AI-based Distribution Network Fault Identification Method and Related Devices

Abstract

The present application provides an AI-based distribution network fault identification method and related devices. The method includes obtaining historical fault types, historical operation data corresponding to the historical fault types, and historical fault times of a plurality of preset distribution network systems; the historical grid types corresponding to the distribution network systems are any one of urban, rural, and industrial areas; extracting historical feature vectors corresponding to each historical operation data; according to the historical grid type and the historical feature vectors corresponding to the historical grid type, performing feature transformation on the historical feature vectors to obtain the transformed historical feature vectors, so as to complete the feature alignment of the plurality of historical feature vectors; inputting the transformed historical feature vectors and the historical fault times into an AI model to be trained, and outputting predicted fault types; completing the training of the AI model according to the predicted fault types and the historical fault types corresponding to each historical feature vector, and completing the fault identification of the running distribution network system according to the trained AI model.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence, and particularly relates to an AI-based distribution network fault identification method and related devices. Background Art

[0002] Distribution network fault identification is a crucial task in the power system, directly affecting the reliability and operation efficiency of the power grid. Traditional fault identification methods mainly rely on physical models and expert experience. Although they can achieve good results in some scenarios, with the expansion of the scale and the increase in complexity of the distribution network, the limitations of these methods become more and more obvious. Specifically, the traditional methods have the following problems:

[0003] 1. Limited data processing ability: Traditional methods can usually only process a small amount of historical data and simple time series data, and are insufficient in processing large-scale and diverse data sets.

[0004] 2. Poor adaptability: Traditional methods are based on fixed physical models and rules, and it is difficult to adapt to different types of distribution networks and new fault modes. Once the operating environment of the distribution network changes, the performance of the model will drop significantly.

[0005] 3. Insufficient real-time performance: Traditional methods take a long time in data processing and fault identification, and it is difficult to achieve real-time fault warning and rapid response.

[0006] 4. High maintenance cost: The methods relying on expert experience and physical models need to be maintained and updated regularly, which not only takes time and effort, but is also prone to errors.

[0007] Therefore, there is an urgent need for a distribution network fault identification method to solve at least one of the above problems. Summary of the Invention

[0008] The present application provides an AI-based distribution network fault identification method and related devices, aiming to solve the problem that traditional fault identification methods mainly rely on physical models and expert experience. Although they can achieve good results in some scenarios, with the expansion of the scale and the increase in complexity of the distribution network, the limitations of these methods have become increasingly obvious. Specifically, the traditional methods have the following problems: 1. Limited data processing ability: Traditional methods can usually only process a small amount of historical data and simple time series data, and are insufficient in processing large-scale and diverse data sets. 2. Poor adaptability: Traditional methods are based on fixed physical models and rules, and it is difficult to adapt to different types of distribution networks and new fault modes. Once the operating environment of the distribution network changes, the performance of the model will drop significantly. 3. Lack of real-time performance: Traditional methods take a long time in data processing and fault identification, and it is difficult to achieve real-time fault warning and rapid response. 4. High maintenance cost: Methods relying on expert experience and physical models need to be maintained and updated regularly, which not only consumes time and effort, but also is prone to errors, etc.

[0009] In a first aspect, the present application provides an AI-based distribution network fault identification method, including:

[0010] Obtain historical fault types of multiple preset distribution network systems, historical operation data corresponding to the historical fault types, and historical fault times; the historical grid types corresponding to the distribution network systems are any one of urban, rural, and industrial areas; the historical operation data includes one or more of voltage, current, temperature, and humidity;

[0011] Extract historical feature vectors corresponding to each of the historical operation data;

[0012] According to the historical grid type and the historical feature vectors corresponding to the historical grid type, perform feature transformation on the historical feature vectors to obtain the transformed historical feature vectors; so that the data features corresponding to different historical grid types are aligned;

[0013] Input the transformed historical feature vectors and historical fault times into an AI model to be trained. The AI model includes an input layer, multiple long short-term memory layers, and an output layer; the historical feature vectors and historical fault times are output to predict fault types through the input layer, multiple long short-term memory layers, and the output layer in sequence;

[0014] Complete the training of the AI model according to the predicted fault types and historical fault types corresponding to each of the historical feature vectors, and complete the fault identification of the running distribution network system according to the trained AI model.

[0015] In some embodiments, performing feature transformation on the historical feature vectors according to the historical power grid type and the historical feature vectors corresponding to the historical power grid type to obtain the transformed historical feature vectors includes: standardizing each of the historical feature vectors according to the historical power grid type; performing feature enhancement on each of the historical feature vectors; performing principal component analysis on each of the enhanced historical feature vectors to obtain a feature alignment matrix for each of the historical feature vectors, and transforming the corresponding historical feature vectors according to the feature alignment matrix to obtain the transformed historical feature vectors, so as to complete the feature alignment of the multiple historical feature vectors.

[0016] Exemplarily, standardizing each of the historical feature vectors according to the historical power grid type includes: obtaining a first mean and a first standard deviation corresponding to each of the historical feature vectors whose historical power grid type is urban; obtaining a second mean and a second standard deviation corresponding to each of the historical feature vectors whose historical power grid type is rural; obtaining a third mean and a third standard deviation corresponding to each of the historical feature vectors whose historical power grid type is industrial area; standardizing each of the historical feature vectors whose historical power grid type is urban according to the first mean and the first standard deviation; standardizing each of the historical feature vectors whose historical power grid type is rural according to the second mean and the second standard deviation; standardizing each of the historical feature vectors whose historical power grid type is industrial area according to the third mean and the third standard deviation.

[0017] Exemplarily, performing feature enhancement on each of the historical feature vectors includes: determining a reference feature vector among the multiple historical feature vectors; calculating the value of the Gaussian kernel function of each of the historical feature vectors and the reference feature vector; completing the feature enhancement of the historical feature vectors according to the value of the Gaussian kernel function corresponding to each of the historical feature vectors; wherein the expression of the Gaussian kernel function includes:

[0018] ;

[0019] wherein, is the Gaussian kernel function, is the th historical feature vector, is the th reference feature vector, is the bandwidth parameter, which determines the distribution of the historical feature vectors in the high-dimensional space. Wherein the expression of the enhanced historical feature vector includes:

[0020] ;

[0021] The enhanced historical feature vector , represents the Gaussian kernel function of the historical feature vector and the th benchmark feature vector represents the Gaussian kernel function of the historical feature vector and the th benchmark feature vector represents the Gaussian kernel function of the historical feature vector and the th benchmark feature vector represents the Gaussian kernel function of the historical feature vector and the th benchmark feature vector, where m is the number of the benchmark feature vectors

[0022] Exemplarily, performing principal component analysis on each of the enhanced historical feature vectors to obtain the feature alignment matrix of each historical feature vector includes: calculating the covariance matrix corresponding to the historical feature vectors of the same enhanced historical power grid type; obtaining the eigenvalues of each covariance matrix and the eigenvectors corresponding to each eigenvalue; determining the target eigenvalue among the multiple eigenvalues corresponding to each historical power grid type, and the eigenvector corresponding to the target eigenvalue constitutes the projection matrix corresponding to the principal component analysis; constructing the feature alignment matrix according to the projection matrix corresponding to each historical power grid type, so as to transform each historical feature vector according to the alignment matrix

