Power distribution network fault partition positioning method, system and device and storage medium

Abstract

The invention discloses a power distribution network fault partition positioning method, system and device and a storage medium, and the method comprises the steps: constructing a data set sample according to node feature data after a power distribution network breaks down, and dividing the data set sample into a model offline training sample and a test sample; obtaining feature importance of data in the data set samples by using a principal component analysis method, selecting related variables, and establishing important feature variable data samples; performing model training on the plurality of machine learning models, and outputting training sample and test sample data prediction results of different machine learning models; training the element classifier, and obtaining a power distribution network fault partition positioning prediction result through the element classifier; and combining a plurality of trained machine learning models with the meta-classifiers to obtain a power distribution network fault partition positioning model, actually collecting node feature data after the power distribution network has a fault, inputting the node feature data into the power distribution network fault partition positioning model, and judging whether the power distribution network has a fault or not and a specific fault position. According to the invention, the fault positioning accuracy can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of distribution network dispatching operation and fault handling, and specifically relates to a distribution network fault zoning positioning method, system, equipment and storage medium. Background Art

[0002] The distribution network is an important link between the transmission network and users. Due to its numerous branches and complex structure, it is prone to failure. Rapidly and accurately locating distribution network faults has become a key research task to improve power grid security and power supply reliability.

[0003] Currently available fault location methods include fault line selection, fault section location, and precise fault location. Fault line selection methods generally include techniques based on transient information, steady-state information, and externally injected current. However, since fault line selection methods can only locate the fault on the feeder where the fault occurred, location accuracy is relatively low, creating further challenges for subsequent troubleshooting. Fault section location methods include matrix methods and artificial intelligence methods. Furthermore, matrix methods use fault information uploaded by feeder terminal units, combined with switch node relationships and fault information characteristics, to calculate a fault discriminant matrix to locate the fault. However, matrix methods suffer from slow computational speed and limited fault tolerance. Artificial intelligence methods construct a fault knowledge model based on historical fault data and train the model to perform fault location. This avoids tedious matrix calculations, enables rapid and accurate fault resolution in complex distribution networks, and offers strong fault tolerance. Artificial intelligence methods primarily include expert systems, fuzzy set methods, Petri nets, and Bayesian networks. Accurate fault location methods include impedance methods, traveling wave methods, and signal injection methods. These methods are all distance-measuring algorithms that can accurately calculate fault distances. Their principles are simple and easy to use, but they suffer from large calculation errors when locating faults in distribution networks with complex branch lines, making it difficult to pinpoint the fault location. Traditional distribution network fault location methods are only feasible in theory but have significant limitations in practical applications. Compared to traditional fault location methods, artificial intelligence algorithms can rapidly process complex data and exhibit good fault tolerance, contributing to improved power supply reliability. However, existing distribution network fault location methods cannot accurately determine the fault segment, resulting in large errors in fault location assessment, which affects the efficiency and accuracy of line status investigation by dispatchers. Relying solely on a single fault location model may be unable to comprehensively and accurately address distribution network fault location due to its inherent limitations, resulting in unstable performance in different scenarios. Furthermore, the lack of appropriate integration techniques to combine the strengths of multiple models makes it difficult to fully leverage the potential of different models in fault location, thus limiting the effectiveness of fault location. Existing fault location methods cannot provide dispatching personnel with sufficiently accurate information to assist them in troubleshooting the status of distribution network lines, resulting in inaccurate and inefficient troubleshooting. Summary of the Invention

[0004] The purpose of the present invention is to address the problems in the above-mentioned prior art and provide a distribution network fault zoning location method, system, device and storage medium. The method adopts an integrated learning method to combine the prediction results of multiple base models to construct a higher-level meta-classifier model for final prediction. It can fully utilize the advantages of different base models, improve the accuracy and stability of the prediction, accurately evaluate the fault location of the distribution network, and assist dispatching personnel to more accurately check the line status of the distribution network.

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

[0006] In a first aspect, a method for locating a distribution network fault zone is provided, comprising:

[0007] Construct a data set sample based on the node characteristic data after a distribution network failure;

[0008] Use principal component analysis to obtain feature importance for the data in the dataset sample, select relevant variables based on feature importance, and form a data sample of important feature variables;

[0009] Use important feature variable data samples to train multiple pre-established machine learning models, obtain multiple trained machine learning models, and output the prediction results of training sample data and test sample data of different machine learning models;

[0010] Using the prediction results of the training sample data and the prediction results of the test sample data to train the pre-established meta-classifier to obtain a trained meta-classifier;

[0011] The trained multiple machine learning models are combined with the meta-classifier to obtain the distribution network fault zoning location model. The node feature data after the distribution network fault occurs is actually collected and input into the distribution network fault zoning location model to determine whether the distribution network fault occurs and output the fault location information.

[0012] As a preferred solution, in the step of constructing a data set sample based on the node feature data after a distribution network fault occurs, a fault point is set in the distribution network simulation system, and the effective values ​​of the three-phase voltages in all stages after the line fault occurs are used as node feature data to construct a data set sample; based on the data set sample, the effective values ​​of the three-phase voltages in all stages under the fault state of different sections of the distribution network are used to establish sample set data features, and the sample set data labels are established based on whether a fault occurs in the corresponding operating state and the fault section number, where 0 indicates no fault occurs and 1 indicates a fault occurs; the data set samples are randomly split into model offline training samples and test samples according to a set ratio.

