully preserves the temporal dynamic characteristics of the sequence. (10) The discriminator and generator computations in the TS-GAN model are both performed in the hidden coding space. The specific computation steps are as follows: (1) The main task of the discriminator is to evaluate the input hidden-coded sequence and determine whether it comes from the real data distribution. The discriminator first uses a function... The hidden-coded sequences are scored, with the output score ranging from generation 0 to generation 1, representing the probability that the sequence belongs to a real fault sequence. The core function of this module is to perform discriminant analysis on the hidden codes of real fault sequences and generated sequences, and output binary classification results through feature comparison. This design enables a quantitative assessment of the spatial distribution differences between the codes of the two types of sequences. (2) The generator adopts a two-layer cascaded architecture, and its operation mechanism is as follows: First, the encoding layer generates a random sequence Convert to implicit coding Secondly, conditional distribution calculation. The temporal coherence of the generated sequence is ensured by modeling stateful Markov dependencies. This design gives the generated fault sequence implicit coding probabilistic characteristics similar to the real sequence. represents the temporal dynamics of the implicitly coded sequence. The second layer further approximates the real conditional distribution through recalculation. This ensures that the generated hidden-coded sequences closely approximate the distribution of actual fault sequences. This design not only guarantees that the model can capture the temporal dynamics of real fault sequences but also improves the effectiveness of model training.

[0077] Step 2: Identify the time of fault occurrence based on the mutation monitoring algorithm; Fault recording devices in power systems record sampled values ​​of signals such as voltage, current, and switch states before and after a fault occurs, providing reliable data for analyzing system accidents. When a short circuit occurs in the system, the fault recording data will exhibit the following characteristics: 1) The voltage waveform will show a significant decrease or disappearance, and the effective voltage value will drop to a certain percentage below the normal value. 2) The current waveform will show a significant increase or fluctuation, and the effective value of the current will rise to several times or even tens of times the normal value. 3) The duration of a short circuit is generally between tens of milliseconds and hundreds of milliseconds. After the short circuit ends, the voltage and current will return to normal or near-normal levels. A sudden change in voltage or current occurs instantaneously in a power system when a fault occurs. The magnitude and direction of the sudden change reflect the type, location, and severity of the fault. By analyzing fault waveform data, information about sudden changes can be extracted, enabling fault and fault timing identification. Sudden change monitoring algorithms only require simple differential operations on the fault waveform data to obtain sudden change information, without the need for complex mathematical transformations or model building. Furthermore, sudden change monitoring algorithms can sensitively reflect the occurrence and elimination of faults with high accuracy, are unaffected by system parameters or noise interference, and can effectively distinguish between fault and non-fault conditions.

[0078] The mutation monitoring algorithm is used to determine the moment a fault occurs, as shown in the following equation: (11) (12) in , Let i(n) and u(n) be the current and voltage abrupt changes per unit cycle, respectively, and i(n) and u(n) be the current and voltage values ​​at sampling point n. The number of times per wave sampling, , The setpoints for the mutation threshold criterion are all greater than 0. Multiple judgments are made using continuous multi-point sampling values. If the conditions of formulas (11) and (12) are met three times consecutively, the first time is considered the fault initiation time. The initiation time criterion is determined according to the following formula. , .

[0079] (13) (14) in, , These are the threshold values ​​for current and voltage surges, both set to 0.5. , , These are the steady-state current amplitudes of phases A, B, and C, respectively. , , These are the steady-state voltage amplitudes of phases A, B, and C, respectively.

[0080] (15) in, This indicates the sampling point corresponding to the start of the fault.

[0081] Step 3: Decompose fault signals based on DOB-VMD; The dung beetle optimization algorithm uses biomimetic modeling to intelligently divide the optimization population into four functionally distinct groups: 1) Rolling Beetles: Responsible for overall exploration and resource transportation. 2) Breeding Balls: Performing localized development and optimal solution breeding. 3) Young Beetles: Preserving the diversity of the solution space 4) Stealing Beetles: Enhancing competitiveness and escaping local optima This biomimetic optimization algorithm effectively solves the problem of exploration-expansion balance in traditional optimization algorithms by simulating the division of labor and cooperation behavior of dung beetle colonies in the process of dung ball processing. The synergistic effect of the four roles gives the algorithm both global search capability and local refinement efficiency, making it particularly suitable for complex nonlinear optimization problems.

