Fault detection method, device and equipment for low-voltage distribution area, and storage medium

Abstract

The invention discloses a low-voltage distribution area fault detection method, device and equipment and a storage medium, which can more accurately identify and position faults and reduce the situations of false report and missing report. The method comprises the following steps: constructing a to-be-predicted feature set; inputting the to-be-predicted feature set into the feeder line detection model to enable the feeder line detection model to output a prediction result; when the prediction result indicates that the fault feeder exists, inputting the to-be-predicted feature set to a branch detection model, so that the branch detection model outputs a fault branch set; inputting the fault branch set into a category detection model, so that the category detection model determines a corresponding fault category for each fault branch and outputs the fault category; inputting each fault branch included in the fault branch set into a distance detection model, so that the distance detection model calculates a fault relative distance corresponding to each fault branch according to the fault point position of each fault branch and outputs the fault relative distance; and generating fault early warning information according to the fault scene and the fault relative distance corresponding to each fault branch.

Description

Technical Field

[0001] The present application relates to the technical field of fault detection of power systems, and in particular to a method, device, equipment and storage medium for detecting faults in a low-voltage distribution station area. Background Art

[0002] As an important part of the power system, the operating status of the distribution network directly affects the reliability and safety of power supply. The low-voltage distribution substation is at the end of the power grid, with a wide distribution range and a complex power supply and use environment. In addition, the equipment and infrastructure of many low-voltage distribution substations may have aged, resulting in the performance of the equipment and increasing the risk of failure. These failures may not only cause power outages, but also damage the equipment and even cause safety accidents. With the continuous growth of society's demand for electricity, higher requirements are placed on the operating stability and power supply reliability of the distribution network.

[0003] However, current fault detection and early warning methods mainly rely on manual inspection and experience judgment. Manual inspection relies heavily on the personal experience and judgment of the inspectors. Different people may have different understandings and judgments, resulting in inaccurate detection results. At the same time, manual inspections in large-scale equipment or complex systems require a lot of time and manpower. An inspection may take hours or even days, resulting in low efficiency of fault detection and early warning. Summary of the invention

[0004] In view of this, the present application provides a low-voltage distribution station area fault detection method, device, equipment and storage medium, the main purpose of which is to solve the current fault detection and early warning methods that mainly rely on manual inspections and experience judgments. Manual inspections rely heavily on the personal experience and judgment ability of inspectors. Different personnel may have different understandings and judgments, resulting in inaccurate detection results. At the same time, manual inspections in large-scale equipment or complex systems require a lot of time and manpower, and an inspection may take hours or even days, resulting in low efficiency of fault detection and early warning.

[0005] According to the first aspect of the present application, a low-voltage distribution station area fault detection method is provided, the method comprising: Determine a low-voltage distribution substation area to be fault-detected, and construct a feature set to be predicted using the operation status data to be predicted of the low-voltage distribution substation area; Inputting the feature set to be predicted into a feeder detection model, so that the feeder detection model uses the feature set to be predicted to determine whether there is a faulty feeder in the low-voltage distribution station area and output a prediction result; When the prediction result indicates that there is a faulty feeder, the feature set to be predicted is input into a branch detection model, so that the branch detection model uses the feature set to be predicted to determine a faulty branch in a branch corresponding to the faulty feeder and output a faulty branch set, wherein the faulty branch set includes at least one faulty branch detected by the branch detection model; Inputting each fault branch included in the fault branch set into a category detection model, so that the category detection model determines a corresponding fault category for each fault branch and outputs it; Input each fault branch included in the fault branch set into a distance detection model, so that the distance detection model calculates the relative fault distance corresponding to each fault branch according to the fault point position of each fault branch and outputs it; According to the fault scenario and relative fault distance corresponding to each of the fault branches, fault warning information is generated and output.

[0006] Optionally, before determining the low-voltage distribution station area to be fault-detected and constructing the feature set to be predicted using the operation status data to be predicted of the low-voltage distribution station area, the method further comprises: Constructing a model architecture for each detection model based on a deep neural network, and initializing the model architecture parameters of the model architecture corresponding to each detection model, wherein the detection model is any one of the feeder detection model, the branch detection model, the category detection model, and the distance detection model, and the model architecture parameters include but are not limited to a batch normalization parameter, a random dropout parameter, a learning rate parameter, an activation function, the number of neural network layers, the number of neurons in each layer, and a normalization parameter; Acquire historical operation status data of the low-voltage distribution station area, and perform feature selection on the historical operation status data using a tree-structured Parzen estimator according to a preset feature selection standard to obtain a sample input feature set, wherein the historical operation status data includes first historical operation status data before a fault and second historical operation status data after a fault, and the data items of the historical operation status data include voltage data, current data, power generation data, and load data; Dividing the sample input feature set according to a preset division ratio to obtain the training data set, the verification data set and the test data set; Using the training data set to perform model training on each of the detection models until the model weight parameters of each of the detection models tend to be stable, thereby obtaining a plurality of detection models to be verified, wherein the model weight parameters are weight parameters of mapping connections between different neurons; The validation data set is used to optimize the multiple detection models to be validated until each of the detection models to be validated meets a preset stop criterion, and the model validation of the multiple detection models to be validated is completed to obtain multiple detection models to be tested; Using the test data set to test the plurality of detection models to be tested, and calculating the prediction accuracy of each of the detection models to be tested; When the prediction accuracy of each of the detection models to be tested meets the preset accuracy threshold, the testing of the multiple detection models to be tested is terminated to obtain the feeder detection model, the branch detection model, the category detection model and the distance detection model.

[0007] Optionally, the training data set is used to perform model training on each of the detection models until the model weight parameters of each of the detection models tend to be stable, thereby obtaining a plurality of detection models to be verified, including: For any detection model, the detection model converts and transmits the data input to the detection model through the mapping equation of the neuron to obtain the predicted data output by the detection model. The mapping equation of the neuron is specifically shown in the following formula: ; in, For the The values ​​of the layer neurons; represents the activation function of the neuron; is the value of the neuron in the previous layer; For the All neurons in the layer Layer of neurons The weight matrix between ; is a neuron bias parameter, wherein the activation function is a sigmoid function, a hyperbolic tangent function or a rectified linear unit; Determine the real data associated with the detection model, calculate the difference value between the corresponding real data and the predicted data based on the cost function associated with each detection model, and optimize the model architecture parameters of the corresponding detection model according to the difference value until the model weight parameters of each detection model tend to be stable.

[0008] Optionally, the data input to the feeder detection model and the branch detection model are training data sets, and the data input to the classification detection model and the distance detection model are fault branch sets output by the branch detection model; The cost function of the feeder detection model, the branch detection model, and the classification detection model is to minimize the binary cross entropy loss function; The cost function of the distance detection model is a standard mean square error function.

[0009] Optionally, the determining of the low-voltage distribution station area to be fault-detected and constructing a feature set to be predicted using the operation status data to be predicted of the low-voltage distribution station area include: A tree-structured Parzen estimator is used to perform feature selection on the operation state data to be predicted, and current data, voltage data, power generation data and load data are determined; The voltage data and the current data are interpolated to obtain an interpolation voltage matrix and an interpolation current matrix, and a sample input feature set is constructed according to the interpolation voltage matrix, the interpolation current matrix, the power generation data and the load data.

[0010] Optionally, the interpolating the voltage data and the current data to obtain an interpolated voltage matrix and an interpolated current matrix includes: Determine the branch voltage data of each branch and the position of the voltage sensor that collects the branch voltage data in the voltage data, generate an equally spaced position matrix corresponding to each branch according to the position of the voltage sensor and a preset interpolation number, calculate the interpolation voltage value of each branch according to the branch voltage data of each branch and the position of the corresponding voltage sensor, and insert the interpolation voltage value of each branch into the branch voltage data of each branch according to the equally spaced position matrix to obtain the interpolation voltage matrix of the low-voltage distribution station area; The branch current data of each branch and the position of the current sensor that collects the branch current data are determined in the current data, and an equally spaced position matrix corresponding to each branch is generated according to the position of the current sensor and a preset interpolation number, and the interpolated current value of each branch is calculated according to the branch current data of each branch and the position of the corresponding current sensor, and the interpolated current value of each branch is inserted into the branch current data of each branch according to the equally spaced position matrix to obtain the interpolated current matrix of the low-voltage distribution station area.

[0011] Optionally, generating fault warning information according to the fault scenario and the relative fault distance corresponding to each fault branch, and outputting the fault warning information includes: For each faulty branch in the faulty branch set, query the actual branch total distance associated with the faulty branch, and calculate the product of the actual branch total distance and the corresponding fault relative distance to obtain the actual fault distance corresponding to the faulty branch; Fault warning information is generated according to the actual fault distance and the fault scenario associated with the fault branch, and the fault warning information is sent to a designated terminal device for visual display.