[0013] As a preferred solution, the steps of using principal component analysis to obtain feature importance from the data in the dataset sample, selecting relevant variables based on feature importance, and forming a data sample of important feature variables include:

[0014] Through data standardization processing, each indicator value is converted into a standardized indicator, and the sample standard deviation and mean are calculated. The calculation expression is as follows:

[0015]

[0016] In the formula, b is the evaluation object of the sample data, represents the normalized index after transformation, S k and are the sample standard deviation and mean of the kth indicator respectively;

[0017] The correlation coefficient matrix W is calculated according to the following expression:

[0018] W=(w ik ) a×b

[0019]

[0020] In the formula, a is the characteristic variable of the sample data, w ik represents the correlation coefficient between the i-th indicator and the k-th indicator, w ik =1;

[0021] Calculate the eigenvalues ​​and eigenvectors and construct the characteristic variables according to the following expressions:

[0022]

[0023] Where, is the correlation coefficient in the correlation coefficient matrix W, as the eigenvalue, v ba is the corresponding eigenvector;

[0024] z a is the ath principal component, z1, z2, ..., z a Constitute characteristic variables;

[0025] Calculate the comprehensive evaluation value according to the following formula, determine the important variables and form the important characteristic variable data sample:

[0026]

[0027] Where c k and β s Represent the information contribution rate and cumulative contribution rate of the principal component, r k is the information contribution of principal component k, r λ is the information contribution of the principal component λ, z k is the kth characteristic variable; when β s When it is close to 1, the first s variables are selected to conduct a comprehensive analysis of the principal components, and s important characteristic variables are selected according to the comprehensive score value Y to form an important characteristic variable data sample.

[0028] As a preferred solution, the steps of using important feature variable data samples to perform model training on multiple pre-established machine learning models, obtaining multiple trained machine learning models, and outputting prediction results of training sample data and test sample data of different machine learning models, the pre-established machine learning models include:

[0029] Set the maximum depth D of the extreme gradient boosting model XGBoost and build the XGBoost model;

[0030] Select the radial basis function (RBF) kernel as the kernel function of the support vector machine, set the hyperparameters, and build the support vector machine model:

[0031]

[0032] Where R(a,b) is the expression of the radial basis function, and η is the hyperparameter of the radial basis function;

[0033] Set the maximum depth D of the decision tree and build a random forest model.

[0034] As a preferred solution, the step of training a pre-established meta-classifier using the prediction results of the training sample data and the prediction results of the test sample data to obtain the trained meta-classifier includes:

[0035] The training samples are input into the XGBoost model, support vector machine model and random forest model respectively, and the prediction results of the training sample data of the three machine learning models are output as Y1, Y2 and Y3; the test samples are input into the XGBoost model, support vector machine model and random forest model respectively, and the prediction results of the test sample data of the three machine learning models are output as T1, T2 and T3;

[0036] The training sample data prediction results Y1, Y2, Y3 and the training set data label Yt are combined into the data set R as follows:

[0037] n{[Y1][Y2][Y3][Yt]→n{[W][Yt]

[0038] In the formula, the correlation coefficient matrix W is obtained by predicting the results Y1, Y2, and Y3 from the training sample data;

[0039] Combine the test sample data prediction results T1, T2, and T3 into the data set S as follows:

[0040] n{[T1][T2][T3]

[0041] The meta-classifier is trained on the dataset R, and the dataset S is input into the trained meta-classifier to obtain the fault location prediction result.

[0042] As a preferred solution, the steps of combining the trained multiple machine learning models with the meta-classifier to obtain a distribution network fault partition location model, actually collecting node feature data after a distribution network fault occurs, inputting the data into the distribution network fault partition location model, judging whether a fault has occurred in the distribution network and outputting the fault location information, reading the actual distribution network node voltage measurement data, inputting the distribution network node voltage measurement data into the distribution network fault partition location model, judging whether a fault has occurred, and if a fault is judged to have occurred, the distribution network fault partition location model outputs the fault location information.

[0043] In a second aspect, a distribution network fault zoning location system is provided, comprising:

[0044] The sample building module is used to build a data set sample based on the node characteristic data after a distribution network failure occurs;

[0045] The principal component analysis module is used to obtain the feature importance of the data in the dataset sample using the principal component analysis method, select relevant variables based on the feature importance, and form a data sample of important feature variables;

[0046] The machine learning model training module is used to use important feature variable data samples to train multiple pre-established machine learning models, obtain multiple trained machine learning models, and output the prediction results of training sample data and test sample data of different machine learning models;

[0047] A meta-classifier training module is used to train a pre-established meta-classifier using the prediction results of the training sample data and the prediction results of the test sample data to obtain a trained meta-classifier;

[0048] The fault location module is used to combine multiple trained machine learning models with meta-classifiers to obtain a distribution network fault partition location model. It actually collects node feature data after a distribution network fault occurs, inputs the distribution network fault partition location model, determines whether a distribution network fault occurs, and outputs the fault location information.

[0049] As a preferred solution, the sample establishment module sets a fault point in the distribution network simulation system, takes the effective values ​​of the three-phase voltage in all stages after the line fault occurs as node feature data, and constructs a data set sample; based on the data set sample, the effective values ​​of the three-phase voltage in all stages under the fault state of different sections of the distribution network are used to establish the sample set data features, and establishes the sample set data label according to whether a fault occurs in the corresponding operating state and the fault section number, where 0 indicates that no fault has occurred and 1 indicates that a fault has occurred; the data set samples are randomly split into model offline training samples and test samples according to a set ratio.