[0082] The rolling dung beetle's position update mechanism is expressed in this algorithm as follows: (16) (17) In equations (16) and (17) t This represents the current iteration number; For the first i Only one dung beetle individual in the first t The position at the next iteration; The natural coefficient is assigned to 1 or -1 by probability; k∈(0,2) is the deflection coefficient; b It is a constant belonging to (0, 1); Used to simulate changes in light intensity; This is the worst bit globally.

[0083] The dung beetle's dance behavior can be represented by the following formula: (18) In the formula, This is the deflection angle. When θ When the value is 0, π / 2, or π, the dung beetle's position will not be updated. For dung beetles, choosing a crucial oviposition site is essential for the survival of their offspring. A novel boundary-based selection method is proposed, defined as follows: (19) (20) (twenty one) In the formula Indicates the lower boundary of the spawning area; Indicates the upper part of the spawning area; Indicates the current local optimal position; Lb Denotes the lower bound of the optimization problem; Ub This represents the upper bound of the optimization problem; This indicates the maximum number of iterations.

[0084] In the dung beetle optimization algorithm, once the female dung beetle selects a suitable oviposition area, it will lay its eggs in that area, and lay only one egg in each iteration. From formulas (19) and (20), we can see that... R The value determines the boundary of the spawning zone, and the position of the reproductive organism is not static; it changes continuously during algorithm iterations, as detailed below: (twenty two) In the formula, This indicates the position of the i-th breeding ball in the t-th iteration; , They are independent random vectors of size 1×D. In nature, some larvae emerge from the soil to forage for food; these are known as dung beetles. To better reflect the foraging behavior of dung beetles under natural conditions, their optimal foraging range has been determined: (twenty three) (twenty four) Indicates the lower bound of the optimal foraging area; Indicates the upper limit of the optimal foraging area; This represents the globally optimal position. Therefore, the dung beetle's position update formula is: (25) In the formula, These are random numbers that follow a normal distribution. Let be a random variable belonging to (0, 1). Stealing dung beetles refer to dung beetles that steal dung balls from other dung beetles. In formulas (23) and (24)... This represents the best food source, assuming... The vicinity is the best foraging ground; the location update formula for dung beetles is defined as follows: (26) In the formula, S Let g be a constant; g is an independent random vector following a normal distribution of 1×D. Variational mode decomposition constructs a constrained variational model based on fault characteristics, which is then transformed into an unconstrained variational problem to obtain the unconstrained optimal solution, thereby achieving the separation of fault signals.

[0085] Step 1: Construct a constraint variational model. Construct a model as shown in equation (27): (27) Step 2: Transform the constrained problem into an unconstrained problem. Introducing a secondary penalty factor With Lagrange multiplication operators , can be represented as: (28) Step 3: Solve the unconstrained model.

[0086] Equation (13) is solved using the alternating direction multiplier method, and is continuously updated during the solution process. , , The specific expression is as follows: (29) Its frequency domain expression is: (30) Similarly, we can obtain: (31) (32) Calculated through the above steps, when the iterative loop satisfies... If the iteration stops, then stop.

[0087] Based on the VMD process and decomposition principles, the decomposition number K and the penalty factor are known. α This has a significant impact on the results. Based on this, a VMD algorithm based on DBO is proposed and applied to the optimization of VMD parameters. The formula is: (33) in, (34) In the formula: a ( j () represents the probability of an event occurring; m The number of samples; a ( j ) is the Hilbert demodulated envelope sequence of VMD modal components.

[0088] In summary, the DBO-optimized VMD process proposed in this invention is as follows: Figure 3 .

[0089] Step 4: Distribution network fault identification based on WOA-LSSVM.

[0090] The basic principle of the whale algorithm is to mimic the hunting process of a whale pod. First, the whale pod searches for prey within a certain area, then gradually narrows the hunting range in a spiraling manner within the locked area. During this process, the whales release bubbles to surround the prey, thus capturing it. This hunting process can be compared to the algorithm's optimization process, where the prey represents the optimal solution to the problem. This algorithm has good global optimization capabilities and the advantage of avoiding local optima, mainly consisting of three parts: surrounding the prey, the predation mechanism, and random search.

