A method for detecting faults of power distribution equipment, electronic equipment and storage medium

Abstract

A method for fault detection of power distribution equipment belongs to the field of deep learning technology and solves the problem of low fault detection accuracy caused by the lack of operation and maintenance data in a power system. The invention collects operation and maintenance data of a power distribution network, performs data segmentation on a state data set based on a time series, extracts spatial features through a graph convolutional network, and uses a multi-layer perceptron to extract and learn the time dependency and nonlinear relationship in the data; performs feature splicing to unify the dimensions of the data, and fuses feature vectors after bit-by-bit addition of the unified dimensions; merges the feature vectors with environmental information, and sends the merged feature vectors to a deep learning network composed of a multi-layer perceptron and a fully connected layer for training; uses a mean square error loss function to evaluate model deviations, and uses back propagation and gradient descent to update and optimize model parameters. The invention improves the stability, reliability and operation efficiency of the power system.

Description

Technical Field

[0001] The present invention belongs to the technical field of deep learning algorithms and relates to a distribution equipment fault detection method, an electronic device and a storage medium. Background Art

[0002] In the operation and maintenance of power distribution networks, it is crucial to monitor the status of equipment such as transformers and circuit breakers. The status data of these equipment usually includes but is not limited to voltage ( )、Current( ) and power ( ) and other parameters. These parameters are collected through a series of sensors and measuring devices, and the quality and integrity of these data are crucial to ensure the stable operation of the power grid. However, in actual operation, due to various reasons, such as intermittent data collection, equipment failures, problems in the data transmission process, etc., these status data often have many missing values ​​or discontinuities, which directly affect the reliability of the stable operation of the system. Summary of the invention

[0003] The technical solution of the present invention is used to solve the problem of low fault detection accuracy caused by the lack of operation and maintenance data in the power system.

[0004] The present invention solves the above technical problems through the following technical solutions:

[0005] A method for detecting a fault in a power distribution device comprises the following steps:

[0006] Step 1: Collect the power distribution network operation and maintenance data with missing values ​​or discontinuities, and represent it as a state data set at a series of time points, denoted as ,in Representative Status data of power distribution equipment at each moment;

[0007] Step 2: Split the state data set into segments based on the order of the time series. Evenly divide the data into multiple subsets, with the first half of the subsets recorded as the left half subset and the second half as the right half subset; the left half subset can reveal the behavior and characteristics of the power system in the early stage of operation, and the right half subset is helpful for analyzing and understanding the changes and trends in the later stage of the time series;

[0008] Step 3: Each data subset is first passed through a graph convolutional network to extract spatial features using the topological structure of the power system; then, the output of the graph convolutional network is fed into a multi-layer perceptron to extract and learn the time dependency and nonlinear relationships in the data;

[0009] Step 4: splice the features of the data stream extracted from the left half subset with the features of the data stream extracted from the corresponding right half subset to ensure time continuity; use zero padding technology to process the spliced ​​feature vectors to unify the dimension of the data, and use the bit-by-bit addition method to fuse the feature vectors after unification of the dimension;

[0010] Step 5: Collect relevant environmental information, convert the environmental information into numerical features, and design an embedding layer to embed the environmental data into the network;

[0011] Step 6: Merge the feature vector with the environmental information, and send the merged feature vector to a deep learning network consisting of a multi-layer perceptron and a fully connected layer for training;

[0012] Step 7: Use the mean square error loss function to evaluate the deviation between the power system state predicted by the model and the actual observed state;

[0013] Step 8: Use back propagation and gradient descent algorithms to update and optimize the parameters of the model and use them to predict the operating status of the power system.

[0014] Furthermore, the power distribution network operation and maintenance data includes: voltage, current and power.

[0015] Furthermore, the relevant environmental information includes: temperature T, humidity H and alarm information A.

[0016] Furthermore, the formula of the mean square error loss function is as follows:

[0017]

[0018] in, represents the predicted value, Represents the true value.

[0019] Furthermore, the formula for updating and optimizing the parameters is as follows:

[0020]

[0021] in, represents the learning rate, represents the mean square error loss, represents the original parameters of the model, Represents new parameters of the model.

[0022] An electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above-mentioned distribution equipment fault detection method, and the processor is configured to execute the program stored in the memory.

[0023] A storage medium stores a computer program, which executes the steps of the above-mentioned distribution equipment fault detection method when executed by a processor.

