Answers are grounded in this paper and cite their sources.
Review analyzes causal inference methods in power systems, highlighting advances and implications for forecasting.
Given the increasingly rapid development and transformation of power and energy systems, forecasting and modeling these systems have become both more important and more challenging. Historically, and continuing to the present, these models have predominantly relied on statistical and traditional machine learning approaches, including the recent emergence of deep neural networks, but they are not necessarily causal. In response to the need for generalization and causal learning beyond correlation, this paper analyzes the two dominant categories of causal inference methods—causal discovery and causal effect estimation—and provides a comprehensive review of their advancement and application in power and energy systems. The study examines causal methods for improving time series forecasting and their progress in counterfactual prediction, which has become increasingly important due to the need for forecasting under unseen scenarios in such rapidly evolving systems. Furthermore, we explore advances in understanding systems, including examining the intersection between policy and market forces on both systems and human behavior. Fault detection and diagnosis are also discussed. Finally, this review investigates evaluation methods for causal models, identifies their shortcomings along with other key issues, and summarizes the future roadmap for incorporating causal learning in power and energy systems. This review offers valuable insights for policymakers, grid operators, and stakeholders to facilitate a smooth transition toward a data-driven and causal-informed AI-powered power and energy system.
No takes yet. Share an insight, caveat, or question.
Jiang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: