New energy power distribution network fault identification method and system based on XGBoost

Abstract

The invention discloses a new energy power distribution network fault identification method and system based on XGBoost, and belongs to the technical field of power distribution network fault identification, and the method comprises the steps: carrying out the frequency domain conversion of collected power distribution network electrical data through wavelet transform, and obtaining the fault feature data of a power distribution network; according to a preset exponential distribution optimizer, optimizing parameters of a preset XGBoost-based fault identification model by simulating the characteristics of exponential distribution by taking optimization of model performance as a target; wherein the exponential distribution optimizer is adjusted based on a sine and cosine disturbance strategy and a Gaussian variation strategy; and according to the optimized fault identification model and the fault feature data, identifying the fault type of the power distribution network, and outputting a fault identification result. Therefore, by implementing the method and the device, the problem of inaccurate fault identification of the power distribution network connected with the new energy in the prior art can be solved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power distribution network fault identification, and particularly relates to a new energy power distribution network fault identification method and system based on XGBoost. BACKGROUND

[0002] As a key link of energy supply, the safe and stable operation of the power distribution network is crucial to achieving the double carbon goal. However, the power distribution network sometimes has faults such as overload and short circuit during operation, which leads to a decline in power supply quality and causes energy waste. In addition, with the rapid development of new energy and distributed power, the structure and operation mode of the power distribution network are also changing, so how to identify faults in new energy power distribution networks in a dynamically changing power distribution network environment is a major problem that needs to be solved.

[0003] Current research has collected different forms of power distribution network equipment data and used databases to store and analyze to establish a fault feature perception model, thereby achieving accurate identification of power distribution network faults. However, the topology of the power distribution network with new energy access is relatively complex, making the fault features more hidden, and the above fault identification method often cannot accurately capture the changes. At the same time, the volatility of new energy may also lead to voltage and frequency stability problems of the power distribution network, further increasing the difficulty of fault identification. SUMMARY

[0004] The application proposes a new energy power distribution network fault identification method and system based on XGBoost, which can solve the problem of inaccurate fault identification of the power distribution network with new energy access in the prior art.

[0005] The first aspect of the application provides a new energy power distribution network fault identification method based on XGBoost, which comprises:

[0006] The collected power distribution network electrical data is subjected to frequency domain conversion by wavelet transform to obtain fault feature data of the power distribution network;

[0007] According to a preset exponential distribution optimizer, the parameters of the preset fault identification model based on XGBoost are optimized by simulating the characteristics of the exponential distribution with the goal of optimizing the model performance; wherein the exponential distribution optimizer is adjusted based on the positive sine perturbation strategy and the Gaussian mutation strategy;

[0008] According to the optimized fault identification model and the fault feature data, the fault type of the power distribution network is identified, and the fault identification result is output.

[0009] The scheme extracts the feature data related to the fault from the power distribution network electrical data through wavelet transform, which is used for subsequent fault identification. In order to improve the accuracy of the fault identification model, an exponential distribution optimizer is introduced to optimize the parameters related to the performance of the fault identification model, so that the model can quickly identify the fault of the power distribution network in the complex and variable power distribution network environment. Moreover, in order to further improve the convergence speed and global search ability of the model, the exponential distribution optimizer is improved by using the sine and cosine disturbance strategy and the Gaussian mutation strategy. Finally, the high-performance fault identification model is used to process the fault feature data to quickly and accurately determine the fault of the power distribution network.

[0010] In a possible implementation method of the first aspect, the collected power distribution network electrical data is subjected to frequency domain conversion through wavelet transform to obtain fault feature data of the power distribution network, specifically as follows:

[0011] The power distribution network electrical data is subjected to multi-scale decomposition in the frequency domain through wavelet transform to extract wavelet coefficients of the power distribution network electrical data in each frequency band;

[0012] The local features of the power distribution network electrical data in each frequency band are identified and extracted based on the wavelet coefficients, and then the extraction results are normalized to obtain the fault feature data; wherein the local features are related to the fault type of the power distribution network.

[0013] The above scheme converts the power distribution network electrical data into data in different frequency bands through wavelet transform, and collects local features obviously related to fault features from different frequency bands to obtain fault feature data that can be used to identify the fault of the power distribution network.

[0014] In a possible implementation method of the first aspect, the exponential distribution optimizer is adjusted based on the sine and cosine disturbance strategy and the Gaussian mutation strategy, specifically as follows:

[0015] An initial population is constructed by randomly generating a plurality of random solutions, and an exponential distribution model is constructed for the initial population; wherein the coordinate information of the random solution is related to the parameters of the exponential distribution model;

[0016] A development stage and an exploration stage are set for the exponential distribution model, and when a preset random variable is less than a first threshold value, the exponential distribution model is in the development stage; when the random variable is greater than or equal to the first threshold value, the exponential distribution model is in the exploration stage;

[0017] According to the random variable, the exponential distribution model is iteratively optimized in the exploration stage and the development stage through the sine and cosine disturbance strategy and the Gaussian mutation strategy, until the number of iterations reaches a maximum iteration number to obtain the exponential distribution optimizer.

[0018] The above scheme uses the characteristics of the exponential distribution to iteratively optimize the constructed initial population through the development and exploration stages to find the optimal solution. The initial population can be regarded as a random variable subject to an exponential distribution, and an exponential distribution model is constructed accordingly. By constantly switching between the development stage and the exploration stage through a predetermined random variable, the optimal solution for the initial population is searched for, and the parameters that can make the performance of the fault identification model optimal are found.

[0019] In a possible implementation method of the first aspect, the exponential distribution model is iteratively optimized in the exploration stage and the development stage through a sine-cosine perturbation strategy and a Gaussian mutation strategy, specifically:

[0020] In the exploration stage, the dispersion degree of the random solution in the initial population is analyzed by the exponential distribution model to determine the search direction, and the global optimal solution is found in the initial population according to the search direction;

[0021] In the development stage, the local optimal solution in the initial population is found through directional search, and based on the local optimal solution and the search range of the initial population, a new random solution is generated through the sine-cosine perturbation strategy and the Gaussian mutation strategy to update the initial population, until the global optimal solution is found in the updated initial population, and the generation of the new random solution is stopped;

[0022] The exponential distribution model is optimized according to the global optimal solution.

[0023] The above scheme introduces the sine-cosine perturbation strategy and the Gaussian mutation strategy to improve the exponential distribution optimizer, thereby widening the search environment corresponding to the initial population and updating the initial population, avoiding premature convergence in the iterative updating process, and improving the probability of finding the global optimal solution and avoiding falling into the local optimal solution in the development stage.

[0024] In a possible implementation method of the first aspect, the new random solution is generated through the sine-cosine perturbation strategy and the Gaussian mutation strategy to update the initial population, specifically:

[0025] A nonlinear factor is set, and the search range is expanded through the nonlinear factor to generate a new random solution, thereby expanding the initial population, in combination with the periodic variation characteristics of the sine and cosine functions; and as the number of iterations increases, the value of the nonlinear factor gradually decreases.

