Power distribution network fault real-time studying and judging and accurate positioning processing method and system

Abstract

The invention relates to the technical field of power distribution network fault research and judgment, discloses a power distribution network fault real-time research and judgment and accurate positioning processing method and system, and aims to improve the operation and maintenance efficiency and the power supply reliability of a power grid. According to the method, historical fault records and environment monitoring data of a power distribution network system are collected and preprocessed to form a basic data set; then performing data analysis, including fault type and environmental factor correlation analysis and fault area clustering analysis, to construct a training data set; the core is to construct a power distribution network fault research and judgment model adopting a dual machine learning model, and the accurate research and judgment of fault types are realized by fitting fault features and fault influences and utilizing a residual error to construct the model to estimate a causal effect. And after training and optimizing the model, studying and judging real-time power distribution network monitoring data, identifying a fault risk, and formulating a targeted disposal strategy. According to the method, various data are comprehensively considered, and the accuracy and the handling efficiency of fault research and judgment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault analysis and judgment, and in particular to a method and system for real-time analysis and judgment and precise positioning of distribution network faults. Background Art

[0002] With the continuous development of power systems and the improvement of their intelligence level, the safe and stable operation of distribution networks, as an important part of power systems, plays a vital role in ensuring social and economic development and people's daily lives. However, due to factors such as complex structure, numerous equipment, and changing operating environment, the distribution network has a relatively high frequency of failures and various types of failures, which brings huge challenges to the operation and maintenance of the power grid. The traditional distribution network fault analysis method mainly relies on experience judgment, which is not only inefficient, but also difficult to accurately and quickly locate the fault point, often resulting in a long fault handling time, affecting the power supply reliability and user satisfaction of the power grid. With the rapid development of big data and artificial intelligence technology, it has become possible to use data analysis and machine learning technology to conduct real-time analysis and precise positioning of distribution network faults.

[0003] Although the existing distribution network fault analysis technology has improved the efficiency and accuracy of fault handling to a certain extent, it still has some shortcomings. For example, some methods only rely on historical fault record data for modeling, ignoring the impact of environmental factors on the occurrence of faults, resulting in limited accuracy of the analysis results; in addition, some methods lack effective model design and algorithm support when dealing with the relationship between fault characteristics and fault effects, making it difficult to accurately estimate the causal effect of fault characteristics on fault effects, thereby affecting the accuracy of the analysis. Therefore, there is an urgent need for a real-time analysis and precise positioning method for distribution network faults that can comprehensively consider historical fault record data and environmental monitoring data, mine the correlation between fault types and environmental factors through data analysis, and accurately judge the fault type and locate the fault point. At the same time, the method also needs to have the ability to quickly respond to and process real-time distribution network monitoring data, so as to formulate targeted fault handling strategies in a timely manner and improve the operation and maintenance efficiency and power supply reliability of the power grid. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for real-time analysis and precise positioning of distribution network faults, so as to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for real-time analysis and accurate positioning of distribution network faults, the method comprising:

[0006] Collect historical fault record data and environmental monitoring data from the distribution network system, and clean and pre-process the collected data to form a basic data set;

[0007] Perform data analysis on the basic data set, including fault type and environmental factor correlation analysis and fault area cluster analysis, and merge the analysis result data with the basic data set to form a training data set;

[0008] Construct a distribution network fault analysis model, which adopts a dual machine learning model design, constructs a fault feature model and a fault impact model respectively, uses a machine learning algorithm to fit the fault features and fault impacts, and uses the residuals of the two models to construct a residual fitting model to estimate the causal effect of the fault features on the fault impact, so as to achieve accurate analysis of the fault type;

[0009] Use the training dataset to train the dual machine learning model and optimize the model's hyperparameters;

[0010] Use the trained model to analyze the real-time distribution network monitoring data and identify potential fault risks; and formulate targeted fault handling strategies based on the analysis results.

[0011] Preferably, the method for implementing the correlation analysis between the fault type and the environmental factor is:

[0012] The association rule mining algorithm is used to conduct association analysis on the fault records and environmental monitoring records in the distribution network system; the minimum support threshold and the minimum confidence threshold are set to screen out the strong association rules between the frequently occurring fault types and environmental factors;

[0013] Calculate the support and confidence of the association rules, where the support represents the frequency of the simultaneous occurrence of faults and environmental factors, and the confidence represents the probability of the fault occurring when the environmental factors occur;

[0014] Identify the types of failures that frequently accompany specific environmental factors.