[0023] In some embodiments, the trained AI model is used to complete the fault identification of the operating distribution network system, including: obtaining the current power grid type, current operation data, and current operation time corresponding to the operating distribution network system; extracting the current feature vector corresponding to each current operation data; performing feature transformation on the current feature vector according to the current power grid type and the current feature vector to obtain the transformed current feature vector; inputting the transformed current feature vector and the current operation time into the trained AI model, and the AI model outputs the identified fault type

[0024] In some embodiments, before extracting the historical feature vector corresponding to each historical operation data, it further includes: identifying each historical operation data based on the interquartile range method, confirming the abnormal data in the historical operation data and removing the abnormal data; filling the missing values in each historical operation data based on the method of polynomial interpolation to complete the preprocessing of the historical operation data

[0025] Second aspect, the present application provides an AI-based distribution network fault identification device, including:

[0026] A fault acquisition unit, configured to acquire historical fault types of a plurality of preset distribution network systems, historical operation data corresponding to the historical fault types, and historical fault times; the historical power grid type corresponding to the distribution network system is any one of urban, rural, and industrial areas; the historical operation data includes one or more of voltage, current, temperature, and humidity;

[0027] A vector extraction unit, configured to extract historical feature vectors corresponding to each of the historical operation data;

[0028] A vector conversion unit, configured to perform feature conversion on the historical feature vectors according to the historical power grid type and the historical feature vectors corresponding to the historical power grid type, so as to obtain the converted historical feature vectors; so that the historical feature vectors corresponding to different historical power grid types are aligned;

[0029] A model input unit, configured to input the converted historical feature vectors and historical fault times into an AI model to be trained, where the AI model includes an input layer, a plurality of long short-term memory layers, and an output layer; the historical feature vectors and historical fault times are output to predict fault types through the input layer, a plurality of long short-term memory layers, and the output layer in sequence;

[0030] A model output unit, configured to complete the training of the AI model according to the predicted fault types and historical fault types corresponding to each of the historical feature vectors, and complete the fault identification of the running distribution network system according to the trained AI model.

[0031] Third aspect, the present application further provides a computer device, including:

[0032] A memory and a processor;

[0033] The memory is used to store a computer program;

[0034] The processor is configured to execute the computer program and implement the steps of an AI-based distribution network fault identification method as described in the first aspect above when executing the computer program.

[0035] Fourth aspect, the present application further provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor is caused to implement the steps of an AI-based distribution network fault identification method as described in the first aspect above.

[0036] An AI-based method for identifying faults in a distribution network and related devices provided by an embodiment of the present application. The present application proposes an AI-based method for identifying faults in a distribution network, aiming to solve the limitations of traditional fault identification methods, including limited data processing capabilities, poor adaptability, insufficient real-time performance, and high maintenance costs. The main steps of this method are as follows:

[0037] 1. Data acquisition: Collect historical fault types, corresponding historical operation data, and historical fault times of multiple preset distribution network systems. The distribution network system can be a city, rural area, or industrial zone, and the historical operation data includes voltage, current, temperature, and humidity, etc.

[0038] 2. Feature extraction: Extract historical feature vectors from the historical operation data of each distribution network system. These feature vectors contain the key information required for fault identification.

[0039] 3. Feature transformation: Perform feature transformation on the historical feature vectors according to different historical power grid types (city, rural area, industrial zone) to ensure the alignment of data features of different types of distribution network systems. This step is to improve the generalization ability of the model so that it can better process data from different environments.

[0040] 4. Model training: Input the transformed historical feature vectors and historical fault times into the AI model to be trained. This model includes an input layer, multiple long short-term memory (LSTM) layers, and an output layer. Through training, the model can learn the relationship between historical data and fault types and output the predicted fault types.

[0041] 5. Model application: The trained AI model can be used for real-time monitoring and fault identification, and can perform real-time fault early warning and rapid response on the operating distribution network system.

[0042] Based on the above steps, the method provided has at least the following beneficial effects:

[0043] 1. Improve data processing capabilities: The AI-based method can process large-scale and diverse data sets. Compared with traditional methods, it can utilize historical data more comprehensively for fault identification, improving the accuracy and reliability of fault identification.

[0044] 2. Enhance adaptability: Through feature transformation, the model can maintain high performance in different types of distribution network systems, has strong adaptability, can identify new fault modes, and is not restricted by fixed physical models and rules.

[0045] 3. Improve real-time performance: The AI model is more efficient in data processing and fault identification, can achieve real-time fault early warning and rapid response, and improves the operation safety and stability of the distribution network system.

[0046] 4. Reduce maintenance costs: By reducing the dependence on expert experience and physical models, the costs of regular maintenance and updates are lowered, making the system operation more economical and efficient.

[0047] 5. Strong generalization ability: Through feature transformation and training on different types of distribution network systems, the model has good generalization ability and can operate stably in various complex environments.

[0048] In summary, by introducing AI technology and feature transformation methods, this application solves the limitations of traditional distribution network fault identification methods in terms of data processing ability, adaptability, real-time performance, and maintenance costs, and provides a more efficient, accurate, and economical fault identification solution. This not only improves the operation safety and stability of the distribution network system but also reduces the complexity of maintenance and management.

[0049] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit this application. Brief Description of the Drawings

[0050] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of this application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0051] Figure 1 It is a schematic flowchart of the steps of a distribution network fault identification method based on AI provided by an embodiment of this application;

[0052] Figure 2 It is a schematic structural diagram of a distribution network fault identification device based on AI provided by an embodiment of this application;

[0053] Figure 3 It is a schematic block diagram of the structure of a computer device provided by an embodiment of this application.

[0054] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit this application. Detailed Embodiments

[0055] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the drawings in the embodiments of this application. Obviously, the described embodiments are some, but not all, of the embodiments of this application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of this application without creative efforts fall within the scope of protection of this application.

[0056] The flowcharts shown in the accompanying drawings are only illustrative examples, and do not necessarily include all content and operations / steps, nor do they necessarily need to be executed in the described order. For example, some operations / steps can be decomposed, combined, or partially merged, so the actual execution order may change according to the actual situation.

[0057] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, terms such as "first" and "second" are used to distinguish identical or similar items with basically the same functions and effects. Those skilled in the art can understand that terms such as "first" and "second" do not limit the quantity and execution order, and "first", "second", etc. do not necessarily mean different.

[0058] It should be understood that the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. As used in the specification of this application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0059] It should also be understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0060] The following will describe in detail some embodiments of this application in conjunction with the accompanying drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0061] Distribution network fault identification is a crucial task in the power system, directly affecting the reliability and operation efficiency of the power grid. Traditional fault identification methods mainly rely on physical models and expert experience. Although they can achieve good results in some scenarios, with the expansion of the scale and the increase in complexity of the distribution network, the limitations of these methods have become increasingly obvious. Specifically, the traditional methods have the following problems:

[0062] 1. Limited data processing ability: Traditional methods can usually only process a small amount of historical data and simple time series data, and are insufficient in processing large-scale and diverse data sets.