[0026] A Gaussian distribution random variable is set, and the initial population is expanded through the Gaussian distribution random variable in each iteration.

[0027] In a possible implementation of the first aspect, the initial population is expanded, specifically as follows:

[0028] The initial population expanded by the cosine-sine perturbation strategy is specifically expressed as follows:

[0029]

[0030] wherein V i t+1 is the initial population expanded, t is the number of iterations of optimization, a is an adaptive parameter, b is the search range of the initial population expanded, is the i-th solution of the non-memory matrix, is the number of connections from the current time to the i-th local optimal solution, f is a random number in the interval [0, 1], r and r1 are random numbers in the interval [0, 2p], w is a non-linear factor, and s is the variance of the random solution.

[0031] The initial population expanded by the Gaussian mutation strategy is specifically expressed as follows:

[0032]

[0033] wherein V i t+1 is the initial population expanded, M t is the average value of the initial population, c is an adjustment parameter, Z1 and Z2 are the area vectors around M t , and Gauus(0, 1) is a Gaussian-distributed random variable.

[0034] In a possible implementation of the first aspect, according to a preset exponential distribution optimizer, the parameters of a preset XGBoost-based fault identification model are optimized by simulating the characteristics of the exponential distribution, with the goal of optimizing the performance of the model, specifically as follows:

[0035] The hyperparameters of the fault identification model are optimized by fusing the exponential distribution optimizer and the fault identification model.

[0036] In the fusion process, the learning rate, decision tree, sample random sampling parameter, and sample column sampling parameter of the fault identification model are optimized by the exponential distribution optimizer; wherein the decision tree includes the maximum tree depth, pruning parameter, and minimum sample number of leaf node.

[0037] The above scheme optimizes the hyperparameters in the fault identification model by using the improved exponential distribution optimizer, and the optimal solution is given by using the hyperparameters to improve the performance of the model, thereby improving the accuracy of fault identification of the power distribution network.

[0038] The second aspect of the application provides a new energy power distribution network fault identification system based on XGBoost, the system comprises: a feature extraction module, a parameter optimization module and a fault identification module.

[0039] The feature extraction module is configured to perform frequency domain conversion on the collected power distribution network electrical data through wavelet transform to obtain fault feature data of the power distribution network.

[0040] The parameter optimization module is configured to optimize the parameters of the preset fault identification model based on XGBoost by simulating the characteristics of the exponential distribution, with the goal of optimizing the performance of the model, wherein the exponential distribution optimizer is adjusted based on the positive sine perturbation strategy and the Gaussian mutation strategy.

[0041] The fault identification module is configured to identify the fault type of the power distribution network according to the optimized fault identification model and the fault feature data, and output a fault identification result.

[0042] The third aspect of the application provides a terminal device, the device comprises: a terminal device comprising a processor and a memory, the memory stores a computer program, and the processor executes the computer program to realize the steps of the new energy power distribution network fault identification method based on XGBoost in any one of the embodiments of the application.

[0043] The fourth aspect of the application provides a storage medium, the storage medium stores computer readable program code, when the computer readable program code is executed, the steps of the new energy power distribution network fault identification method based on XGBoost in any one of the embodiments of the application are realized. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the application, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0045] Figure 1 is a specific flowchart of a new energy power distribution network fault identification method based on XGBoost provided by an embodiment of the application;

[0046] Figure 2 is a model training cycle diagram of a new energy power distribution network fault identification method based on XGBoost provided by an embodiment of the application;

[0047] Figure 3is a model performance evaluation result diagram of a new energy power distribution network fault identification method based on XGBoost provided by an embodiment of the present application.

[0048] Figure 4 is a structural diagram of a new energy power distribution network fault identification system based on XGBoost provided by an embodiment of the present application.

[0049] Figure 5 A structural diagram of a terminal device is provided by an embodiment of the present application. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0051] It should be understood that the step numbers used herein are only for the convenience of description, and are not limited to the execution sequence of the steps.

[0052] First embodiment

[0053] Due to the increase of the complexity of the power distribution network topology structure caused by the access of new energy, the fault features are more concealed, and the traditional fault identification method often cannot accurately capture the changes. At the same time, the volatility of new energy may also cause voltage and frequency stability problems of the power distribution network, further increasing the difficulty of fault identification. The embodiments of the present application aim at the above problems, and improve the fault identification model by introducing adaptive sine and cosine strategy and Gaussian mutation strategy, improve the convergence speed and calculation efficiency of the model, so as to accurately identify the fault defects in the new energy power distribution network.

[0054] As shown in Figure 1 To solve the problem of inaccurate fault identification of the power distribution network with access to new energy in the prior art, the first embodiment of the present application provides a specific flowchart of a new energy power distribution network fault identification method based on XGBoost. The new energy power distribution network fault identification method based on XGBoost of the present embodiment includes steps S1 to S3, which are described in detail as follows.

[0055] Step S1, the collected power distribution network electrical data is converted in frequency domain by wavelet transform, and the fault feature data of the power distribution network is obtained.

[0056] In the step S1, the current collected power grid electrical data is first decomposed into sub-signals of different frequencies through wavelet transform, and then the position and nature of the data containing fault features are identified by analyzing the energy distribution and time-frequency characteristics of the sub-signals, and the fault feature data of the power grid is extracted.

[0057] Specifically, the wavelet transform is used to perform multi-scale decomposition on the power grid electrical data, and the feature information of the data in different frequency bands is extracted, and the wavelet coefficients of the power grid electrical data in each frequency band are obtained. The number of decomposition layers of the wavelet transform is generally determined according to the complexity and fluctuation of the data, and is generally 3 to 5 layers. The more the number of layers, the finer the frequency band after decomposition, and the more detailed signal features can be captured.

[0058] The wavelet transform formula is:

[0059]

[0060] In the formula, W x (j,k) is the wavelet coefficient, representing the feature of the data at scale j and position k; x(n) is the discrete power grid electrical data; j is the scale parameter, and k is the displacement parameter, both of which are related to the number of decomposition layers; is the conjugate of the wavelet function;

[0061] The expression of the wavelet function ψ j,k (n) is:

[0062] ψ j,k (n)=2 -j / 2 ψ(2 -j n-k);

[0063] The wavelet coefficient can be used for feature enhancement to distinguish the data related to the fault type of the power grid in the power grid electrical data, and to accurately extract the local feature E(j,k) in each frequency band. The specific expression is:

[0064] E(j,k)=|W x (j,k)| 2 ;

[0065] In order to adapt the extracted local feature to the subsequent fault recognition model, the mean-standard deviation strategy is used to normalize the data to obtain the fault feature data of the power grid. The specific expression of the mean-standard deviation strategy is:

[0066]

[0067] In the formula, μ F is the mean of the data, σ F is the standard deviation of the data, F, F normThe data before and after normalization, respectively.