[0012] According to a second aspect of the present application, a low-voltage distribution station area fault detection device is provided, the device comprising: A determination module, used to determine a low-voltage distribution substation area to be detected for faults, and to construct a feature set to be predicted using the operation status data to be predicted of the low-voltage distribution substation area; A first detection module, used for inputting the feature set to be predicted into a feeder detection model, so that the feeder detection model uses the feature set to be predicted to determine whether there is a faulty feeder in the low-voltage distribution station area and output a prediction result; a second detection module, configured to input the feature set to be predicted into a branch detection model when the prediction result indicates that there is a faulty feeder, so that the branch detection model uses the feature set to be predicted to determine a faulty branch in a branch corresponding to the faulty feeder and output a faulty branch set, wherein the faulty branch set includes at least one faulty branch detected by the branch detection model; A third detection module, configured to input each faulty branch included in the faulty branch set into a category detection model, so that the category detection model determines a corresponding fault category for each faulty branch and outputs the result; A fourth detection module, used for inputting each fault branch included in the fault branch set into a distance detection model, so that the distance detection model calculates and outputs the relative fault distance corresponding to each fault branch according to the fault point position of each fault branch; A generation module is used to generate fault warning information according to the fault scenario and relative fault distance corresponding to each of the fault branches, and output the fault warning information.

[0013] Optionally, the device further comprises: An initialization module, used to construct a model architecture for each detection model based on a deep neural network, and initialize the model architecture parameters of the model architecture corresponding to each detection model, wherein the detection model is any one of the feeder detection model, the branch detection model, the category detection model and the distance detection model, and the model architecture parameters include but are not limited to batch normalization parameters, random dropout parameters, learning rate parameters, activation functions, number of neural network layers, number of neurons in each layer and normalization parameters; An acquisition module is used to acquire the historical operation status data of the low-voltage distribution station area, and perform feature selection on the historical operation status data using a tree-structured Parzen estimator according to a preset feature selection standard to obtain a sample input feature set, wherein the historical operation status data includes first historical operation status data before a fault and second historical operation status data after a fault, and the data items of the historical operation status data include voltage data, current data, power generation data, and load data; A splitting module is used to divide the sample input feature set according to a preset division ratio to obtain the training data set, the verification data set and the test data set; A training module, used to perform model training on each of the detection models using the training data set until the model weight parameters of each of the detection models tend to be stable, thereby obtaining a plurality of detection models to be verified, wherein the model weight parameters are weight parameters for mapping connections between different neurons; A verification module, used to optimize the multiple detection models to be verified by using the verification data set until each of the detection models to be verified meets a preset stop criterion, complete the model verification of the multiple detection models to be verified, and obtain multiple detection models to be tested; A testing module, used to test the plurality of detection models to be tested using the test data set, and calculate the prediction accuracy of each of the detection models to be tested; A judgment module is used to end the testing of multiple detection models to be tested when the prediction accuracy of each detection model to be tested meets a preset accuracy threshold, and obtain the feeder detection model, the branch detection model, the category detection model and the distance detection model.

[0014] Optionally, the training module is used for any detection model, wherein the detection model converts and transmits the data input to the detection model through a mapping equation of a neuron to obtain the predicted data output by the detection model, and the mapping equation of the neuron is specifically shown in the following formula: ; in, For the The values ​​of the layer neurons; represents the activation function of the neuron; is the value of the neuron in the previous layer; For the All neurons in the layer Layer of neurons The weight matrix between ; is a neuron bias parameter, wherein the activation function is a sigmoid function, a hyperbolic tangent function or a rectified linear unit; determining the real data associated with the detection model, calculating the difference between the corresponding real data and the predicted data based on the cost function associated with each detection model, and optimizing the model architecture parameters of the corresponding detection model according to the difference value until the model weight parameters of each detection model tend to be stable.

[0015] Optionally, in the training module, the data input to the feeder detection model and the branch detection model are training data sets, and the data input to the classification detection model and the distance detection model are the set of faulty branches output by the branch detection model; the cost function of the feeder detection model, the branch detection model, and the classification detection model is the minimization of the binary cross entropy loss function; and the cost function of the distance detection model is the standard mean square error function.

[0016] Optionally, the determination module is used to use a tree-structured Parzen estimator to perform feature selection on the operating status data to be predicted, determine current data, voltage data, power generation data and load data; interpolate the voltage data and the current data to obtain an interpolated voltage matrix and an interpolated current matrix, and construct a sample input feature set based on the interpolated voltage matrix, the interpolated current matrix, the power generation data and the load data.

[0017] Optionally, the determination module is used to determine the branch voltage data of each branch and the position of the voltage sensor that collects the branch voltage data in the voltage data, generate an equally spaced position matrix corresponding to each branch according to the position of the voltage sensor and a preset interpolation number, calculate the interpolation voltage value of each branch according to the branch voltage data of each branch and the position of the corresponding voltage sensor, and insert the interpolation voltage value of each branch into the branch voltage data of each branch according to the equally spaced position matrix to obtain the interpolation voltage matrix of the low-voltage distribution station area; determine the branch current data of each branch and the position of the current sensor that collects the branch current data in the current data, generate an equally spaced position matrix corresponding to each branch according to the position of the current sensor and a preset interpolation number, calculate the interpolation current value of each branch according to the branch current data of each branch and the position of the corresponding current sensor, and insert the interpolation current value of each branch into the branch current data of each branch according to the equally spaced position matrix to obtain the interpolation current matrix of the low-voltage distribution station area.

[0018] Optionally, the generation module is used to query the actual branch total distance associated with each fault branch in the fault branch set, and calculate the product of the actual branch total distance and the corresponding fault relative distance to obtain the actual fault distance corresponding to the fault branch; generate fault warning information according to the actual fault distance and the fault scenario associated with the fault branch, and send the fault warning information to a designated terminal device for visual display.

[0019] According to a third aspect of the present application, a device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the methods described in the first aspect when executing the computer program.

[0020] According to a fourth aspect of the present application, a storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the methods described in the first aspect are implemented.

[0021] By means of the above technical scheme, the present application provides a method, device, equipment and storage medium for fault detection in a low-voltage distribution station area. The present application first determines the low-voltage distribution station area to be fault detected, and uses the operating status data to be predicted of the low-voltage distribution station area to construct a feature set to be predicted. Further, the feature set to be predicted is input into the feeder detection model, so that the feeder detection model uses the feature set to be predicted to determine whether there is a faulty feeder in the low-voltage distribution station area and outputs the prediction result. When the prediction result indicates the existence of a faulty feeder, the feature set to be predicted is input into the branch detection model, so that the branch detection model uses the feature set to be predicted to determine the faulty branch in the branch corresponding to the faulty feeder and output the faulty branch set, and the faulty branch set includes at least one faulty branch detected by the branch detection model. Next, each faulty branch included in the faulty branch set is input into the category detection model, so that the category detection model determines the corresponding fault category for each faulty branch and outputs it. Subsequently, each fault branch included in the fault branch set is input into the distance detection model, so that the distance detection model calculates and outputs the relative fault distance corresponding to each fault branch according to the fault point position of each fault branch. Finally, according to the fault scenario and relative fault distance corresponding to each fault branch, fault warning information is generated, and the fault warning information is output. In an embodiment of the present application, through a multi-level detection model, from feeder to branch to fault category judgment, faults can be more accurately identified and located, and false alarms and missed alarms can be reduced. At the same time, through the detailed information of the predicted fault category and relative fault distance, maintenance personnel and equipment can be arranged more reasonably, maintenance efficiency and power supply stability can be improved, and user satisfaction with power services can be improved.