[0050] As a preferred solution, the principal component analysis module converts each index value into a standardized index through data standardization processing, and calculates the sample standard deviation and mean. The calculation expression is as follows:

[0051]

[0052] In the formula, b is the evaluation object of the sample data, represents the normalized index after transformation, S k and are the sample standard deviation and mean of the kth indicator respectively;

[0053] The correlation coefficient matrix W is calculated according to the following expression:

[0054] W=(w ik ) a×b

[0055]

[0056] In the formula, a is the characteristic variable of the sample data, w ik represents the correlation coefficient between the i-th indicator and the k-th indicator, w ik =1;

[0057] Calculate the eigenvalues ​​and eigenvectors and construct the characteristic variables according to the following expressions:

[0058]

[0059] Where, is the correlation coefficient in the correlation coefficient matrix W, as the eigenvalue, v ba is the corresponding eigenvector;

[0060] z a is the ath principal component, z1, z2, ..., z a Constitute characteristic variables;

[0061] Calculate the comprehensive evaluation value according to the following formula, determine the important variables and form the important characteristic variable data sample:

[0062]

[0063] Where c k and β s Represent the information contribution rate and cumulative contribution rate of the principal component, r k is the information contribution of principal component k, r λ is the information contribution of the principal component λ, z k is the kth characteristic variable; when β sWhen it is close to 1, the first s variables are selected to conduct a comprehensive analysis of the principal components, and s important characteristic variables are selected according to the comprehensive score value Y to form an important characteristic variable data sample.

[0064] As a preferred solution, the machine learning model training module uses important feature variable data samples to perform model training on multiple pre-established machine learning models, obtains multiple trained machine learning models, and outputs the training sample data prediction results and test sample data prediction results of different machine learning models. The pre-established machine learning models include:

[0065] Set the maximum depth D of the extreme gradient boosting model XGBoost and build the XGBoost model;

[0066] Select the radial basis function (RBF) kernel as the kernel function of the support vector machine, set the hyperparameters, and build the support vector machine model:

[0067]

[0068] Where R(a,b) is the expression of the radial basis function, and η is the hyperparameter of the radial basis function;

[0069] Set the maximum depth D of the decision tree and build a random forest model.

[0070] As a preferred solution, the meta-classifier training module inputs the training samples into the XGBoost model, the support vector machine model and the random forest model respectively, and outputs the training sample data prediction results Y1, Y2 and Y3 of the three machine learning models; inputs the test samples into the XGBoost model, the support vector machine model and the random forest model respectively, and outputs the test sample data prediction results T1, T2 and T3 of the three machine learning models;

[0071] The training sample data prediction results Y1, Y2, Y3 and the training set data label Yt are combined into the data set R as follows:

[0072] n{[Y1][Y2][Y3][Yt]→n{[W][Yt]

[0073] In the formula, the correlation coefficient matrix W is obtained by predicting the results Y1, Y2, and Y3 from the training sample data;

[0074] Combine the test sample data prediction results T1, T2, and T3 into the data set S as follows:

[0075] n{[T1][T2][T3]

[0076] The meta-classifier is trained on the dataset R, and the dataset S is input into the trained meta-classifier to obtain the fault location prediction result.

[0077] As a preferred solution, the fault location module reads the actual distribution network node voltage measurement data, inputs the distribution network node voltage measurement data into the distribution network fault partition location model, determines whether a fault occurs, and if it is determined that a fault occurs, the distribution network fault partition location model outputs the fault location information.

[0078] In a third aspect, an electronic device is provided, comprising a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the distribution network fault zoning location method.

[0079] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the distribution network fault partition location method is implemented.

[0080] Compared with the prior art, the first aspect of the present invention has at least the following beneficial effects:

[0081] By using principal component analysis to obtain feature importance from the data in the model's offline training samples and test samples, relevant variables are selected based on feature importance, important feature variable data samples are formed, and model training is performed on multiple pre-established machine learning models. The prediction results of the training sample data and the prediction results of the test sample data of different machine learning models are output. The present invention integrates different machine learning models to complete the segment location of the fault, which can accurately evaluate the fault location of the distribution network and improve the accuracy of fault location. The use of principal component analysis greatly reduces the computational cost, is easy to understand, and is also easier to obtain data samples that are easy to use. The present invention uses the prediction results of the training sample data and the test sample data to train a meta-classifier, and obtains the distribution network fault partition location prediction results through the meta-classifier. The meta-classifier integrates the basic learning algorithm through ensemble learning to obtain a distribution network fault partition location model with better performance and better prediction effect. The present invention can assist dispatching and operation personnel to more accurately check the line status of the distribution network, and has great engineering application value and promotion prospects.

[0082] Furthermore, the pre-established machine learning models of the present invention include an XGBoost model, a support vector machine model, and a random forest model. The radial basis RBF kernel is selected as the kernel function of the support vector machine. The radial basis RBF kernel can map samples to a higher-dimensional space and can process samples when the relationship between class labels and features is nonlinear.

