AI-Optimized Decision-Making in Energy Systems: Toward a Decision-Aware Machine Learning Framework

Machine learning (ML) and mathematical programming (MP) models are essential for predicting uncertain system parameters and optimizing decision-making in energy systems. However, incorporating the impact of prediction errors into ML models to inform MP models remains a significant challenge. To address this issue, we propose an artificial intelligence (AI)-native optimization (AIOpti) system that enables AI to be aware of the impact of predictions on decision-making through a deep fusion of ML and MP models. This AI-native deep fusion distinguishes AIOpti from existing research, as the latter separates accuracy-oriented learning from problem-oriented learning. When applied to scenarios such as virtual power plant operations and distributed energy management based on real-world data, AIOpti achieves lower computational costs, improved convergence, and robust performance in complex systems, enabling ML models to effectively account for the impact of prediction errors on MP models. Furthermore, our enhanced AIOpti system not only maximizes predictive accuracy but also improves decision-making quality.