[0024] The advantages of the present invention are:

[0025] The technical solution of the present invention can not only complete the missing data, but also timely discover the abnormal status of the equipment through data analysis, thereby greatly improving the stability and reliability of the power system, not only helping to improve the operating efficiency of the power grid, but also enhancing the ability to prevent potential problems in the future; it can identify and report potential faults and achieve early warning, thereby allowing operators to take timely maintenance and repair measures to reduce the impact of power outages and faults. It reduces operation and maintenance costs and enhances the overall stability and reliability of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a model structure diagram of the power distribution equipment fault detection method according to the first embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are 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.

[0028] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments:

[0029] Embodiment 1

[0030] like Figure 1 As shown, the power distribution equipment fault detection method of this embodiment includes the following contents:

[0031] 1. Data Collection

[0032] Collect power distribution network operation and maintenance data with missing values ​​or discontinuities, and represent the data in this case as a state data set at a series of time points ,in Representative Device status data at a given moment.

[0033] 2. Data segmentation and feature extraction

[0034] The operation and maintenance data of the power distribution network, such as voltage, current and power, are segmented to deal with data discontinuity and missing data caused by data transmission problems or insufficient signal acquisition frequency. By segmenting each type of data into multiple subsets, dynamic features at different time scales can be captured, which is difficult to achieve with traditional methods. On each segmented subset, a combination of graph convolutional network (GCN) and multi-layer perceptron (MLP) is used to extract spatial-temporal features, which can not only capture the structural information of the power distribution network, but also learn the nonlinear dependencies of the data through MLP.

[0035] 2.1 Data segmentation

[0036] A detailed time series data segmentation method is introduced to deal with discontinuities and missing issues in power distribution network operation and maintenance data (such as voltage, current and power). This data segmentation technique allows the dynamic changes of data to be captured at different time scales, greatly enhancing the sensitivity and accuracy of detecting potential faults.

[0037] The specific data segmentation process is as follows:

[0038] First, the dataset The data is evenly divided into six subsets to ensure that each subset can represent a different part of the time series and can capture data fluctuations in different periods.

[0039] The six subsets are divided into two parts: the left half subset and the right half subset, each of which contains three subsets; this segmentation strategy is designed based on the order of the time series to ensure that the continuity and changes in the time dimension of the data can be captured in detail.

[0040] Left half subset: This part contains three subsets of the first half of the time series, which are recorded as These subsets can reveal the behavior and characteristics of the power system operation in the early stages, which is particularly important for understanding the initiation and development patterns of the entire time series.

[0041] The right half subset: This part contains three subsets of the second half of the time series, which are recorded as These subsets help analyze and understand the later changes and trends in the time series, especially in capturing possible anomalies and failure modes.

[0042] By segmenting the data in this way, the behavior of the power system at each time point can be analyzed more accurately, and fluctuations from long to short periods can be carefully observed and analyzed. This not only helps to more fully understand the patterns and trends in the data, but also provides more effective support in fault prevention and diagnosis.

[0043] 2.2 Feature Extraction

[0044] In order to extract valuable features from these segmented data subsets, this embodiment adopts a combination method of graph convolutional network (GCN) and multi-layer perceptron (MLP), which can not only capture the structural information in the power distribution network, but also learn the nonlinear dependencies of the data through MLP, thereby providing more comprehensive data analysis.

[0045] The feature extraction process is as follows: Each data subset is first processed by GCN, which can effectively exploit the topological structure of the power system and extract spatial features. Subsequently, the output of GCN is fed into MLP, which further processes these features to extract and learn the temporal dependencies and nonlinear relationships in the data.

[0046] For each left half subset, feature extraction is:

[0047]

[0048] For each right half subset, feature extraction is:

[0049]

[0050] Through this advanced data processing and feature extraction method, the present invention can extract the most valuable information from incomplete or discontinuous data, greatly improving the efficiency and accuracy of power system fault detection. The innovation of this process lies in that it combines the spatial structure and temporal dynamics of the power system, providing strong technical support for fault detection in complex and changeable power environments.

[0051] 3. Feature splicing and fusion

[0052] After feature extraction, the present invention concatenates the corresponding left and right data stream features to form a new set of feature vectors. In order to unify the dimensions of these feature vectors, the zero padding technology is used to ensure that each feature vector has the same size. Subsequently, these feature vectors are fused by bit-by-bit addition. This method not only improves the ability of data completion, but also provides a richer feature representation for anomaly detection.

