Short-and mid-term load forecasting using machine learning models

Abstract

With the ever-increasing load demand for diversified users, load forecasting emerges as an integral part in the energy management system (EMS). Improving the load prediction accuracy is of great significance to the construction and development of smart grid. This paper focuses on forecasting short and medium terms of electrical load using three machine learning models as follows: Linear Regression (LR), Support Vector Regression (SVR), Gradient Boosting Regression Trees (GBRT). The input features contain the correlation between the weather information and the electrical load data. The proposed models are tested with the data acquired from New York Independent System Operator (NYISO) data set. The simulation results show that although all models achieve satisfactory performance on prediction accuracy. Gradient Boosting Regression Trees model yields the most promising results on both short-and mid-term load forecasting with higher accuracy. A hybrid method of AdaBoost ensemble algorithm based on GBRT is proposed in this paper, which shows an improvement in load forecasting accuracy compared with the above three methods.