Strategic enhancements in electricity price forecasting: The role of XGBoost and error correction features
Lilia Tightiz, Joon Yoo, Wedad Khamis Al-Shibli
Abstract
Since integrating renewable energy sources now drives electricity markets, they experience very high price volatility, making accurate forecasting essential. This paper presents an enhanced forecasting model to manage such volatility using an optimized XGBoost algorithm with integrated error correction features. The proposed model outperforms traditional forecasting methods, achieving lower symmetric mean absolute percentage error (sMAPE), which lies in the range from 1.77% to 11.41% throughout the year, compared with standard XGBoost and autoregressive integrated moving average (ARIMA)+ long short-term memory (LSTM). Besides, residual analysis using the autocorrelation function (ACF) and partial autocorrelation function (PACF) confirms that the enhanced XGBoost model effectively minimizes residual dependencies, outperforming ARIMA+LSTM and the standard XGBoost model. The proposed model increases computational efficiency and accuracy due to Bayesian optimization, while Explainable AI (XAI) techniques-SHAP and LIME-enable global and local interpretability of feature importance to be transparent with respect to the decisions made by the model. This model includes essential features such as temperature, dew point, load demand, fuel prices, seasonal trends, and holiday effects, all driving significant changes in electricity prices. Results confirm that integration of error correction and XAI improves forecasting performance while providing a robust and interpretable solution for volatile electricity markets.