[0063] 2. Poor adaptability: Traditional methods are based on fixed physical models and rules, and it is difficult to adapt to different types of distribution networks and new fault modes. Once the operating environment of the distribution network changes, the performance of the model will drop significantly.

[0064] 3. Insufficient real-time performance: Traditional methods take a long time in data processing and fault identification, and it is difficult to achieve real-time fault warning and rapid response.

[0065] 4. High maintenance cost: The methods relying on expert experience and physical models need to be maintained and updated regularly, which is not only time-consuming and laborious but also error-prone.

[0066] Therefore, there is an urgent need for a method for identifying distribution network faults to solve at least one of the above problems.

[0067] To solve the above problems, please refer to Figure 1 , Figure 1 which is a schematic flowchart of an AI-based method for identifying distribution network faults provided by an embodiment of the present application. The AI-based method for identifying distribution network faults can be implemented by a computer device, and the computer device can be deployed on a single server or a server cluster. It can also be deployed on a handheld terminal, a laptop, a wearable device, a robot, etc.

[0068] To solve the above problems, please refer to Figure 1 . Specifically, as Figure 1 shown, an AI-based method for identifying distribution network faults provided includes steps S101 to S105. Details are as follows:

[0069] Step S101. Obtain historical fault types, historical operation data corresponding to the historical fault types, and historical fault times of a plurality of preset distribution network systems; the historical power grid types corresponding to the distribution network systems are any one of urban, rural, and industrial areas; the historical operation data includes one or more of voltage, current, temperature, and humidity.

[0070] Specifically, at this stage, it is necessary to collect historical fault types, historical operation data, and historical fault times from the distribution network system. The historical power grid type specifically refers to the regional type (urban, rural, or industrial area) served by the power grid, and the historical operation data includes information such as voltage, current, temperature, and humidity recorded during the operation of the distribution network. By collecting these data, a basis is provided for subsequent feature extraction and model training.

[0071] Suppose a fault identification model is being developed for a distribution network system in the suburbs of a large city. First, the method obtains the fault records of the past few years from the local power monitoring center, including fault types (such as short circuit, overload, ground fault, etc.), the specific time when the fault occurred, and the operation data before and after the fault. The operation data may include but is not limited to the following: Voltage: Measurement data of different voltage levels such as 110 kV, 35 kV, 10 kV, etc. Current: Standard operating current, abnormal current during the fault. Temperature: Ambient temperature, equipment temperature. Humidity: Ambient humidity.

[0072] Furthermore, by collecting distribution network data of different regional types, a wider range of fault modes can be covered, improving the generalization ability of the model. Containing multiple operating parameters (voltage, current, temperature, humidity), it helps to analyze the causes of faults from multiple dimensions.

[0073] Step S102. Extract the historical feature vectors corresponding to each piece of historical operating data.

[0074] Specifically, extract feature vectors from the obtained historical operating data. The feature vectors can be extracted through statistical methods, signal processing techniques, or machine learning methods, and are used to describe the key features of the data. These feature vectors will be used for subsequent model training and prediction.

[0075] For example, for each piece of historical operating data, the following features can be extracted: Voltage features: maximum value, minimum value, mean value, variance, crest factor, etc. Current features: maximum value, minimum value, mean value, variance, harmonic components, etc. Temperature features: maximum value, minimum value, mean value, change rate, etc. Humidity features: maximum value, minimum value, mean value, change rate, etc.

[0076] Suppose a data record is as follows: Voltage: 105kV, 110kV, 108kV Current: 200A, 210A, 205A Temperature: 25°C, 27°C, 26°C Humidity: 70%, 72%, 71%

[0077] The extracted feature vectors may be: Voltage features: maximum value 110kV, minimum value 105kV, mean value 107.67kV, variance 2.67 Current features: maximum value 210A, minimum value 200A, mean value 205A, variance 10 Temperature features: maximum value 27°C, minimum value 25°C, mean value 26°C, change rate 0.67°C / minute Humidity features: maximum value 72%, minimum value 70%, mean value 71%, change rate 0.67% / minute

[0078] Through feature extraction, the high-dimensional original data is converted into low-dimensional feature vectors, reducing the computational complexity. The extracted data features can more accurately reflect the system state when a fault occurs, improving the recognition accuracy of the model.

[0079] Step S103. According to the historical grid type and the historical feature vectors corresponding to the historical grid type, perform feature transformation on the historical feature vectors to obtain the transformed historical feature vectors; so as to align the data features corresponding to different historical grid types.

[0080] Specifically, perform feature transformation on the historical feature vectors corresponding to different historical power grid types (urban, rural, industrial areas) to ensure the alignment of the data features of different types of distribution networks. This step can be achieved through standardization, normalization, or other feature transformation methods, enabling the model to better process data from different regions.

[0081] Suppose the method has extracted feature vectors from the distribution network systems in urban and rural areas, but it is found that the voltage fluctuations in urban areas are large, while the temperature changes frequently in rural areas. To align these features, the following methods can be adopted: Standardization: Standardize all feature vectors so that their mean is 0 and the standard deviation is 1. Normalization: Convert the values of all feature vectors to the interval [0, 1]. Feature alignment: By introducing weights or adjusting the order of features, make the feature vectors in different regions have consistency in some key features.

[0082] Through feature transformation, ensure that the feature vectors in different regions are comparable, and avoid inaccurate model training caused by data differences. Improve the generalization ability of the model in distribution networks of different regional types, enabling it to be more widely applied.

[0083] Step S104. Input the transformed historical feature vectors and historical fault times into the AI model to be trained. The AI model includes an input layer, multiple long short-term memory layers, and an output layer; the historical feature vectors and historical fault times are output through the input layer, multiple long short-term memory layers, and the output layer to predict the fault type.

[0084] Specifically, input the transformed historical feature vectors and historical fault times into the AI model to be trained. The model includes an input layer, multiple long short-term memory (LSTM) layers, and an output layer. The LSTM layer can capture the long-term and short-term dependencies in time series data and is suitable for processing the time characteristics of distribution network operation data. The model gradually extracts features through multiple LSTM layers and finally predicts the fault type through the output layer.

[0085] Suppose the method uses an AI model with 3 LSTM layers, and the specific structure is as follows:

[0086] Input layer: Receive input data with a shape of [time step, number of features]. For example, [10, 8] means that each piece of data has 10 time steps and 8 features for each time step.

[0087] LSTM layer:

[0088] The first layer: Contains 128 units for initially extracting time features.

[0089] The second layer: Contains 64 units for further extracting intermediate features.

[0090] Layer 3: It contains 32 units and finally extracts high-order features.

[0091] Output layer: It uses a fully connected layer to output the probability distribution of the fault types.

[0092] The training process includes:

[0093] Forward propagation: The input data is passed through each layer, and finally the predicted fault type is obtained.

[0094] Backward propagation: Calculate the loss function based on the predicted fault type and the actual fault type, and adjust the model parameters through backward propagation.

