A Review on Machine Learning Applications in Chance-Constrained Power System Optimization

The growing integration of renewable energy sources increases uncertainty in power systems, exposing the limits of deterministic and chance-constrained optimization. Although chance constraints balance risk and efficiency, their adoption is restricted by computational complexity, conservatism, and distributional assumptions. Machine learning offers promising solutions for addressing chance-constrained programming challenges. This paper reviews machine learning-enhanced approaches in power systems, classifying methods into uncertainty modeling, constraint reduction, surrogate modeling, and reformulation strategies. Key challenges of generalization, data quality, and interpretability are discussed, along with opportunities such as reinforcement learning, physics-informed learning, federated learning, and digital twin integration.