Neural network based model predictive control performance monitoring-data-driven approach

A data-driven neural network based approach for model predictive control performance diagnosis was proposed. Considering four common MPC degradation factors, namely noise variance change, model mismatch, control variables constraint saturation, and manipulated variables constraint saturation, MPC performance patterns were divided into four categories. Performance signatures are extracted from the process input and output variables directly, and classifier is constructed via neural network. The effectiveness of the proposed method was demonstrated on NIAT platform by a two tank liquid level process.