[0015] Preferably, the implementation method of the fault area cluster analysis is:

[0016] Based on the multi-dimensional data of geographical location, fault type and fault frequency in historical fault records, the fault area feature vector is constructed; the cluster analysis algorithm DBSCAN is used to perform cluster analysis on the fault area feature vector;

[0017] By calculating the density reachability between fault areas, the fault areas are divided into different clusters;

[0018] Based on the clustering results, high-failure areas and low-risk areas are identified.

[0019] Preferably, the method for constructing a distribution network fault analysis model is:

[0020] S1: A dual machine learning model design is adopted to construct a fault feature model and a fault impact model respectively;

[0021] S2: The fault feature model uses current anomaly, voltage fluctuation, equipment aging and weather conditions as input features, uses the support vector machine algorithm for fitting, and outputs the predicted value of the fault feature; the fault impact model uses the fault range, fault level, power outage duration and number of users affected as output targets, and also uses the support vector machine algorithm for fitting, combining the impact of fault features and other covariates, and outputs the predicted value of the result variable;

[0022] S3: Construct a residual fitting model, use the residuals of the fault feature model and the fault impact model, and use the ridge regression algorithm to estimate the causal effect of the fault feature on the fault impact; the calculation formula of the ridge regression algorithm is:

[0023] β= ( X T X+λI ) -1 X T Y

[0024] Among them, Y is the residual target value, X is the residual eigenvalue, β is the regression coefficient, λ is the ridge parameter, and I is the unit matrix;

[0025] S4: Integrate the fault feature model, fault impact model and residual fitting model to form a complete distribution network fault analysis model.

[0026] Preferably, the distribution network fault analysis model is trained by using leave-one-out cross validation, and the specific method is as follows:

[0027] Step 1: Take each sample in the training data set D as a validation set and the remaining samples as training sets;

[0028] Step 2: For each sample, use the training set to train the distribution network fault analysis model to obtain model parameters;

[0029] Step 3: Apply the trained model to the validation set and use the absolute error algorithm to calculate the error between the predicted value and the true value as the validation error of the current sample;

[0030] Step 4: Repeat steps 2 to 3 until all samples have been used as validation sets once;

[0031] Step 5: Calculate the average value of all sample validation errors as an evaluation indicator of model performance;

[0032] Step 6: Select the model or hyperparameter combination with the best performance according to the average verification error as the final distribution network fault analysis model.

[0033] Preferably, based on the output of the distribution network fault analysis model, areas or equipment with potential fault risks are identified; for the identified risk points, how the adjustment of different fault characteristics affects the fault impact is analyzed; based on the analysis results, targeted fault handling strategies are formulated, including adjusting equipment maintenance plans to prevent faults, optimizing fault response processes to reduce power outage duration, and adjusting the distribution network operation mode.

[0034] Preferably, the method of cleaning the collected data includes:

[0035] For missing values ​​in the data set, the mean filling algorithm is used to fill them;

[0036] For outliers in the data set, the box plot algorithm is used to identify and remove them.

[0037] Preferably, the method of preprocessing the collected data includes:

[0038] Standardize the continuous data to eliminate the dimension difference. Use the Z-score method to standardize the data. Subtract the mean of each cleaned numerical data and divide it by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: Among them, X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data;

[0039] Perform label encoding conversion on categorical data.

[0040] Preferably, the operating status of the distribution network system is monitored in real time based on the output of the distribution network fault analysis model, and when it is predicted that the potential fault risk exceeds a preset threshold, the early warning mechanism is automatically triggered.

[0041] Preferably, a distribution network fault real-time analysis and accurate positioning and disposal system comprises:

[0042] The data collection module is used to collect historical fault record data and environmental monitoring data from the distribution network system, and clean and pre-process the collected data to form a basic data set;

[0043] The data analysis module is used to perform data analysis on the basic data set, including the correlation analysis between fault types and environmental factors and the cluster analysis of fault areas, and merge the analysis result data with the basic data set to form a training data set;

[0044] A model building module is used to build a distribution network fault analysis model. The distribution network fault analysis model adopts a dual machine learning model design, respectively constructs a fault feature model and a fault impact model, uses a machine learning algorithm to fit the fault features and fault impacts, and uses the residuals of the two models to build a residual fitting model to estimate the causal effect of the fault features on the fault impact, so as to achieve accurate analysis of the fault type; and uses a training data set to train the dual machine learning model and optimize the model's hyperparameters;

[0045] The real-time analysis module is used to use the trained model to analyze the real-time distribution network monitoring data and identify potential fault risks; and formulate targeted fault handling strategies based on the analysis results.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention comprehensively considers historical fault record data and environmental monitoring data, and uses data analysis technology to explore the correlation between fault types and environmental factors, thereby effectively improving the accuracy of fault analysis. The design of the dual machine learning model, especially the construction of the fault feature model and the fault impact model, and the application of the residual fitting model, can more accurately estimate the causal effect of fault characteristics on fault impact, thereby achieving accurate analysis of fault types. Traditional methods rely on manual inspections and empirical judgments, and the efficiency of fault location is low. The present invention can quickly identify potential fault risks and accurately locate fault points by real-time analysis of distribution network monitoring data combined with a trained dual machine learning model. This greatly shortens the fault handling time and improves the operation and maintenance efficiency of the power grid.

