[4Rin1] Interactive 2

[4Rin1-06] Application of machine learning in the power demand forecasting work

〇Shinta Watanabe¹, Kentaro Koyama¹, Kazuhiro Aoyagi¹, Ryuzo Tachikawa¹, Yukiko Ono¹ (1.East Japan Railway Company)

Keywords:regression analysis, error detection, re-training

JR East supplies electric power for trains from its own power plant. In order to determine the amount of power generation, the power operators manually predicted power demand from past performance. The work took about an hour, and the prediction accuracy varied. The purpose of this effort is to establish a method that allows anyone to predict power demand with high accuracy by machine learning. In this effort, we propose a prediction model based on multiple regression analysis. Abnormal data is excluded from the learning data by using an anomaly detection method by machine learning. The forecast model is divided into several according to the trend of power demand. Parameters affecting the prediction are adopted in the prediction model. With this method, it became possible to instantly make predictions with the same accuracy as a veteran operator.