[0083] It can be understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0085] Figure 1 A schematic flow chart of a method for locating a distribution network fault zone according to an embodiment of the present invention;

[0086] Figure 2 Wiring diagram of the IEEE 34-node power distribution system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0087] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0088] See also Figure 1 , an embodiment of the present invention proposes a distribution network fault zoning location method, comprising:

[0089] S1, constructing a data set sample based on the node characteristic data after the distribution network fails;

[0090] S2. Use principal component analysis to obtain feature importance for the data in the dataset sample, select relevant variables based on feature importance, and form a data sample of important feature variables;

[0091] S3. Use important feature variable data samples to perform model training on multiple pre-established machine learning models, obtain multiple trained machine learning models, and output prediction results of training sample data and test sample data for different machine learning models;

[0092] S4. Using the prediction results of the training sample data and the prediction results of the test sample data, a pre-established meta-classifier is trained to obtain a trained meta-classifier;

[0093] S5. Combine the trained multiple machine learning models with the meta-classifier to obtain a distribution network fault zoning location model. Actually collect node feature data after the distribution network fault occurs, input the data into the distribution network fault zoning location model, determine whether the distribution network fault occurs and output the fault location information.

[0094] In one possible implementation, step S1 sets a fault point in the distribution network simulation system and calculates the system's operating status after the fault. The effective three-phase voltage values ​​at all stages after the line fault occur are used as node features of the distribution network system and stored as data to construct a dataset sample. When dividing the model's offline training and testing samples, based on the dataset samples, the effective three-phase voltage values ​​at all stages under fault conditions in different sections of the distribution network are used to establish sample set data features. Sample set data labels are established based on whether a fault has occurred in the corresponding operating state and the fault section number, where 0 indicates no fault has occurred and 1 indicates a fault has occurred. The dataset samples are randomly split into model offline training and testing samples in a ratio of 3:7.

[0095] In one possible implementation, step S2 performs standardization processing on the original data, converts each indicator value into a standardized indicator, and calculates the sample standard deviation and mean. The calculation expression is as follows:

[0096]

[0097] In the formula, b is the evaluation object of the sample data, represents the normalized index after transformation, S k and are the sample standard deviation and mean of the kth indicator respectively; It is a standardized indicator variable.

[0098] The correlation coefficient matrix W is calculated according to the following expression:

[0099] W=(w ik ) a×b

[0100]

[0101] In the formula, a is the characteristic variable of the sample data, w ik represents the correlation coefficient between the i-th indicator and the k-th indicator, w ik =1;

[0102] Calculate the eigenvalues ​​and eigenvectors and construct the characteristic variables according to the following expressions:

[0103]

[0104] Where, is the correlation coefficient in the correlation coefficient matrix W, as the eigenvalue, v ba is the corresponding eigenvector;

[0105] z a is the ath principal component, z1, z2, ..., z a Constitute characteristic variables;

[0106] Calculate the comprehensive evaluation value according to the following formula, determine the important variables and form the important characteristic variable data sample:

[0107]

[0108] Where c k and β s Represent the information contribution rate and cumulative contribution rate of the principal component, r k is the information contribution of principal component k, r λ is the information contribution of the principal component λ, z k is the kth characteristic variable; when β s When it is close to 1, the first s variables are selected to conduct a comprehensive analysis of the principal components, and s important characteristic variables are selected according to the comprehensive score value Y to form an important characteristic variable data sample.

[0109] In one possible implementation, the machine learning model pre-established in step S3 includes:

[0110] Set the maximum depth D of the extreme gradient boosting model XGBoost and build the XGBoost model;

[0111] Select the radial basis function (RBF) kernel as the kernel function of the support vector machine, set the hyperparameters, and build the support vector machine model:

[0112]

[0113] Where R(a,b) is the expression of the radial basis function, and η is the hyperparameter of the radial basis function;

[0114] Set the maximum depth D of the decision tree and build a random forest model.

[0115] In one possible implementation, step S4 inputs the training samples into the XGBoost model, the support vector machine model, and the random forest model, respectively, and outputs the prediction results Y1, Y2, and Y3 of the training sample data of the three machine learning models; similarly, inputs the test samples into the XGBoost model, the support vector machine model, and the random forest model, respectively, and outputs the prediction results T1, T2, and T3 of the test sample data of the three machine learning models;

[0116] The training sample data prediction results Y1, Y2, Y3 and the training set data label Yt are combined into the data set R as follows:

[0117] n{[Y1][Y2][Y3][Yt]→n{[W][Yt]

[0118] In the formula, the correlation coefficient matrix W is obtained by predicting the results Y1, Y2, and Y3 from the training sample data;

[0119] Combine the test sample data prediction results T1, T2, and T3 into the data set S as follows:

[0120] n{[T1][T2][T3]

[0121] The meta-classifier is trained on the dataset R, and the dataset S is input into the trained meta-classifier to obtain the fault location prediction result.

[0122] In one possible implementation, step S5 reads actual distribution network node voltage measurement data from the distribution automation system and inputs the distribution network node voltage measurement data into the distribution network fault zoning localization model to determine whether a fault has occurred. If a fault is determined to have occurred, the distribution network fault zoning localization model outputs fault location information.

[0123] The distribution network fault zoning method of this embodiment uses busbar voltage amplitude measurements to predict the location of three-phase grounding faults. The training data includes voltage measurements of all bus phases in the system, and the three-phase measured voltages are used to train a neural network. The training data contains 500 fault location samples, labeled with line names. The method uses an integrated learning algorithm, including an XGBoost model, a support vector machine model, and a random forest model, to accurately assess the location of distribution network faults.