[0053] 3.1 Feature stitching

[0054] After the feature extraction stage of the present invention, the present invention adopts an innovative feature splicing technology to make full use of the information extracted from different time windows to enhance the model's understanding and prediction capabilities of power system behavior. This feature splicing operation not only strengthens the dimension of the data, but also enables the model to more comprehensively capture the front-end dependencies in the time series data.

[0055] The detailed feature stitching process is as follows:

[0056] After the feature extraction is completed, the present invention concatenates the features extracted from the left half subset and the right half subset of the time series. This method is based on the recognition that the change of the power system state is not only related to the system conditions at the current moment, but also closely related to the previous state.

[0057] By concatenating the features of the data stream extracted from the left subset with the features of the data stream extracted from the corresponding right subset, a new and richer set of feature vectors can be formed. This concatenation can be regarded as a horizontal expansion of features, thus providing a broader data perspective and helping the model capture more complex patterns and dependencies.

[0058] For the first set of feature vectors, concatenate the first set of features extracted from the left half The first set of features extracted from the right half , forming a new feature vector Similarly, the second set of features and After splicing , the third set of features and After splicing .

[0059] The specific formula is as follows:

[0060] For the first set of features:

[0061]

[0062] For the second set of features:

[0063]

[0064] For the third set of features:

[0065]

[0066] in, Represents the concatenation operation of feature vectors.

[0067] Through this refined feature stitching method, we can take into account more comprehensive factors when analyzing and diagnosing potential problems in the power system, thereby improving the accuracy of predictions and the efficiency of operations. This strategy significantly improves the depth and breadth of data analysis, providing strong technical support for the stable operation and fault prevention of the power system.

[0068] 3.2 Dimensional unification and integration

[0069] In order to ensure that the concatenated feature vectors can be effectively combined, the dimensions of these feature vectors need to be unified. After the dimensions are unified, the feature vector fusion operation can be performed to further improve the information richness and accuracy of fault detection.

[0070] (1) Dimension unification: Zero padding is used to process each concatenated feature vector to ensure that all feature vectors have the same length. This step is necessary because different feature extractions may result in outputs of different lengths. Zero padding can supplement the missing dimensions and align the feature vectors.

[0071] Unify the dimensions of each feature vector:

[0072]

[0073] (2) Feature vector fusion: The feature vectors with unified dimensions are fused by bit-by-bit addition. This fusion method utilizes the features of all time windows and provides a comprehensive feature representation, which helps capture more complex data patterns and anomalies.

[0074] All unified feature vectors are added bit by bit and fused:

[0075]

[0076] in, Represents the bit-wise addition operation of eigenvectors.

[0077] Through these detailed steps, the present invention not only improves the data completion capability, but also provides a richer feature representation for anomaly detection, thereby significantly improving the efficiency and accuracy of power system fault detection. This feature splicing and fusion process provides a powerful analysis tool for complex and dynamic power systems.

[0078] 4. Multimodal environmental information fusion

[0079] Taking into account the environmental factors affecting the operation of the power system, the present invention introduces multimodal environmental information fusion, which integrates environmental data from different modes, such as temperature, humidity, and alarm information. After embedding processing, these data are input into a joint learning network together with the characteristics of the power system operation data, further improving the accuracy and response speed of fault detection. This in-depth environmental analysis is crucial to understanding and predicting the behavior of the power system under complex and changing operating conditions.

[0080] 4.1. Environmental Information Embedding

[0081] In this invention, special attention is paid to the processing and application of environmental information, because this information is crucial to understand and predict the behavior of the power system. In order to effectively use this data, an advanced embedding processing step is adopted to convert all relevant environmental information into a numerical feature form suitable for processing by machine learning models. This conversion process is achieved through a technology called "embedding", which can convert non-numerical data into useful numerical features, laying the foundation for subsequent data fusion and analysis.

[0082] The specific process of embedding environmental information is as follows:

[0083] First, all relevant environmental information is collected, such as temperature (T), humidity (H), and alarm information (A). These information are indispensable external factors for power system operation status analysis and may have a significant impact on system performance.

[0084] These environmental variables are then processed through a specially designed embedding layer. The embedding layer is a structure used in many neural network models that can learn how to transform these diverse environmental variables into a rich and expressive numerical feature representation. .

