A DUAL-SEQUENCE HYBRID STACKING MODEL FOR ENHANCED SHORT-TERM LOAD FORECASTING

Abstract

 Accurate short-term electric load forecasting is essential for effective power system planning and operational reliability, particularly in the face of demand variability introduced by intermittent energy sources. This paper introduces a novel Dual-Sequence hybrid model for short-term load forecasting (STLF) that integrates LSTM/BiLSTM networks with Extreme Gradient Boosting (XGBoost) using stacking technique. Initially, LSTM, BiLSTM, and XGBoost models are developed independently to forecast electricity load. Subsequently, a combined structure is constructed by embedding LSTM and BiLSTM generated features into the XGBoost model, allowing the integration of temporal learning capabilities with robust feature selection and optimization. While LSTM networks are proficient in modeling nonlinear temporal dependencies, their performance can be hindered by increased computational complexity as model parameters scale. To address this, XGBoost is employed to reduce input dimensionality and mitigate overfitting risks. The proposed hybrid model was validated using real-world datasets from Malaysia. Experimental results demonstrate that the combined model significantly outperforms individual models, as well as single-hybrid models, reducing MAPE by 5.7% and 4.3%, respectively, achieving a mean absolute percentage error (MAPE) of 1.15%. This notable reduction in forecasting error highlights the potential of the Dual-Sequence hybrid approach for advancing short-term load forecasting applications.