[0095] Optimization: Use optimization algorithms such as gradient descent to gradually optimize the model parameters.

[0096] The AI model constructed by building 3 LSTM layers can capture the long-term and short-term dependencies in time series data and is suitable for processing the time characteristics of the operation data of the distribution network. The model can automatically learn the characteristics of different fault modes through training and adapt to the complex and changeable distribution network environment. The multi-layer LSTM structure can gradually extract and refine the features, improving the accuracy of fault identification.

[0097] Step S105. Complete the training of the AI model according to the predicted fault type and the historical fault type corresponding to each historical feature vector, and complete the fault identification of the running distribution network system according to the trained AI model.

[0098] Specifically, complete the training of the AI model according to the predicted fault type and the actual historical fault type corresponding to each historical feature vector. By evaluating the performance of the model (such as accuracy, recall rate, F1 score, etc.), ensure the effectiveness and reliability of the model. After the training is completed, apply the model to the running distribution network system to achieve real-time fault identification and early warning.

[0099] For example, using the cross-validation method, divide the dataset into a training set and a test set. Train the model on the training set and evaluate the performance of the model on the test set. Assume that the accuracy of the model on the test set is 95%, the recall rate is 90%, and the F1 score is 92%. Deploy the trained model to the distribution network monitoring system, receive the operation data from each sensor in real time, extract the feature vectors, and input them into the model for fault prediction.

[0100] Through strict model evaluation, ensure the reliability and effectiveness of the model in practical applications. The AI model can process new data in real time, detect faults in time and issue early warnings, reducing the fault response time. Once the model is trained, it can be used for a long time without frequent updates and maintenance, reducing the maintenance cost.

[0101] Meanwhile, during the training of the AI model, if the number of long short-term memory (LSTM) layers is 3, the first 2 LSTM layers can be frozen, and only the last LSTM layer and the fully connected layer are fine-tuned. By using a lower learning rate (such as 0.0001), overfitting of the model on the new dataset can be prevented. Using a smaller batch size (such as 8), the model parameters are gradually adjusted to better adapt to the target dataset. Since the training process of the LSTM network is usually very time-consuming, especially on large datasets. Freezing some network layers can significantly reduce the training time because there is no need to perform backpropagation and weight updates on these layers. Training a deep neural network requires a large amount of memory to store gradients and intermediate results. Freezing some layers can reduce memory occupancy, which is very useful especially in resource-constrained environments (such as edge devices).

[0102] Meanwhile, freezing the first two LSTM layers means that the parameters of these layers remain unchanged, and only the subsequent layers are fine-tuned. This reduces the total amount of parameter updates and the risk of overfitting of the model on the new dataset. The pre-trained LSTM model has learned general time series features from a large amount of data. By freezing the first two layers, the advantages of these pre-trained models can be retained, making its performance more stable on the new dataset.

[0103] The first two LSTM layers usually learn low-level features of time series data, such as short-term fluctuations and basic patterns. These features are relatively stable across different datasets. Freezing these layers can ensure that these low-level features are not disturbed by the noise in the new dataset. The subsequent LSTM layers and the fully connected layer learn more advanced features, such as fault patterns and long-term dependencies. These features are more specific and need to be adjusted for a specific dataset. Freezing the first two layers can concentrate resources on the adjustment of these advanced features. By freezing the first two LSTM layers and gradually fine-tuning the subsequent layers, the model can adapt to the new dataset more gradually. This gradual adaptation method helps the model better understand the features of the new dataset without performance degradation due to sudden parameter changes. After each fine-tuning, the model can gradually find local optimal solutions, avoiding the computational complexity brought by searching for the global optimal solution.

[0104] If the entire model is retrained from scratch, the model may forget the general features learned in the pre-training stage, which is called catastrophic forgetting. Freezing some layers can prevent this from happening and maintain the generalization ability of the model.

[0105] In summary, through the above steps, the AI-based distribution network fault identification method provided by this application can effectively solve the limitations of traditional methods in terms of data processing capabilities, adaptability, real-time performance, and maintenance costs, and improve the reliability and operation efficiency of the distribution network.

[0106] In some embodiments, according to the historical power grid type and the historical feature vector corresponding to the historical power grid type, the historical feature vector is subjected to feature transformation to obtain the transformed historical feature vector, including: standardizing each historical feature vector according to the historical power grid type; performing feature enhancement on each historical feature vector; performing principal component analysis on each enhanced historical feature vector to obtain the feature alignment matrix of each historical feature vector, and transforming the corresponding historical feature vector according to the feature alignment matrix to obtain the transformed historical feature vector, so as to complete the feature alignment of multiple historical feature vectors.

[0107] Specifically, in this embodiment, the historical feature vector is subjected to feature transformation through the following three main steps to achieve the feature alignment of data of different historical power grid types: Standardization: Standardize each historical feature vector according to the historical power grid type. Feature enhancement: Perform feature enhancement on each historical feature vector. Principal component analysis (PCA): Perform principal component analysis on each enhanced historical feature vector to obtain the feature alignment matrix of each historical feature vector, and transform the corresponding historical feature vector according to the feature alignment matrix to obtain the transformed historical feature vector, so as to complete the feature alignment of multiple historical feature vectors.

[0108] Standardization is to transform each feature vector in the dataset into a form with a mean of 0 and a standard deviation of 1. This can eliminate the dimensional differences between different features and make it easier for the model to process. Suppose the current feature vector in a certain urban area is [200A, 210A, 205A], its mean is 205A, and the standard deviation is 5A. The standardized current feature vector is: [-1, 1, 0].

[0109] Feature enhancement is to enrich the dataset by generating new features and improve the recognition ability of the model. Common feature enhancement methods include polynomial features, cross features, window features, etc. For example, polynomial features enhance the current feature vector [200A, 210A, 205A] to [200A, 210A, 205A, 200^2A^2, 210^2A^2, 205^2A^2]. Cross features enhance the current and voltage feature vectors [200A, 210A, 205A] and [105kV, 110kV, 108kV] to [200A, 210A, 205A, 105kV, 110kV, 108kV, 200A * 105kV, 210A * 110kV, 205A * 108kV].

[0110] Principal component analysis is a dimensionality reduction technique that transforms the original features into a new feature space through linear transformation, such that the new features represent the main directions of data variation. This helps reduce the dimensionality of the data while retaining most of the information. Suppose the current and voltage feature vectors after feature enhancement are [200A, 210A, 205A, 105kV, 110kV, 108kV, 200A * 105kV, 210A * 110kV, 205A * 108kV]. Through PCA, we may select the first two principal components to reduce the feature vector to [0.9, 0.2]. Based on the principal components obtained from PCA, a feature alignment matrix is constructed to align the historical feature vectors of different grid types into the same feature space. Suppose the feature alignment matrix for the urban grid is M 城市 , and the feature alignment matrix for the rural grid is M 农村 . Let the feature vector of the urban grid be X 城市 . After being transformed by M 城市 , it is converted into the aligned feature vector X 对齐 . Then the relationship is: X 对齐 = M 城市 * X 城市 .