[0085] The embedding operation can be expressed as:

[0086]

[0087] Here, "Embedding" refers to a function or a neural network layer that automatically learns how to map the input environmental data into a high-dimensional space based on the characteristics of the data, thereby extracting the features that are most helpful for model prediction.

[0088] This method not only effectively integrates environmental information into the data analysis and fault prediction model of the power system, but also ensures that this information plays the greatest role in the entire machine learning process. This environmental information embedding strategy greatly enhances the expressiveness of data and improves the adaptability and prediction accuracy of the model in the face of complex and changing environments.

[0089] 4.2 Deep Fusion of Feature and Environmental Information

[0090] In this invention, an advanced data processing strategy is adopted to comprehensively analyze and fuse the processed features and environmental information through deep learning technology. This step is carried out after ensuring that each data stream has been finely processed, with the aim of improving the accuracy of power system fault prediction and the overall response speed of the system.

[0091] The detailed fusion process is as follows:

[0092] (1) Information processing and integration: First, the key characteristics of the power system and environmental variables are embedded and characterized through the previous steps. This includes converting the dynamic characteristics of the power system and environmental factors such as temperature, humidity, and alarm information into useful numerical features through a specially designed embedding layer.

[0093] (2) Feature merging: Subsequently, these processed power system features (denoted as ) and environmental characteristics (denoted as ) to merge. This merge uses a specific operation (such as bitwise addition, etc.), marked as to ensure the effective combination of information from both parties.

[0094] (3) Application of deep learning network: The merged feature vector is then fed into a deep learning network consisting of a multi-layer perceptron (MLP) and a fully connected layer (FC) for further processing. This network structure is specially designed to process complex data relationships and can learn and extract the most critical features for fault detection and system response.

[0095] The fusion operation expression can be described as:

[0096]

[0097] Through this comprehensive approach, the present invention can not only fully utilize the information provided by multiple data sources, but also significantly improve the diagnostic accuracy and prediction ability of possible power system faults through the powerful processing power of deep learning technology. This in-depth environmental analysis and data fusion strategy is extremely critical for understanding and predicting the behavior of power systems in a changing operating environment, especially when environmental factors have a significant impact on system operation. Such technological advances provide important technical support for the stable and safe operation of power systems.

[0098] 5. Loss calculation

[0099] In the model training process of the present invention, the mean square error (MSE) loss function is used to evaluate and optimize the performance of the power system state prediction model. This loss function measures the accuracy of the prediction by calculating the square error between the model prediction output and the actual system state, thereby providing a basis for the model parameter tuning.

[0100] The detailed model optimization and fault detection process includes the following key steps:

[0101] Application of loss function:

[0102] The loss function is mainly used to evaluate the deviation between the power system state predicted by the model and the actual observed state. Specifically, the mean square error loss function calculates the predicted value With the true value The square of the difference between , thus quantifying the prediction error of the model. This calculation can be expressed as:

[0103]

[0104] This loss function not only helps identify errors in model predictions, but also helps us understand to what extent the model deviates from the true system state, which is a key indicator for optimizing model performance.

[0105] 6. Optimization of model parameters:

[0106] In order to reduce the prediction error and improve the accuracy of the model, the model parameters are optimized using back propagation and gradient descent algorithms. The gradient of the loss function with respect to these parameters will be updated to reduce the overall loss value.

[0107] The parameter update formula is as follows:

[0108]

[0109] in, Represents the learning rate, which is a key hyperparameter that controls the update amplitude of the model at each step in the learning process. Proper selection of the learning rate can ensure that the model converges to the optimal solution quickly and stably. represents the mean square error loss, represents the original parameters of the model, Represents new parameters of the model.

[0110] Through such loss calculation and parameter optimization process, our model can effectively identify and predict potential faults in the power system under various operating conditions, even in the absence of data. This approach not only improves the accuracy of fault detection, but also optimizes the operational stability and safety of the power system.

[0111] In the power system, data incompleteness and discontinuity often occur due to unstable communication, limited acquisition frequency and complex working conditions. This poses a challenge to the accuracy and timeliness of fault detection. The present invention provides an efficient framework that integrates data completion and anomaly detection, and uses advanced machine learning and pattern recognition techniques to synchronously process these problems in a real-time environment, ensuring data integrity while improving fault prevention and rapid response capabilities.