[0068] In step S2, the parameters of the preset XGBoost-based fault identification model are optimized by simulating the characteristics of the exponential distribution, with the goal of optimizing the model performance, according to a preset exponential distribution optimizer; wherein the exponential distribution optimizer is adjusted based on a positive sine perturbation strategy and a Gaussian variation strategy.

[0069] The embodiments of the present application use XGBoost as the basis of the fault identification model. XGBoost uses a Classification And Regression Tree (CART) as the base classifier, and constructs multiple base classifiers by iteration and combination to improve the prediction accuracy and generalization ability of the model. The core idea is to gradually optimize the model by continuously adding decision trees, and to fit the residual error of the previous round of prediction to gradually reduce the error of the prediction model. XGBoost minimizes the objective function by gradient descent method, thereby gradually optimizing the model parameters. In addition, in the process of constructing the decision tree, XGBoost adopts a gradient boosting strategy, which can better capture the non-linear relationships and complex patterns in the data.

[0070] Therefore, the fault identification model is constructed by XGBoost, and the model parameters are optimized as follows:

[0071] For a given data set D={(x i ,y i )}, with n samples and m features, the prediction result of the fault identification model after the tth iteration training is:

[0072]

[0073] In the formula, is the prediction result of the data set in the tth iteration, f t (x i ) is the prediction result of the tth CART.

[0074] Using the second-order Taylor expansion can more accurately capture the local properties of the loss function of the model, thereby providing more detailed gradient information in the optimization process. Therefore, the second-order Taylor expansion is used as the objective function of the fault identification model to quantify the loss residual obj (t) of the optimized model, which is specifically shown in the following formula:

[0075]

[0076] In the formula, is the loss function, y i is the i th sample data value, is a regularity term; gi , h i is a regularization term, used to control the model complexity to avoid overfitting, g i is the first-order derivative of the loss function, h i is the second-order derivative.

[0077] In each iteration, according to the calculated gradient information g i and the second-order derivative information h i , the objective function is then optimized to determine the decision tree f t (x) that should be added in this round. Finally, all the decision trees generated in the iteration process are accumulated to build a complete fault identification model.

[0078] The complexity of the decision tree Ω is usually measured by the depth of the tree, the number of internal nodes, the leaf node data, and the proportion of the leaf node in the whole, and the expression is as follows:

[0079]

[0080] In the formula, γ and λ are regularization factors, is the weight vector of node j, T t is the number of leaf nodes of the decision tree t.

[0081] After building a complete fault identification model by the XGBoost decision tree described above, in order to further improve the performance of the model and obtain more accurate fault identification results, the exponential distribution identifier is introduced to optimize the hyperparameters of the fault identification model, so that the model can better adapt to data fluctuations and complexity, still show high accuracy in the case of connecting new energy to the power distribution network, and show good stability in the fault identification process, and can adapt to the complex and changeable power distribution network environment.

[0082] The exponential distribution identifier provided by the embodiment of the present application is constructed based on the EDO algorithm, and on this basis, the exponential distribution identifier is adjusted by introducing the positive sine disturbance strategy and the Gaussian variation strategy, so that it has faster convergence speed and global search ability.

[0083] The EDO algorithm, which stands for Exponential Distribution Optimizer, is used to solve complex optimization problems, allowing a group of "individuals" to constantly explore the search space. Each individual updates its own position according to the random characteristics of the exponential distribution, gradually approaching the global optimal solution.

[0084] Based on the above characteristics, the hyperparameters of the fault identification model are processed by using the exponential distribution identifier, and the global optimal solution is found for these hyperparameters, and then the fault identification model is optimized, effectively improving the model performance.

[0085] The exponential distribution optimizer mainly optimizes the learning rate, decision tree, sample random sampling parameter and sample column sampling parameter of the fault identification model. The decision tree includes maximum tree depth, pruning parameter, and minimum sample number of leaf node.

[0086] Therefore, the exponential distribution optimizer generates a plurality of random solutions as an initial population based on the characteristics of the exponential distribution, and constructs an exponential distribution model for the initial population. The coordinate information of the random solution is related to the parameters of the exponential distribution model, so the exponential distribution optimizer updates the position of the random solution by updating the exponential distribution model. The random solutions herein are randomly generated and have great differences from each other.

[0087] The exponential distribution optimizer needs to go through two processes, a development stage and an exploration stage, to find the global optimal solution. These two stages are mainly used to iteratively update the position of the random solution until the global optimal solution that meets the requirements is obtained. A random variable a is set in the embodiment of the present application to distinguish between the two stages, and the value range of the random variable a is [0, 1]. When a is less than 0.5, the exponential distribution model is in the development stage, and when a is greater than or equal to 0.5, the exponential distribution model is in the exploration stage.

[0088] In the development stage and the exploration stage, the coordinate information of each random solution in the initial population is regarded as the parameter of the exponential distribution model, and the exponential distribution of the parameter is used to reflect the change and trend of the search space. The variable distribution of the random solution is:

[0089]

[0090] Xwinners i,j = lb + rand(ub - lb);

[0091] In the formula, Xwinners is the initial population, Xwinners i,j is the i-th row and j-th column random solution, n and d are the dimensions of the initial population; ub and lb are the upper and lower boundaries of the iterative update set, and rand is a random number in [0, 1].

[0092] When the set a is less than 0.5, the exponential distribution model enters the development stage. The development stage mainly approaches the global optimal solution by directional search, and at the same time, in order to ensure that the entire search range can be covered, an attraction mechanism is used to guide the poor performing solution to its adjacent area to expand the search range, and the search center is concentrated in the vicinity of the guide solution, until the global optimal solution is obtained. In this process, the guide solution is dynamically updated, and the average value of the top three optimal solutions is taken as the new guide solution. The expression of the guide solution is as follows:

[0093]

[0094] where Xguide t is the guide solution updated at the tth iteration, is the global optimal solution calculated at the current time.

[0095] The guide solution is the historical global optimal solution determined in the initial population. When a new global optimal solution is found in the initial population through iterative updating, the original global optimal solution will serve as the guide solution. In the iterative updating process, other individuals in the initial population will conduct local fine search around the guide solution.

[0096] In the development phase, a memoryless attribute matrix memoryless is also constructed to simulate an exponential distribution to record newly generated solutions in the initial population. The goal of this matrix is to use the guide solution found at the current time as a basis to explore the global optimal solution in the neighborhood.

[0097] When the initial population updated at the tth iteration is equal to the memoryless attribute matrix, the updating process will explore around the guide solution, with the purpose of finding a better solution to guide the subsequent population development; otherwise, if the condition is not met, the poor solution will move closer to the local optimal solution, which is expressed as follows:

[0098]

[0099] where is the ith solution of the memoryless attribute matrix, φ is a random number in the interval [0, 1], is the connection number from the current time to the ith local optimal solution, a is an adaptive parameter, b is an expanded search space, f is a random number in the interval [-1, 1], and σ is the variance of the random solution.