[0048] The present invention can respond to and process distribution network monitoring data in real time, and formulate targeted fault handling strategies in a timely manner. This helps to quickly restore power supply, reduce the duration and scope of power outages, and thus significantly improve the power supply reliability of the power grid. This is crucial for social and economic development and people's daily lives. Through data analysis and machine learning technology, the present invention can provide more accurate and comprehensive fault analysis results, and provide powerful decision-making support for power grid operation and maintenance managers. This helps to optimize resource allocation and ensure that the necessary human and material resources can be quickly mobilized and efficiently handled when a fault occurs.

[0049] The method and system of the present invention have strong adaptability and scalability. With the continuous development of the distribution network and the improvement of the intelligent level, new data sources and algorithm models can be easily incorporated to further improve the ability of fault analysis and location. At the same time, the method is also applicable to different types of distribution networks and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1This is a working principle diagram of the intelligent voice interaction method of the present invention;

[0051] Figure 2 This is a diagram of the steps to implement the correlation analysis between fault types and environmental factors;

[0052] Figure 3 Workflow diagram for building a distribution network fault analysis model;

[0053] Figure 4 Workflow diagram for training the distribution network fault analysis model using leave-one-out cross validation. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] See also Figure 1-4 The present invention provides a technical solution: a method for real-time analysis and accurate positioning of distribution network faults, the method comprising:

[0056] Collect historical fault record data from the distribution network system, including but not limited to fault type, fault occurrence time, fault location, fault recovery time, etc. At the same time, collect environmental monitoring data, such as weather conditions (temperature, humidity, wind speed, precipitation, etc.), equipment operating environment (such as equipment temperature, load conditions), etc. Clean the collected data to remove duplicate, erroneous or incomplete data records. Preprocess the data, such as data normalization, missing value filling, etc., to form a basic data set.

[0057] Perform association analysis on the basic data set between fault types and environmental factors to explore the potential association between fault types and environmental factors. Perform cluster analysis on fault areas to divide the distribution network into different fault areas based on the location, frequency and other information of the faults, providing a basis for subsequent fault location. Combine the result data of association analysis and cluster analysis with the basic data set to form a training data set.

[0058] The design of dual machine learning models is adopted to construct fault feature models and fault impact models respectively. The fault feature model is used to fit the relationship between fault features and fault occurrence, and the fault impact model is used to fit the prediction of fault features on fault impact (such as fault range, fault duration, etc.). The machine learning algorithm is used to fit the fault features and fault impact. The residual fitting model is constructed using the residuals of the fault feature model and the fault impact model to estimate the causal effect of the fault features on the fault impact. Through the design of dual machine learning models, accurate judgment of fault types can be achieved.

[0059] Use the training data set to train the dual machine learning model and iteratively optimize the model parameters to improve the model's prediction accuracy. Optimize the model's hyperparameters to obtain the best model performance.

[0060] Use the trained model to analyze the real-time distribution network monitoring data and identify potential fault risks. According to the analysis results, such as fault type, fault location, fault impact, etc., formulate targeted fault handling strategies. Communicate the fault handling strategies to the operation and maintenance personnel in a timely manner to guide them in troubleshooting and repairing faults, and improve the operation and maintenance efficiency and power supply reliability of the power grid.

[0061] The present invention will be further described below in conjunction with Examples 1 to 5:

[0062] Embodiment 1:

[0063] The method for analyzing the association between the fault type and the environmental factors specifically includes:

[0064] Historical fault record data and environmental monitoring data are extracted from the distribution network system. Fault record data includes information such as fault type, fault occurrence time, and fault location; environmental monitoring data includes weather conditions (such as temperature, humidity, wind speed, etc.), equipment load, equipment operating environment parameters, etc.

[0065] Mining uses association rule mining algorithms (such as Apriori algorithm or FP-Growth algorithm) to perform association analysis on fault records and environmental monitoring records. The algorithm scans the data set to find all frequently occurring item sets and further generates association rules.

