Physics-Informed Learning for Economic Dispatch in Sustainable Power Systems

The variable and intermittent nature of renewable energy sources and the uncertain behavior of electrical loads pose a critical challenge for real-time economic dispatch (ED) in modern power grids. ED aims to optimize the dispatch of generation units to minimize production costs while satisfying operational constraints. However, solving this problem in real time is computationally demanding, especially in renewable-dominated power systems. This paper examines two approaches to address the ED problem: conventional solvers and artificial intelligence-based methods. Conventional solvers provide accurate and reliable solutions but are often too slow for real-time applications. Traditional learning techniques improve computational efficiency but may compromise accuracy and fail to obey physical system constraints. A physics-informed neural network (PINN) can address these shortcomings. The proposed PINN method in this paper has achieved a 3.88x speedup in computation compared to traditional solvers, reducing execution time from 1.215 seconds to 0.313 seconds on average. This AI-based model also produced results with minimal constraint violations, with 3.36 percent overall violations across the test data.