Explainable Singular Spectrum Analysis deep learning model for half-hourly electricity price prediction

Electricity price (EP) prediction is crucial for efficient operations and cost management in the power industry. However, the inherent complexity and nonlinearity of EP series pose challenges for accurate energy management. This study introduces an innovative prediction system for half-hourly EP, integrating Singular Spectrum Analysis (SSA) with Neural Basis Expansion Analysis for Time Series (NBEATS). Bayesian optimization is employed to optimize NBEATS’ hyperparameters. SSA decomposes the EP series into sub-series, enabling NBEATS to predict each sub-series individually. Real-world data from New South Wales (NSW) and Queensland (QLD), Australia, spanning from January 2016 to October 2022, validates the model’s effectiveness. Evaluation using deterministic metrics (Root Mean Square Error, Mean Absolute Error, Symmetric Mean Absolute Percentage Error) demonstrates that SSA-NBEATS outperforms other decomposition-based models in prediction accuracy, precision, and stability. Furthermore, the Global Performance Indicator (GPI), which integrates these metrics, positions the proposed SSA-NBEATS model at the forefront with a GPI of ≈ 2.787 (QLD) and ≈ 2.117 (NSW), surpassing benchmark models. Statistical tests including Diebold-Mariano and Giacomini-White confirm the superior accuracy of EP predictions by SSA-NBEATS over benchmark models. Additionally, eXplainable Artificial Intelligence (xAI) techniques—SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME)—are employed to interpret NBEATS model predictions effectively.