[0124] See also Figure 2 ,by Figure 2 Taking the IEEE 34-node power distribution system wiring diagram shown in FIG as an example, the method for locating a distribution network fault zone according to an embodiment of the present invention includes the following steps:

[0125] S1. Build a distribution system dataset.

[0126] S101. Set a fault point in the distribution network simulation system and prepare 13,000 sample data;

[0127] S102. Taking the effective value of the three-phase voltage at all stages after the line fault occurs as the node feature of the distribution system;

[0128] S103. Store the node characteristics of the power distribution system in the form of data as a data set.

[0129] S2. Build model offline training samples and test samples.

[0130] S201. Construct a sample set based on historical fault data;

[0131] S202. The sample data feature X is the effective value of the three-phase voltage at all stages under fault conditions in different sections of the distribution network, which is a 13000 × 93 matrix.

[0132] S203. The sample set data label Y is whether a fault occurs in the corresponding operating state and the fault segment number, 0 means no fault occurs, 1 means a fault occurs, and Y is a 13000×2 matrix;

[0133] S204. Randomly split the training set and test set into a ratio of 3:7;

[0134] S3. Use principal component analysis (PCA) to obtain feature importance, select relevant variables, and form a new data sample.

[0135] S301. Standardize the original data and convert each indicator value into a standardized indicator. The sample standard deviation and mean are calculated as follows:

[0136]

[0137] In the formula, b is the evaluation object of the sample data, represents the normalized index after transformation, S k and are the sample standard deviation and mean of the kth indicator respectively;

[0138] S302. Calculate the correlation coefficient matrix W according to the following expression:

[0139] W=(w ik ) a×b

[0140]

[0141] In the formula, a is the characteristic variable of the sample data, w ik represents the correlation coefficient between the i-th indicator and the k-th indicator, w ik =1;

[0142] S303. Calculate the eigenvalue and eigenvector, and construct the characteristic variable according to the following expression:

[0143]

[0144] Where, is the correlation coefficient in the correlation coefficient matrix W, as the eigenvalue, v ba is the corresponding eigenvector;

[0145] z a is the ath principal component, z1, z2, ..., z a Constitute characteristic variables;

[0146] S304. Calculate the comprehensive evaluation value according to the following formula, determine the important variables and form the important characteristic variable data sample:

[0147]

[0148] Where c k and β s Represent the information contribution rate and cumulative contribution rate of the principal component, r k is the information contribution of principal component k, r λ is the information contribution of the principal component λ, z k is the kth characteristic variable; when β s When it is close to 1, 10 variables are selected for comprehensive analysis of the principal components. The voltage values ​​of the 10 columns of data, 1, 2, 3, 11, 27, 29, 50, 57, 74, and 89, are selected according to the comprehensive score to form the important characteristic variable data sample.

[0149] S4. Build an integrated learning algorithm model.

[0150] S401. Set the maximum depth D of the extreme gradient boosting model XGBoost and build the XGBoost model;

[0151] S402. Select the radial basis function (RBF) kernel as the kernel function of the support vector machine (SVM) and set the hyperparameters to build the SVM model:

[0152]

[0153] Where R(a,b) is the expression of the radial basis function, and η is the hyperparameter of the radial basis function;

[0154] S403. Set the maximum depth D of the decision tree and build a random forest model.

[0155] S5. Model training based on meta-classifier ensemble learning.

[0156] S501. Input the training samples into the XGBoost model, support vector machine model and random forest model respectively, and output the prediction results Y1, Y2 and Y3 of the training sample data of the three machine learning models;

[0157] The test samples are input into the XGBoost model, support vector machine model and random forest model respectively, and the prediction results of the test sample data of the three machine learning models are output as T1, T2 and T3;

[0158] S502. Combine the training sample data prediction results Y1, Y2, Y3 and the training set data label Yt into a data set R according to the following formula:

[0159] n{[Y1][Y2][Y3][Yt]→n{[W][Yt]

[0160] In the formula, the correlation coefficient matrix W is obtained by predicting the results Y1, Y2, and Y3 from the training sample data;

[0161] S503. Combine the test sample data prediction results T1, T2, and T3 into a data set S according to the following formula:

[0162] n{[T1][T2][T3]

[0163] S504. Train a meta-classifier on the dataset R;

[0164] S505. Input the data set S to the trained meta-classifier to obtain the fault location prediction result.

[0165] S6. Build an integrated learning algorithm model for distribution network fault location deployment and online application.

[0166] S601. Read the distribution network node voltage measurement data from the distribution automation system;

[0167] S602. Input the measurement data into the integrated algorithm model to determine whether a fault has occurred;

[0168] S603. If a fault is determined to have occurred, the integrated model is used to output the fault location information and the corresponding fault location test results, which are three groups of 1300-dimensional vectors. Table 1 shows the accuracy comparison of the integrated model compared with the individual models.

[0169] Table 1

[0170] Algorithm Model Recall Accuracy Model of the present invention 96％ 97％ XGBoost Model 85％ 87％ Support Vector Machine 87％ 86％ Random Forest 87％ 87％

[0171] The present invention provides a distribution network fault zoning location method based on artificial intelligence algorithm integration technology. Based on the stacking method, three models of XGBoost, support vector machine and random forest are used to complete the fault segment location. It can accurately evaluate the fault location of the distribution network and assist dispatching operators to more accurately check the line status of the distribution network.

