Interpretable Deep Learning for Strategic Behavior Analysis in Joint Electricity and Carbon Markets

The integration of carbon markets with the electricity sector represents a crucial development in aligning economic activities with environmental goals, particularly in China. Generation companies (GenCos) navigate these coupled markets through strategic bidding behaviors influenced by various carbon emission allowance (CEA) allocation methods. This study explores how different market conditions and regulatory policies shape GenCo bidding strategies, employing deep learning models like GAN to model these behaviors. To enhance interpretability, we utilize Local Interpretable Model-agnostic Explanations (LIME) and introduce a Bayesian extension (BayLIME), which leverages empirical data and prior knowledge for a more robust analysis. The results demonstrate significant correlations between operational features and strategic dimensions, providing insights that could aid in optimizing market performance and ensuring compliance. The application of BayLIME particularly enhances the reliability and consistency of model explanations, offering a clearer understanding of market dynamics for stakeholders. The findings contribute to both strategic planning and policy-making efforts aimed at fostering a sustainable energy market.