Research And Application Of Distribution Transformer Line Loss Analysis Method Based On Data Mining

The complex and diverse topology of the low-voltage distribution network and the large supply area make it difficult to calculate the theoretical line loss in the station area.The lack of an effective and convenient way to diagnose the causes of abnormal line loss in a station area is not conducive to refining station area line loss management and achieving accurate loss reduction.Therefore,this thesis takes station loss as the research object,based on data mining technology,to study the problems of station theoretical line loss calculation and power theft detection,and to develop a station line loss visualisation management system to help staff make decisions.The specific research contents and results are as follows:(1)An appropriate pre-processing method is given according to the data characteristics,and a sample data expansion method based on generative adversarial networks is proposed.Firstly,a decision tree induction method is established to check the quality of the data,and then the problems of missing values and outliers are dealt with to improve the usability of the data;given the shortage of available station sample data,the pre-processed station sample data is expanded by using generative adversarial networks.The experiments show that the generated data can effectively preserve the similarity and diversity of the original samples and increase the capacity of the dataset.The "raw data" are transformed into high quality "cooked data",which provide a good data basis for station area loss analysis.(2)In order to achieve accurate calculation of line loss,a theoretical line loss calculation method based on an improved neural network for the station area is proposed.Firstly,based on the fluctuation characteristics of the line loss data,the discrimination coefficient is determined,and the improved grey correlation method is used to analyse the correlation between the feature parameters and the theoretical line loss of the station area,and to construct a suitable set of variables;then,a one-dimensional convolutional neural network is used to enhance the features of the feature variable sequence,a BP neural network is established to calculate the theoretical line loss of the station area,and the chaotic adaptive grey wolf algorithm is used to optimise the model network structure parameters;finally,the performance of the algorithm is verified by example.The experiments show that the mean squared error of the algorithm proposed in this paper is 0.34 on the training set and the average relative error of calculation on the test set is6.32%,which can effectively improve the calculation accuracy compared with BP neural network,GA-BP neural network and GWO-BP neural network models.(3)A CNN-SVM model-based method is proposed to identify customers’ electricity theft behaviour,which enables efficient detection of customers’ electricity theft behaviour.The traditional convolutional neural network structure is then improved by replacing the softmax layer in the network with SVM,using CNN to extract key features of data sequences,and using SVM to achieve electricity theft discrimination.The Dropout layer is introduced to reduce neuron coupling and prevent overfitting,and the model hyperparameters are optimised based on Bayesian to improve the model performance;finally,after experimental validation,the established CNN-SVM model is compared with 1D-CNN and SVM methods,with an accuracy of 95.08%,a recall of 96.77% and a harmonic mean of 95.24% in the performance evaluation index.The overall performance is optimal.(4)Development of a station area line loss visualisation management system.Based on the actual functional requirements of enterprises,this paper develops a station area line loss visualization management system based on the research results of this paper.It adopts a B/S architecture combined with technical methods such as station area theoretical line loss calculation,line loss abnormality cause analysis and data visualization to realize the functions of real-time station area operation status monitoring,station area equipment monitoring and detection,and line loss data visualization.The digital twin model of station equipment operation status is built using "condition-state-event" to realise the visualisation and management of station line loss in the process and time.The system can provide timely warning and quick location of abnormal line loss areas,store information on abnormal line loss areas,perform data management work in the background of the system,analyse possible causes of abnormalities,and help staff make decisions.