CDI-MPC framework detects grid contingencies to slash prediction mismatch

A team led by Erfan Mehdipour Abadi and Hamid Varmazyari (with collaborators from Wayne State and the University of Jordan) has released a preprint on arXiv (2608.04370) that tackles a critical weakness in model predictive control (MPC) for power grids. Traditional MPC-based load frequency control (LFC) assumes a known system model, but sudden contingencies — like generator trips or line faults — change the grid's dynamics, causing prediction mismatches that can destabilize frequency regulation. The proposed CDI-MPC framework integrates contingency detection directly into the MPC loop, treating contingencies as stochastic discrete events within a stochastic hybrid system (SHS). A disturbance-aware residual formulation jointly identifies which operating mode the grid is in and estimates unknown disturbances, then updates the MPC's prediction model in real time — all within the fast LFC time scale.

The paper reports simulations demonstrating accurate contingency detection and substantial improvements in closed-loop LFC performance across multiple contingency scenarios and unknown disturbances, with 6 pages and 4 figures. Unlike approaches that rely on protection systems to provide post-contingency models (which often aren't available quickly enough), CDI-MPC works with the observed residuals alone. This makes the method practical for existing grid infrastructure and opens the door to more resilient, adaptive frequency control that can maintain stability during unexpected events. The framework could be a key building block for future grids with high renewable penetration, where contingencies are more frequent and less predictable.