Analysis of Feature Engineering in Deep Learning Architectures for Dynamic Electricity Price Forecasting

Forecasting is critical for the electricity market and is particularly challenging with the introduction of Intermittent Energy Sources (IES), such as solar or wind. Due to the inherent volatility of IES, electricity price forecasting (EPF) models often fail to deliver the expected performance. Notably, Deep Neural Networks (DNN), assisted with feature engineering, demonstrated good forecasting performance in deregulated electricity markets for the EPF application domain, even in highly volatile scenarios. In this work, we demonstrate the impact of diverse data transformation approaches for a DNN-based EPF, which is crucial for explainable artificial intelligence (xAI). Results indicate that feature engineering through data transformations and normalization enhances the performance and explainability of EPF. Performance evaluation via error metrics shows the efficacy of the proposed approach for Nordpool.