Probabilistic electricity price forecasting by integrating interpretable model

Abstract

The establishment of a high-quality and efficient interpretable probability prediction model is crucial for the development of the electricity market. However, challenges related to prediction instability and interpretability limit electricity price probability forecasting. To address these issues, we propose a novel interpretable electricity price probability prediction model, L-NBeatsX, which incorporates a multifactor pathway. Initially, by adaptively fusing NBeatsX and LassoNet models, we effectively handle the multifactor nature of electricity price prediction. The fusion mechanism enables L-NBeatsX to utilize a subset of features, thereby enhancing both accuracy and interpretability. Furthermore, the integration of skip connections from input to output in the fusion process enhances the robustness and flexibility of L-NBeatsX predictions. Additionally, we introduce unstable correction factors into the loss function to improve the model’s adaptability in probability prediction. By mitigating the impact of instability factors, we effectively reduce the cost of prediction instability while improving the accuracy and reliability of results. Empirical studies conducted across four distinct electricity markets demonstrate the superior performance of L-NBeatsX in electricity price probability forecasting, providing valuable insights for decision-making in the electricity market.