[0100] The mean μ and variance σ of the random solution are expressed as follows:

[0101]

[0102] where λ' is an exponential decay parameter, the value of which is inversely proportional to the mean and variance values.

[0103] During the iterative updating period, the initial population presents a relatively dispersed distribution at the initial stage, and the aggregation between solutions is not obvious. However, when the iterative process enters the middle and later stages, with the continuous increase of the number of iterations, the initial population gradually converges, and the difference between individuals decreases, leading to a rapid decrease in diversity. Not only does this limit the search ability of the algorithm, but it can also cause the algorithm to prematurely converge to a certain local optimal solution and fail to continue exploring other potential global optimal solutions.

[0104] In order to solve the problem of insufficient global search ability and lack of solution diversity of the EDO algorithm, the embodiments of the present application update the initial population in the development stage by using the sine-cosine disturbance strategy and the Gaussian mutation strategy. The sine-cosine disturbance strategy aims to effectively interfere with the update of the initial population, and the purpose is to broaden the search range that a single random solution can explore in the iteration process, so as to alleviate the phenomenon of premature or excessive aggregation of the initial population.

[0105] The core of the sine-cosine disturbance strategy is to add a nonlinear factor for regulation. In the early stage of algorithm running, the factor makes the disturbance amplitude generated larger, which is helpful for extensive search. With the continuous advancement of algorithm iteration, the value of the factor gradually decreases and the disturbance intensity weakens, that is, with the increase of the number of iteration optimization, the value of the nonlinear factor gradually decreases. Thus, a dynamic adjustment process from large to small is generated, which can establish a balance between global exploration and local development of the algorithm, and finally improve the search and convergence performance of the algorithm.

[0106] Specifically, the nonlinear factor is combined with the periodic variation characteristics of the sine and cosine functions to expand the search range of each random solution to generate new random solutions, and then expand the initial population.

[0107] The initial population expanded by the sine-cosine disturbance strategy is specifically expressed as:

[0108]

[0109] V i t+1 For the expanded initial population, t is the number of iteration optimization, a is the adaptive parameter, b is the expanded search range, is the ith solution of the memoryless matrix, is the connection number from the current time to the ith local optimal solution, φ is a random number in the interval [-1, 1], f is a random number in the interval [0, 1], r and r1 are both random numbers in the interval [0, 2π], w is the nonlinear factor, and σ is the variance of the random solution.

[0110] The EDO algorithm generates new solutions by simulating the characteristics of exponential distribution, which can effectively explore the solution space in the initial stage, but as the iteration proceeds, its search behavior gradually becomes conservative and tends to fine-tune in the neighborhood of the better solution found, resulting in that the EDO algorithm is prone to fall into local optimal solution in the development stage.

[0111] As an improvement of the above scheme, the Gaussian mutation strategy is introduced in the embodiments of the present application, so that the initial population can produce certain random changes in each iteration, which can further expand the search range of the algorithm and avoid the problem of premature convergence. At the same time, due to the characteristics of Gaussian distribution, the addition of Gaussian distributed random variables will not have a great impact on the overall trend of the initial population, thereby ensuring the stability of the algorithm on a macro level. The expression is as follows:

[0112]

[0113] In the formula, V i t+1 is the initial population, M t is the average value of the initial population, c is the adjustment parameter, Z1 and Z2 are the region vectors around M t , and Gauus(0, 1) is a Gaussian distributed random variable.

[0114] Therefore, by introducing the sine-cosine disturbance strategy and the Gaussian mutation strategy, the convergence speed and global search ability of the exponential distribution optimizer are improved. Therefore, the method for constructing the fault identification model in the embodiments of the present application is different from the conventional EDO and XGBoost algorithms, but is an IEDO-XGBoost algorithm.

[0115] When the set α is greater than or equal to 0.5, the exponential distribution model enters the exploration stage. The purpose of the exploration stage is to locate the region with the global optimal solution in the search range, so as to provide high-quality hyperparameters for the subsequent optimization process. By continuously updating the position of the solution, the dispersion degree of the random solution is analyzed by using the memoryless property of the exponential distribution to determine the search direction and step length. This makes it possible to flexibly adjust the search strategy during the exploration process, thereby maintaining strong robustness in complex and multi-modal optimization problems. In this process, the update expression of the initial population is as follows:

[0116]

[0117] In the formula, V i t+1 is the initial population, M t is the average value of the initial population, c is the adjustment parameter, Z1 and Z2 are the region vectors around M t , D1 and D2 are the distances between M t and Xwinners randl and Xwinners rand2 , respectively.

[0118] After generating a new random solution for the initial population, it is verified whether it meets the given boundary conditions. Then the solution is recorded and stored in a matrix that does not retain historical information, and the solution in the initial population is updated using a greedy strategy, and the newly generated solution is further adjusted and optimized. If the new solution has a better fitness function, the optimal solution in the initial population is updated. Otherwise, the algorithm will guide it to a reference solution that performs better, with the purpose of exploring whether there is a solution that surpasses the current best individual. As the optimization process progresses, all solutions in the population will be highly concentrated around the global optimal solution.

[0119] In summary, by determining the random variable, the development stage or the exploration node is used to iterate the exponential distribution model to update the initial population, until the number of iterations reaches the maximum number of iterations, the exponential distribution optimizer outputs the global optimal solution, and the above XGBoost-based fault identification model is optimized by the global optimal solution, and the update of the model hyperparameters is realized.

[0120] Specifically, by fusing the exponential distribution optimizer and the fault identification model, the hyperparameters of the fault identification model are optimized, and the accuracy of the power distribution network defect identification is improved.

[0121] Step S3, according to the optimized fault identification model and the fault feature data, identifying the fault type of the power distribution network, and outputting the fault identification result.

[0122] By optimizing the hyperparameters of the fault identification model, the input fault feature data is processed, the fault place in the power distribution network is identified and the corresponding fault type is determined, and finally the fault identification result is output for subsequent fault maintenance.

[0123] Implementing the embodiments of the present application has the following beneficial effects:

[0124] The embodiments of the present application first extract the feature data related to the fault from the power distribution network electrical data through wavelet transform, which is used for subsequent fault identification. In order to improve the accuracy of the fault identification model, an exponential distribution optimizer is introduced to optimize the parameters related to the performance of the model in the fault identification model, so that the model can quickly identify the power distribution network fault in the complex and variable power distribution network environment. Moreover, in order to further improve the convergence speed and global search ability of the model, the exponential distribution optimizer is improved by using the sine-cosine disturbance strategy and the Gaussian mutation strategy. Finally, the fault feature data is processed by the high-performance fault identification model to quickly and accurately determine the fault condition of the power distribution network.

[0125] Second embodiment

[0126] Further, in order to show the difference between the XGBoost-based new energy power distribution network fault identification method provided by the above method embodiment and the conventional technology.Figure 2 A model training cycle diagram under different power distribution network fault types is provided.