[0066] Set the minimum support threshold and the minimum confidence threshold. Support indicates the frequency of the simultaneous occurrence of faults and environmental factors, that is, the proportion of the number of simultaneous occurrences of the two to the total number of occurrences; confidence indicates the probability of the fault occurring when the environmental factors occur, that is, the probability of the fault occurring when the environmental factors occur.

[0067] According to the set threshold, the strong association rules between the frequently occurring fault types and environmental factors are screened out. For example, if it is found that "when the temperature exceeds 35°C, the incidence of transformer failure increases significantly", and the support and confidence of the rule exceed the set threshold, it is considered to be a strong association rule.

[0068] Calculate support and confidence. For each association rule, calculate its support and confidence. Support can be calculated by counting the number of times the fault and environmental factors appear at the same time; confidence can be obtained by calculating the ratio of the number of faults that occur when the environmental factor appears to the total number of times the environmental factor appears.

[0069] Identify the fault types that are frequently accompanied by specific environmental factors. According to the support and confidence of the association rules, identify the fault types that are frequently accompanied by specific environmental factors. These fault types are more likely to occur under specific environmental conditions, so they can be the focus of fault warning and disposal.

[0070] Example: By analyzing historical data, it is found that when the humidity exceeds 80%, the incidence of cable failure increases significantly, with a support of 0.2 (i.e., 20% of the fault records also have humidity exceeding 80%) and a confidence of 0.7 (i.e., when the humidity exceeds 80%, 70% of the faults are cable failures). Therefore, "humidity exceeding 80%" can be used as one of the early warning conditions for cable failure.

[0071] The methods of fault area cluster analysis include:

[0072] Construct fault region feature vector: Construct fault region feature vector based on multi-dimensional data of geographical location, fault type and fault frequency in historical fault records. For example, the number of faults, fault type distribution, average fault duration and other information in a certain area can be used as the feature vector of the area.

[0073] Application of clustering analysis algorithm: Select DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering analysis algorithm to cluster the fault area feature vector. The DBSCAN algorithm performs clustering based on density accessibility, can find clusters of any shape, and has good robustness to noise data.

[0074] Calculate density reachability: In the DBSCAN algorithm, the density reachability between fault regions is calculated to determine whether they belong to the same cluster. Density reachability means that if the distance from one region to another is less than a certain threshold and the density of all regions on the path between the two regions is greater than a certain minimum value, they are considered density reachable.

[0075] Clustering: The fault areas are divided into different clusters based on density accessibility. Each cluster represents a group of areas with similar fault characteristics. For example, areas with frequent faults and similar fault types can be divided into a cluster as a high-incidence fault area.

[0076] Identify high-failure areas and low-risk areas: Based on the clustering results, identify high-failure areas and low-risk areas. High-failure areas are areas where failures occur frequently and have a greater impact, and require special attention and strengthened operation and maintenance management; low-risk areas are areas where failures are less frequent and have a smaller impact, and the investment in operation and maintenance resources can be appropriately reduced.

[0077] For example, through cluster analysis of historical fault data, it was found that several adjacent areas in a certain urban area frequently experienced faults, and the fault types were mainly cable faults and transformer faults. These areas were divided into a high-fault area cluster as key maintenance targets. At the same time, it was also found that other remote areas had fewer faults and were divided into low-risk area clusters, so the inspection frequency could be appropriately reduced.

[0078] Embodiment 2:

[0079] This embodiment is used to describe in detail the specific implementation method of building a distribution network fault analysis model:

[0080] S1: Using dual machine learning model design

[0081] In order to more accurately judge the distribution network fault, the present invention adopts a dual machine learning model design to construct a fault feature model and a fault impact model respectively. This design can capture the characteristic information of the fault and the impact of the fault on the system respectively, thereby improving the accuracy and reliability of the judgment.

[0082] S2: Construct fault feature model and fault impact model

[0083] Fault signature models include:

[0084] Input features: Current anomalies, voltage fluctuations, equipment aging, and weather conditions are selected as input features of the fault feature model. These features can fully reflect the operating status of the distribution network system and the impact of the external environment on the system.

[0085] Algorithm selection: Use the support vector machine (SVM) algorithm to fit the input features. The SVM algorithm has excellent performance in processing high-dimensional data and nonlinear problems, and can accurately predict fault characteristics.

[0086] Output: The fault signature model outputs the predicted values ​​of the fault signatures, which will serve as one of the inputs of the fault impact model.

[0087] Fault impact models include:

[0088] Output target: Fault scope, fault level, power outage duration and number of users affected are selected as the output targets of the fault impact model. These targets can fully reflect the impact of the fault on the distribution network system.