[0022] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings: Figure 1 A schematic diagram of a low-voltage distribution station area fault detection method provided in an embodiment of the present application is shown; Figure 2 A schematic diagram of an interpolation method for obtaining equally spaced measurements in a low-voltage distribution station area fault detection method provided in an embodiment of the present application is shown; Figure 3 A schematic diagram of a fault detection process of a low-voltage distribution station area fault detection method provided in an embodiment of the present application is shown; Figure 4 A schematic diagram of a general deep neural network with two hidden layers for a low-voltage distribution station area fault detection method provided in an embodiment of the present application is shown; Figure 5 A schematic diagram of the structure of a low-voltage distribution station area fault detection device provided in an embodiment of the present application is shown; Figure 6 A schematic diagram of the structure of a low-voltage distribution station area fault detection device provided in an embodiment of the present application is shown; Figure 7 A schematic diagram of the device structure of a computer device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0024] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0025] As an important part of the power system, the operating status of the distribution network directly affects the reliability and safety of power supply. The low-voltage distribution substation is at the end of the power grid, with a wide distribution range and a complex power supply and use environment. In addition, the equipment and infrastructure of many low-voltage distribution substations may have aged, resulting in the degradation of equipment performance and increasing the risk of failure. These failures may not only cause power outages, but also damage equipment and even cause safety accidents. With the continuous growth of society's demand for electricity, higher requirements are placed on the operating stability and power supply reliability of the distribution network. However, the current fault detection and early warning methods mainly rely on manual inspections and experience judgments. Manual inspections rely heavily on the personal experience and judgment ability of inspectors. Different personnel may have different understandings and judgments, resulting in inaccurate detection results. At the same time, manual inspections in large-scale equipment or complex systems require a lot of time and manpower, and an inspection may take hours or even days, resulting in low efficiency in fault detection and early warning.

[0026] Therefore, the present application proposes a method for fault detection in a low-voltage distribution station area. First, the low-voltage distribution station area to be detected for fault is determined, and the predicted operating state data of the low-voltage distribution station area is used to construct a feature set to be predicted. Further, the feature set to be predicted is input into the feeder detection model, so that the feeder detection model uses the feature set to be predicted to determine whether there is a faulty feeder in the low-voltage distribution station area and outputs the prediction result. When the prediction result indicates that there is a faulty feeder, the feature set to be predicted is input into the branch detection model, so that the branch detection model uses the feature set to be predicted to determine the faulty branch in the branch corresponding to the faulty feeder and output the faulty branch set, and the faulty branch set includes at least one faulty branch detected by the branch detection model. Next, each faulty branch included in the faulty branch set is input into the category detection model, so that the category detection model determines the corresponding fault category for each faulty branch and outputs it. Subsequently, each faulty branch included in the faulty branch set is input into the distance detection model, so that the distance detection model calculates the relative distance of the fault corresponding to each faulty branch according to the fault point position of each faulty branch and outputs it. Finally, according to the fault scenario and relative distance of each fault branch, fault warning information is generated and output.

[0027] In the embodiment of the present application, through the multi-level detection model, from feeder to branch to fault category judgment, the fault can be identified and located more accurately, reducing false alarms and missed alarms. At the same time, through the detailed information of the predicted fault category and relative distance of the fault, maintenance personnel and equipment can be arranged more reasonably, improving maintenance efficiency and power supply stability, thereby improving user satisfaction with power services.

[0028] The present application embodiment provides a method for detecting faults in a low-voltage distribution station area, such as Figure 1As shown, the method includes: 101. Determine the low-voltage distribution substation area to be fault-detected, and construct a feature set to be predicted using the operation status data to be predicted of the low-voltage distribution substation area.

[0029] With the continuous progress and development of sensor technology, data processing technology, artificial intelligence technology and other fields, data-driven decision-making has become an important means. By using various dimensions of operating data for big data mining and analysis, potential problems and optimization opportunities can be effectively discovered. This data-driven approach can achieve real-time monitoring, data analysis, intelligent early warning and rapid positioning of the operating status of the distribution network, so as to better understand the operating status of the substation and provide strong support for decision-making. Specifically, sensor technology can collect various data in the distribution network in real time, such as key parameters such as voltage, current, and temperature. Data processing technology can clean, integrate and analyze these massive data to extract valuable information. Artificial intelligence technology, especially machine learning and deep learning algorithms, can perform pattern recognition and predictive analysis on this basis, thereby realizing intelligent early warning and fault diagnosis. The application of these technologies has greatly improved the accuracy of fault detection and the timeliness of early warning. For example, through real-time monitoring and analysis, the system can quickly identify abnormal data patterns, issue early warning signals in time, and remind operation and maintenance personnel to check and handle them. This can not only reduce power outage time and improve power supply reliability, but also effectively prevent the occurrence of major faults, providing a strong guarantee for the safe and stable operation of the distribution network.

[0030] The core goal of this application is to achieve fault detection in low-voltage distribution areas through deep learning (DL) technology and deep neural networks (DNN). This process not only includes early detection of faults, but also extends to accurate analysis of fault types, effective identification of faulty feeders, and precise positioning of the specific location of the fault.

[0031] In the specific implementation process of this application, this application constructs four independent deep neural network detection models for these four different functional requirements, and systematically trains these models. In order to ensure the applicability and accuracy of these models in practical applications, this application adopts an interpolation method in the data preprocessing stage to perform detailed interpolation preprocessing on the voltage data and current data in the input feature set. The purpose of this processing step is to enhance the continuity and integrity of the data, so that the model can better capture the subtle changes and features in the data. In addition, considering the noise problem of the data, a filtering algorithm can be introduced to further optimize the data quality, ensure the efficiency of model training and the reliability of the detection results. After the model training is completed, this application can also perform cross-validation and testing to evaluate the performance of each model to ensure that the four independent deep neural network detection models can provide accurate and stable detection results when facing various practical application scenarios.

[0032] During the training process, this application uses two different training methods, classification and regression, according to different task requirements. The classification method is mainly used for fault type analysis and identification of faulty feeders, while the regression method is mainly used for accurate positioning of fault distances. Through this targeted training strategy, this application can ensure that each model can perform at its best performance on its specific task, thereby providing a comprehensive and efficient solution for fault detection in low-voltage distribution substations.

[0033] In actual operation, it is necessary to first identify and determine the low-voltage distribution area where fault detection is required. The so-called low-voltage distribution area usually refers to the power supply range or area from the low-voltage side of a transformer to the electric meter side of the user it ultimately serves. The main function of this area is to distribute electrical energy from the low-voltage side of the transformer to each user, thereby ensuring a stable supply of electricity.

[0034] In an embodiment of the present application, after determining the low-voltage distribution station area that needs to be fault detected, and using the operating status data to be predicted of the low-voltage distribution station area to construct the feature set to be predicted, a series of preparatory work needs to be performed. First, it is necessary to build the model architecture of the detection model, which includes determining the key elements such as the structure, number of layers, and activation function of the model. Secondly, it is necessary to initialize the model architecture parameters of the model architecture, which include weights, biases, etc., which will directly affect the performance of the model. Finally, it is necessary to construct sample input feature sets, which will be used to train, verify and test the detection model. Through these steps, a detection model with excellent performance can be trained, and then put into practical use to achieve fault detection of low-voltage distribution stations.

[0035] It should be noted that the detection model includes a feeder detection model, a branch detection model, a category detection model and a distance detection model. The specific process of constructing the model architecture of the detection model for model training is shown in the following steps 11 to 15: 11. Build a model architecture for each detection model based on the deep neural network, and initialize the model architecture parameters of the model architecture corresponding to each detection model.

[0036] In this step, deep neural network technology is used to build a corresponding model architecture for each detection model, and the model architecture parameters of these model architectures are initialized, which helps to speed up the learning speed of the model and ultimately improve the accuracy of the model and its ability to adapt to different situations. The model architecture parameters include but are not limited to batch normalization parameters, random dropout parameters, learning rate parameters, activation functions, number of neural network layers, number of neurons in each layer, and normalization parameters.

[0037] The random dropout parameter randomly drops a part of neurons during the training process to prevent overfitting and improve the generalization ability of the model. The learning rate parameter determines the step size of the parameter update during the optimization process, which has a significant impact on the convergence speed and final performance of the model. The activation function introduces nonlinear factors to the neural network, enabling the network to learn and simulate complex function mappings. The number of neural network layers and the number of neurons in each layer jointly determine the capacity and complexity of the model, affecting the learning and generalization capabilities of the model. Normalization parameters, which help to accelerate the training process of the model and improve the convergence speed and performance of the model. The specific model architecture parameters are shown in Table 1 below: Table 1 Model architecture parameters

[0038] In most cases, a neural network can be viewed as a of Model, input feature value To predict a specific variable of interest Therefore, for a given task, using a neural network, i.e. predicting , only one dataset needs to be collected , which is the sample input feature set. Then use this data set to estimate the optimal parameters that best suit the data set , so that the neural network can make accurate predictions. Define the input of the neural network as , the output is In the structure of a neural network, there are usually one or more hidden layers, each of which contains a certain number of neurons. In order to describe this structure, the concept of state vector is introduced, which represents the state of the neural network at a certain moment. For each hidden layer, the number of neurons in the kth hidden layer is used to represent the complexity of the layer. The number of neurons in the kth hidden layer is And the state vector It should be noted that in practical applications, different detection models require different types of data as input.

