An MPC-based nonlinear data-driven model for cascading failure prediction in large-scale infrastructure networks

Abstract

With the continuously increase of network scale and complexity, cascading failures have become the main cause of large-scale infrastructure network paralysis. Network modeling is an important technique for simulating and understanding the cascading failure process. However, linear network modeling methods cannot accurately account for the dynamic characteristics of network systems, while nonlinear network modeling approaches tend to incur high computational costs. To tackle these challenges, we propose a nonlinear data-driven cascading failure prediction model based on Model Predictive Control (MPC) for large-scale infrastructure networks. We consider the influence of noise resulted from dynamic network characteristics on the input dataset and leverage Gaussian Process Regression (GPR) to filter it out. Then, we create a linearized model for the network system using the Koopman operator. We solve the convex quadratic optimization problems by employing the MPC algorithm in the closed-loop verification under a controlled cost model. Finally, we validate the superiority of our proposal by rigorously testing the optimized input datasets across four commonly used cascading failure propagation methods. Experimental results have demonstrated that the proposed approach minimizes 80 % of the input datasets while accurately predicting cascading failures. To our best knowledge, we are the first to apply MPC to a nonlinear data-driven model for cascading failure prediction in large-scale infrastructure networks.