A method for intelligent management of power distribution system based on machine learning

Abstract

The present invention discloses a method for intelligent management of a power distribution system based on machine learning, the method comprising data collection, data preprocessing, establishing a distribution system abnormality monitoring model and intelligent management of the distribution system. The present invention belongs to the technical field of power system management, and specifically refers to a method for intelligent management of a power distribution system based on machine learning. The present scheme introduces feature optimization and deviation optimization layers to enhance the capture of feature independence and correlation, improve the ability to handle diversified equipment, thereby improving the accuracy of prediction of the state of distribution system equipment, and strengthening the management of the distribution system; designing an abnormality monitoring loss function, capturing important abnormal points, and paying more attention to key abnormal features, thereby improving the accuracy of abnormality monitoring; by designing a dynamic learning rate adjustment, responding to unbalanced data in the distribution system, achieving rapid convergence, improving the efficiency of intelligent management of the distribution system, and promoting the stable operation and management of the distribution system.

Description

Technical Field

[0001] The present invention relates to the technical field of power system management, and in particular to an intelligent management method for a power distribution system based on machine learning. Background Art

[0002] Intelligent management methods for distribution systems refer to a series of methods and processes that use advanced technical means, algorithms and information technology to conduct real-time monitoring, analysis and optimization management of power distribution systems. Its purpose is to improve the operating efficiency, reliability and safety of the distribution system, while reducing maintenance costs and accident risks. However, general intelligent management methods for distribution systems have weak generalization capabilities for different equipment and states, which leads to low prediction accuracy of the status of distribution system equipment; general intelligent management methods for distribution systems pay little attention to key abnormal features and have poor ability to capture important abnormal points in the distribution system, which leads to poor accuracy of abnormal monitoring and low monitoring efficiency. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a distribution system intelligent management method based on machine learning. In view of the fact that the general distribution system intelligent management method has weak generalization ability for different devices and states, which leads to low prediction accuracy of the distribution system equipment state, this scheme introduces feature optimization and deviation optimization layers to enhance the capture of feature independence and correlation, improve the ability to handle diversified equipment and unpredictable operating environments, thereby improving the prediction accuracy of the distribution system equipment state and strengthening the management of the distribution system; in view of the fact that the general distribution system intelligent management method has little attention to key abnormal features and poor ability to capture important abnormal points of the distribution system, which leads to poor accuracy of abnormal monitoring and low monitoring efficiency, this scheme designs an abnormal monitoring loss function by considering errors, confidence weights and change rates, captures important abnormal points, and pays more attention to key abnormal features, thereby improving the accuracy of abnormal monitoring; by designing dynamic learning rate adjustment, it can deal with unbalanced data in the distribution system, achieve rapid convergence, improve the efficiency of intelligent management of the distribution system, and promote stable operation and management of the distribution system.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides a distribution system intelligent management method based on machine learning, the method comprising the following steps:

[0005] Step S1: data collection;

[0006] Step S2: data preprocessing;

[0007] Step S3: Establishing a distribution system abnormality monitoring model;

[0008] Step S4: Intelligent management of power distribution system.

[0009] Furthermore, in step S1, the data collection is to collect historical power distribution system management data; the historical power distribution system management data includes voltage, current, active power, reactive power, load, operating time, grid frequency, meteorological conditions and equipment status; the equipment status includes normal operation, slight abnormality, moderate abnormality and severe abnormality; the equipment status is used as a data label.

[0010] Furthermore, in step S2, the data preprocessing is to perform data cleaning, data conversion and data set division on the collected data; the data cleaning is to process missing values, duplicate values ​​and outliers; the data conversion is to convert the data into vector form and perform normalization processing; the data set division is to divide the processed data into a test set and a training set.

[0011] Furthermore, in step S3, the establishment of the power distribution system abnormality monitoring model specifically includes the following steps:

[0012] Step S31: Data stacking: Divide the input into windows by stacking, each window contains K time steps of data, and generate new input;

[0013] Step S32: Model architecture design; the established distribution system abnormality monitoring model includes: a causal convolution layer, which is used to extract local features and ensure the causal relationship of the time series; a bidirectional LSTM layer, which uses the bidirectional LSTM layer to capture the bidirectional long dependency features in the time series data and handle the abnormality monitoring task across multiple time steps in the distribution system; finally, the outputs of the two layers are merged in the fully connected layer to output the equipment status of the distribution system; expressed as: ; Among them, y output is the final output of the model, which is used to predict the equipment status of the power distribution system; FC is the fully connected layer, which produces the final prediction result; Concat is the connection operation; y CNN is the output of the causal convolutional layer; y LSTM is the output of the bidirectional LSTM layer;

