A progressive method to solve large-scale AC optimal power flow with discrete variables and control of the feasibility

This paper introduces a study on power system networks which aims to produce a dynamic simulation based security assessment taking into account uncertainties. An extended Optimal Power Flow without any guarantee on feasibility leads to the resolution of a Mixed-Integer NonLinear Problem (MINLP). Solving a (MINLP) is very challenging and even harder when the problem is not convex and could be unfeasible. A custom filtering method which tries to explain infeasibilities and uses the nonlinear solver KNITRO to reformulate discrete variables into nonlinear constraints is proposed. Preliminary results on the French and European power systems network demonstrate the interest of this approach on real-life cases.