[0172] Another embodiment of the present invention further provides a distribution network fault zone location system, comprising:

[0173] The sample building module is used to build a data set sample based on the node characteristic data after a distribution network failure occurs;

[0174] The principal component analysis module is used to obtain the feature importance of the data in the dataset sample using the principal component analysis method, select relevant variables based on the feature importance, and form a data sample of important feature variables;

[0175] The machine learning model training module is used to use important feature variable data samples to train multiple pre-established machine learning models, obtain multiple trained machine learning models, and output the prediction results of training sample data and test sample data of different machine learning models;

[0176] A meta-classifier training module is used to train a pre-established meta-classifier using the prediction results of the training sample data and the prediction results of the test sample data to obtain a trained meta-classifier;

[0177] The fault location module is used to combine multiple trained machine learning models with meta-classifiers to obtain a distribution network fault partition location model. It actually collects node feature data after a distribution network fault occurs, inputs the distribution network fault partition location model, determines whether a distribution network fault occurs, and outputs the fault location information.

[0178] In one possible implementation, a sample establishment module sets a fault point in a distribution network simulation system, uses the effective values ​​of the three-phase voltage at all stages after a line fault occurs as node feature data, and constructs a data set sample. Based on the data set sample, the effective values ​​of the three-phase voltage at all stages under fault conditions in different sections of the distribution network are used to establish sample set data features, and establishes sample set data labels based on whether a fault occurs in the corresponding operating state and the fault section number, where 0 indicates no fault occurs and 1 indicates a fault occurs. The data set samples are randomly split into model offline training samples and test samples according to a set ratio.

[0179] In one possible implementation, the principal component analysis module converts each index value into a standardized index through data standardization processing, and calculates the sample standard deviation and mean. The calculation expression is as follows:

[0180]

[0181] In the formula, b is the evaluation object of the sample data, represents the normalized index after transformation, S k and are the sample standard deviation and mean of the kth indicator respectively;

[0182] The correlation coefficient matrix W is calculated according to the following expression:

[0183] W=(w ik ) a×b

[0184]

[0185] In the formula, a is the characteristic variable of the sample data, w ik represents the correlation coefficient between the i-th indicator and the k-th indicator, w ik =1;

[0186] Calculate the eigenvalues ​​and eigenvectors and construct the characteristic variables according to the following expressions:

[0187]

[0188] Where, is the correlation coefficient in the correlation coefficient matrix W, as the eigenvalue, v ba is the corresponding eigenvector;

[0189] z a is the ath principal component, z1, z2, ..., z a Constitute characteristic variables;

[0190] Calculate the comprehensive evaluation value according to the following formula, determine the important variables and form the important characteristic variable data sample:

[0191]

[0192] Where c k and β s Represent the information contribution rate and cumulative contribution rate of the principal component, r k is the information contribution of principal component k, r λ is the information contribution of the principal component λ, z k is the kth characteristic variable; when β s When it is close to 1, the first s variables are selected to conduct a comprehensive analysis of the principal components, and s important characteristic variables are selected according to the comprehensive score value Y to form an important characteristic variable data sample.

[0193] In one possible implementation, the machine learning model training module uses important feature variable data samples to perform model training on multiple pre-established machine learning models, obtains multiple trained machine learning models, and outputs prediction results of training sample data and test sample data of different machine learning models. The pre-established machine learning models include:

[0194] Set the maximum depth D of the extreme gradient boosting model XGBoost and build the XGBoost model;

[0195] Select the radial basis function (RBF) kernel as the kernel function of the support vector machine, set the hyperparameters, and build the support vector machine model:

[0196]

[0197] Where R(a,b) is the expression of the radial basis function, and η is the hyperparameter of the radial basis function;

[0198] Set the maximum depth D of the decision tree and build a random forest model.

[0199] In one possible implementation, the meta-classifier training module inputs the training samples into the XGBoost model, the support vector machine model, and the random forest model, respectively, and outputs the prediction results Y1, Y2, and Y3 of the training sample data of the three machine learning models; inputs the test samples into the XGBoost model, the support vector machine model, and the random forest model, respectively, and outputs the prediction results T1, T2, and T3 of the test sample data of the three machine learning models;

[0200] The training sample data prediction results Y1, Y2, Y3 and the training set data label Yt are combined into the data set R as follows:

[0201] n{[Y1][Y2][Y3][Yt]→n{[W][Yt]

[0202] In the formula, the correlation coefficient matrix W is obtained by predicting the results Y1, Y2, and Y3 from the training sample data;

[0203] Combine the test sample data prediction results T1, T2, and T3 into the data set S as follows:

[0204] n{[T1][T2][T3]

[0205] The meta-classifier is trained on the dataset R, and the dataset S is input into the trained meta-classifier to obtain the fault location prediction result.

[0206] In one possible implementation, the fault location module reads actual distribution network node voltage measurement data, inputs the distribution network node voltage measurement data into the distribution network fault partition location model, determines whether a fault occurs, and if a fault is determined to have occurred, the distribution network fault partition location model outputs fault location information.

[0207] Another embodiment of the present invention further provides an electronic device, including a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the distribution network fault zone location method.

[0208] Another embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the distribution network fault partition location method is implemented.