[0127] The data set used by the embodiments of the application is composed of normal scenarios and six fault types, and distributed photovoltaic units are added to the nodes in the power distribution network line. Table 1 below shows the power distribution network fault setting situation:

[0128] Table 1 Power distribution network fault types

[0129]

[0130] The three-phase currents at both ends of the fault in the data set are collected at a sampling rate of 20 kHz, and the number of samples of each type of fault is 1800. The data set is divided into a training set, a validation set and a test set according to a ratio of 3:1:1.

[0131] The data set after division is used to train, test and verify the fault recognition model based on the XGBoost improved by the exponential distribution optimizer provided by the method embodiments multiple times, and a model training cycle diagram as shown in Figure 2 is obtained. Figure 2 In the model training cycle diagram, each colored line segment represents the fault type corresponding to the serial number in Table 1 above, the horizontal coordinate is the training cycle, and the total number is 50; the vertical coordinate is the model accuracy, which mainly shows the relationship between the model accuracy and the training cycle under the data set of each fault type.

[0132] In the model training cycle diagram, Figure 2 shows that as the training cycle increases, the model accuracy also gradually increases. After 50 rounds of training cycle, the fault positioning accuracy of each line in the training set data tends to 100%. Further, the average value of the experimental results is taken as the final diagnosis result, and 3 groups of data misjudgment occur between faults 2 and 3, which is mainly due to the similar characteristics of short circuit faults. In addition, the training set data of other fault conditions maintains a nearly 100% accurate recognition, and the test set data accuracy maintains at 98.6% and above, indicating that the proposed algorithm basically meets the requirements of the dispatch center for defect recognition accuracy.

[0133] Then, the power distribution network fault data collected in the actual scenario is used to verify the trained fault recognition model, and the fault recognition model based on IEDO-XGBoost of the application is compared and analyzed with the first model based on EDO-XGBoost and the second model based on PSO-XGBoost to determine the accuracy of these models in identifying different fault types, and the average accuracy is taken as the standard for evaluating the performance of the model. The performance evaluation results of the three models are as shown in Figure 3(a) shows that the abscissa is the type of fault identified, the ordinate is the accuracy of identification, purple represents the fault identification model, cyan represents the first model, and yellow represents the second model.

[0134] In Figure 3 In (a), in the defect identification scene of the power distribution network containing new energy, the fault identification model exhibits significant superiority in accuracy compared to the first model and the second model. The fault identification model can more effectively adjust the model parameters by introducing an improved dynamic optimization strategy to optimize the model parameters, thereby improving the convergence speed and accuracy of the model. This dynamic optimization strategy not only adapts to the data fluctuations and complexity brought by the integration of new energy, but also maintains stable performance when processing high-dimensional data, with an average identification accuracy close to 100%. In contrast, the first model and the second model rely more on manual intervention in feature selection, making it difficult to adapt to the data diversity brought by the integration of new energy, with an average identification accuracy of about 95% and 90%, respectively.

[0135] To further verify the effect of the method provided by the embodiment of the application on improving the stability of the fault identification model, the standard deviation of accuracy is used to evaluate the stability of the three deep learning algorithms in defect identification. As shown in Figure 3 (b), the accuracy standard deviations of the first model and the second model are relatively high, and when the training tends to be stable, the standard deviations are maintained at 0.51% and 0.93%, respectively. The standard deviation of the fault identification model is the lowest, and when the training tends to be stable, the standard deviation is maintained within 0.3%. This is mainly due to the optimization of the training parameters, which makes the model more stable when extracting features, reducing the impact of noise and disturbance on the performance of the model. Therefore, it can be concluded that the fault identification model has good stability in the application of power distribution network fault identification.

[0136] Third embodiment

[0137] Further, in order to execute the XGBoost-based new energy power distribution network fault identification system corresponding to the above-mentioned method embodiment, to realize the corresponding functions and technical effects, Figure 4 A structural diagram of an XGBoost-based new energy power distribution network fault identification system is provided. For ease of illustration, only the part related to the present embodiment is shown. The XGBoost-based new energy power distribution network fault identification system provided by the embodiment of the application comprises:

[0138] The feature extraction module 201 is configured to perform frequency domain conversion on the collected power distribution network electrical data by wavelet transform to obtain fault feature data of the power distribution network.

[0139] In the embodiment of the present application, the current collected power distribution network electrical data is first decomposed into sub-signals of different frequencies through wavelet transform, and then the position and nature of the data containing fault characteristics are identified by analyzing the energy distribution and time-frequency characteristics of the sub-signals, and the fault characteristic data of the power distribution network is extracted.

[0140] Specifically, the power distribution network electrical data is multi-scale decomposed through wavelet transform to extract feature information of the data in different frequency bands, and wavelet coefficients of the power distribution network electrical data in each frequency band are obtained. The number of decomposition layers of wavelet transform is generally determined according to the complexity and fluctuation of the data, and is generally 3 to 5 layers. The more the number of layers is, the finer the frequency band after decomposition is, and the more detailed signal characteristics can be captured.

[0141] The wavelet transform formula is:

[0142]

[0143] In the formula, W x (j, k) is the wavelet coefficient, representing the feature of the data at scale j and position k; x(n) is the discrete power distribution network electrical data; j is the scale parameter, and k is the displacement parameter, both of which are related to the number of decomposition layers; is the conjugate of the wavelet function;

[0144] The expression of the wavelet function ψ j,k (n) is:

[0145] ψ j,k (n) = 2 -j / 2 ψ(2 -j n-k);

[0146] The wavelet coefficient can be used for feature enhancement to distinguish the data related to the fault type of the power distribution network in the power distribution network electrical data, and to realize accurate extraction of the local feature E(j, k) in each frequency band. The specific expression is:

[0147] E(j, k) = |W x (j, k)| 2 ;

[0148] In order to adapt the extracted local feature to the subsequent fault identification model, the mean-standard deviation strategy is used to normalize the data to obtain the fault characteristic data of the power distribution network. The specific expression of the mean-standard deviation strategy is:

[0149]

[0150] In the formula, μ F is the mean of the data, σ F is the standard deviation of the data, F, F normThe data before and after normalization, respectively.

[0151] The parameter optimization module 202 is configured to optimize parameters of a preset XGBoost-based fault identification model by simulating characteristics of an exponential distribution, so as to optimize model performance, according to a preset exponential distribution optimizer; wherein the exponential distribution optimizer is adjusted based on a positive sine perturbation strategy and a Gaussian variation strategy.

[0152] In the embodiments of the present application, a number of random solutions are randomly generated to construct an initial population, and an exponential distribution model is constructed for the initial population; wherein coordinate information of the random solutions is related to parameters of the exponential distribution model.

[0153] The exponential distribution model is set with a development stage and an exploration stage; when a preset random variable is less than a first threshold value, the exponential distribution model is in the development stage; when the random variable is greater than or equal to the first threshold value, the exponential distribution model is in the exploration stage.

