Research On Line Loss Causes Of Distribution Network Based On Association Relationship And Deep Learning Methods

With the continuous maturity of big data,artificial intelligence,Internet of Things and other technologies,the development of smart grid has also been rapidly promoted,but at the same time,the increasing line loss has also attracted wide attention.In the process of building smart power grid,each link of power grid operation produces a large amount of multi-source heterogeneous data,including line loss data and line loss causerelated data,which constitute the big data of line loss.Distribution network is a region with high incidence of line loss.The complexity of distribution network structure also determines the diversity and complexity of the causes of line loss.The analysis of the cause of line loss is an important part of line loss analysis,which is beneficial to the accurate location of line loss and the exploration of line loss correlation.Therefore,it is of great significance to research the cause of line loss in distribution network.The cause of line loss in distribution network is complex and difficult to locate,and it is difficult to accurately predict line loss.In this paper,a line loss cause analysis method based on correlation relationship and deep learning prediction is proposed.Taking the station area of distribution network as the analysis object,from the perspective of nontechnical line loss,line loss features are mined from the four abnormal causes,namely,file anomaly,tabulation anomaly,acquisition anomaly,and station operation anomaly.Due to different courts may show different line loss characteristics and associated characteristics,this paper combined with the feature of load curve and other area to differentiate area clustering,and then for each type of area of line loss feature dimension reduction optimization,including eliminate similar characteristics,excluding irrelevant features and extracting principal component,while retain the original characteristics of the main information will streamline line loss characteristics of the data set.The line loss feature data sets,which were not optimized,only partially optimized with information entropy,and completely optimized with information entropy and principal component analysis,were input into BP neural network for line loss prediction training,it is verified that the model which features optimized has the best prediction effect.The optimized characteristic data set will be used for the correlation analysis between the causes of line loss and the correlation analysis between the causes of line loss and line loss.Firstly,considering the mining efficiency in big data,the FP-Growth algorithm of association rules was selected to search the frequent item sets of line loss features,and the correlation between the causes of line loss was analyzed with support,confidence and promotion degree as evaluation indexes.Secondly,a line loss prediction model based on deep learning is established.By eliminating the influence of line loss characteristics in turn,the correlation contribution degree of line loss causes to line loss is calculated,and the line loss caused by the line loss causes is quantified.It has been verified that the deep confidence network and BP deep neural network as the prediction model of deep learning method are better than the shallow artificial neural network model in the prediction effect,and the accuracy of prediction means the reliability of contribution calculation.Finally,combined with the analysis of the above two aspects,the causes of line loss in the station area are comprehensively evaluated,and the guidance and suggestions are given to assist the power enterprises to make decisions.