[0091] (1) Surround the prey In the WOA algorithm, the area where each whale is located is represented as a solution, and whales use sound waves for localization to determine the location of their prey. The position of each whale is determined based on the positional relationships between them, and the corresponding mathematical model is as follows: (35) (36) t This represents the current iteration number; This represents the whale's position in the current iteration. This indicates the location of a randomly selected individual whale. Indicates the distance between other whales and the target location; and The calculation formula is: (37) (38) In the formula, and All are random numbers between [0, 1]. Let... This represents the maximum number of algorithm iterations. a The expression is: (39) (2) Encirclement and contraction This algorithm simulates the strategy of whales, which achieve precise capture by continuously approaching their prey. Its effect is to improve hunting efficiency and ensure that the prey has little chance of escape. The position update formula is as follows: (40) (41) In the formula, It is the best one in the current population. This represents the distance between other whales and the best individual.

[0092] (3) Spiral predation Once the food location is determined, whales can use a spiraling ascent trajectory to quickly approach and successfully hunt their prey. This spiraling ascent strategy allows them to effectively control the distance, thus increasing their capture success rate. The value range is set to ≤1, the corresponding mathematical model is as follows: (42) (43) In the formula, k It is a constant; w It is a random number between [-1, 1]. w When the value is 1, the whale is furthest from its food; w When the value is -1, the whale is closest to its food; This indicates the distance between the whale and its food.

[0093] As a whale spirals upwards toward its prey, it also performs a circling and contracting motion. This allows the whale to flexibly respond to the prey's movements, increasing its hunting success rate. Therefore, the whale's position update formula can be expressed as: (44) In the formula, rand It is a random number, and its value range is [0, 1].

[0094] 4.2 Least Squares Support Vector Machine (LSSVM) Support Vector Machine (SVM) is applied to the power load model, with the following training samples set: (45) It is the input quantity. Here, l represents the target output value, and l represents the total training volume. Regression is applied here. arrive R The mapping, and thus achieve f ( x )= y . The model for electricity load forecasting is set as follows: (46) We can efficiently fit the covered samples using a linear function, provided that the Euclidean space norm is minimized. Then, the linear problem is transformed into the following optimization problem, which we then solve further: (47) Constraints: (48) Taking error factors into account, a slack variable needs to be added: (49) An insensitive loss function can be defined. Definition: (50) The corresponding optimization problem is transformed into: (51) The constraints are: (52) The main differences between least squares support vector machines and nonlinear load forecasting models are twofold: (1) Transform inequalities into equations and quadratic programming into a class of linear equations to facilitate problem-solving. (2) Modifying the loss function to the squared error reduces the complexity of the calculation and speeds up the calculation. The objective function can be reorganized as follows: (53) In the formula: This is the error value; γ This is the regularization parameter.

[0095] By introducing the Lagrange function to solve, we obtain: (54) The least squares regression prediction model is: (55) Least squares support vector machines (LSVMs), based on support vector machine theory, employ a least squares loss function and have equality constraints, simplifying the load forecasting process and significantly improving the model's computational efficiency and predictive performance. LSVMs exhibit significant advantages in generalization ability, nonlinear fitting, and robustness, and can substantially reduce training time.

[0096] Kernel parameters and penalty factors are two crucial components of LSSVM, determining the grouping of data points and the model's fit. If the kernel parameter is set too small, classification performance will be poor; conversely, if it's set too large, all data points may be grouped into the same category, failing to achieve the desired outcome. Regarding the penalty factor, a larger value generally leads to a better model fit, but this can also negatively impact prediction performance. Relatively speaking, a smaller penalty factor may yield more reasonable results because the model performs better in handling noise, balancing empirical and structural risks. The whale optimization algorithm optimizes key parameters such as... Figure 4 As shown, through the above process, the identification of various short-circuit faults such as single-phase ground faults and three-phase ground faults can be achieved.

[0097] Example 2: An electronic device, comprising a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the method described in Embodiment 1 when executing the computer program.