[0154] According to the random variable, the exponential distribution model is iteratively optimized by the positive sine perturbation strategy and the Gaussian variation strategy in the exploration stage and the development stage, until the number of iterations reaches a maximum iteration number, to obtain the exponential distribution optimizer.

[0155] The fault identification module 203 is configured to identify a fault type of the power distribution network according to the optimized fault identification model and the fault feature data, and output a fault identification result.

[0156] In the embodiments of the present application, the input fault feature data is processed by the fault identification model with optimized hyperparameters, the place where a fault exists in the power distribution network is identified and the corresponding fault type is determined, and finally the fault identification result is output for subsequent fault repair.

[0157] In some embodiments, the feature extraction module 201 specifically includes:

[0158] In the embodiments of the present application, XGBoost is used as the basis of the fault identification model. XGBoost uses a Classification And Regression Tree (CART) as a base classifier, iteratively constructs multiple base classifiers and combines them to improve the prediction accuracy and generalization ability of the model. The core idea is to gradually optimize the model by continuously adding decision trees, and to fit the residual of the previous round of prediction to gradually reduce the error of the prediction model. XGBoost minimizes the objective function by gradient descent method, thereby gradually optimizing the model parameters. In addition, in the process of constructing the decision tree, XGBoost adopts a gradient boosting strategy, which can better capture the non-linear relationships and complex patterns in the data.

[0159] Therefore, the fault identification model is built by XGBoost, and the model parameters are optimized, as follows:

[0160] For a given data set D = {(x i ,y i )}, with n samples and m features, the prediction result of the fault identification model after the tth iteration training is:

[0161]

[0162] In the formula, is the prediction result of the data set in the tth iteration, f t (x i ) is the prediction result of the tth classification and regression tree.

[0163] Using the second-order Taylor expansion can more accurately capture the local properties of the loss function of the model, thereby providing more detailed gradient information in the optimization process. Therefore, the second-order Taylor expansion is used as the objective function of the fault identification model to quantify the loss residual obj (t) of the optimized model, as shown in the following formula:

[0164]

[0165] In the formula, is the loss function, y i is the i th sample data value, is the regularization term; g i , h i is the regularization term, which is used to control the model complexity to prevent overfitting, g i is the first-order derivative of the loss function, and h i is the second-order derivative.

[0166] In each iteration, according to the calculated gradient information g i and the second-order derivative information h i , the objective function is then optimized to determine the decision tree f t (x) that should be added in this round. Finally, all the decision trees generated in the iteration process are accumulated to build a complete fault identification model.

[0167] The complexity of the decision tree Ω is usually measured by considering the depth of the tree, the number of internal nodes, the leaf node data, and the proportion of the leaf node in the whole, etc. The expression is as follows:

[0168]

[0169] In the formula, γ and λ are both regularization factors, is the weight vector of node j, T t is the number of leaf nodes of decision tree t.

[0170] After constructing a complete fault identification model through the XGBoost decision tree described above, in order to further improve the performance of the model and obtain more accurate fault identification results, the exponential distribution identifier is introduced to optimize the hyperparameters of the fault identification model in the embodiment of the application, so that the model can better adapt to data fluctuations and complexity, still exhibit high accuracy in the case of connecting new energy to the power distribution network, and show good stability in the fault identification process, and can adapt to the complex and changeable power distribution network environment.

[0171] The exponential distribution identifier provided by the embodiment of the application is constructed based on the EDO algorithm, and on this basis, the exponential distribution identifier is adjusted by introducing the positive sine disturbance strategy and the Gaussian variation strategy, so that it has faster convergence speed and global search capability.

[0172] The EDO algorithm is full name for Exponential Distribution Optimizer, which is used to solve complex optimization problems, so that a group of "individuals" constantly explore in the search space, and each individual updates its own position according to the random characteristics of the exponential distribution, and gradually approaches the global optimal solution.

[0173] Based on the above characteristics, the hyperparameters of the fault identification model are processed by using the exponential distribution identifier, the global optimal solution is found for these hyperparameters, and then the fault identification model is optimized, so as to effectively improve the model performance.

[0174] Among them, the exponential distribution optimizer mainly optimizes the learning rate, decision tree, sample random sampling parameter and sample column sampling parameter of the fault identification model. The decision tree includes the maximum tree depth, the pruning parameter, and the minimum sample number of the leaf node.

[0175] Therefore, the exponential distribution optimizer first generates a plurality of random solutions as an initial population based on the exponential distribution characteristics, and constructs an exponential distribution model for the initial population. Among them, the coordinate information of the random solution is related to the parameters of the exponential distribution model, so the exponential distribution optimizer updates the position of the random solution by updating the exponential distribution model. The random solutions here have great difference between each other and are randomly generated.

[0176] The exponential distribution optimizer needs to go through two processes to find the global optimal solution, the development stage and the exploration stage. These two stages are mainly used to iteratively update the position of the random solution until the global optimal solution that meets the requirements is obtained. The embodiments of the present application set a random variable a to distinguish the two stages, and the value range of the random variable a is [0, 1]. When a is less than 0.5, the exponential distribution model is in the development stage, and when a is greater than or equal to 0.5, the exponential distribution model is in the exploration stage.

[0177] In the development stage and the exploration stage, the coordinate information of each random solution in the initial population is regarded as the parameter of the exponential distribution model, and the exponential distribution of the parameter is used to reflect the change and trend of the search space. The variable distribution of the random solution is:

[0178]

[0179] Xwinners i,j = lb + rand (ub-lb) ;

[0180] In the formula, Xwinners is the initial population, Xwinners i,j is the i-th row and j-th column random solution, n and d are the dimensions of the initial population; ub and lb are the upper and lower boundaries of the iterative update set, and rand is a random number in [0, 1].

[0181] When the set a is less than 0.5, the exponential distribution model enters the development stage. The development stage mainly approaches the global optimal solution through directional search, and at the same time, in order to ensure that the entire search range can be covered, an attraction mechanism is used to guide the poor performing solution to its adjacent area to expand the search range, and the search center is concentrated around the guide solution, until the global optimal solution is obtained. In this process, the guide solution is dynamically updated, and the average value of the top three optimal solutions in the current global is taken as the new guide solution. The expression of the guide solution is as follows:

[0182]

[0183] In the formula, Xguide t is the guide solution after the t-th iterative update, is the current calculated global optimal solution.

[0184] The guide solution is the current determined historical global optimal solution in the initial population, and when a new global optimal solution is found in the initial population through iterative updating, the original global optimal solution will be used as the guide solution. In the iterative updating process, other individuals in the initial population will perform local fine search around the guide solution.

[0185] In the development stage, a memoryless attribute matrix memoryless is also constructed to simulate an exponential distribution to record newly generated solutions in the initial population. The goal of this matrix is to use the current found guiding solution as a basis to explore the global optimal solution in the neighborhood.