[0039] For example, the data input to the feeder detection model and the branch detection model are usually collected and preprocessed training data sets. These data sets contain sample features that need to be analyzed so that the model can learn and recognize potential patterns. For the classification detection model and the distance detection model, the data they need is different. These models usually also need the set of fault branches output by the branch detection model as input data for further analysis and prediction. This design ensures that each model can receive the data that best suits its task, thereby improving the overall prediction accuracy and efficiency.

[0040] 12. Obtain the historical operating status data of the selected low-voltage distribution station area, and use the tree-structured Parzen estimator to perform feature selection on the historical operating status data according to the preset feature selection criteria to obtain a sample input feature set.

[0041] When performing feature selection, it is usually necessary to comprehensively consider multiple key factors to ensure that the selected features can meet the needs of the model and improve its performance. First, since the main task of this model is to detect faults in the branches of the power system, the input feature set must contain the state data of each branch. These state data include but are not limited to parameters such as voltage, current, and temperature, so that the operating status of the branch can be fully reflected. Secondly, in order to enhance the applicability and generalization ability of the model, the selection of feature sets should try to avoid including features specific to a certain branch. To this end, a predefined interpolation method can be used to interpolate the measured values ​​of each branch, so that the irregularly distributed measured values ​​are equivalently converted into voltage data measured by equally spaced sensors. The purpose of this is to enable the model to better adapt to the measurement data of different branches and improve its robustness in practical applications. In addition, the input feature set also needs to be able to model various states of the branch in normal operation and fault conditions.

[0042] This means that in order to effectively identify faults, the feature set needs to include not only the current operating status data, but also the historical operating status data. Specifically, the historical operating status data includes the first historical operating status data before the fault occurs and the second historical operating status data after the fault occurs. These historical data items cover multiple aspects such as voltage data, current data, power generation data, and load data. In order to construct the input feature set, it is necessary to interpolate the voltage data and current data based on these historical data to obtain the interpolation voltage matrix and the interpolation current matrix. This process can be achieved through the following steps: first, extract the measured values ​​of voltage and current from the historical data; second, according to the predefined interpolation method, interpolate these measured values ​​to generate equally spaced voltage and current data; then, organize the interpolated data into a matrix form to form an interpolation voltage matrix and an interpolation current matrix; finally, use these matrices as part of the input feature set for subsequent model training and fault detection.

[0043] Through the above steps, it can be ensured that the input feature set includes both the current state of the branch and its historical state, thereby providing comprehensive data support for the model and improving the accuracy and reliability of fault detection. The process of interpolating the voltage data and the current data to obtain the interpolated voltage matrix and the interpolated current matrix is ​​shown in the following steps 121 to 124: 121. When processing voltage data and current data, interpolation is usually required to obtain more accurate results. Specifically, it is necessary to first determine the voltage data of each branch in the voltage data, and to clearly identify the specific location of the voltage sensor that collects the voltage data of these branches. The branches here refer to the branches in the circuit, and the total number of nodes refers to the number of components connected in each branch. Assume that there is branches, each branch has nodes.

[0044] Interpolate the voltage and current data, that is, determine the voltage data for each branch The branch voltage data and the location of the voltage sensor for collecting the branch voltage data , the location of the voltage sensor , ,in, is the number of branches, For branch The total number of nodes. According to the location of the voltage sensor and preset interpolation number , generate an equally spaced position matrix corresponding to each branch , .

[0045] This matrix can be used to evenly insert the voltage and current measurements on each branch. Specifically, the number of rows in this equally spaced position matrix is , represents the number of interpolated voltages and currents, and the number of columns is the actual length l of each branch. is the number of interpolated voltages and currents, For branch The actual length of

[0046] For example, the preset interpolation number Can be set to 5 (corresponding to Figure 2 The equivalent branches in the circuit are divided into four equal parts), which means that each branch is equivalently divided into four equal parts. In this way, the voltage and current data of each branch can be interpolated at these equally spaced positions to obtain a more detailed voltage and current distribution. In this way, the working state of the circuit can be analyzed and understood more accurately, thereby optimizing the circuit design or performing fault diagnosis.

[0047] 122. According to each branch Branch voltage data , and the corresponding voltage sensor locations Calculate the interpolated voltage value for each branch , ,in, is the number of interpolated voltages and currents. According to the equally spaced position matrix Insert the interpolated voltage value of each branch into the branch voltage data of each branch. In this way, a complete interpolated voltage matrix of the low-voltage distribution station area can be obtained. This matrix can reflect in detail the voltage distribution at each location of the low-voltage distribution station area.

[0048] 123. Determine each branch in the current data Branch current data and the location of the current sensor for collecting branch current data , , depending on the location of the current sensor and preset interpolation number , generate an equally spaced position matrix corresponding to each branch , .

[0049] 124. According to each branch Branch current data , and the corresponding current sensor locations Calculate the interpolated current value for each branch , , and insert the interpolation current value of each branch into the branch current data of each branch according to the equally spaced position matrix to obtain the interpolation current matrix of the low-voltage distribution station area .

[0050] 13. If Figure 3 As shown in the figure, firstly, a sample input feature set is constructed based on the interpolated voltage matrix, interpolated current matrix, power generation data and load data. This feature set contains all the key information required for training and testing the model. Next, the sample input feature set is divided according to the preset division ratio to ensure the reasonable distribution of the data set. For example, in the actual operation process, 60% of the sample data in the sample input feature set can be divided into the training data set for the initial learning and parameter adjustment of the model; 20% of the sample data can be divided into the validation data set for evaluating the performance of the model and adjusting the hyperparameters during the training process; and the other 20% of the sample data can be divided into the test data set for evaluating the final performance of the model after the model training is completed. The model architecture (initial DNN model) of the pre-built detection model is trained using the training data set and the validation data set. Through continuous iteration and optimization, the model can better capture the characteristics and laws in the data. At the same time, the model architecture parameters of the model are optimized using the efficient hyperparameter optimization algorithm TPE (Tree-structured Parzen Estimator) to find the most suitable parameter combination, so as to obtain the most optimized and usable DNN detection model. In this way, it can be ensured that the model has high accuracy and robustness in practical applications.

[0051] 14. In the detection model, the interpolation voltage matrix Sum difference current matrix The data in is used as the input of the neural network and is converted and transmitted through the neuron mapping equation. It is the weight of establishing the mapping connection between different neurons in the network. Therefore, the training data set is used to train each detection model until the model weight parameter of each detection model is All tend to be stable, and multiple detection models to be verified can be obtained. Specifically, the neuron In the The mapping equation of the layer is shown in the following formula 1: Formula 1:

[0052] in, For the The values ​​of the layer neurons; represents the activation function of the neuron; is the value of the neuron in the previous layer; For the All neurons in the layer Layer of neurons The weight matrix between ; is the neuron bias parameter, where the activation function is the sigmoid function, the hyperbolic tangent function, or the rectified linear unit. Figure 4 is a general deep neural network with two hidden layers, including an input layer (containing , … nodes), the first hidden layer (containing , … neurons), and the second hidden layer (containing , … neurons), the output layer (containing , … Node). Determine the real data associated with the detection model, calculate the difference between the corresponding real data and the predicted data based on the cost function associated with each detection model, and optimize the model architecture parameters of the corresponding detection model according to the difference value until the model weight parameters of each detection model tend to be stable. Specifically, network training can be completed by solving a general optimization problem as shown in the following formula 2: Formula 2:

[0053] in, As the cost function, the cost function (also known as the loss function) plays a vital role in the training process of deep learning and neural networks. It is used to measure the difference between the model's predicted value and the actual value, thereby evaluating the performance of the model. Specifically, the cost function is calculated by comparing the difference between the output of the neural network (based on the mapping equation of the neuron) and the true label (for example, whether a fault has occurred, faulty feeder, etc.). For the specific task of power grid fault diagnosis, the choice of cost function varies depending on the specific task. Specifically, in the feeder detection model, the branch detection model, and the classification detection model, the cost function is usually minimized as the binary cross entropy loss function. The binary cross entropy loss function is a commonly used loss function, especially suitable for binary classification problems. It evaluates the model performance by calculating the difference between the probability distribution predicted by the model and the probability distribution of the actual label. By minimizing this loss function, the model can be effectively trained to make its prediction results closer to the true label. In the distance detection model, the cost function uses the standard mean square error function. The mean square error function measures the model performance by calculating the average of the squared difference between the model prediction value and the actual value. This method is particularly suitable for regression problems and can effectively measure the difference between the predicted value and the actual value. In order to solve this cost function, the Adam (Adaptive Moment Estimation) optimizer can be used for optimization.