[0089] Algorithm selection: The support vector machine (SVM) algorithm is also used to fit the fault characteristics and the impact of other covariates. By considering the influence of multiple factors, the fault impact model can more accurately predict the impact of the fault.

[0090] Output: The predicted values ​​of the output result variables of the fault impact model. These predicted values ​​will be used for subsequent residual fitting and fault analysis.

[0091] S3: Constructing residual fitting model

[0092] Residual calculation: First, calculate the residuals of the fault feature model and the fault impact model. The residual is the difference between the actual observation value and the model prediction value, reflecting the information that the model fails to fully capture.

[0093] Algorithm selection: The ridge regression algorithm is used to fit the residuals. Ridge regression is an extension of linear regression. It prevents overfitting and improves the generalization ability of the model by adding L2 regularization terms to the regression coefficients. The calculation formula of the ridge regression algorithm is:

[0094] β= ( X T X+λI ) -1 X T Y

[0095] Among them, Y is the residual target value, X is the residual eigenvalue, β is the regression coefficient, λ is the ridge parameter, and I is the unit matrix.

[0096] Causal effect estimation: The ridge regression model can be used to estimate the causal effect of fault characteristics on fault impact. This estimation helps to understand how fault characteristics affect the impact of faults, thus providing a stronger basis for fault analysis.

[0097] S4: Integrate the models to form a complete distribution network fault analysis model

[0098] The fault feature model, fault impact model and residual fitting model are integrated together to form a complete distribution network fault analysis model. This model can comprehensively consider the fault feature information, impact degree and residual information, and conduct real-time analysis and precise positioning of distribution network faults.

[0099] Specific example: Assume that a distribution network system has abnormal current and voltage fluctuations, and the weather conditions are bad and the equipment is aged. By inputting this information into the fault feature model, the predicted value of the fault feature can be obtained. Then, the predicted value of the fault feature and other covariates are input into the fault impact model to predict the scope, level, power outage duration and number of users affected by the fault. Finally, by calculating the residuals and fitting the residual fitting model, the causal effect of the fault feature on the fault impact can be estimated. Based on this information, operation and maintenance personnel can formulate targeted fault handling strategies in a timely manner to improve the operation and maintenance efficiency and power supply reliability of the power grid.

[0100] Embodiment 3:

[0101] The present invention adopts leave-one-out cross-validation to train the distribution network fault analysis model, and the specific implementation method includes:

[0102] Step 1: Take each sample in the training data set D as a validation set, and the remaining samples as training sets. Assuming that the training data set D contains N samples, when performing leave-one-out cross-validation, each sample will be selected as a validation set in turn, and the remaining N-1 samples will be used as training sets.

[0103] Step 2: For each sample, use its corresponding training set to train the distribution network fault analysis model. During the training process, according to the type and structure of the model, use appropriate algorithms (such as support vector machines, random forests, neural networks, etc.) to fit the training set and obtain the model parameters. These parameters determine the mapping relationship between the model and the input data, that is, how to convert the input features into output prediction values.

[0104] Step 3: Apply the trained model to the currently selected validation set, that is, the sample that was left out. Use the absolute error algorithm to calculate the error between the predicted value and the true value. The absolute error is the absolute value of the difference between the predicted value and the true value, which reflects the accuracy of the model's prediction of the current sample. This error is used as the validation error of the current sample.

[0105] Step 4: Repeat steps 2 to 3 until all samples have been used as validation sets once. This means that for each sample in the training dataset D, the model is validated once as a validation set, and the corresponding validation error is calculated.

[0106] Step 5: Calculate the average validation error of all samples as an evaluation indicator of model performance. This average reflects the average prediction accuracy of the model on the entire training data set. The smaller the average validation error, the better the performance of the model.

[0107] Step 6: Select the model or hyperparameter combination with the best performance based on the average validation error. During the training process, different model types, structures, or hyperparameter combinations may be tried. Through leave-one-out cross-validation, the model performance under these different choices can be compared, and the one with the smallest average validation error can be selected as the final distribution network fault analysis model.

[0108] Suppose there is a distribution network fault analysis model whose input features include current, voltage, equipment status, etc., and the output is the fault type. The training data set D contains 100 samples. When performing leave-one-out cross-validation, each sample will be selected as the validation set in turn, and the remaining 99 samples will be used as the training set. For each validation set, the corresponding training set is used to train the model, and the absolute error between the predicted value and the true value is calculated. Finally, the average of the validation errors of all samples is calculated to obtain the average prediction accuracy of the model. By comparing the average validation errors under different models or hyperparameter combinations, the one with the best performance is selected as the final distribution network fault analysis model. This method can ensure that the model has good generalization ability on the entire training data set, thereby improving the accuracy and reliability of fault analysis.