[0186] When the initial population after the tth iteration update is equal to the memoryless attribute matrix, the update process is centered on the guiding solution to explore the purpose of finding better solutions to guide the subsequent population development. Conversely, if the condition is not met, the poor solution will converge to the local optimal solution, which is expressed as follows:

[0187]

[0188] In the formula, is the ith solution of the memoryless attribute matrix, φ is a random number in the interval [0, 1], is the connection number from the current time to the ith local optimal solution, a is an adaptive parameter, b is an expanded search space, f is a random number in the interval [-1, 1], and σ is the variance of the random solution.

[0189] The mean μ and variance σ of the random solution are expressed as:

[0190]

[0191] In the formula, λ' is an exponential decay parameter, which is inversely proportional to the mean and variance values.

[0192] During the iterative update, the initial population presents a relatively dispersed distribution in the early stage, and the aggregation between solutions is not obvious. However, when the iterative process enters the middle and late stages, with the continuous increase of the number of iterations, the initial population gradually converges, and the difference between individuals decreases, resulting in a rapid decrease in diversity. Not only does this limit the search ability of the algorithm, but it can also cause the algorithm to prematurely converge to a certain local optimal solution and fail to continue exploring other potential global optimal solutions.

[0193] In order to solve the problems of insufficient global search ability and lack of solution set diversity in the EDO algorithm, the sine-cosine perturbation strategy and the Gaussian mutation strategy are used to update the initial population in the development stage. The sine-cosine perturbation strategy aims to effectively interfere with the update of the initial population, and the purpose is to broaden the search range that a single random solution can explore in the iterative process, so as to alleviate the phenomenon of premature or excessive aggregation of the initial population.

[0194] The core of the sine-cosine perturbation strategy is to add a nonlinear factor for regulation. In the initial stage of algorithm operation, the factor makes the generated perturbation amplitude larger, which helps to conduct extensive search. As the algorithm iteration proceeds, the value of the factor gradually decreases and the perturbation intensity weakens, that is, the value of the nonlinear factor gradually decreases with the increase of the number of iteration optimization. Thus, a dynamic adjustment process from large to small is generated, which can establish a balance between global exploration and local development of the algorithm, and finally improve the search and convergence performance of the algorithm.

[0195] Specifically, the nonlinear factor is combined with the periodic variation characteristics of the sine and cosine functions to expand the search range of each random solution to generate a new random solution, thereby expanding the initial population.

[0196] The expanded initial population by the sine-cosine perturbation strategy is specifically expressed as:

[0197]

[0198] V i t+1 For the expanded initial population, t is the number of iteration optimization, a is an adaptive parameter, b is the expanded search range, is the ith solution of the memoryless matrix, is the connection number from the current time to the ith local optimal solution, φ is a random number in the interval [-1, 1], f is a random number in the interval [0, 1], r and r1 are both random numbers in the interval [0, 2π], w is the nonlinear factor, and σ is the variance of the random solution.

[0199] The EDO algorithm generates new solutions by simulating the characteristics of exponential distribution, which can effectively explore the solution space in the initial stage, but as the iteration proceeds, its search behavior gradually becomes conservative and tends to fine-tune in the neighborhood of the better solution found, which leads to the EDO algorithm easily falling into a local optimal solution in the development stage.

[0200] As an improvement of the above scheme, the embodiment of the present application introduces a Gaussian mutation strategy, so that the initial population can produce certain random changes in each iteration, which can further expand the search range of the algorithm and avoid the problem of premature convergence. At the same time, due to the characteristics of Gaussian distribution, the addition of Gaussian distributed random variables will not have a great impact on the overall trend of the initial population, thereby ensuring the stability of the algorithm in the macroscopic aspect. Its expression is as follows:

[0201]

[0202] In the formula, V i t+1 For the expanded initial population, M tis the average value of the initial population, c is an adjustment parameter, Z1, Z2 are M t the surrounding area vector, Gauus (0, 1) is a Gaussian distribution random variable.

[0203] Therefore, by introducing the cosine perturbation strategy and Gaussian mutation strategy, the convergence speed and global search ability of the exponential distribution optimizer are improved. Therefore, the method for constructing the fault identification model in the embodiment of the application is different from the conventional EDO and XGBoost algorithm, but is an IEDO-XGBoost algorithm.

[0204] When the set alpha is greater than or equal to 0.5, the exponential distribution model enters the exploration stage. The purpose of the exploration stage is to locate the area with the global optimal solution in the search range, so as to provide high-quality hyperparameters for the subsequent optimization process. By continuously updating the position of the solution, the dispersion degree of the random solution is analyzed by using the memoryless characteristic of the exponential distribution, and the search direction and step are determined. The search strategy can be flexibly adjusted in the exploration process, so that the robustness can be maintained in the complex and multi-modal optimization problem. In this process, the update expression of the initial population is as follows:

[0205]

[0206] In the formula, V i t+1 is the initial population after expansion, M t is the average value of the initial population, c is an adjustment parameter, Z1, Z2 are M t the surrounding area vector; D1 and D2 are M t the distance between Xwinners randl and Xwinners rand2 .

[0207] After generating a new random solution for the initial population, it is verified whether it meets the given boundary condition. Then the solution is recorded and stored in a matrix that does not retain historical information, and the solution in the initial population is updated using a greedy strategy, and the newly generated solution is further adjusted and optimized. If the fitness function of the new solution is better, the optimal solution in the initial population is updated. Otherwise, the algorithm will guide it to a reference solution that performs better, the purpose is to explore whether there is a solution that exceeds the current best individual. With the advancement of the whole optimization process, all solutions in the population will be highly concentrated around the global optimal solution.

[0208] In summary, by using random variables to determine whether to use the development phase or the exploration node to iterate the exponential distribution model to update the initial population, the exponential distribution optimizer outputs the global optimal solution when the maximum number of iterations is reached. The global optimal solution is then used to optimize the aforementioned fault identification model based on XGBoost, thereby updating the model's hyperparameters.

[0209] Specifically, by integrating the exponential distribution optimizer with the fault identification model, the hyperparameters of the fault identification model are optimized, thereby improving the accuracy of power distribution network defect identification.

[0210] Implementing the embodiments of this application has the following beneficial effects:

[0211] This application first extracts fault-related feature data from the electrical data of the distribution network using wavelet transform for subsequent fault identification. To improve the accuracy of the fault identification model, an exponential distribution optimizer is introduced to optimize the parameters related to model performance, enabling the model to quickly identify distribution network faults in complex and ever-changing environments. Furthermore, to further improve the model's convergence speed and global search capability, sine and cosine perturbation strategies and Gaussian mutation strategies are employed to improve the exponential distribution optimizer. Finally, a high-performance fault identification model processes the fault feature data to quickly and accurately determine the fault status of the distribution network.

[0212] Furthermore, Figure 5 This is a structural diagram of a terminal device provided in one embodiment of this application. Figure 5 As shown, the terminal device 3 of this embodiment includes: at least one processor 30 (in... Figure 5 The present invention includes only one of the following: a memory 31 and a computer program 32 stored in the memory 31 and executable on the at least one processor. When the processor 30 executes the computer program 32, it can implement the steps of the XGBoost-based new energy distribution network fault identification method according to any one of the embodiments of this application.

