Short-Term Load Forecasting Using GRU-LGBM Fusion

Abstract

The paper proposes an ensemble model of Light Gradient Boosting Machines (LGBM) and Gated Recurrent Units (GRU) for Short-Term Load Forecasting (STLF). The utilization of GRU's sequential learning combined with the ensemble power of LGBM is seen to outperform the traditional load forecasting approaches, which are not capable of capturing complicated load patterns. For predicting the load of the chosen geographic area, the time series of the load data is utilized along with the hourly weather data. The proposed ensemble model is applied to the hourly load data of the Western region from January 1, 2012, to December 31, 2015, by the Electric Reliability Council of Texas (ERCOT). The accuracy of the forecasting algorithm is seen to increase significantly after the addition of a holiday flag, day of the week, and time of the day index as categorical feature vectors.