[0112] Traditional fault detection methods are often difficult to accurately predict faults in complex environments due to the limitations of a single data source or simple algorithms. This invention introduces an analysis module for multimodal environmental information fusion and uses a block-based ladder design to process and integrate multimodal data from different sources, including meteorological, geographical, and equipment operating status data.

[0113] Through advanced data fusion analysis modules, using deep learning networks such as graph convolutional networks (GCN) and multi-layer perceptrons (MLP), the model can not only process and analyze different types of environmental information, but also more accurately predict and identify potential power system faults. Combining exogenous embeddings, such as time and location information, can enhance the model's predictive ability and provide a more comprehensive perspective for power system fault diagnosis.

[0114] The power system needs to adapt to changing operating conditions and emerging fault types. The present invention introduces an adaptive learning mechanism and feedback system to enable the fault detection model to be continuously optimized based on historical and real-time data. This mechanism not only improves the accuracy of detection, but also enhances the system's ability to adapt to new challenges, ensuring the optimal operation of the system in a changing power environment.

[0115] The power system is analyzed by using horizontal fusion (long-term features) and vertical fusion (short-term features). Horizontal fusion processes long-term operating data to identify persistent changes and trends in the system. Vertical fusion focuses on short-term emergencies and fluctuations. The combination of these two fusions allows the model to automatically extract and utilize remote and local features from time series, enhancing the ability to monitor system stability.

[0116] The ability to automatically exploit long-range and local features further enables the model to identify and fill in missing parts in the data. Through advanced algorithms, the model can infer missing information from existing data, which is especially important for real-world applications where data may be incomplete or contain anomalies.

[0117] Embodiment 2

[0118] An electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the distribution equipment fault detection method in embodiment 1, and the processor is configured to execute the program stored in the memory.

[0119] Embodiment 3

[0120] A storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the power distribution equipment fault detection method in embodiment 1 are executed.

[0121] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting a fault in a power distribution device, characterized in that: The following steps are involved: Step 1: Collect the power distribution network operation and maintenance data with missing values ​​or discontinuities, and represent it as a state data set at a series of time points, denoted as ,in Representative Status data of power distribution equipment at each moment; Step 2: Split the state data set into segments based on the order of the time series. Evenly divide the data into multiple subsets, with the first half of the subsets recorded as the left half subset and the second half as the right half subset; the left half subset can reveal the behavior and characteristics of the power system in the early stage of operation, and the right half subset is helpful for analyzing and understanding the changes and trends in the later stage of the time series; Step 3: Each data subset is first passed through a graph convolutional network to extract spatial features using the topological structure of the power system; Subsequently, the output of the graph convolutional network is fed into a multi-layer perceptron to extract and learn the temporal dependencies and nonlinear relationships in the data; Step 4: splice the features of the data stream extracted from the left half subset with the features of the data stream extracted from the corresponding right half subset to ensure time continuity; use zero padding technology to process the spliced ​​feature vectors to unify the dimension of the data, and use the bit-by-bit addition method to fuse the feature vectors after unification of the dimension; Step 5: Collect relevant environmental information, convert the environmental information into numerical features, and design an embedding layer to embed the environmental data into the network; Step 6: Merge the feature vector with the environmental information, and send the merged feature vector to a deep learning network consisting of a multi-layer perceptron and a fully connected layer for training; Step 7: Use the mean square error loss function to evaluate the deviation between the power system state predicted by the model and the actual observed state; Step 8: Use back propagation and gradient descent algorithms to update and optimize the parameters of the model and use them to predict the operating status of the power system.

2. The method for detecting a fault in a power distribution device according to claim 1, characterized in that: The power distribution network operation and maintenance data includes: voltage, current and power.

3. The method for detecting a fault in a power distribution device according to claim 1, characterized in that: The relevant environmental information includes: temperature, humidity and alarm information.

4. The method for detecting a fault in a power distribution device according to claim 1, characterized in that: The formula of the mean square error loss function is as follows: in, represents the predicted value, Represents the true value.

5. The method for detecting a fault in a power distribution device according to claim 4, characterized in that: The formula for updating and optimizing the parameters is as follows: in, represents the learning rate, represents the mean square error loss, represents the original parameters of the model, Represents new parameters of the model.

6. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the power distribution equipment fault detection method according to any one of claims 1 to 5, and the processor is configured to execute the program stored in the memory.

7. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the power distribution equipment fault detection method according to any one of claims 1 to 5 are executed.

Images