[0109] Embodiment 4:

[0110] The present invention formulates a targeted fault handling strategy based on the output of the distribution network fault analysis model, and the specific steps include:

[0111] Step 1: Identify potential failure risk areas or equipment

[0112] The trained distribution network fault analysis model is used to monitor and analyze the distribution network system in real time. The output of the model includes the fault risk probability or fault characteristic index of each area or device. By setting a certain threshold or risk level classification standard, areas or devices with potential fault risks can be identified.

[0113] For example, in a distribution network system, the model may output that the probability of current abnormality in a certain area is high, or the voltage fluctuation of a certain device exceeds the normal range. These output information can be used as a basis for identifying potential fault risks.

[0114] Step 2: Analyze the impact of fault feature adjustments on fault impact

[0115] For the identified risk points, further analysis is conducted on how the adjustment of different fault characteristics affects the fault impact. This includes analyzing the potential impact of changes in fault characteristics (such as current, voltage, equipment temperature, etc.) on fault impact indicators such as fault scope, fault level, and power outage duration.

[0116] Specifically, the fault characteristic values ​​of risk points can be adjusted through simulation or experiment, and the changes in fault impact indicators can be observed. For example, the degree of current anomaly in a certain area can be simulated to reduce the fault range, the fault level, the power outage duration, etc.

[0117] Step 3: Develop a targeted troubleshooting strategy

[0118] Based on the analysis results of the fault characteristics and the impact of the fault, a targeted fault handling strategy is formulated. This includes adjusting the equipment maintenance plan to prevent faults, optimizing the fault response process to reduce the duration of power outages, and adjusting the distribution network operation mode. Specifically, it includes:

[0119] Adjust equipment maintenance plan: For equipment with potential failure risk, maintenance or overhaul work can be arranged in advance to prevent failure. According to the failure characteristics and maintenance history of the equipment, a personalized maintenance plan is formulated to ensure that the equipment is in good condition.

[0120] Optimize the fault response process: Develop detailed fault response processes and emergency plans for possible fault types. Verify the effectiveness of the fault response process through simulation exercises and other methods, and make necessary optimizations and adjustments.

[0121] Adjust the operation mode of the distribution network: According to the size and distribution of the fault risk, adjust the operation mode of the distribution network, such as adjusting the load distribution, changing the power supply path, etc. By adjusting the operation mode of the distribution network, the impact of faults on the system can be reduced and the reliability and stability of the system can be improved.

[0122] Specific example: Assume that the output of the distribution network fault analysis model shows that the probability of current abnormality in a certain area is high, and through analysis, it is found that reducing the current load in this area can effectively reduce the fault level and shorten the power outage duration. Based on this analysis result, the following targeted fault handling strategies can be formulated:

[0123] Perform preventive maintenance on key equipment in this area, such as cleaning, tightening connections, checking insulation, etc., to ensure the equipment is in good condition.

[0124] Develop a fault response process for abnormal current, including fault alarm, on-site investigation, fault isolation, emergency repair and restoration, and clearly define the person in charge and time requirements for each step.

[0125] Without affecting the stability of the system, appropriately reduce the current load in the area, such as by adjusting the transformer taps, transferring part of the load to other areas, etc., to reduce the risk of failure.

[0126] Embodiment 5:

[0127] This embodiment is used to describe a specific implementation method for cleaning and preprocessing the collected data.

[0128] The data cleaning includes:

[0129] Missing value processing: For missing values ​​in the data set, the mean filling algorithm is used to fill them.

[0130] The specific implementation steps are: calculate the mean of each feature in the data set; then, for the missing values ​​in the data set, fill them with the mean of the corresponding feature. For example, in a feature column, if there are missing values, calculate the mean of the non-missing values ​​in the feature column and use the mean to fill all the missing values ​​in the feature column.

[0131] Outlier processing: For outliers in the data set, the box plot algorithm is used to identify and remove them.

[0132] The specific implementation steps are: draw a box plot for each feature in the data set; then, according to the definition of the box plot, identify outliers that are 1.5 times the interquartile range of the upper quartile or 1.5 times the interquartile range of the lower quartile; finally, remove these outliers from the data set. For example, in a feature column, after outliers are identified according to the box plot algorithm, these outliers are deleted from the feature column.

