Line Loss Prediction In Low-voltage Districts Based On Gradient Boosting Decision Tree

Loss reduction and energy saving is one of the important strategies for our national energy saving and emission reduction.Line loss rate,as a comprehensive indicator of loss reduction and energy conservation,not only reflects the efficiency of transmission and distribution,but also reflects the planning,operation and management level of power grid companies.In view of the fact that line loss is related to the economic benefits of power grid companies,it is also one of the key indicators for grid company assessment.The low-voltage districts is one of the key points of the "four-separate" management of line loss,its line loss electricity accounts for 20% of the total,which means the heavy loss situation is more common and has a greater potential for loss reduction,hence the low-voltage districts should receive the attention of relevant departments.At present,the grid structure of the low-voltage districts is relatively complex,with its large number,weak management and poor data quality.Therefore,the quality and level of lean management of line loss in the low-voltage districts cannot be guaranteed.At present,for the calculation of line loss rate in low-voltage districts,Guangxi Power Grid adopts the method of total districts and typical districts based on measured line loss rate.Although it can achieve the purpose of quickly calculating line loss rate,its assumptions cannot be matched with the actual situation.In addition,the calculation result is still in the category of statistical line loss,hence the calculation accuracy is not high,which is not convenient for the analysis of line loss in low-voltage districts.Therefore,fast and accurate line loss calculation has become an urgent need.To this end,this paper proposes a line loss prediction method for low-voltage districts based on gradient boosting decision tree.Firstly,anomaly detection and standardization of the original data in low-voltage districts are performed to improve data availability and reliability;secondly,the Spearman correlation analysis and the relative importance of the gradient boosting decision tree are used to screen the electrical feature indicators to form a feature index system for low-voltage districts;Considering the characteristics of large difference in grid structure and scattered value of line loss rate,density-based clustering algorithm i.e.DBSCAN is used to classify the low-voltage districts.Finally,train and build a gradient boosting decision tree model for each type of low-voltage districts to mine the relationship between electrical parameters and line loss rate to realize line loss prediction for low-voltage districts.The analysis of 3518 sample data in the actual low-voltage districts and the comparison with the prediction effect of random forest and multiple linear regression models verify that the method proposed in this paper can effectively improve the prediction accuracy of the line loss rate,which is scientific,reasonable and effective.Because the data required in this paper is easy to obtain and the modeling is relatively simple,the method proposed in this paper not only can realize the fast and high-accuracy prediction of the line loss rate for the low-voltage districts,but also can provide a practical reference for the grid company to formulate the loss reduction plan.