Clustering Technology And Evaluation Of Line Loss In Transformer Districts With Or Without DG Based On Machine Learning

With the continuous development of power grids,the country has begun to attach importance to energy-saving and emission-reducing strategies.The management of line loss is a significant technical-economic index that reflects power companies.The line loss in the transformer district accounts for a great proportion of the power loss and has a great potential for loss reduction.However,because the transformer district has the characteristics of complex structure,many branch lines,and difficulty in data collection,the accuracy of predicting the line loss rate using traditional calculation methods is not well.Therefore,it is very necessary to seek an accurate and effective line loss rate evaluation method.The development of artificial intelligence and machine learning provides a new method for the evaluation of the line loss rate of the transformer district.This thesis proposes a line loss evaluation method of the transformer district based on K-Means++clustering algorithm and Stacking ensemble learning.Firstly,in order to improve the quality of the transformer district data,it is necessary to preprocess the data.Bad data is handled by methods such as mutual information.Establish the best electrical characteristic index system based on grey correlation analysis.Secondly,considering the large differences in the scale of power grids and power consumption structure of the transformer district,a K-Means++clustering model based on t-SNE dimension reduction was established to classify the transformer district.The line loss rate is evaluated in the same class of the transformer district.It effectively solves the problem of unsatisfactory evaluation effect caused by the large difference in the line loss rate of the transformer district.Then,the model based on multi-algorithm combination of Stacking ensemble learning was built.Considering the difference of algorithms,linear regression,random forest and GBDT were involved in Stacking base-learner layer.The model fits the complex relationship between the line loss rate of the transformer district and the electrical characteristic parameters.Finally,selecting 2070 transformer districts and 9073 transformer districts with DG as examples,accuracy and effectiveness of the proposed method is confirmed based on the data of transformer districts.The results demonstrate that the proposed method has better evaluation accuracy and higher generalization compared with the methods of linear regression,random forest and GDBT in existing literatures.The evaluation method proposed in this thesis can quickly and effectively calculate the line loss rate of each class of transformer districts,which has practical significance.