[0213] The terminal device 3 may be a computing device such as a desktop computer, a cloud server, or a laptop computer, and the computing device may include, but is not limited to, a processor 30 and a memory 31. Figure 5 This is merely an example of terminal device 3 and does not constitute a limitation on terminal device 3. It may include more or fewer components than those shown in the figure.

[0214] This application provides a storage medium that stores computer-readable program code. When the computer-readable program code is executed, it implements the steps of the above-described method for identifying faults in a new energy distribution network based on XGBoost.

[0215] The above-described specific embodiments have further detailed the purposes, technical solutions and beneficial effects of the present application. It should be understood that the above-described specific embodiments are merely for the purpose of illustrating the present application, and are not intended to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A fault identification method for new energy distribution networks based on XGBoost, characterized in that, include: By performing frequency domain transformation on the collected electrical data of the distribution network using wavelet transform, fault characteristic data of the distribution network can be obtained. Based on a preset exponential distribution optimizer, with the goal of optimizing model performance, the parameters of a preset XGBoost-based fault identification model are optimized by simulating the characteristics of an exponential distribution; wherein, the exponential distribution optimizer is adjusted based on a sine and cosine perturbation strategy and a Gaussian mutation strategy. Based on the optimized fault identification model and the fault feature data, the fault types of the distribution network are identified, and the fault identification results are output.

2. The fault identification method for new energy distribution networks based on XGBoost according to claim 1, characterized in that, The process of performing frequency domain transformation on the collected electrical data of the distribution network using wavelet transform to obtain fault characteristic data of the distribution network is as follows: The distribution network electrical data is decomposed into multiple scales in the frequency domain by wavelet transform, and the wavelet coefficients of the distribution network electrical data in each frequency band are extracted. Based on the wavelet coefficients, local features of the electrical data of the distribution network in each frequency band are identified and extracted, and then the extraction results are normalized to obtain the fault feature data; wherein, the local features are related to the fault type of the distribution network.

3. The fault identification method for new energy distribution networks based on XGBoost according to claim 1, characterized in that, The exponential distribution optimizer is adjusted based on a sine / cosine perturbation strategy and a Gaussian mutation strategy, specifically as follows: An initial population is constructed by randomly generating several random solutions, and an exponential distribution model is constructed for the initial population; wherein the coordinate information of the random solutions is related to the parameters of the exponential distribution model; The exponential distribution model is configured with a development phase and an exploration phase. When the preset random variable is less than a first threshold, the exponential distribution model is in the development phase; when the random variable is greater than or equal to the first threshold, the exponential distribution model is in the exploration phase. Based on the random variables, the exponential distribution model is iteratively optimized using sine and cosine perturbation strategies and Gaussian mutation strategies during the exploration and development phases until the maximum number of iterations is reached, thus obtaining the exponential distribution optimizer.

4. The fault identification method for new energy distribution networks based on XGBoost according to claim 3, characterized in that, The iterative optimization of the exponential distribution model during the exploration and development phases using sine and cosine perturbation strategies and Gaussian mutation strategies specifically involves: During the exploration phase, the degree of dispersion of the random solutions in the initial population is analyzed by the exponential distribution model to determine the search direction, and the global optimal solution is found in the initial population according to the search direction; During the development phase, a local optimum is found in the initial population through directional search. Based on the local optimum and the search range of the initial population, a new random solution is generated through a sine and cosine perturbation strategy and a Gaussian mutation strategy to update the initial population. The generation of a new random solution stops when a global optimum is found in the updated initial population. The exponential distribution model is optimized based on the global optimal solution.

5. The fault identification method for new energy distribution networks based on XGBoost according to claim 3, characterized in that, The process of generating new random solutions using sine and cosine perturbation strategies and Gaussian mutation strategies to update the initial population specifically involves: A nonlinear factor is set, and combined with the periodic variation characteristics of sine and cosine functions, the search range is expanded through the nonlinear factor to generate new random solutions, thereby expanding the initial population; wherein, as the number of iterations increases, the value of the nonlinear factor gradually decreases; A Gaussian distributed random variable is set, and the initial population is expanded through the Gaussian distributed random variable in each iteration of optimization.

6. The fault identification method for new energy distribution networks based on XGBoost according to claim 5, characterized in that, The expansion of the initial population specifically involves: The initial population expanded using the sine and cosine perturbation strategy is specifically expressed as follows: In the formula, V i t+1 The initial population is defined as follows: t is the number of iterations, a is the adaptive parameter, and b is the expanded search range. The i-th solution to the memoryless matrix. φ is the number of connections from the current time to the i-th local optimum, φ is a random number in the interval [-1,1], f is a random number in the interval [0,1], r and r1 are both random numbers in the interval [0,2π], w is a nonlinear factor, and σ is the variance of the random solution. The initial population expanded using the Gaussian mutation strategy is specifically expressed as follows: In the formula, V i t+1 For the expanded initial population, M t The initial population mean is given by c, the adjustment parameter is given by Z1 and Z2, and M is given by M. t The surrounding region vector, Gauus(0,1), is a Gaussian distributed random variable.

7. The fault identification method for new energy distribution networks based on XGBoost according to claim 1, characterized in that, The step involves optimizing the parameters of a pre-defined XGBoost-based fault identification model using a pre-defined exponential distribution optimizer, with the goal of improving model performance. Specifically, this optimization is achieved by simulating the characteristics of an exponential distribution. The hyperparameters of the fault identification model are optimized by fusing the exponential distribution optimizer with the fault identification model. During the fusion process, the learning rate, decision tree, random sampling parameters, and sample column sampling parameters of the fault identification model are optimized by the exponential distribution optimizer; wherein, the decision tree includes the maximum tree depth, pruning parameters, and the minimum number of samples in the leaf nodes.

8. A fault identification method for new energy distribution networks based on XGBoost, characterized in that, include: Feature extraction module, parameter optimization module, and fault identification module; Among them, the feature extraction module is used to perform frequency domain transformation on the collected electrical data of the distribution network through wavelet transform to obtain the fault feature data of the distribution network; The parameter optimization module is used to optimize the parameters of a preset XGBoost-based fault identification model by simulating the characteristics of an exponential distribution, with the goal of improving model performance, based on a preset exponential distribution optimizer. The exponential distribution optimizer is adjusted based on a sine and cosine perturbation strategy and a Gaussian mutation strategy. The fault identification module is used to identify the fault type of the distribution network based on the optimized fault identification model and the fault feature data, and output the fault identification result.

9. A terminal device, characterized in that, It includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the steps of the XGBoost-based new energy distribution network fault identification method according to any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores computer-readable program code, which, when executed, implements the steps of the XGBoost-based new energy distribution network fault identification method according to any one of claims 1 to 7.

Images