Hybrid Framework with Machine Learning and Mixed Integer Linear Programming for Resilience in Electric Power Distribution Systems in the Face of Extreme Weather Events

The resilience of electric power distribution systems in the face of extreme weather events requires rapid, optimized decision-making. This paper proposes a hybrid methodology for resilience analysis in electric power distribution networks, integrating machine learning (ML) and Mixed-Integer Linear Programming (MILP). An XGBoost classifier estimates bus failure probabilities based on topological and operational characteristics. These parameters feed into an MILP model that determines the optimal network reconfiguration and load-shedding plan under contingencies, while adhering to physical and topological constraints. Validated on the IEEE 69-bus system with a simulated fault on line 61, the approach achieved a ROC AUC of 0.61. The MILP optimizer converged to the optimal solution in 0.84 seconds, maintaining full service to 59 buses (with 64 connected to the substation) and reducing unsupplied energy to 0.0465 MW (46.5 kW), representing approximately 1.2% of the system's total demand.