[0014] Step S33: feature optimization: introducing a deviation optimization layer in each layer of the network structure to enhance the capture of feature independence and correlation. The deviation optimization layer is expressed as: ; ;in, is the output feature of the kth layer; μ k is the mean of the input features; is the variance of the input feature; α k is the scale adjustment parameter; x k is the input feature; β k is the offset adjustment parameter; γ k is the adaptive adjustment factor; is an element-wise product operation;

[0015] Step S34: Design an abnormal monitoring loss function; define the operation abnormality including; in order to capture the abnormal points including voltage mutation and current overload, an abnormal monitoring loss function is designed, which is expressed as: ; ; Where DL is the anomaly monitoring loss function; α1 and α2 are weight parameters that control the impact of the loss term; N is the total number of time steps, and t is the current time step; is the confidence weight; is the true label; is the predicted label; is the confidence weight of the rate of change; is the true label change rate; is the predicted label change rate; is the time weighting factor; β is the adjustment parameter; is the number of anomalies detected at time step t; is the total number of data at time step t;

[0016] Step S35: Learning rate adjustment; the learning rate of the deviation optimization layer is expressed as: ; ;in, and are the scale adjustment parameters before and after updating, respectively; is the gradient of the loss function, is the learning rate of the bias optimization layer at layer k; is the initial learning rate; is the gradient at time step t, is the amount of change;

[0017] Step S36: model determination; the model is selected to train the distribution system abnormality monitoring model based on the training set. When the model converges on the loss, the distribution system abnormality monitoring model training is completed; the prediction limit is set in advance. When the prediction accuracy of the trained distribution system abnormality monitoring model for the test set is higher than the prediction limit, the distribution system abnormality monitoring model is established; if the maximum number of training times is reached or the prediction amount of the trained distribution system abnormality monitoring model for the test set is not higher than the prediction limit, the initial parameters of the model are optimized based on the particle swarm search algorithm, and the data set is re-divided to train the distribution system abnormality monitoring model.

[0018] Furthermore, in step S4, the intelligent management of the distribution system is based on the established distribution system abnormality monitoring model, and voltage, current, active power, reactive power, load and operating time are collected in real time, and input into the distribution system abnormality monitoring model after preprocessing. When the model outputs the device status that the operation is abnormal, an early warning is given and the management personnel are notified.

[0019] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0020] (1) In view of the problem that general intelligent management methods for distribution systems have weak generalization capabilities for different devices and states, which leads to low prediction accuracy of distribution system equipment states, this solution introduces feature optimization and deviation optimization layers to enhance the capture of feature independence and correlation, improve the ability to handle diverse equipment and unpredictable operating environments, thereby improving the prediction accuracy of distribution system equipment states and strengthening the management of distribution systems.

[0021] (2) In view of the fact that general intelligent management methods of distribution systems pay little attention to key abnormal features and have poor ability to capture important abnormal points of distribution systems, which leads to poor accuracy and low monitoring efficiency of abnormal monitoring, this scheme designs an abnormal monitoring loss function by considering errors, confidence weights and change rates, captures important abnormal points, and pays more attention to key abnormal features, thereby improving the accuracy of abnormal monitoring; by designing dynamic learning rate adjustment, it can cope with unbalanced data of distribution systems and achieve rapid convergence, thereby improving the efficiency of intelligent management of distribution systems and promoting stable operation and management of distribution systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flow chart of a method for intelligent management of a power distribution system based on machine learning provided by the present invention;

[0023] Figure 2 It is a schematic diagram of the process of step S3.

[0024] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0026] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0027] Example 1, see Figure 1 The present invention provides a method for intelligent management of a power distribution system based on machine learning, the method comprising the following steps:

[0028] Step S1: data collection, collecting historical power distribution system management data;

[0029] Step S2: Data preprocessing, data cleaning, data conversion and data set division of the collected data;

[0030] Step S3: Establish a distribution system abnormality monitoring model, introduce feature optimization and deviation optimization layers; design an abnormality monitoring loss function by considering errors, confidence weights and change rates; and implement the establishment of a distribution system abnormality monitoring model in combination with dynamic learning rate adjustment;

[0031] Step S4: Intelligent management of the power distribution system, realizing intelligent management of the power distribution system based on real-time data based on the established power distribution system abnormality monitoring model.

[0032] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the historical distribution system management data includes voltage, current, active power, reactive power, load, operating time, grid frequency, meteorological conditions and equipment status; the equipment status includes normal operation, slight abnormality, moderate abnormality and severe abnormality; the equipment status is used as a data label.