[0133] The data preprocessing includes:

[0134] Continuous data standardization: Standardize continuous data to eliminate dimensional differences. Use the Z-score method to standardize the data, subtract the mean from each cleaned numerical data and divide it by its standard deviation, so that the processed data conforms to the standard normal distribution. The formula for standardization is: Among them, X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data. For example, for a continuous feature column, first calculate the mean and standard deviation of the feature column; then, subtract the mean from each value in the feature column and divide it by the standard deviation to get the standardized value.

[0135] Categorical data label encoding conversion: perform label encoding conversion on categorical data. The specific implementation steps are: determine the unique identifier of each category in the categorical data; then, map each category to a unique numeric label. For example, for a categorical feature column, if it contains three categories of "high", "medium", and "low", "high" can be mapped to 1, "medium" to 2, and "low" to 3, thus completing the label encoding conversion.

[0136] Suppose there is a distribution network fault dataset, which contains continuous features such as "current value" and "voltage value", as well as categorical features such as "equipment status" (including three categories: "normal", "abnormal" and "fault").

[0137] Data cleaning: For missing values ​​in the "current value" feature column, calculate the mean of the feature column and fill the missing values ​​with the mean. For the "voltage value" feature column, draw a box plot to identify and remove outliers that exceed 1.5 times the interquartile range of the upper quartile or are lower than 1.5 times the interquartile range of the lower quartile.

[0138] Data preprocessing: Standardize the "current value" and "voltage value" feature columns. Calculate the mean and standard deviation of each feature column, then subtract the mean from the value in each feature column and divide it by the standard deviation to get the standardized value. Perform label encoding conversion on the "device status" feature column. Map "normal" to 1, "abnormal" to 2, and "fault" to 3.

[0139] The present invention also includes a distribution network fault real-time analysis and accurate positioning and disposal system, the system comprising:

[0140] The data collection module is used to collect historical fault record data and environmental monitoring data from the distribution network system, and clean and pre-process the collected data to form a basic data set;

[0141] The data analysis module is used to perform data analysis on the basic data set, including the correlation analysis between fault types and environmental factors and the cluster analysis of fault areas, and merge the analysis result data with the basic data set to form a training data set;

[0142] A model building module is used to build a distribution network fault analysis model. The distribution network fault analysis model adopts a dual machine learning model design, respectively constructs a fault feature model and a fault impact model, uses a machine learning algorithm to fit the fault features and fault impacts, and uses the residuals of the two models to build a residual fitting model to estimate the causal effect of the fault features on the fault impact, so as to achieve accurate analysis of the fault type; and uses a training data set to train the dual machine learning model and optimize the model's hyperparameters;

[0143] The real-time analysis module is used to use the trained model to analyze the real-time distribution network monitoring data and identify potential fault risks; and formulate targeted fault handling strategies based on the analysis results.

[0144] The implementation of the system refers to the above embodiment and will not be described in detail in the specification.

[0145] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0146] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for real-time analysis and accurate positioning of distribution network faults, characterized in that: The method comprises: Collect historical fault record data and environmental monitoring data from the distribution network system, and clean and pre-process the collected data to form a basic data set; Perform data analysis on the basic data set, including fault type and environmental factor correlation analysis and fault area cluster analysis, and merge the analysis result data with the basic data set to form a training data set; Construct a distribution network fault analysis model, which adopts a dual machine learning model design, constructs a fault feature model and a fault impact model respectively, uses a machine learning algorithm to fit the fault features and fault impacts, and uses the residuals of the two models to construct a residual fitting model to estimate the causal effect of the fault features on the fault impact, so as to achieve accurate analysis of the fault type; Use the training dataset to train the dual machine learning model and optimize the model's hyperparameters; Use the trained model to analyze the real-time distribution network monitoring data and identify potential fault risks; and formulate targeted fault handling strategies based on the analysis results.

2. The method for real-time analysis and accurate positioning of distribution network faults according to claim 1 is characterized in that: The implementation method of the correlation analysis between the fault type and the environmental factor is as follows: The association rule mining algorithm is used to conduct association analysis on the fault records and environmental monitoring records in the distribution network system; the minimum support threshold and the minimum confidence threshold are set to screen out the strong association rules between the frequently occurring fault types and environmental factors; Calculate the support and confidence of the association rules, where the support represents the frequency of the simultaneous occurrence of faults and environmental factors, and the confidence represents the probability of the fault occurring when the environmental factors occur; Identify the types of failures that frequently accompany specific environmental factors.

3. The method for real-time analysis and accurate positioning of distribution network faults according to claim 1 is characterized in that: The implementation method of the fault area cluster analysis is: Construct the fault area feature vector based on the multi-dimensional data of geographical location, fault type and fault frequency in the historical fault records; The clustering analysis algorithm DBSCAN is used to perform cluster analysis on the fault area feature vectors; By calculating the density reachability between fault areas, the fault areas are divided into different clusters; Based on the clustering results, high-failure areas and low-risk areas are identified.