[0054] Adam optimizer is an optimization algorithm widely used in the field of deep learning. It combines the advantages of momentum gradient descent (Momentum) and adaptive learning rate (such as AdaGrad, RMSprop). Momentum gradient descent helps to accelerate the training process of the model and avoid falling into the local minimum, while the adaptive learning rate can dynamically adjust the learning rate according to the gradient history of the parameters, thereby improving the training efficiency and stability of the model. Adam optimizer can effectively balance global and local searches by calculating the first-order moment estimate (i.e., mean) and second-order moment estimate (i.e., uncentered variance) of the gradient, thereby performing well in various tasks and accelerating the convergence process of the model.

[0055] 15. Use the validation data set to optimize multiple detection models to be verified until each detection model to be verified meets the preset stopping criteria, complete the model verification of multiple detection models to be verified, and obtain multiple detection models to be tested. Use the test data set to test multiple detection models to be tested, and calculate the prediction accuracy of each detection model to be tested. When the prediction accuracy of each detection model to be tested meets the preset accuracy threshold, end the test of multiple detection models to be tested, and obtain the feeder detection model, branch detection model, category detection model and distance detection model. In other words, use the training data set and the validation data set to optimize the hyperparameters of the model, and use the test data set to evaluate the performance of the optimized network.

[0056] In the embodiment of the present application, by using the verification data set, the detection model to be verified can be subjected to a meticulous optimization process. This process will continue until each detection model to be verified reaches a preset stop criterion. In this way, the comprehensive model verification work of these detection models to be verified is completed, and a plurality of optimized detection models to be tested are finally obtained. Next, these detection models to be tested are rigorously tested using the test data set. During the test process, further testing of these models will only be stopped when the prediction accuracy of all detection models to be tested reaches a preset accuracy threshold. Through this series of tests and verifications, a feeder detection model, a branch detection model, a category detection model, and a distance detection model with excellent performance can finally be obtained.

[0057] After successfully building a detection model suitable for practical applications, the next step is to obtain the operating state data to be predicted in the low-voltage distribution station area in order to construct a feature set to be predicted. It should be emphasized that the process of constructing the feature set to be predicted is essentially the same as the process of constructing the sample input feature set previously. Specifically, the embodiment of the present application needs to use a tree-structured Parzen estimator to perform in-depth feature extraction on these operating state data to be predicted. Through this process, the embodiment of the present application can effectively extract the voltage data to be predicted, the current data to be predicted, the power generation data in the distribution network, and the load data in the distribution network. In order to further improve the accuracy and credibility of the data, the embodiment of the present application also needs to interpolate these voltage data to be predicted and the current data to be predicted. Through interpolation processing, a more refined and complete interpolation voltage matrix and interpolation current matrix can be obtained. These matrices can more accurately reflect the distribution of voltage and current at different time points and spatial positions, thereby providing a more solid foundation for subsequent analysis and prediction. Finally, a complete feature set to be predicted will be constructed based on the interpolation voltage matrix, the interpolation current matrix, the power generation data, and the load data. This feature set will contain all necessary information so that the embodiment of the present application can use the developed detection model to perform accurate prediction analysis. In this way, the embodiment of the present application can effectively monitor and manage the operating status of the low-voltage distribution station area to ensure the stability and reliability of the power system.

[0058] After the feature set to be predicted is constructed, these features are input into the trained and verified detection model for prediction. Each detection model will process and analyze the input feature set according to its specific network architecture and trained weight parameters. That is, the following steps 102 to 105 are executed, in which the feeder detection model will predict whether the fault occurs on a specific feeder, the branch detection model will identify the branch location where the fault occurs, the classification detection model will classify the fault type, and the distance detection model will calculate the distance between the fault point and the detection point. These models will receive operating status data in real time, quickly and accurately detect and locate faults, thereby improving the operating efficiency and reliability of the power grid and reducing the losses and impacts caused by faults.

[0059] 102. Input the feature set to be predicted into the feeder detection model, so that the feeder detection model uses the feature set to be predicted to determine whether there is a faulty feeder in the low-voltage distribution station area and output the prediction result.

[0060] In the embodiment of the present application, the feeder detection model can be modeled using a classification network. For the feeder detection model, for each input related to a branch , defined as the interpolation voltage matrix and interpolation current matrix of each branch in the feature set to be predicted. is defined as 1 or 0, indicating that the branch has a fault or that the branch has no fault and is in a healthy feeder. It should be noted that when training the feeder detection model in step 101, the neural network only needs to minimize the binary cross entropy loss (corresponding to the cost function mentioned in step 101 ), the cost function is specifically shown in the following formula 3. The network is applied to all branches of the feeder to identify whether there is a fault on the feeder.

[0061] Formula 3:

[0062] in: It is The binary label value of each sample is 0 or 1. The model is The predicted value of the sample, that is, the model predicts the The probability that the sample label value is 1, Represents the number of training samples.

[0063] 103. When the prediction result indicates the existence of a faulty feeder, the feature set to be predicted is input into the branch detection model, so that the branch detection model uses the feature set to be predicted to determine the faulty branch in the branch corresponding to the faulty feeder and output a faulty branch set, wherein the faulty branch set includes at least one faulty branch detected by the branch detection model.

[0064] In an embodiment of the present application, when the prediction result indicates that there is no faulty feeder, the subsequent detection model will be stopped and the next detection node will be waited for. However, when the prediction result indicates that there is a faulty feeder, the feature set to be predicted will be input into the branch detection model. In this way, the branch detection model can use these feature sets to be predicted to analyze the branches corresponding to the faulty feeder to determine the specific faulty branches, and finally output a set containing the faulty branches. Specifically, the task of identifying the faulty branch in the faulty feeder is essentially similar to the task of identifying whether the entire feeder has a fault. It is no longer necessary to mark the branches inside the faulty feeder and the branches outside the faulty feeder as 1 and 0 respectively, but only focus on the branches inside the faulty feeder. For these branches, the output of the faulty branch in the faulty feeder is defined as 1, and the output of the healthy branch is defined as 0. By aggregating all branches with an output result of 1, these branches can be stored as faulty branches in a special set of faulty branches.

[0065] 104. Input each faulty branch included in the faulty branch set into a category detection model, so that the category detection model determines and outputs a corresponding fault category for each faulty branch.

[0066] In an embodiment of the present application, in order to carry out effective model training for the category detection model, it is first necessary to configure the fault scenario category in detail. Specifically, it is necessary to set the load conditions of the distribution network, and record in detail the various key parameters before the fault occurs. These parameters include the power generation and load conditions of the distribution network, and the values ​​of the A-phase current, B-phase current, C-phase current and zero-sequence current on each feeder. In addition, it is also necessary to record the voltage conditions of each branch node, including the A-phase voltage, the B-phase voltage, the C-phase voltage, and the values ​​of the positive-sequence voltage, the negative-sequence voltage and the zero-sequence voltage. At the same time, for the current conditions of each branch node, it is also necessary to record the values ​​of the A-phase current, the B-phase current, the C-phase current and the zero-sequence current.

[0067] After completing the data recording before the fault, the next step is to simulate various fault scenarios. In this process, it is necessary to record the power generation and load of the distribution network after the fault occurs, as well as the values ​​of phase A current, phase B current, phase C current and zero-sequence current on each feeder. Similarly, it is also necessary to record the voltage conditions of each branch node, including phase A voltage, phase B voltage, phase C voltage, and the values ​​of positive sequence voltage, negative sequence voltage and zero sequence voltage. For the current conditions of each branch node, it is also necessary to record the values ​​of phase A current, phase B current, phase C current and zero sequence current.

[0068] After collecting the data before and after the fault, the next task is to build a branch-independent data set. In order to make the data set richer and more representative, 5 equally spaced voltage values ​​and 5 equally spaced current values ​​are interpolated in each branch. In this way, a more detailed and comprehensive sample input feature set can be obtained, thereby providing more sufficient and accurate data support for the training of the category detection model. This sample input feature set will contain a large number of distribution network parameters before and after the fault, so that the model can better learn and identify the features under different fault scenarios, thereby improving its detection accuracy and reliability.

[0069] In the actual fault detection process, each specific fault branch contained in the fault branch set is first input into the category detection model one by one. The category detection model will perform detailed analysis and judgment on each fault branch according to the feature set to be predicted, so as to determine its corresponding fault category. This category information will then be output for further processing and analysis.

