Near real time load forecasting in power system

Abstract

Near real time load forecasting is the prediction of power system load for the duration of next few hours. It is becoming an increasingly important topic because of the process of deregulation and introduction of competition in the wholesale electricity markets. Further spinning reserves, unit commitment, contingency analysis and economic load dispatch functions of the EMS rely heavily on near real-time load forecasts. This paper uses SVR machine as a base and presents other nascent methods. Different models are made for holidays and weekends which are based on SVR method. A self - learning weekly window is applied which automatically trains the models for three weeks and predicts the load for next week. To evaluate the performance of the developed methods, the models has been tested and trained for data provided by the Rajasthan SLDC using evaluation metrics such as MAPE and RMSE. It has been found that the models by proposed methods significantly outperforms the traditional SVR model by a lesser MAPE. © 2020 IEEE