Dispatch Decision in New Power Systems: From Mathematical Programming to Artificial Intelligence

The talk by Prof. Li Zuyi (Zhejiang University) introduces the topic "Dispatch Decision in New Power Systems: From Mathematical Programming to Artificial Intelligence," outlining the transition from traditional power systems to new, renewable‑rich systems and the resulting changes in dispatch strategies.

Traditional dispatch relies on mathematical programming, primarily the Security‑Constrained Unit Commitment (SCUC) and Security‑Constrained Economic Dispatch (SCED) problems, modeled as mixed‑integer or linear programs with constraints on generation, network flows, and operational limits; solution methods include Lagrangian relaxation, mixed‑integer programming, Benders decomposition, and heuristic approaches.

High‑penetration renewable energy introduces significant uncertainty and variability, prompting the use of stochastic optimization (scenario sampling) and robust optimization (worst‑case planning). These methods face challenges such as conservatism, computational scalability, and difficulty handling integer variables.

Artificial‑intelligence‑based solutions are explored, notably the CPA framework (Check‑Predict‑Accommodate), which first uses a deep neural network to assess feasibility, then predicts a near‑optimal solution, and finally adapts the solution via an optimization layer to satisfy all physical constraints, achieving feasibility, efficiency, and near‑optimality.

The presentation concludes with a summary of the limitations of pure mathematical or AI methods and outlines future research directions: embedding constraints directly into neural networks, hybrid AI‑optimization approaches, handling integer variables in learning‑based solvers, and leveraging large‑model techniques for scenario generation and extreme‑event detection.