4. The method for real-time analysis and accurate positioning of distribution network faults according to claim 1 is characterized in that: The method of constructing the distribution network fault analysis model is as follows: S1: A dual machine learning model design is adopted to construct a fault feature model and a fault impact model respectively; S2: The fault feature model uses current anomaly, voltage fluctuation, equipment aging and weather conditions as input features, uses the support vector machine algorithm for fitting, and outputs the predicted value of the fault feature; The fault impact model takes fault scope, fault level, power outage duration and number of users affected as output targets. It also uses the support vector machine algorithm for fitting, combines the fault characteristics and the influence of other covariates, and outputs the predicted value of the result variable. S3: Construct a residual fitting model, use the residuals of the fault feature model and the fault impact model, and use the ridge regression algorithm to estimate the causal effect of the fault feature on the fault impact; the calculation formula of the ridge regression algorithm is: β=(X T X+λI) -1 X T Y Among them, Y is the residual target value, X is the residual eigenvalue, β is the regression coefficient, λ is the ridge parameter, and I is the unit matrix; S4: Integrate the fault feature model, fault impact model and residual fitting model to form a complete distribution network fault analysis model.

5. The method for real-time analysis and accurate positioning of distribution network faults according to claim 4 is characterized in that: The leave-one-out cross validation method is used to train the distribution network fault analysis model. The specific method is as follows: Step 1: Take each sample in the training data set D as a validation set and the remaining samples as training sets; Step 2: For each sample, use the training set to train the distribution network fault analysis model to obtain model parameters; Step 3: Apply the trained model to the validation set and use the absolute error algorithm to calculate the error between the predicted value and the true value as the validation error of the current sample; Step 4: Repeat steps 2 to 3 until all samples have been used as validation sets once; Step 5: Calculate the average value of all sample validation errors as an evaluation indicator of model performance; Step 6: Select the model or hyperparameter combination with the best performance according to the average verification error as the final distribution network fault analysis model.

6. The method for real-time analysis and accurate positioning of distribution network faults according to claim 1 is characterized in that: Based on the output of the distribution network fault analysis model, identify areas or equipment with potential fault risks; for the identified risk points, analyze how the adjustment of different fault characteristics affects the fault impact; Based on the analysis results, targeted fault handling strategies are formulated, including adjusting equipment maintenance plans to prevent faults, optimizing fault response processes to reduce power outage duration, and adjusting distribution network operation modes.

7. The method for real-time analysis and precise positioning of distribution network faults according to claim 1 is characterized in that: Methods for cleaning the collected data include: For missing values ​​in the data set, the mean filling algorithm is used to fill them; For outliers in the data set, the box plot algorithm is used to identify and remove them.

8. The method for real-time analysis and accurate positioning of distribution network faults according to claim 1 is characterized in that: The methods for preprocessing the collected data include: Standardize the continuous data to eliminate the dimension difference. Use the Z-score method to standardize the data. Subtract the mean of each cleaned numerical data and divide it by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: Among them, X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data; Perform label encoding conversion on categorical data.

9. The method for real-time analysis and accurate positioning of distribution network faults according to claim 1 is characterized in that: The operating status of the distribution network system is monitored in real time based on the output of the distribution network fault analysis model. When the potential fault risk is predicted to exceed the preset threshold, the early warning mechanism is automatically triggered.

10. A distribution network fault real-time analysis and accurate positioning system, characterized in that: The system comprises: The data collection module is used to collect historical fault record data and environmental monitoring data from the distribution network system, and clean and pre-process the collected data to form a basic data set; The data analysis module is used to perform data analysis on the basic data set, including the correlation analysis between fault types and environmental factors and the cluster analysis of fault areas, and merge the analysis result data with the basic data set to form a training data set; A model building module is used to build a distribution network fault analysis model. The distribution network fault analysis model adopts a dual machine learning model design, respectively constructs a fault feature model and a fault impact model, uses a machine learning algorithm to fit the fault features and fault impacts, and uses the residuals of the two models to build a residual fitting model to estimate the causal effect of the fault features on the fault impact, so as to achieve accurate analysis of the fault type; and uses a training data set to train the dual machine learning model and optimize the model's hyperparameters; The real-time analysis module is used to use the trained model to analyze the real-time distribution network monitoring data and identify potential fault risks; and formulate targeted fault handling strategies based on the analysis results.

Images