[0098] Example 3: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0099] Example 4: A computer program product includes a computer program that, when executed by a processor, implements the method described in Example 1.

[0100] In the above embodiments, the reference to "this embodiment" in the specification indicates that a specific feature, structure, or characteristic described in connection with the embodiment is included in at least some embodiments, but not necessarily all embodiments. Multiple appearances of "this embodiment" do not necessarily all refer to the same embodiment.

[0101] In the above embodiments, although the invention has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory structures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed. The embodiments of the invention are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims.

[0102] As will be understood by those skilled in the art, the computer-readable storage medium described in this embodiment allows for the implementation of all or part of the steps in the above method embodiments by computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0103] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication between them. The memory is used to store computer programs, the communication interface is used to perform communication, and the processor and the transceiver are used to run the computer programs, so that the electronic terminal performs the steps of the above method.

[0104] In this embodiment, the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0105] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0106] This invention can be used in a wide range of general-purpose or special-purpose computing system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.

[0107] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for distribution network fault identification based on the whale optimization algorithm LSSVM, characterized in that: Includes the following steps: S1: Fault recording data is generated based on the improved TS-GAN network model; S2: Identify the time of fault occurrence based on mutation monitoring algorithm; S3: Decomposing fault signals based on the DOB-VMD network model; S4: Identify distribution network faults based on the WOA-LSSVM network model.

2. The distribution network fault identification method based on the whale optimization algorithm LSSVM according to claim 1, characterized in that: The improved TS-GAN network model described in step S1 incorporates the global attention mechanism GLFM into the traditional TS-GAN structure model, combines residual connections, and integrates the information captured by each layer; it introduces conditional WGAN to optimize and adjust the initial synthetic image generated by standard WGAN; it captures the conditional dependency relationship between the current electrical quantity state and historical data, and optimizes the generation model to make the approximate conditional distribution approximate the real conditional distribution.

3. The distribution network fault identification method based on the whale optimization algorithm LSSVM as described in claim 2, characterized in that: The calculation steps of the Global Attention Mechanism (GLFM) are as follows: First, calculate the query matrix. and The key matrix To obtain the similarity, then divide by and through softmax The function normalizes the fraction matrix to obtain the fraction matrix. : Then, the fraction matrix AND-value matrix Multiply by this to obtain the attention matrix. : Input to the attention matrix The output expression of this layer is obtained after linear transformation as follows. i =1, 2, ..., s ; in These are learnable network weight parameters; The computation process of a convolutional self-attention layer is as follows: First, the input from the previous layer is... X The whole is divided into multiple local blocks Then, one-dimensional convolution operations are used to learn each local block. of Q , K , V ,in: In the formula Represents the convolution kernel ( p , q Weights at positions; k Indicates the kernel size; Indicates the size of the input / output channels; Indicates the feature location on the local block ( i , j The characteristic tensor of ); calculate Q and K The dot product, then divided by the key vector K Square root of dimension ,use softmax The function is normalized and compared with the value matrix. V Multiplying them together yields the attention matrix between the slices and the slices. : Convolutional self-attention layers use the input sequence h Convolution operations are performed on groups of distinct convolutional kernels, and their attention matrices are computed. ,Will Z The final output of this layer is obtained after linear transformation. Z : 。 4. The distribution network fault identification method based on the whale optimization algorithm LSSVM as described in claim 3, characterized in that: Step S1, which involves generating fault recording data based on the improved TS-GAN network model, includes the following steps: In the data generation process, the generator first generates an image by combining a random noise vector with a text feature vector; then the generated image is continuously improved during the training process, gradually approaching the real image; the discriminator's function is to determine whether the input image comes from a real sample or is an image generated by the generator; The network structure includes: an input layer that receives real and generated images and combines them with the input text features; a convolutional layer that extracts image features through multiple convolutions and uses activation functions to avoid gradient vanishing; feature fusion that combines text features with temporal features; weighted fusion of the two through concatenation or attention mechanisms; a fully connected layer that outputs the final judgment result after a series of convolutional and pooling layers; and finally, an output layer that outputs a probability value indicating that the sample is a real sample. The training process of TS-GAN is adversarial, with the generator and discriminator being trained alternately in two stages. For a given electrical quantity sequence The model focuses on capturing the current state. Compared with historical data Conditional dependency According to probability model theory, the joint distribution is decomposed into: ;Model m ( φ The optimization objective of this generation is to approximate the conditional distribution. At any time Both approximate the true conditional distribution Fault generation model m ( φ )satisfy: The implicit coding space in the TS-GAN framework is used to design and characterize the conditional distribution characteristics of fault sequences, providing the generation model with the ability to model time series information. The modules are connected by multi-level gated cyclic units, and the information transmission path is dynamically adjusted through an improved gating system. Each module uses a multi-level GRU to process the sequence data. The improved TS-GAN network model includes: encoder e Establishing a temporal feature space To include space The mapping is used to encode the original sequence into a low-dimensional representation through a function; State recursive computation: at each time step, the state inheritance history information is combined with the current input; decoder r The inverse transformation from the spatial domain to the feature space is achieved through computation, and the reconstructed sequence is output. The discriminator and generator computations in the TS-GAN model are both performed in the hidden coding space. The specific computation steps are as follows: The discriminator evaluates the input hidden-coded sequence and determines whether it comes from the real data distribution; the discriminator first uses a function... The hidden-coded sequence is scored, and the output score ranges from generation 0 to generation 1, representing the probability that the sequence belongs to a real fault sequence. The generator employs a two-layer cascaded architecture, and its operating mechanism is as follows: First, the encoding layer generates a random sequence... Convert to implicit coding Secondly, conditional distribution calculation. The first layer ensures the temporal coherence of the generated sequence by modeling state-dependent Markov dependencies; the second layer further approximates the true conditional distribution through recalculation. .