[0111] The above embodiments eliminate the dimensional differences between different features through standardization processing, making the model easier to process and converge. The standardized data helps accelerate the training process of the model and improve the training efficiency. At the same time, by generating new features, the information volume of the dataset is increased, which helps the model capture fault patterns more accurately. Feature enhancement enables the model to better adapt to different types of fault patterns and improve the generalization ability. By reducing the high-dimensional feature vector to a low-dimensional feature vector through PCA, the computational complexity is reduced and the running efficiency of the model is improved. PCA retains the main directions of data variation by selecting the principal components, reducing the loss of information. Finally, through the feature alignment matrix, the historical feature vectors of different grid types are aligned into the same feature space, ensuring that the model can process data from different regions. Feature alignment helps improve the recognition performance of the model in distribution networks in different regions, making it more stable and reliable.

[0112] Through the technical content of this embodiment, the problem of data feature differences in distribution networks in different regions can be effectively solved, the training effect and application effect of the AI model can be improved, and the accuracy and real-time performance of distribution network fault recognition can be further enhanced

[0113] Exemplarily, standardizing each historical feature vector according to the historical power grid type includes: obtaining the first mean and the first standard deviation corresponding to each historical feature vector with the historical power grid type of urban; obtaining the second mean and the second standard deviation corresponding to each historical feature vector with the historical power grid type of rural; obtaining the third mean and the third standard deviation corresponding to each historical feature vector with the historical power grid type of industrial area; standardizing each historical feature vector with the historical power grid type of urban according to the first mean and the first standard deviation; standardizing each historical feature vector with the historical power grid type of rural according to the second mean and the second standard deviation; standardizing each historical feature vector with the historical power grid type of industrial area according to the third mean and the third standard deviation.

[0114] The expression for standardizing each corresponding historical feature vector may include:

[0115] ; ; ;

[0116] Wherein, is the vector after standardizing the historical feature vector with the historical power grid type of urban, is the historical feature vector with the historical power grid type of urban, and are the first mean and the first standard deviation; is the vector after standardizing the historical feature vector with the historical power grid type of rural, is the historical feature vector with the historical power grid type of rural, and are the second mean and the second standard deviation; is the vector after standardizing the historical feature vector with the historical power grid type of industrial area, is the historical feature vector with the historical power grid type of industrial area, and are the third mean and the third standard deviation. Through the standardization process, the dimensional difference between different features is eliminated, making the model easier to process and converge. At the same time, the standardized data helps to accelerate the training process of the model and improve the training efficiency.

[0117] Exemplarily, feature enhancement of each historical feature vector includes: determining a reference feature vector among multiple historical feature vectors; calculating the value of the Gaussian kernel function of each historical feature vector and the reference feature vector; completing the feature enhancement of the historical feature vector according to the value of the Gaussian kernel function corresponding to each historical feature vector; wherein the expression of the Gaussian kernel function includes:

[0118] ;

[0119] Wherein, is the Gaussian kernel function, is the th historical feature vector, is the th reference feature vector, is the bandwidth parameter, which determines the distribution of the historical feature vectors in the high-dimensional space. The expression of the enhanced historical feature vector includes:

[0120] ;

[0121] is the enhanced historical feature vector , represents the Gaussian kernel function of the historical feature vector and the th reference feature vector, represents the Gaussian kernel function of the historical feature vector and the th reference feature vector represents the Gaussian kernel function of the historical feature vector and the th reference feature vector represents the Gaussian kernel function of the historical feature vector and the th reference feature vector, where m is the number of reference feature vectors.

[0122] By selecting one of multiple historical feature vectors as the reference feature vector. For example, the average value of the historical feature vectors of the urban power grid can be selected as the reference feature vector of the urban power grid, the median of the historical feature vectors of the rural power grid can be selected as the reference feature vector of the rural power grid, and the best-performing feature vector of the historical feature vectors of the industrial area power grid can be selected as the reference feature vector of the industrial area power grid. For each historical feature vector, calculate the value of the Gaussian kernel function between it and the reference feature vector. Use the value of the Gaussian kernel function as a weight to perform weighted processing on the historical feature vectors, or use the value of the Gaussian kernel function as a new feature to add to the original feature vector. By introducing the value of the Gaussian kernel function, the representation ability of the feature vector is enhanced, making it more capable of reflecting the similarities and differences between data. The Gaussian kernel function can effectively capture the non-linear relationship between feature vectors, enabling the model to better process complex non-linear data. Through weighted processing or adding new features, the noise in the original feature vector can be reduced, improving the cleanliness and reliability of the data. The enhanced feature vector can provide more information, helping to improve the classification and prediction performance of the model, making it more accurate and reliable in practical applications. Selecting the reference feature vectors of different regions enables the feature enhancement process to better adapt to the data characteristics of different regions, improving the generalization ability of the model. Furthermore, the representation ability of each historical feature vector can be effectively enhanced, improving the performance and reliability of the model in distribution network fault identification and other applications.

[0123] At the same time, the value of the bandwidth parameter can be selected by performing cross-validation on the training data to choose the bandwidth parameter value that optimizes the model performance. Or an empirical value can be selected according to the scale and distribution of the feature vectors. For example, the bandwidth parameter can be set to the average distance between historical feature vectors.

[0124] Exemplarily, perform principal component analysis on each enhanced historical feature vector to obtain the feature alignment matrix of each historical feature vector, including: calculating the covariance matrix corresponding to the historical feature vectors of the same historical power grid type after enhancement; obtaining the eigenvalues of each covariance matrix and the eigenvectors corresponding to each eigenvalue; determining the target eigenvalue among the multiple eigenvalues corresponding to each historical power grid type, and the eigenvectors corresponding to the target eigenvalue form the projection matrix corresponding to the principal component analysis; constructing the feature alignment matrix according to the projection matrix corresponding to each historical power grid type to transform each historical feature vector according to the alignment matrix.

[0125] The high-dimensional feature vectors are reduced to low-dimensional feature vectors through PCA, which reduces the data complexity and improves the computational efficiency and the running speed of the model. The eigenvectors corresponding to the larger eigenvalues are selected to ensure that the feature vectors after dimensionality reduction still retain the main change directions and information of the data, reducing the information loss. Through the feature alignment matrix, the historical feature vectors of different power grid types are aligned into the same feature space, ensuring that the model can process data from different regions and improving the generalization ability of the model. The aligned feature vectors are more comparable and consistent, which helps to improve the recognition performance of the model in distribution networks in different regions, making it more stable and reliable. Through the aligned feature vectors, the input data of the model is simplified, making the model easier to train and converge, and improving the training efficiency.

[0126] In some embodiments, the trained AI model is used to complete the fault identification of the running distribution network system, including: obtaining the current power grid type, current operation data, and current operation time corresponding to the running distribution network system; extracting the current feature vectors corresponding to each current operation data; performing feature transformation on the current feature vectors according to the current power grid type and the current feature vectors to obtain the transformed current feature vectors; inputting the transformed current feature vectors and the current operation time into the trained AI model, and the AI model outputs the identified fault type.