[0070] Specifically, the fault categories cover a variety of common power system fault types. These types include but are not limited to open circuit faults, short circuit faults, overload faults, ground faults, and ground faults. Each fault type has its own specific manifestations and causes, so when performing fault detection, these types need to be carefully distinguished and identified.

[0071] By inputting the fault branches into the category detection model one by one and performing detailed analysis and judgment, various fault types can be identified quickly and accurately. This method not only improves the efficiency of fault detection, but also ensures the accuracy of the detection results. Once the fault type is accurately identified, corresponding treatment measures can be taken according to the specific fault type, thereby providing strong support and guidance for subsequent fault handling and maintenance work. In this way, not only can the time for fault handling be shortened, but also the stability and reliability of the power system can be improved, ensuring the continuity and safety of power supply.

[0072] 105. Input each faulty branch included in the faulty branch set into the distance detection model, so that the distance detection model calculates and outputs the relative fault distance corresponding to each faulty branch according to the fault point position of each faulty branch.

[0073] In the embodiment of the present application, the importance of transforming the distance estimation problem of the fault occurrence from a classification problem to a regression problem is fully considered. This means that the output of the distance detection model needs to be redefined so that it can directly reflect the specific distance of the fault occurrence. In addition, given that in the previous processing, the interpolation method has been used for the voltage value in the input feature set, thereby ignoring the actual length of each branch, in this case, the estimated value of the fault distance will be limited to an interval of [0,1]. In order to ensure the accuracy and reliability of the distance detection model, the standard mean square error is selected as the cost function in the model. In this way, the difference between the model prediction value and the actual value can be effectively measured and optimized. Finally, in the process of real-time detection, once a branch is identified as having a fault by the first two networks, the network will immediately display the specific value of the fault distance, thereby providing timely and accurate fault location information for maintenance personnel. This real-time fault detection and location function not only improves the response speed of the system, but also greatly enhances the work efficiency of maintenance personnel, enabling them to take measures quickly, repair faults in time, and ensure the stable operation of the system.

[0074] 106. Generate fault warning information according to the fault scenario and relative fault distance corresponding to each fault branch, and output the fault warning information.

[0075] In an embodiment of the present application, for each specific fault branch in the fault branch set, a query operation is first performed to obtain the actual branch total distance associated with the fault branch. Then, the actual fault distance corresponding to the fault branch is obtained by calculating the product of the actual branch total distance and the corresponding fault relative distance. On this basis, the corresponding fault warning information is further generated according to the actual fault distance and the fault scenario associated with the fault branch. The generated fault warning information will be sent to the designated terminal device and displayed visually on the terminal device so that relevant personnel can understand and handle the fault situation in a timely manner.

[0076] In addition, the technical solution provided in this application has excellent scalability, and the detection model can be flexibly added or adjusted according to actual application requirements and scenarios. This flexibility enables the technical solution to adapt to low-voltage distribution substations of different sizes and types, thereby meeting the diverse power system monitoring and management needs. In this way, the reliability and safety of the power system can be effectively improved, and the stability and continuity of power supply can be ensured. Specifically, the technical solution can adapt to different power system environments and needs by adding or adjusting the parameters and algorithms of the detection model according to different application scenarios. For example, in some specific application scenarios, it may be necessary to increase the sensitivity of the detection model to improve the detection capability of abnormal conditions in the power system; while in other application scenarios, it may be necessary to adjust the computational complexity of the detection model to adapt to the monitoring requirements of large-scale power systems. Through this flexible adjustment and optimization, the technical solution can better adapt to various power system environments, thereby improving the reliability and safety of the power system. In addition, the technical solution also has good compatibility and scalability, and can be seamlessly integrated with other power system monitoring and management technologies to further improve the overall performance and efficiency of the power system. In summary, the technical solution provided in this application can effectively respond to various power system monitoring and management needs through flexible scalability and optimization, improve the reliability and safety of the power system, and ensure the stability and continuity of power supply.

[0077] In this embodiment, once the system successfully predicts the fault point, a detailed fault warning report will be generated immediately. This report will describe the specific location of the fault point in detail, analyze the possible causes of the fault, and provide a series of recommended repair measures. This information will provide strong support for operation and maintenance personnel, helping them to quickly and accurately locate and solve problems. In addition, the system will also record the entire process and results of this fault handling, and save this information as part of the historical data. This historical data will be used to continuously optimize and improve the prediction and detection functions of the system, and improve the efficiency and accuracy of future fault handling. In general, through this series of operations and processes, it is possible to quickly and accurately identify and handle faulty feeders and faulty branches in the power system, thereby ensuring the stable operation and power supply quality of the power system.

[0078] Specifically, after predicting the fault point, the system can immediately start an automated process to generate a report containing the detailed location of the fault point. This report will not only point out the specific location of the fault, but also deeply analyze the various causes that may have caused the fault, thereby providing comprehensive reference information for operation and maintenance personnel. The report will also list a series of system-optimized and recommended repair measures, which are based on the system's analysis and learning of historical fault data, and can help operation and maintenance personnel solve problems more efficiently. In addition, the system will record every step and final result of the fault handling in detail and save this information as part of the historical data. These historical data not only help operation and maintenance personnel review and summarize the lessons learned from the fault handling, but also provide the system with valuable learning materials for continuous optimization and improvement of its prediction and detection functions. Through the accumulation and analysis of these historical data, the system can gradually improve its prediction accuracy and processing efficiency for future faults. Finally, through this series of operations and processes, the system can quickly and accurately identify and handle faulty feeders and faulty branches in the power system, thereby ensuring the stable operation and power supply quality of the power system, and bringing more reliable and efficient power services to users.

[0079] The method provided in the embodiment of the present application first determines the low-voltage distribution station area to be detected for faults, and constructs a feature set to be predicted using the operating state data to be predicted of the low-voltage distribution station area. Further, the feature set to be predicted is input into the feeder detection model so that the feeder detection model uses the feature set to be predicted to determine whether there is a faulty feeder in the low-voltage distribution station area and outputs the prediction result. When the prediction result indicates the existence of a faulty feeder, the feature set to be predicted is input into the branch detection model so that the branch detection model uses the feature set to be predicted to determine the faulty branch in the branch corresponding to the faulty feeder and output the faulty branch set, and the faulty branch set includes at least one faulty branch detected by the branch detection model. Next, each faulty branch included in the faulty branch set is input into the category detection model so that the category detection model determines the corresponding fault category for each faulty branch and outputs it. Subsequently, each faulty branch included in the faulty branch set is input into the distance detection model so that the distance detection model calculates the fault relative distance corresponding to each faulty branch according to the fault point position of each faulty branch and outputs it. Finally, according to the fault scenario and relative distance of each fault branch, fault warning information is generated and output. In the embodiment of the present application, through a multi-level detection model, from feeder to branch to fault category judgment, faults can be identified and located more accurately, reducing false alarms and missed alarms. At the same time, through the detailed information of the predicted fault category and relative distance of the fault, maintenance personnel and equipment can be arranged more reasonably, improving maintenance efficiency and power supply stability, thereby improving user satisfaction with power services.

[0080] Further, as Figure 1 In a specific implementation of the method, the present application provides a low-voltage distribution station area fault detection device, such as Figure 5 As shown, the device includes: a determination module 501, a first detection module 502, a second detection module 503, a third detection module 504, a fourth detection module 505, and a generation module 506.

[0081] The determination module 501 is used to determine the low-voltage distribution station area to be fault-detected, and to construct a feature set to be predicted using the operation status data to be predicted of the low-voltage distribution station area; The first detection module 502 is used to input the feature set to be predicted into the feeder detection model, so that the feeder detection model uses the feature set to be predicted to determine whether there is a faulty feeder in the low-voltage distribution station area and output a prediction result; The second detection module 503 is used for inputting the feature set to be predicted into the branch detection model when the prediction result indicates that there is a faulty feeder, so that the branch detection model uses the feature set to be predicted to determine the faulty branch in the branch corresponding to the faulty feeder and output a faulty branch set, wherein the faulty branch set includes at least one faulty branch detected by the branch detection model; The third detection module 504 is used to input each fault branch included in the fault branch set into a category detection model, so that the category detection model determines a corresponding fault category for each fault branch and outputs it; The fourth detection module 505 is used to input each fault branch included in the fault branch set into a distance detection model, so that the distance detection model calculates the relative fault distance corresponding to each fault branch according to the fault point position of each fault branch and outputs it; The generating module 506 is used to generate fault warning information according to the fault scenario and relative fault distance corresponding to each fault branch, and output the fault warning information.

[0082] In specific application scenarios, such as Figure 6 As shown, the device also includes: an initialization module 601, an acquisition module 602, a splitting module 603, a training module 604, a verification module 605, a testing module 606, and a judgment module 607.