5. The distribution network fault identification method based on the whale optimization algorithm LSSVM according to claim 1, characterized in that: Step S2, which identifies the time of fault occurrence based on the mutation monitoring algorithm, specifically includes the following steps: The mutation monitoring algorithm is shown in the following formula: in , Let i(n) and u(n) be the current and voltage abrupt changes per unit cycle, respectively, and i(n) and u(n) be the current and voltage values ​​at sampling point n. The number of times per wave sampling, , The setpoints for the mutation threshold criterion are all greater than 0; Multiple judgments are made using continuous multi-point sampling values. If the above two formulas are satisfied three times consecutively, the first time is considered the fault initiation time. The initiation time criterion is determined according to the following formula. , : in, , These are the threshold values ​​for current and voltage surges, both set to 0.

5. , , These are the steady-state current amplitudes of phases A, B, and C, respectively. , , These are the steady-state voltage amplitudes of phases A, B, and C, respectively. in, This indicates the sampling point corresponding to the start of the fault.

6. The distribution network fault identification method based on the whale optimization algorithm LSSVM according to claim 1, characterized in that: Step S3, which involves decomposing the fault signal based on the DOB-VMD network model, specifically includes the following steps: S31: Initialize the DBO parameters of the dung beetle optimization algorithm, and decompose the variational mode decomposition (VMD) into its decomposition numbers. K Penalty Factor α Set the location to the dung beetle's position; S32: Calculate the local minimum envelope entropy of the IMF component, compare and update the local minimum envelope entropy, global minimum, and dung beetle position information, repeat this process until the maximum number of iterations is met, and obtain the global minimum and optimal parameters; S33: Perform VMD decomposition on the fault signal using optimal parameters to obtain K IMF components; S34: Select the IMF component with the maximum kurtosis, perform envelope demodulation analysis to extract fault features, reconstruct the signal, and complete the fault signal preprocessing.