[0209] The computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form. The computer-readable storage medium may include: any entity or device, medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electric carrier signal, telecommunication signal and software distribution medium that can carry the computer program code. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals. For ease of explanation, the above content only shows the part related to the embodiment of the present invention. For specific technical details not disclosed, please refer to the method part of the embodiment of the present invention. The computer-readable storage medium is non-transitory and can be stored in a storage device formed by various electronic devices, and can implement the execution process recorded in the method of the embodiment of the present invention.

[0210] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0211] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0212] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0213] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0214] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A method for locating distribution network fault zones, characterized in that: include: Construct a data set sample based on the node characteristic data after a distribution network failure; Use principal component analysis to obtain feature importance for the data in the dataset sample, select relevant variables based on feature importance, and form a data sample of important feature variables; Use important feature variable data samples to train multiple pre-established machine learning models, obtain multiple trained machine learning models, and output the prediction results of training sample data and test sample data of different machine learning models; Using the prediction results of the training sample data and the prediction results of the test sample data to train the pre-established meta-classifier to obtain a trained meta-classifier; The trained multiple machine learning models are combined with the meta-classifier to obtain the distribution network fault zoning location model. The node feature data after the distribution network fault occurs is actually collected and input into the distribution network fault zoning location model to determine whether the distribution network fault occurs and output the fault location information.

2. The method for locating distribution network fault zones according to claim 1, characterized in that: In the step of constructing a data set sample based on the node feature data after a distribution network fault occurs, a fault point is set in the distribution network simulation system, and the effective values ​​of the three-phase voltages in all stages after the line fault occurs are used as node feature data to construct a data set sample; based on the data set sample, the effective values ​​of the three-phase voltages in all stages under the fault state of different sections of the distribution network are used to establish sample set data features, and the sample set data labels are established based on whether a fault occurs in the corresponding operating state and the fault section number, where 0 indicates that no fault has occurred and 1 indicates that a fault has occurred; and the data set samples are randomly split into model offline training samples and test samples according to a set ratio.

3. The method for locating distribution network fault zones according to claim 1, characterized in that: The steps of using principal component analysis to obtain feature importance from the data in the dataset sample, selecting relevant variables based on feature importance, and forming a data sample of important feature variables include: Through data standardization processing, each indicator value is converted into a standardized indicator, and the sample standard deviation and mean are calculated. The calculation expression is as follows: In the formula, b is the evaluation object of the sample data, represents the normalized index after transformation, S k and are the sample standard deviation and mean of the kth indicator respectively; The correlation coefficient matrix W is calculated according to the following expression: In=(in ik ) a×b In the formula, a is the characteristic variable of the sample data, w ik represents the correlation coefficient between the i-th indicator and the k-th indicator, w ik =1; Calculate the eigenvalues ​​and eigenvectors and construct the characteristic variables according to the following expressions: Where, is the correlation coefficient in the correlation coefficient matrix W, as the eigenvalue, v ba is the corresponding eigenvector; z a is the ath principal component, z1, z2, ..., z a Constitute characteristic variables; Calculate the comprehensive evaluation value according to the following formula, determine the important variables and form the important characteristic variable data sample: Where c k and β s Represent the information contribution rate and cumulative contribution rate of the principal component, r k is the information contribution of principal component k, r λ is the information contribution of the principal component λ, z k is the kth characteristic variable; when β s When it is close to 1, the first s variables are selected to conduct a comprehensive analysis of the principal components, and s important characteristic variables are selected according to the comprehensive score value Y to form an important characteristic variable data sample.

4. The method for locating distribution network fault zones according to claim 3, characterized in that: The steps of using important feature variable data samples to perform model training on multiple pre-established machine learning models, obtaining multiple trained machine learning models, and outputting prediction results of training sample data and test sample data of different machine learning models, the pre-established machine learning models include: Set the maximum depth D of the extreme gradient boosting model XGBoost and build the XGBoost model; Select the radial basis function (RBF) kernel as the kernel function of the support vector machine, set the hyperparameters, and build the support vector machine model: Where R(a,b) is the expression of the radial basis function, and η is the hyperparameter of the radial basis function; Set the maximum depth D of the decision tree and build a random forest model.

5. The method for locating distribution network fault zones according to claim 4, characterized in that: The step of training a pre-established meta-classifier using the prediction results of the training sample data and the prediction results of the test sample data to obtain a trained meta-classifier comprises: The training samples are input into the XGBoost model, support vector machine model and random forest model respectively, and the prediction results of the training sample data of the three machine learning models are output as Y1, Y2 and Y3; the test samples are input into the XGBoost model, support vector machine model and random forest model respectively, and the prediction results of the test sample data of the three machine learning models are output as T1, T2 and T3; The training sample data prediction results Y1, Y2, Y3 and the training set data label Yt are combined into the data set R as follows: n{[Y1][Y2][Y3][Yt]→n{[W][Yt] In the formula, the correlation coefficient matrix W is obtained by predicting the results Y1, Y2, and Y3 from the training sample data; Combine the test sample data prediction results T1, T2, and T3 into the data set S as follows: n{[T1][T2][T3] The meta-classifier is trained on the dataset R, and the dataset S is input into the trained meta-classifier to obtain the fault location prediction result.