[0033] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, data cleaning is to process missing values, duplicate values ​​and outliers; data conversion is to convert data into vector form and perform normalization processing; data set partitioning is to divide the processed data into a test set and a training set.

[0034] Example 4, see Figure 1 and Figure 2 Based on the above embodiment, in step S3, establishing a distribution system abnormality monitoring model specifically includes the following steps:

[0035] Step S31: Data stacking: Divide the input into windows by stacking, each window contains K time steps of data, and generate new input;

[0036] Step S32: Model architecture design; the established distribution system abnormality monitoring model includes: a causal convolution layer, which is used to extract local features and ensure the causal relationship of the time series; a bidirectional LSTM layer, which uses the bidirectional LSTM layer to capture the bidirectional long dependency features in the time series data and handle the abnormality monitoring task across multiple time steps in the distribution system; finally, the outputs of the two layers are merged in the fully connected layer to output the equipment status of the distribution system; expressed as: ; Among them, y output is the final output of the model, which is used to predict the equipment status of the power distribution system; FC is the fully connected layer, which produces the final prediction result; Concat is the connection operation; y CNN is the output of the causal convolutional layer; y LSTM is the output of the bidirectional LSTM layer;

[0037] Step S33: feature optimization;

[0038] Step S34: Design an abnormal monitoring loss function; define the operation abnormality including; in order to capture the abnormal points including voltage mutation and current overload, an abnormal monitoring loss function is designed, which is expressed as: ; ; Where DL is the anomaly monitoring loss function; α1 and α2 are weight parameters that control the impact of the loss term; N is the total number of time steps, and t is the current time step; is the confidence weight; is the true label; is the predicted label; is the confidence weight of the rate of change; is the true label change rate; is the predicted label change rate; is the time weighting factor; β is the adjustment parameter; is the number of anomalies detected at time step t; is the total number of data at time step t;

[0039] Step S35: Learning rate adjustment; the learning rate of the deviation optimization layer is expressed as: ; ;in, and are the scale adjustment parameters before and after updating, respectively; is the gradient of the loss function, is the learning rate of the bias optimization layer at layer k; is the initial learning rate; is the gradient at time step t, is the amount of change;

[0040] Step S36: model determination; the model is selected to train the distribution system abnormality monitoring model based on the training set. When the model converges on the loss, the distribution system abnormality monitoring model training is completed; the prediction limit is set in advance. When the prediction accuracy of the trained distribution system abnormality monitoring model for the test set is higher than the prediction limit, the distribution system abnormality monitoring model is established; if the maximum number of training times is reached or the prediction amount of the trained distribution system abnormality monitoring model for the test set is not higher than the prediction limit, the initial parameters of the model are optimized based on the particle swarm search algorithm, and the data set is re-divided to train the distribution system abnormality monitoring model.

[0041] By performing the above operations, in view of the fact that the general intelligent management method of the distribution system pays little attention to key abnormal features and has poor ability to capture important abnormal points of the distribution system, which leads to poor accuracy of abnormal monitoring and low monitoring efficiency, this solution designs an abnormal monitoring loss function by considering errors, confidence weights and change rates, captures important abnormal points, and pays more attention to key abnormal features, thereby improving the accuracy of abnormal monitoring; by designing dynamic learning rate adjustment, it can cope with unbalanced data of the distribution system and achieve rapid convergence, thereby improving the efficiency of intelligent management of the distribution system and promoting stable operation and management of the distribution system.

[0042] Example 5, see Figure 2 This embodiment is based on the above embodiment. In step S33, feature optimization is specifically to introduce a deviation optimization layer in each layer of the network structure to enhance the capture of feature independence and correlation. The deviation optimization layer is expressed as: ; ;in, is the output feature of the kth layer; μ k is the mean of the input features; is the variance of the input feature; α k is the scale adjustment parameter; x k is the input feature; β k is the offset adjustment parameter; γ k is the adaptive adjustment factor; is an element-wise product operation.

[0043] By performing the above operations, the general intelligent management method of the power distribution system has a weak generalization ability for different equipment and states, which leads to low prediction accuracy of the power distribution system equipment status. This solution introduces feature optimization and deviation optimization layers to enhance the capture of feature independence and correlation, improve the ability to handle diversified equipment and unpredictable operating environments, thereby improving the prediction accuracy of the power distribution system equipment status and strengthening the management of the power distribution system.

[0044] Example 6, see Figure 1This embodiment is based on the above embodiment. In step S4, the intelligent management of the distribution system is based on the established distribution system abnormality monitoring model. The voltage, current, active power, reactive power, load and operating time are collected in real time and input into the distribution system abnormality monitoring model after preprocessing. When the model outputs the device status that is abnormal, an early warning is given and the management personnel are notified.