[0127] By obtaining the current power grid type corresponding to the running distribution network system (such as urban power grid, rural power grid, industrial area power grid, etc.). Collect the current operation data, which may include key electrical parameters such as voltage, current, power factor, and frequency. Record the current operation time, which may include specific date, timestamp, or time period information. Extract the current feature vectors from the current operation data. These feature vectors are numerical representations of the data and can reflect the key information of the power grid state. According to the current power grid type and the current feature vectors, use the previously determined feature alignment matrix or other transformation methods to perform feature transformation on the current feature vectors. The purpose of the transformation is to make the current feature vectors in the same feature space as the feature vectors in the training data to ensure that the model can accurately perform fault identification. Input the transformed current feature vectors and the current operation time into the trained AI model. The AI model outputs the identified fault type according to the input feature vectors and time information. The fault types may include short circuit, overload, ground fault, etc.

[0128] Furthermore, the provided method allows for real-time fault detection in an operating distribution network system, enabling timely response and reducing potential losses caused by faults. Through feature transformation, it is ensured that the current feature vector is in the same feature space as the feature vector used during model training, thereby improving the accuracy of fault identification. This method can perform feature transformation according to different grid types (such as urban, rural, industrial areas), enabling the model to adapt to various grid environments and enhancing the generality and practicality of the model. Introducing the current operating time as one of the inputs enables the model to consider time-related factors (such as seasonal variations, daily load curves, etc.), further improving the accuracy of fault identification. Automatically identifying the fault type reduces the need for manual intervention, speeds up the fault response, and improves the operational efficiency of the entire distribution network system.

[0129] Through the above embodiments, distribution network operators can more effectively monitor and manage their grid assets, respond promptly to potential fault situations, and thus ensure the stable operation of the grid and power supply reliability. In some embodiments, before extracting the historical feature vector corresponding to each historical operation data, it further includes: identifying each historical operation data based on the interquartile range method, confirming the abnormal data in the historical operation data and removing the abnormal data; filling the missing values in each historical operation data based on the polynomial interpolation method to complete the preprocessing of the historical operation data.

[0130] Calculate the quartiles (the first quartile Q1 and the third quartile Q3) and the interquartile range IQR = Q3 - Q1 of each historical operation data through the interquartile range method (Interquartile Range, IQR) to determine the range of abnormal data. Points outside the range [Q1 - 1.5IQR, Q3 + 1.5IQR] are regarded as abnormal data and removed. Then, for the missing values in each historical operation data, use the polynomial interpolation method to fill them. Select an appropriate interpolation order (such as a first-order, second-order, or third-order polynomial), and fit a polynomial function according to the known data points. Use the fitted polynomial function to predict and fill the missing values.

[0131] Identifying and removing abnormal data through the interquartile range method reduces the noise and errors in the data, improving the data quality. Filling the missing values through polynomial interpolation makes the data more complete and continuous, improving the data reliability. High-quality data can make the trained AI model more accurate and stable, reducing the model bias and overfitting problems caused by abnormal data and missing values. A complete dataset helps the model better learn the distribution and variation rules of features, improving the generalization ability of the model.

[0132] Through the preprocessing step, the historical feature vectors are more in line with the actual power grid operation conditions, thus showing higher accuracy and robustness in fault identification. The model can more effectively identify and locate faults, improving the timeliness and effectiveness of fault handling. After removing abnormal data and filling in missing values, the consistency of each data point in the dataset is better, which helps the model to be applied in different time periods and different power grid types. A dataset with strong consistency can improve the applicability and reliability of the model in different scenarios. The steps of abnormal data identification and missing value filling can be automated, reducing the complexity and time cost of manual processing. The preprocessed dataset is more standardized, providing convenience for subsequent feature extraction and model training.

[0133] Through the above content, the quality of historical operation data can be effectively improved, ensuring that the trained AI model is more accurate and stable, and thus showing higher fault identification performance and reliability in practical applications.

[0134] The method provided by the embodiment of the present application first collects the historical fault types, corresponding historical operation data, and historical fault times of multiple preset distribution network systems. The distribution network systems can be cities, rural areas, or industrial areas, and the historical operation data includes voltage, current, temperature, humidity, etc. Then, historical feature vectors are extracted from the historical operation data of each distribution network system, and these feature vectors contain the key information required for fault identification. Then, according to different historical power grid types (cities, rural areas, industrial areas), the historical feature vectors are subjected to feature transformation to ensure the alignment of data features of different types of distribution network systems. This step is to improve the generalization ability of the model so that it can better process data from different environments. Further, the transformed historical feature vectors and historical fault times are input into the AI model to be trained, which includes an input layer, multiple long short-term memory layers (LSTM), and an output layer. Through training, the model can learn the relationship between historical data and fault types and output the predicted fault types. Finally, the trained AI model can be used for real-time monitoring and fault identification, and can give real-time fault warnings and quick responses to the running distribution network system.

[0135] The method provided by the embodiment of the present application has at least the following beneficial effects:

[0136] 1. Improve data processing ability: The AI-based method can process large-scale and diverse datasets. Compared with traditional methods, it can make more comprehensive use of historical data for fault identification, improving the accuracy and reliability of fault identification.

[0137] 2. Enhance adaptability: Through feature transformation, the model can maintain high performance in different types of distribution network systems, has strong adaptability, can identify new fault modes, and is not restricted by fixed physical models and rules.

[0138] 3. Improve real-time performance: The AI model is more efficient in data processing and fault identification, enabling real-time fault warning and rapid response, thus enhancing the operational safety and stability of the distribution network system.

[0139] 4. Reduce maintenance costs: By reducing the dependence on expert experience and physical models, the costs of regular maintenance and updates are lowered, making the system operation more economical and efficient.

[0140] 5. Strong generalization ability: Through feature transformation and training on different types of distribution network systems, the model has good generalization ability and can operate stably in various complex environments.

[0141] In summary, this application solves the limitations of traditional distribution network fault identification methods in terms of data processing capacity, adaptability, real-time performance, and maintenance costs by introducing AI technology and feature transformation methods, and provides a more efficient, accurate, and economical fault identification solution. This not only improves the operational safety and stability of the distribution network system but also reduces the complexity of maintenance and management.

[0142] Please refer to Figure 2 shown in Figure 2 FIG. 200 is a schematic structural diagram of a distribution network fault identification device 200 based on AI provided by an embodiment of this application. The distribution network fault identification device 200 based on AI is used to execute the steps of a distribution network fault identification method based on AI shown in the above embodiments. The distribution network fault identification device 200 based on AI can be a single server or a server cluster, or the distribution network fault identification device 200 based on AI can be a terminal, and the terminal can be a handheld terminal, a laptop, a wearable device, or a robot, etc.

[0143] As Figure 2 shown in FIG., the distribution network fault identification device 200 based on AI includes:

[0144] A fault acquisition unit 201, configured to acquire historical fault types of a plurality of preset distribution network systems, historical operation data corresponding to the historical fault types, and historical fault times; the historical power grid types corresponding to the distribution network systems are any one of urban, rural, and industrial areas; the historical operation data includes one or more of voltage, current, temperature, and humidity;

[0145] A vector extraction unit 202, configured to extract historical feature vectors corresponding to each of the historical operation data;

[0146] A vector conversion unit 203, configured to perform feature conversion on the historical feature vector according to the historical power grid type and the historical feature vector corresponding to the historical power grid type, so as to obtain the converted historical feature vector, and align the historical feature vectors corresponding to different historical power grid types.