[0083] The initialization module 601 is used to construct a model architecture for each detection model based on a deep neural network, and initialize the model architecture parameters of the model architecture corresponding to each detection model, wherein the detection model is any one of the feeder detection model, the branch detection model, the category detection model and the distance detection model, and the model architecture parameters include but are not limited to batch normalization parameters, random dropout parameters, learning rate parameters, activation functions, number of neural network layers, number of neurons in each layer and normalization parameters; The acquisition module 602 is used to acquire the historical operation status data of the low-voltage distribution station area, and select the features of the historical operation status data using a tree-structured Parzen estimator according to a preset feature selection standard to obtain a sample input feature set, wherein the historical operation status data includes first historical operation status data before the fault and second historical operation status data after the fault, and the data items of the historical operation status data include voltage data, current data, power generation data and load data; The splitting module 603 is used to divide the sample input feature set according to a preset division ratio to obtain the training data set, the verification data set and the test data set; The training module 604 is used to perform model training on each of the detection models using the training data set until the model weight parameters of each of the detection models tend to be stable, thereby obtaining a plurality of detection models to be verified, wherein the model weight parameters are weight parameters of mapping connections between different neurons; The verification module 605 is used to optimize the multiple detection models to be verified by using the verification data set until each of the detection models to be verified meets the preset stop criterion, complete the model verification of the multiple detection models to be verified, and obtain multiple detection models to be tested; The testing module 606 is used to test the plurality of detection models to be tested using the test data set, and calculate the prediction accuracy of each detection model to be tested; The judgment module 607 is used to end the testing of multiple detection models to be tested when the prediction accuracy of each detection model to be tested meets the preset accuracy threshold, and obtain the feeder detection model, the branch detection model, the category detection model and the distance detection model.

[0084] In a specific application scenario, the training module 604 is used for any detection model, wherein the detection model converts and transmits the data input to the detection model through a mapping equation of a neuron to obtain the predicted data output by the detection model. The mapping equation of the neuron is specifically shown in the following formula: ; in, For the The values ​​of the layer neurons; represents the activation function of the neuron; is the value of the neuron in the previous layer; For the All neurons in the layer Layer of neurons The weight matrix between ; is a neuron bias parameter, wherein the activation function is a sigmoid function, a hyperbolic tangent function or a rectified linear unit; determining the real data associated with the detection model, calculating the difference between the corresponding real data and the predicted data based on the cost function associated with each detection model, and optimizing the model architecture parameters of the corresponding detection model according to the difference value until the model weight parameters of each detection model tend to be stable.

[0085] In a specific application scenario, the training module 604 inputs the data of the feeder detection model and the branch detection model as a training data set, and the data input to the classification detection model and the distance detection model is the set of fault branches output by the branch detection model; the cost function of the feeder detection model, the branch detection model, and the classification detection model is the minimization of the binary cross entropy loss function; the cost function of the distance detection model is the standard mean square error function.

[0086] In a specific application scenario, the determination module 501 is used to use a tree-structured Parzen estimator to perform feature selection on the operating status data to be predicted, determine current data, voltage data, power generation data and load data; interpolate the voltage data and the current data to obtain an interpolated voltage matrix and an interpolated current matrix, and construct a sample input feature set based on the interpolated voltage matrix, the interpolated current matrix, the power generation data and the load data.

[0087] In a specific application scenario, the determination module 501 is used to determine the branch voltage data of each branch and the position of the voltage sensor that collects the branch voltage data in the voltage data, generate an equally spaced position matrix corresponding to each branch according to the position of the voltage sensor and a preset interpolation number, calculate the interpolation voltage value of each branch according to the branch voltage data of each branch and the position of the corresponding voltage sensor, and insert the interpolation voltage value of each branch into the branch voltage data of each branch according to the equally spaced position matrix to obtain the interpolation voltage matrix of the low-voltage distribution station area; determine the branch current data of each branch and the position of the current sensor that collects the branch current data in the current data, generate an equally spaced position matrix corresponding to each branch according to the position of the current sensor and a preset interpolation number, calculate the interpolation current value of each branch according to the branch current data of each branch and the position of the corresponding current sensor, and insert the interpolation current value of each branch into the branch current data of each branch according to the equally spaced position matrix to obtain the interpolation current matrix of the low-voltage distribution station area.

[0088] In a specific application scenario, the generation module 506 is used to query the actual branch total distance associated with each fault branch in the fault branch set, and calculate the product of the actual branch total distance and the corresponding fault relative distance to obtain the actual fault distance corresponding to the fault branch; generate fault warning information according to the actual fault distance and the fault scenario associated with the fault branch, and send the fault warning information to a designated terminal device for visual display.

[0089] The device provided in the embodiment of the present application first determines the low-voltage distribution station area to be detected for faults, and uses the operating state data to be predicted of the low-voltage distribution station area to construct a feature set to be predicted. Further, the feature set to be predicted is input into the feeder detection model, so that the feeder detection model uses the feature set to be predicted to determine whether there is a faulty feeder in the low-voltage distribution station area and outputs the prediction result. When the prediction result indicates that there is a faulty feeder, the feature set to be predicted is input into the branch detection model, so that the branch detection model uses the feature set to be predicted to determine the faulty branch in the branch corresponding to the faulty feeder and output the faulty branch set, and the faulty branch set includes at least one faulty branch detected by the branch detection model. Next, each faulty branch included in the faulty branch set is input into the category detection model, so that the category detection model determines the corresponding fault category for each faulty branch and outputs it. Subsequently, each faulty branch included in the faulty branch set is input into the distance detection model, so that the distance detection model calculates the fault relative distance corresponding to each faulty branch according to the fault point position of each faulty branch and outputs it. Finally, according to the fault scenario and relative distance of each fault branch, fault warning information is generated and output. In the embodiment of the present application, through a multi-level detection model, from feeder to branch to fault category judgment, faults can be identified and located more accurately, reducing false alarms and missed alarms. At the same time, through the detailed information of the predicted fault category and relative distance of the fault, maintenance personnel and equipment can be arranged more reasonably, improving maintenance efficiency and power supply stability, thereby improving user satisfaction with power services.

[0090] It should be noted that for other corresponding descriptions of the functional units involved in the low-voltage distribution station area fault detection device provided in the embodiment of the present application, reference can be made to Figures 1 to 4 The corresponding description in will not be repeated here.

[0091] In an exemplary embodiment, see Figure 7 , and also provides a device, which includes a communication bus, a processor, a memory and a communication interface, and may also include an input and output interface and a display device, wherein each functional unit can communicate with each other through the bus. The memory stores a computer program, and the processor is used to execute the program stored in the memory and execute the low-voltage distribution station area fault detection method in the above embodiment.

[0092] Optionally, the physical device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a WI-FI interface), etc.

[0093] Those skilled in the art will appreciate that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or a combination of certain components, or different arrangements of components.

[0094] A storage medium stores a computer program, which implements the steps of the low-voltage distribution station area fault detection method when executed by a processor.

[0095] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware of the physical device and the software resources to be identified, and supports the operation of the information processing program and other software and / or programs to be identified. The network communication module is used to realize the communication between the components inside the storage medium, and the communication with other hardware and software in the information processing physical device.

[0096] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform, or by hardware. By applying the technical solution of the present application, the low-voltage distribution station area to be detected for fault is first determined, and the predicted operating state data of the low-voltage distribution station area is used to construct a feature set to be predicted. Further, the feature set to be predicted is input into the feeder detection model so that the feeder detection model uses the feature set to be predicted to determine whether there is a faulty feeder in the low-voltage distribution station area and outputs the prediction result. When the prediction result indicates that there is a faulty feeder, the feature set to be predicted is input into the branch detection model so that the branch detection model uses the feature set to be predicted to determine the faulty branch in the branch corresponding to the faulty feeder and output the faulty branch set, and the faulty branch set includes at least one faulty branch detected by the branch detection model. Next, each faulty branch included in the faulty branch set is input into the category detection model so that the category detection model determines the corresponding fault category for each faulty branch and outputs it. Subsequently, each fault branch included in the fault branch set is input into the distance detection model, so that the distance detection model calculates and outputs the relative fault distance corresponding to each fault branch according to the fault point position of each fault branch. Finally, according to the fault scenario and relative fault distance corresponding to each fault branch, fault warning information is generated, and the fault warning information is output. Compared with the prior art, in the embodiment of the present application, through a multi-level detection model, from feeder to branch to fault category judgment, faults can be more accurately identified and located, and false alarms and missed alarms can be reduced. At the same time, through the detailed information of the predicted fault category and relative fault distance, maintenance personnel and equipment can be arranged more reasonably, maintenance efficiency and power supply stability can be improved, and user satisfaction with power services can be improved.

