Variational Representation for Risk Assessment in Aggregated Supply Curve Forecasting: A Case Study in Singapore's Electricity Market

This study underscores the potential of variational representation as a tool for uncertainty estimation in the energy market, which would help Independent System Operators (ISO) in risk assessment and demand response. Forecasts of future supply curves are also very valuable to energy suppliers seeking to optimize bidding strategies in an oligopolistic power market characterized by information privacy and volatility of individual bidding behaviors and variability in renewable energy generation. However, traditional deterministic modelling methods may not adequately capture the inherent uncertainty and volatility of the electricity market. Our variational encoding technique produces a resultant embedding space that is stochastic and exhibits substantial interpretability, which we exploit to demonstrate the model's transparency. By using the learned posterior probability distribution of the embedded aggregated supply curves (ASCs) to determine the statistical dispersion, we identify and study periods of high uncertainty in the Singapore electricity market between December 2021 and March 2023. We seek to provide insights into the origins of these uncertainties, showing that the method can capture changes in the market useful for risk modelling in decision-making.