Research on short-term line loss rate prediction method of distribution network based on RF-CNN-LSTM

Lin Jiang¹Chen Li¹Wei Qiu¹Caili Xiang^2*Jiawei Yang³Jun Shu³

• ¹Zhuhai Power Supply Bureau of Guangdong Power Grid Co., Ltd., Zhuhai, Guangdong Province, China
• ²School of Automation, Wuhan University of Technology, Wuhan, China
• ³Oriental Electric Group Science and Technology Research Institute Co., Ltd., Sichuan, China

Due to the complex relationship between line loss rate and various characteristic parameters, there is no mature method to achieve accurate prediction of short-term line loss rate, mainly relying on manual experience. In addition, under the background of the new distribution network, the power fluctuation on the line is increasing, which will lead to more uncertainties in the predicted line loss rate, thus affecting the economic benefits of the power grid. How to use intelligent algorithms to optimize the existing line loss rate prediction model to improve the prediction accuracy has become a current research hotspot. In this paper, the influence of the uncertain factors faced by the line loss rate during the prediction period on the accuracy of the prediction results is comprehensively considered. In view of the high-dimensional timing characteristics of the line loss rate data, a random forest (RF) algorithm is proposed to analyze the importance of multiple feature variables affecting the line loss rate. Remove the less influential feature variables and select the more important features and the line loss rate data are input into the prediction model together, so as to reduce the data dimension and improve the prediction efficiency and prediction accuracy. After screening important features, this paper constructs a combined model of convolutional neural network and long short-term memory network (CNN-LSTM) to predict line loss rate. In order to verify the accuracy of the prediction results, this paper sets up a support vector machine algorithm for synchronous prediction as a comparative experiment. The results show that the prediction results of the proposed prediction method are more accurate.