[0097] Those skilled in the art will appreciate that the accompanying drawings are merely schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required for implementing the present application.

[0098] Those skilled in the art will appreciate that the modules in the device in the implementation scenario can be distributed in the device in the implementation scenario according to the implementation scenario description, or can be changed accordingly and located in one or more devices different from the implementation scenario. The modules in the above implementation scenario can be combined into one module, or can be further split into multiple sub-modules.

[0099] The above application serial numbers are for description only and do not represent the advantages or disadvantages of the implementation scenarios.

[0100] The above disclosure only discloses several specific implementation scenarios of the present application. However, the present application is not limited thereto, and any changes that can be conceived by technicians in this field should fall within the scope of protection of the present application.

Claims

1. A method for detecting faults in a low voltage distribution station area, characterized in that: include: Determine a low-voltage distribution substation area to be fault-detected, and construct a feature set to be predicted using the operation status data to be predicted of the low-voltage distribution substation area; Inputting the feature set to be predicted into a feeder detection model, so that the feeder detection model uses the feature set to be predicted to determine whether there is a faulty feeder in the low-voltage distribution station area and output a prediction result; When the prediction result indicates that there is a faulty feeder, the feature set to be predicted is input into a branch detection model, so that the branch detection model uses the feature set to be predicted to determine a faulty branch in a branch corresponding to the faulty feeder and output a faulty branch set, wherein the faulty branch set includes at least one faulty branch detected by the branch detection model; Inputting each fault branch included in the fault branch set into a category detection model, so that the category detection model determines a corresponding fault category for each fault branch and outputs it; Input each fault branch included in the fault branch set into a distance detection model, so that the distance detection model calculates the relative fault distance corresponding to each fault branch according to the fault point position of each fault branch and outputs it; According to the fault scenario and relative fault distance corresponding to each of the fault branches, fault warning information is generated and output.

2. The method according to claim 1, characterized in that: Before determining the low-voltage distribution station area to be fault-detected and constructing the feature set to be predicted using the operation status data to be predicted of the low-voltage distribution station area, the method further includes: Constructing a model architecture for each detection model based on a deep neural network, and initializing the model architecture parameters of the model architecture corresponding to each detection model, wherein the detection model is any one of the feeder detection model, the branch detection model, the category detection model, and the distance detection model, and the model architecture parameters include but are not limited to a batch normalization parameter, a random dropout parameter, a learning rate parameter, an activation function, the number of neural network layers, the number of neurons in each layer, and a normalization parameter; Acquire historical operation status data of the low-voltage distribution station area, and perform feature selection on the historical operation status data using a tree-structured Parzen estimator according to a preset feature selection standard to obtain a sample input feature set, wherein the historical operation status data includes first historical operation status data before a fault and second historical operation status data after a fault, and the data items of the historical operation status data include voltage data, current data, power generation data, and load data; Dividing the sample input feature set according to a preset division ratio to obtain a training data set, a verification data set, and a test data set; Using the training data set to perform model training on each of the detection models until the model weight parameters of each of the detection models tend to be stable, thereby obtaining a plurality of detection models to be verified, wherein the model weight parameters are weight parameters of mapping connections between different neurons; The validation data set is used to optimize the multiple detection models to be validated until each of the detection models to be validated meets a preset stop criterion, and the model validation of the multiple detection models to be validated is completed to obtain multiple detection models to be tested; Using the test data set to test the plurality of detection models to be tested, and calculating the prediction accuracy of each of the detection models to be tested; When the prediction accuracy of each of the detection models to be tested meets the preset accuracy threshold, the testing of the multiple detection models to be tested is terminated to obtain the feeder detection model, the branch detection model, the category detection model and the distance detection model.

3. The method according to claim 2, characterized in that The method of using the training data set to perform model training on each of the detection models until the model weight parameters of each of the detection models tend to be stable, thereby obtaining a plurality of detection models to be verified, including: For any detection model, the detection model converts and transmits the data input to the detection model through the mapping equation of the neuron to obtain the predicted data output by the detection model. The mapping equation of the neuron is specifically shown in the following formula: ； in, For the The values ​​of the layer neurons; represents the activation function of the neuron; is the value of the neuron in the previous layer; For the All neurons in the layer Layer of neurons The weight matrix between ; is a neuron bias parameter, wherein the activation function is a sigmoid function, a hyperbolic tangent function or a rectified linear unit; Determine the real data associated with the detection model, calculate the difference value between the corresponding real data and the predicted data based on the cost function associated with each detection model, and optimize the model architecture parameters of the corresponding detection model according to the difference value until the model weight parameters of each detection model tend to be stable.

4. The method according to claim 3, characterized in that The data input to the feeder detection model and the branch detection model are training data sets, and the data input to the classification detection model and the distance detection model are the fault branch set output by the branch detection model; The cost function of the feeder detection model, the branch detection model, and the classification detection model is to minimize the binary cross entropy loss function; The cost function of the distance detection model is a standard mean square error function.

5. The method according to claim 1, characterized in that The step of determining a low-voltage distribution station area to be fault-detected and constructing a feature set to be predicted using the operation status data to be predicted of the low-voltage distribution station area comprises: A tree-structured Parzen estimator is used to perform feature selection on the operation state data to be predicted, and current data, voltage data, power generation data and load data are determined; The voltage data and the current data are interpolated to obtain an interpolation voltage matrix and an interpolation current matrix, and a sample input feature set is constructed according to the interpolation voltage matrix, the interpolation current matrix, the power generation data and the load data.

6. The method according to claim 5, characterized in that The interpolation processing is performed on the voltage data and the current data to obtain an interpolation voltage matrix and an interpolation current matrix, including: Determine the branch voltage data of each branch and the position of the voltage sensor that collects the branch voltage data in the voltage data, generate an equally spaced position matrix corresponding to each branch according to the position of the voltage sensor and a preset interpolation number, calculate the interpolation voltage value of each branch according to the branch voltage data of each branch and the position of the corresponding voltage sensor, and insert the interpolation voltage value of each branch into the branch voltage data of each branch according to the equally spaced position matrix to obtain the interpolation voltage matrix of the low-voltage distribution station area; The branch current data of each branch and the position of the current sensor that collects the branch current data are determined in the current data, and an equally spaced position matrix corresponding to each branch is generated according to the position of the current sensor and a preset interpolation number, and the interpolated current value of each branch is calculated according to the branch current data of each branch and the position of the corresponding current sensor, and the interpolated current value of each branch is inserted into the branch current data of each branch according to the equally spaced position matrix to obtain the interpolated current matrix of the low-voltage distribution station area.

7. The method according to claim 1, characterized in that The generating fault warning information according to the fault scenario and the relative fault distance corresponding to each fault branch, and outputting the fault warning information, comprises: For each faulty branch in the faulty branch set, query the actual branch total distance associated with the faulty branch, and calculate the product of the actual branch total distance and the corresponding fault relative distance to obtain the actual fault distance corresponding to the faulty branch; Fault warning information is generated according to the actual fault distance and the fault scenario associated with the fault branch, and the fault warning information is sent to a designated terminal device for visual display.

8. A low voltage distribution station area fault detection device, characterized in that: include: A determination module, used to determine a low-voltage distribution substation area to be detected for faults, and to construct a feature set to be predicted using the operation status data to be predicted of the low-voltage distribution substation area; A first detection module, used for inputting the feature set to be predicted into a feeder detection model, so that the feeder detection model uses the feature set to be predicted to determine whether there is a faulty feeder in the low-voltage distribution station area and output a prediction result; a second detection module, configured to input the feature set to be predicted into a branch detection model when the prediction result indicates that there is a faulty feeder, so that the branch detection model uses the feature set to be predicted to determine a faulty branch in a branch corresponding to the faulty feeder and output a faulty branch set, wherein the faulty branch set includes at least one faulty branch detected by the branch detection model; A third detection module, configured to input each faulty branch included in the faulty branch set into a category detection model, so that the category detection model determines a corresponding fault category for each faulty branch and outputs the result; A fourth detection module, used for inputting each fault branch included in the fault branch set into a distance detection model, so that the distance detection model calculates and outputs the relative fault distance corresponding to each fault branch according to the fault point position of each fault branch; A generation module is used to generate fault warning information according to the fault scenario and relative fault distance corresponding to each of the fault branches, and output the fault warning information.

9. A device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. 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 method according to any one of claims 1 to 7 are implemented.

Images