Low-voltage Tripping Prediction Of Distribution Transformer Area Based On Generative Adversarial Networks And Ensemble Learning

As an important part of the distribution network,the distribution substation area undertakes the major task of supplying power to users in the area.Its safe and stable operation is of great significance to the safe and reliable operation of the power system.With the surge of residential,commercial and production power load,low-voltage tripping and power failure often occur in the distribution transformer area,and customer complaints occur frequently.If the fault is not handled in time,it will further cause more serious consequences.At present,most scholars mainly control the operation situation of the distribution transformer station area through the load and heavy overload prediction of the distribution transformer station area,and there is little research on the low-voltage trip prediction of the distribution transformer station area directly.In this thesis,with the help of large-scale data resources in the distribution transformer area,the probability of low-voltage tripping in the distribution transformer area is simulated and modeled.On the basis of summarizing the existing relevant research and aiming at the characteristics of low-voltage trip prediction in distribution transformer area,this thesis proposes a low-voltage trip prediction model in distribution transformer area based on generation countermeasure network and ensemble learning,which includes the following work:Firstly,the data of distribution transformer area used for simulation modeling in this thesis are preprocessed,such as outlier elimination,normalization and One-Hot feature conversion,and then the main factors affecting low-voltage tripping in distribution transformer area are preliminarily analyzed by pearson correlation coefficient method;Then the pearson correlation coefficient method is used to preliminarily analyze the main factors affecting the low-voltage tripping in the distribution transformer area.Then,in order to solve the problem that it is difficult for the prediction model to identify a small number of samples due to the imbalance of samples in the original data set,this thesis establishes the WGAN-div model based on wasserstein divergence,skillfully adds residual blocks to WGAN-div to alleviate the problem of gradient disappearance in the training process.A method of lowvoltage trip data expansion in the distribution transformer area based on WGAN-div is proposed.Then,in order to better process the data of distribution transformer area with large amount of data and a large number of features,based on the deep learning model of residual network（ResNet）,this thesis innovatively proposes a low-voltage trip prediction model of distribution transformer area based on Bagging-ResNet which aims to enhance the generalization performance of the model by integrating the prediction results of multiple ResNets,so as to obtain higher prediction accuracy.In view of the defects of traditional statistical analysis methods such as large influence by data samples,weak expression ability of nonlinear relationship and difficult to explain the spatial contribution direction of characteristic factors in different value ranges,this thesis proposes a novel analysis method of influencing factors of low-voltage tripping in distribution transformer area based on shapley additive interpretation（SHAP）interpretation framework,which is used to deeply quantify the marginal contribution of explanatory characteristic variables to the final prediction model.Finally,the simulation modeling is carried out based on the measured data of a distribution transformer station in Guangdong Province.The experimental results show that the WGAN-div-Bagging-Resnet model proposed in this thesis shows better prediction performance than other models;At the same time,by explaining the influencing factors and risk division mechanism through SHAP,it can provide a reliable basis for the operation and maintenance personnel in the distribution transformer substation area and assist the operation and maintenance personnel to make reasonable decisions.