6. The method for locating distribution network fault zones according to claim 5, characterized in that: The steps of combining the trained multiple machine learning models with the meta-classifier to obtain a distribution network fault partition location model, actually collecting node feature data after a distribution network fault occurs, inputting the data into the distribution network fault partition location model, determining whether a fault has occurred in the distribution network and outputting the fault location information, reading the actual distribution network node voltage measurement data, inputting the distribution network node voltage measurement data into the distribution network fault partition location model, determining whether a fault has occurred, and if a fault is determined to have occurred, outputting the fault location information by the distribution network fault partition location model.

7. A distribution network fault zoning location system, characterized in that: include: The sample building module is used to build a data set sample based on the node characteristic data after a distribution network failure occurs; The principal component analysis module is used to obtain the feature importance of the data in the dataset sample using the principal component analysis method, select relevant variables based on the feature importance, and form a data sample of important feature variables; The machine learning model training module is used to use important feature variable data samples to train multiple pre-established machine learning models, obtain multiple trained machine learning models, and output the prediction results of training sample data and test sample data of different machine learning models; A meta-classifier training module is used to train a pre-established meta-classifier using the prediction results of the training sample data and the prediction results of the test sample data to obtain a trained meta-classifier; The fault location module is used to combine multiple trained machine learning models with meta-classifiers to obtain a distribution network fault partition location model. It actually collects node feature data after a distribution network fault occurs, inputs the distribution network fault partition location model, determines whether a distribution network fault occurs, and outputs the fault location information.

8. The distribution network fault zoning location system according to claim 7, characterized in that: The sample establishment module sets a fault point in the distribution network simulation system, uses the effective values ​​of the three-phase voltage at all stages after the line fault occurs as node feature data, and constructs a data set sample; based on the data set samples, the effective values ​​of the three-phase voltage at all stages under the fault state of different sections of the distribution network are used to establish the sample set data features, and establishes the sample set data label according to whether a fault occurs in the corresponding operating state and the fault section number, where 0 indicates no fault occurs and 1 indicates a fault occurs; the data set samples are randomly split into model offline training samples and test samples according to a set ratio.

9. The distribution network fault zoning location system according to claim 7, characterized in that: The principal component analysis module converts each index value into a standardized index through data standardization processing, and calculates the sample standard deviation and mean. The calculation expression is as follows: In the formula, b is the evaluation object of the sample data, represents the normalized index after transformation, S k and are the sample standard deviation and mean of the kth indicator respectively; The correlation coefficient matrix W is calculated according to the following expression: In=(in ik ) a×b In the formula, a is the characteristic variable of the sample data, w ik represents the correlation coefficient between the i-th indicator and the k-th indicator, w ik =1; Calculate the eigenvalues ​​and eigenvectors and construct the characteristic variables according to the following expressions: Where, is the correlation coefficient in the correlation coefficient matrix W, as the eigenvalue, v ba is the corresponding eigenvector; z a is the ath principal component, z1, z2, ..., z a Constitute characteristic variables; Calculate the comprehensive evaluation value according to the following formula, determine the important variables and form the important characteristic variable data sample: Where c k and β s Represent the information contribution rate and cumulative contribution rate of the principal component, r k is the information contribution of principal component k, r λ is the information contribution of the principal component λ, z k is the kth characteristic variable; when β s When it is close to 1, the first s variables are selected to conduct a comprehensive analysis of the principal components, and s important characteristic variables are selected according to the comprehensive score value Y to form an important characteristic variable data sample.

10. The distribution network fault zone location system according to claim 9, characterized in that: The machine learning model training module uses important feature variable data samples to perform model training on multiple pre-established machine learning models, obtains multiple trained machine learning models, and outputs prediction results of training sample data and test sample data of different machine learning models. The pre-established machine learning models include: Set the maximum depth D of the extreme gradient boosting model XGBoost and build the XGBoost model; Select the radial basis function (RBF) kernel as the kernel function of the support vector machine, set the hyperparameters, and build the support vector machine model: Where R(a,b) is the expression of the radial basis function, and η is the hyperparameter of the radial basis function; Set the maximum depth D of the decision tree and build a random forest model.

11. The distribution network fault zone location system according to claim 10, characterized in that: The meta-classifier training module inputs the training samples into the XGBoost model, the support vector machine model, and the random forest model respectively, and outputs the training sample data prediction results Y1, Y2, and Y3 of the three machine learning models; inputs the test samples into the XGBoost model, the support vector machine model, and the random forest model respectively, and outputs the test sample data prediction results T1, T2, and T3 of the three machine learning models; The training sample data prediction results Y1, Y2, Y3 and the training set data label Yt are combined into the data set R as follows: In the formula, the correlation coefficient matrix W is obtained by predicting the results Y1, Y2, and Y3 from the training sample data; Combine the test sample data prediction results T1, T2, and T3 into the data set S as follows: n{[T1][T2][T3] The meta-classifier is trained on the dataset R, and the dataset S is input into the trained meta-classifier to obtain the fault location prediction result.

12. The distribution network fault zone location system according to claim 11, characterized in that: The fault location module reads the actual distribution network node voltage measurement data, inputs the distribution network node voltage measurement data into the distribution network fault partition location model, determines whether a fault occurs, and if a fault is determined to have occurred, the distribution network fault partition location model outputs the fault location information.

13. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the distribution network fault zone location method according to any one of claims 1 to 6.

14. A computer-readable storage medium, characterized in that The computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the method for locating a distribution network fault zone according to any one of claims 1 to 6 is implemented.

Images