[0147] A model input unit 204, configured to input the converted historical feature vector and the historical fault time into an AI model to be trained, where the AI model includes an input layer, a plurality of long short-term memory layers, and an output layer; the historical feature vector and the historical fault time are output through the input layer, the plurality of long short-term memory layers, and the output layer in sequence to predict the fault type.

[0148] A model output unit 205, configured to complete the training of the AI model according to the predicted fault type and the historical fault type corresponding to each historical feature vector, and complete the fault identification of the operating distribution network system according to the trained AI model.

[0149] It should be noted that those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described AI-based distribution network fault identification device and each module can refer to the corresponding processes in the above-described embodiments of the AI-based distribution network fault identification method, and will not be elaborated herein.

[0150] The above-described AI-based distribution network fault identification method can be implemented in the form of a computer program, and the computer program can run on a device as shown in Figure 2 shown.

[0151] Please refer to Figure 3 , Figure 3 , which is a schematic block diagram of the structure of a computer device provided by an embodiment of the present application. The computer device includes a processor, a memory, and a network interface connected through a device bus. Among them, the memory may include a storage medium and an internal memory.

[0152] The storage medium can store an operating device and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any AI-based distribution network fault identification method.

[0153] The processor is used to provide computing and control capabilities to support the operation of the entire computer device.

[0154] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any AI-based distribution network fault identification method.

[0155] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that Figure 3 The structure shown in Figure 3 is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the terminal to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0156] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0157] Wherein, in one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0158] Obtain historical fault types of multiple preset distribution network systems, historical operation data corresponding to the historical fault types, and historical fault times; the historical power grid type corresponding to the distribution network system is any one of urban, rural, and industrial areas; the historical operation data includes one or more of voltage, current, temperature, and humidity;

[0159] Extract historical feature vectors corresponding to each of the historical operation data;

[0160] According to the historical power grid type and the historical feature vectors corresponding to the historical power grid type, perform feature transformation on the historical feature vectors to obtain the transformed historical feature vectors; so that the data features corresponding to different historical power grid types are aligned;

[0161] Input the transformed historical feature vectors and historical fault times into the AI model to be trained. The AI model includes an input layer, multiple long short-term memory layers, and an output layer; the historical feature vectors and historical fault times are output through the input layer, multiple long short-term memory layers, and the output layer to predict the fault type;

[0162] Complete the training of the AI model according to the predicted fault type and historical fault type corresponding to each historical feature vector, and complete the fault identification of the running distribution network system according to the trained AI model.

[0163] In some embodiments, the obtaining of the transformed historical feature vectors by performing feature transformation on the historical feature vectors according to the historical power grid type and the historical feature vectors corresponding to the historical power grid type includes: standardizing each historical feature vector according to the historical power grid type; performing feature enhancement on each historical feature vector; performing principal component analysis on each enhanced historical feature vector to obtain the feature alignment matrix of each historical feature vector, and transforming the corresponding historical feature vector according to the feature alignment matrix to obtain the transformed historical feature vector, so as to complete the feature alignment of the multiple historical feature vectors.

[0164] Exemplarily, the standardizing of each historical feature vector according to the historical power grid type includes: obtaining the first mean and the first standard deviation corresponding to each historical feature vector with the historical power grid type being urban; obtaining the second mean and the second standard deviation corresponding to each historical feature vector with the historical power grid type being rural; obtaining the third mean and the third standard deviation corresponding to each historical feature vector with the historical power grid type being industrial area; standardizing each historical feature vector with the historical power grid type being urban according to the first mean and the first standard deviation; standardizing each historical feature vector with the historical power grid type being rural according to the second mean and the second standard deviation; standardizing each historical feature vector with the historical power grid type being industrial area according to the third mean and the third standard deviation.

[0165] Exemplarily, the performing of feature enhancement on each historical feature vector includes: determining a reference feature vector among the multiple historical feature vectors; calculating the value of the Gaussian kernel function of each historical feature vector and the reference feature vector; completing the feature enhancement of the historical feature vector according to the value of the Gaussian kernel function corresponding to each historical feature vector; wherein the expression of the Gaussian kernel function includes:

[0166] ;

[0167] wherein, is the Gaussian kernel function, is the th historical feature vector, is the th reference feature vector, is a bandwidth parameter that determines the distribution of the historical feature vector in the high-dimensional space. The expression of the enhanced historical feature vector includes:

[0168] ;

[0169] is the enhanced historical feature vector , represents the historical feature vector and the th benchmark feature vector's Gaussian kernel function represents the historical feature vector and the th benchmark feature vector's Gaussian kernel function represents the historical feature vector and the th benchmark feature vector's Gaussian kernel function represents the historical feature vector and the th benchmark feature vector's Gaussian kernel function, where m is the number of the benchmark feature vectors.

[0170] Exemplarily, performing principal component analysis on each of the enhanced historical feature vectors to obtain the feature alignment matrix of each historical feature vector includes: calculating the covariance matrix corresponding to the historical feature vectors of the same historical power grid type after enhancement; obtaining the eigenvalues of each covariance matrix and the eigenvectors corresponding to each eigenvalue; determining the target eigenvalue among the multiple eigenvalues corresponding to each historical power grid type, and the eigenvectors corresponding to the target eigenvalue constitute the projection matrix corresponding to the principal component analysis; constructing the feature alignment matrix according to the projection matrix corresponding to each historical power grid type, so as to transform each historical feature vector according to the alignment matrix.

[0171] In some embodiments, the trained AI model is used to complete the fault identification of the running distribution network system, including: obtaining the current power grid type, current operation data, and current operation time corresponding to the running distribution network system; extracting the current feature vector corresponding to each current operation data; performing feature transformation on the current feature vector according to the current power grid type and the current feature vector to obtain the transformed current feature vector; inputting the transformed current feature vector and the current operation time into the trained AI model, and the AI model outputs the identified fault type.

[0172] In some embodiments, before extracting the historical feature vector corresponding to each piece of the historical operation data, the method further includes: identifying each piece of the historical operation data based on the interquartile range method, confirming the abnormal data in the historical operation data and removing the abnormal data; filling the missing values in each piece of the historical operation data based on the method of polynomial interpolation, and completing the preprocessing of the historical operation data.

[0173] The present application also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the processor is caused to implement the steps of an AI-based distribution network fault identification method as described in the first aspect above.

[0174] Among them, the computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a SmartMedia Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the computer device.