7. The distribution network fault identification method based on the whale optimization algorithm LSSVM according to claim 6, characterized in that: In the dung beetle optimization algorithm, the rolling dung beetle position update mechanism is expressed as follows: t This represents the current iteration number; For the first i Only one dung beetle individual in the first t The position at the next iteration; The natural coefficient is assigned to 1 or -1 by probability; k∈(0,2) is the deflection coefficient; b It is a constant belonging to (0, 1); Used to simulate changes in light intensity; This is the worst bit globally; The dung beetle's dance behavior can be represented by the following formula: In the formula, For deflection angle; when θ When the value is 0, π / 2, or π, the dung beetle's position will not be updated; The method of selecting oviposition sites for dung beetles based on boundary-based oviposition site selection is defined as follows: In the formula Indicates the lower boundary of the spawning area; Indicates the upper part of the spawning area; Indicates the current local optimal position; Lb Denotes the lower bound of the optimization problem; Ub This represents the upper bound of the optimization problem; Indicates the maximum number of iterations; The position of the reproductive body changes continuously during the iteration, as follows: In the formula, This indicates the position of the i-th breeding ball in the t-th iteration; , These are independent random vectors of size 1×D; The optimal feeding range for dung beetles is as follows: Indicates the lower bound of the optimal foraging area; Indicates the upper limit of the optimal foraging area; This represents the globally optimal position; the dung beetle's position update formula is: In the formula, These are random numbers that follow a normal distribution. Let be a random variable belonging to (0, 1); This represents the best food source, assuming... The vicinity is the best foraging ground; the location update formula for dung beetles is defined as follows: In the formula, S Let g be a constant; g is an independent random vector following a normal distribution of 1×D. The submodal decomposition constructs a constrained variational model based on fault characteristics, which is then transformed into an unconstrained variational problem to obtain an unconstrained optimal solution, thereby achieving fault signal separation. Specifically, it includes the following steps: Step 1: Construct a constraint-conditional variational model, as shown in the following equation: Step 2: Transform the constrained problem into an unconstrained problem by introducing a quadratic penalty factor. With Lagrange multiplication operators , represented as: Step 3: Solve the unconstrained model using the alternating direction multiplier method, continuously updating the solution during the process. , , The specific expression is as follows: Its frequency domain expression is: Similarly, we can obtain: Calculated through the above steps, when the iterative loop satisfies... If the iteration stops, then stop. The formula for the DOB-VMD algorithm is: in, In the formula: a ( j () represents the probability of an event occurring; m The number of samples; a ( j ) is the Hilbert demodulated envelope sequence of VMD modal components.

8. The distribution network fault identification method based on the whale optimization algorithm LSSVM according to claim 1, characterized in that: Step S4, which involves identifying distribution network faults based on the WOA-LSSVM network model, specifically includes the following steps: S41: Using the Whale Algorithm (WOA), calculate the individual fitness, take the minimum fitness as the optimal position, and update and iterate through prey encirclement and spiral search operations to output the optimal kernel parameters and penalty factor of the Least Squares Support Vector Machine (LSSVM). S42: Initialize LSSVM, calculate the total kurtosis of multiple intrinsic mode functions, and input the training and test sets; S43: Train LSSVM with optimal parameters to identify various short-circuit faults and output the results.

9. The distribution network fault identification method based on the whale optimization algorithm LSSVM according to claim 8, characterized in that: The Whale Algorithm (WOA) includes the following steps: Surrounding the prey: The position of each whale is determined based on the positional relationships between them, using the following mathematical model: t This represents the current iteration number; This represents the whale's position in the current iteration. This indicates the location of a randomly selected individual whale. Indicates the distance between other whales and the target location; and The calculation formula is: In the formula, and All are random numbers between [0, 1]; let This represents the maximum number of algorithm iterations. a The expression is: Enclosure contraction: The formula for updating the whale's enclosure position is as follows: In the formula, It is the best one in the current population. This represents the distance between other whales and the optimal individual; Spiral hunting: Approaching and successfully capturing prey using a spiral-shaped upward trajectory. The value range is set to ≤1, the corresponding mathematical model is as follows: In the formula, k It is a constant; w It is a random number between [-1, 1]. w When the value is 1, the whale is furthest from its food; w When the value is -1, the whale is closest to its food; Indicates the distance between the whale and its food; The formula for updating the whale's position is expressed as follows: In the formula, rand The result is a random number, and its value ranges from [0, 1]. The objective function for the least squares support vector machine (LSSVM) optimization is summarized as follows: In the formula: Indicates the error value; γ Represents the regularization parameter; By introducing the Lagrange function to solve, we obtain: The least squares regression prediction model is: 。

Images