[0045] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0046] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0047] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for intelligent management of power distribution system based on machine learning, characterized in that: The method comprises the following steps: Step S1: data collection, collecting historical power distribution system management data; Step S2: Data preprocessing, data cleaning, data conversion and data set division of the collected data; Step S3: Establish a distribution system abnormality monitoring model, introduce feature optimization and deviation optimization layers; design an abnormality monitoring loss function by considering errors, confidence weights and change rates; and implement the establishment of a distribution system abnormality monitoring model in combination with dynamic learning rate adjustment; Step S4: Intelligent management of the power distribution system, realizing intelligent management of the power distribution system based on real-time data based on the established power distribution system abnormality monitoring model; In step S1, the historical power distribution system management data includes voltage, current, active power, reactive power, load, operating time, grid frequency, meteorological conditions and equipment status; the equipment status includes normal operation, slight abnormality, moderate abnormality and severe abnormality; the equipment status is used as a data label; In step S3, the establishment of the power distribution system abnormality monitoring model further includes the following steps: Step S31: Data stacking: Divide the input into windows by stacking, each window contains K time steps of data, and generate new input; Step S32: Model architecture design; the established distribution system abnormality monitoring model includes: a causal convolution layer, which is used to extract local features and ensure the causal relationship of the time series; a bidirectional LSTM layer, which uses the bidirectional LSTM layer to capture the bidirectional long dependency features in the time series data and handle the abnormality monitoring task across multiple time steps in the distribution system; finally, the outputs of the two layers are merged in the fully connected layer to output the equipment status of the distribution system; expressed as: ; Among them, y output is the final output of the model, which is used to predict the equipment status of the power distribution system; FC is the fully connected layer, which produces the final prediction result; Concat is the connection operation; y CNN is the output of the causal convolutional layer; y LSTM is the output of the bidirectional LSTM layer; Step S33: feature optimization: introducing a deviation optimization layer in each layer of the network structure to enhance the capture of feature independence and correlation. The deviation optimization layer is expressed as: ; ;in, is the output feature of the kth layer; μ k is the mean of the input features; is the variance of the input feature; α k is the scale adjustment parameter, which is corrected to a vector based on the broadcast mechanism; x k is the input feature; β k is the offset adjustment parameter; γ k is the adaptive adjustment factor; is an element-wise product operation; Step S34: Design an abnormal monitoring loss function; define the operation abnormality including; in order to capture the abnormal points including voltage mutation and current overload, an abnormal monitoring loss function is designed, which is expressed as: ; ; Where DL is the anomaly monitoring loss function; α1 and α2 are weight parameters that control the impact of the loss term; N is the total number of time steps, and t is the current time step; is the confidence weight; is the true label; is the predicted label; is the confidence weight of the rate of change; is the true label change rate; is the predicted label change rate; is the time weighting factor; β is the adjustment parameter; is the number of anomalies detected at time step t; is the total number of data at time step t; Step S35: Learning rate adjustment; the learning rate of the deviation optimization layer is expressed as: ; ;in, and are the scale adjustment parameters before and after updating, respectively; is the gradient of the loss function; is the learning rate of the bias optimization layer at layer k; is the initial learning rate; is the gradient at time step t, is the amount of change; Step S36: model determination; the model is selected to train the distribution system abnormality monitoring model based on the training set. When the model converges to the loss, the distribution system abnormality monitoring model training is completed; the prediction limit is pre-set. When the prediction accuracy of the trained distribution system abnormality monitoring model for the test set is higher than the prediction limit, the distribution system abnormality monitoring model is established; if the maximum number of training times is reached or the prediction amount of the trained distribution system abnormality monitoring model for the test set is not higher than the prediction limit, the initial parameters of the model are optimized based on the particle swarm search algorithm, and the data set is re-divided to train the distribution system abnormality monitoring model; In step S4, the intelligent management of the distribution system is based on the established distribution system abnormality monitoring model, and the voltage, current, active power, reactive power, load and operating time are collected in real time. After preprocessing, they are input into the distribution system abnormality monitoring model. When the model output device status is abnormal, an early warning is given and the management personnel are notified.

2. The method for intelligent management of a power distribution system based on machine learning according to claim 1, characterized in that: In step S2, the data preprocessing is to clean the collected data, convert the data and divide the data set; the data cleaning is to process missing values, duplicate values ​​and outliers; the data conversion is to convert the data into vector form and normalize it; the data set division is to divide the processed data into a test set and a training set.

Images