[0175] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. An AI-based method for identifying faults in a distribution network, characterized in that, Including: Obtain the historical fault types of multiple preset distribution network systems, the historical operation data corresponding to the historical fault types, and the historical fault times; the corresponding historical power grid types of the distribution network systems are cities, rural areas, and industrial zones; the historical operation data includes one or more of voltage, current, temperature, and humidity; Extract the historical feature vectors corresponding to each of the historical operation data; According to the historical power grid type and the historical feature vectors corresponding to the historical power grid type, perform feature transformation on the historical feature vectors to obtain the transformed historical feature vectors, including: standardize each historical feature vector according to the historical power grid type, and perform feature enhancement on each historical feature vector, including: determine a reference feature vector among multiple historical feature vectors; calculate the values of the Gaussian kernel function of each historical feature vector and the reference feature vector; complete the feature enhancement of the historical feature vectors according to the values of the Gaussian kernel function corresponding to each historical feature vector; perform principal component analysis on each enhanced historical feature vector to obtain the feature alignment matrix of each historical feature vector, and transform the corresponding historical feature vector according to the feature alignment matrix to obtain the transformed historical feature vector, so as to complete the feature alignment of multiple historical feature vectors; so that the data features corresponding to different historical power grid types are aligned; where the expression of the Gaussian kernel function includes: ； Among them, is the Gaussian kernel function, is the th historical feature vector, is the th reference feature vector, is the bandwidth parameter, which is the average distance between historical feature vectors and determines the distribution of historical feature vectors in the high-dimensional space; the reference feature vector of the urban power grid is the average value of the historical feature vectors of the urban power grid, the reference feature vector of the rural power grid is the median of the historical feature vectors of the rural power grid, and the reference feature vector of the industrial area power grid is the best performance feature vector of the historical feature vectors of the industrial area power grid; Input the transformed historical feature vectors and historical fault times into the AI model to be trained. The AI model includes an input layer, multiple long short-term memory layers, and an output layer; the historical feature vectors and historical fault times are output to predict the fault type through the input layer, multiple long short-term memory layers, and the output layer in sequence; Complete the training of the AI model according to the predicted fault type and historical fault type corresponding to each historical feature vector, and complete the fault identification of the running distribution network system according to the trained AI model.

2. The method according to claim 1, characterized in that, The standardizing each historical feature vector according to the historical power grid type includes: Obtain the first mean and the first standard deviation corresponding to each historical feature vector with the historical power grid type of the city; Obtain the second mean and the second standard deviation corresponding to each historical feature vector with the historical power grid type of the rural area; Obtain the third mean and the third standard deviation corresponding to each historical feature vector with the historical power grid type of the industrial zone; Standardize each historical feature vector with the historical power grid type of the city according to the first mean and the first standard deviation; Standardize each historical feature vector with the historical power grid type of the rural area according to the second mean and the second standard deviation; Standardize each historical feature vector with the historical power grid type of the industrial zone according to the third mean and the third standard deviation.

3. The method according to claim 1, wherein The expression of the enhanced historical feature vector includes: ； For the enhanced historical feature vector , denote the Gaussian kernel function of the historical feature vector and the th benchmark feature vector denote the Gaussian kernel function of the historical feature vector and the th benchmark feature vector denote the Gaussian kernel function of the historical feature vector and the th benchmark feature vector denote the Gaussian kernel function of the historical feature vector and the th benchmark feature vector, where m is the number of the benchmark feature vectors.

4. The method according to claim 1, wherein The performing principal component analysis on each enhanced historical feature vector to obtain the feature alignment matrix of each historical feature vector includes: Calculate the covariance matrix corresponding to the historical feature vectors of the same historical power grid type after enhancement; Obtaining the eigenvalues of each of the covariance matrices and the eigenvectors corresponding to each of the eigenvalues; Determining a target eigenvalue among the plurality of eigenvalues corresponding to each of the historical power grid types, wherein the eigenvectors corresponding to the target eigenvalues constitute a projection matrix corresponding to the principal component analysis; The feature alignment matrix is constructed according to the projection matrix corresponding to each of the historical power grid types, so as to transform each of the historical feature vectors according to the feature alignment matrix.

5. The method according to claim 1, wherein The AI model that has been trained completes fault identification of the running distribution network system, including: Obtain the current grid type, current operating data, and current operating time corresponding to the running distribution network system; Extracting a current feature vector corresponding to each of the current running data; Performing feature conversion on the current feature vector according to the current power grid type and the current feature vector to obtain the converted current feature vector; The converted current feature vector and current running time are input into the trained AI model, and the AI model output identifies the fault type.

6. The method according to claim 1, wherein Before extracting the historical feature vector corresponding to each of the historical operation data, the method further includes: Identify each of the historical operating data based on the interquartile range method, confirm abnormal data in the historical operating data, and eliminate the abnormal data; The missing values in each of the historical operation data are filled based on a polynomial interpolation method to complete the preprocessing of the historical operation data.

7. An AI-based distribution network fault identification device, characterized in that, include: A fault acquisition unit is configured to acquire historical fault types of a plurality of preset distribution network systems, historical operating data corresponding to the historical fault types, and historical fault times; the historical power grid types corresponding to the distribution network systems are urban, rural, and industrial areas; the historical operating data includes one or more of voltage, current, temperature, and humidity; A vector extraction unit, configured to extract a historical feature vector corresponding to each of the historical operation data; A vector conversion unit is configured to perform feature conversion on the historical feature vectors according to the historical power grid type and the historical feature vectors corresponding to the historical power grid type, and obtain the converted historical feature vectors, including: standardizing each historical feature vector according to the historical power grid type, and performing feature enhancement on each historical feature vector, including: determining a baseline feature vector among multiple historical feature vectors; calculating the value of the Gaussian kernel function of each historical feature vector and the baseline feature vector; completing feature enhancement of the historical feature vector according to the value of the Gaussian kernel function corresponding to each historical feature vector; performing principal component analysis on each enhanced historical feature vector to obtain a feature alignment matrix for each historical feature vector, converting the corresponding historical feature vector according to the feature alignment matrix, and obtaining the converted historical feature vector to complete feature alignment of multiple historical feature vectors; so that the historical feature vectors corresponding to different historical power grid types are aligned; wherein the expression of the Gaussian kernel function includes: ； Among them, is the Gaussian kernel function, is the th historical feature vector, is the th reference feature vector, is the bandwidth parameter, which is the average distance between historical feature vectors and determines the distribution of historical feature vectors in the high-dimensional space. The reference feature vector of the urban power grid is the average value of the historical feature vectors of the urban power grid, the reference feature vector of the rural power grid is the median of the historical feature vectors of the rural power grid, and the reference feature vector of the industrial area power grid is the best performance feature vector of the historical feature vectors of the industrial area power grid; A model input unit, configured to input the converted historical feature vector and historical fault time into an AI model to be trained, where the AI model includes an input layer, a plurality of long short-term memory layers, and an output layer; the historical feature vector and historical fault time are sequentially output through the input layer, a plurality of long short-term memory layers, and the output layer to predict a fault type. A model output unit, configured to complete the training of the AI model according to the predicted fault type and historical fault type corresponding to each historical feature vector, and complete the fault identification of the running distribution network system according to the trained AI model.

8. A computer device, characterized in that, The computer device includes a memory and a processor. The memory is used to store a computer program. The processor is configured to execute the computer program and implement the method according to any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the processor is caused to implement the method according to any one of claims 1 to 6.

Images