Uncertainty Quantification with Conformal Prediction in Energy Data

Overview

The increasing availability of high-resolution energy data, combined with the growing penetration of distributed energy resources, has intensified the need for reliable uncertainty quantification to support operational and planning decisions in power systems. This seminar presents conformal prediction (CP) as a flexible, model-agnostic uncertainty quantification framework for generating statistically valid uncertainty estimates in energy applications. The talk will introduce the fundamentals of CP and demonstrate how it can be layered on top of machine learning models to produce reliable prediction intervals. Two representative applications are then explored. First, CP is applied to photovoltaic (PV) power forecasting to support decision-making of market participants in the day-ahead electricity market. By combining CP-based uncertainty estimates with different bidding strategies, the framework illustrates how market participants can improve profitability while limiting energy imbalance. Second, the seminar presents the use of CP in the probabilistic disaggregation of PV generation from net smart meter data, a task of growing importance for distribution system operators. Different CP variants are discussed to enhance interval reliability. Case studies using real-world datasets demonstrate that CP-based methods can outperform classical probabilistic approaches, such as quantile regression, in both applications. The seminar concludes with a discussion on the broader potential of CP for uncertainty-aware decision-making across energy systems.