Analytic Input Convex Neural Networks-based Model Predictive Control for Power System Transient Stability Enhancement

Transient stability-constrained optimal power flow (TSC-OPF) is a non-convex optimization problem and also computationally intensive to solve. To find the global optimum and improve the convergence, this paper proposes an analytic input convex neural networks (ICNN)-based model predictive control (MPC) for transient stability enhancement. ICNN is developed to predict power system transient stability that has the particular property of ensuring its output is convex to the input. This capability allows ICNN to optimize the convex inputs, such as the active power output of generators, while keeping other values constant, such as variables that do not participate in optimization. In addition, non-convex static constraints are built by decoupled linearized power flow model to form a fully convex TSC-OPF with a global optimum. The proposed method is embedded into MPC to enhance transient stability. Numerical results show that the proposed method can greatly improve convergence, find the global optimal solution, and speed up the calculation.