Research On Transformer Heavy And Overload Prediction Of Regional Distribution Network Based On Neural Network

Electric energy plays an important role in supporting the national economy.With the rapid development of the economy,the increase of the loads has led to the phenomenon that the distribution network transformers often run heavily and overloaded.Traditionally,the staff used the methods of on-the-spot monitoring and post-processing in the heavy and overload management of distribution transformers,which made it difficult to reduce or even eliminate power outages and accidents caused by heavy loads and overloads.With the extensive use of emerging IT technologies such as the Internet of Things and cloud computing in the power industry and the long-term construction and development of strong smart grids and distribution network automation,the power grid deposits abundant amounts of valuable data in various aspects such as power operation,condition monitoring of equipment,collection of customer power information,and marketing systems.Combined with this resource,this paper applies the big data theory to build a model for predicting the heavy and overload operation state of the distribution network transformer.Firstly,transformer ledger data,user file data,network topological relationship,transformer transformation records,meteorological information and holiday label data needed in the prediction modeling process are extracted from multiple data sources.A variety of visualization models,such as bubble diagram model,pie diagram model,radar diagram model,strip-line diagram model and highlight table model,are established to analyze the relationship between heavy and overload of distribution transformer and power consumption region,power consumption type,transformer capacity and season and month.Data transformation is applied to transform the data form of some field attributes into a more suitable and convenient form for modeling and analysis,and the field attribute value of "load rate" is constructed.The data size can be appropriately compressed by deleting irrelevant attribute fields and sampling.The box plot is used to identify the outlier data and delete it.Based on MATLAB software,the missing values of multiple field attributes are filtered out.These values areprocessed by different methods.Then,the k-means clustering algorithm model was established with SPSS Modeler software,and three different "load rate" values were selected as the initial clustering center.After repeated training of the model,the clustering center was adjusted iteratively to achieve the optimal value of contour coefficient.The distribution transformer’s historical heavy overload data were effectively divided into three categories.The Apriori association rule model was established,appropriate parameters of support degree and confidence degree were set,and the main influencing factors for heavy overload of distribution transformer contained in each clustering result were mined.The "degree of improvement" evaluation index was applied to verify the effectiveness of association rules.Finally,80% of the sample data is divided into training samples,and the other 20% is used as the test sample.Radial basis neural network model and multi-layer perceptron neural network model are established respectively.The radial basis neural network model and the multi-layer perceptron neural network model were established respectively.According to the clustering results and correlation analysis results,different initial weights were set for each input field attribute and the number of neurons and hidden layers of the multi-layer perceptron neural network were constantly adjusted to improve the accuracy of prediction as much as possible.According to the running results the multilayer perceptron neural network model is superior to the radial basis neural network model and the multilayer perceptron neural network is trained again to ensure its stability.The data of 535 distribution transformers in Hanzhong area from from November 1st,2018 to February 28 th,2019 were extracted,and the new data was entered into the multi-layer perceptron neural network prediction model.The prediction results were consistent with the actual occurrence of heavy overload.The correct rate reaches 95%.The stability and accuracy of the model are proved by the effective indicators such as matrix,performance evaluation,AUC and Gini coefficient.The results show that the heavy overload prediction model of the distribution transformer can realize the heavy overload prediction requirements of distribution transformers in the Hanzhong area in the coming week.The research work basically completes the content requirements of thedistribution transformer heavy overload prediction,which can provide reference opinions for the enterprise staff to carry out the heavy overload forecast management of distribution transformer to a certain extent.