Self-Evolving Offset-Free Model Predictive Control with Model-Plant Mismatch for Dynamic Working-Point Change Tasks in Industrial Processes

This study proposes a self-evolving offset-free model predictive control (MPC) algorithmfor dynamic working-point change (DWPC) tasks in industrial processes.The algorithm mitigates disturbance impacts caused by model-plantmismatch (MPM) and enhances the dynamic performance of MPC by locatingsequences (scenarios) similar to the current operational scenariofrom historical DWPC tasks and using them for multistep-ahead disturbanceprediction. First, a disturbance-augmented state-space modelguarantees the basic offset-free control behavior of MPC with MPM.Next, to enhance the MPC performance, a direct multistep-ahead disturbanceprediction approach is proposed by combining historically similarDWPC task scenarios. Specifically, a dynamic autoencoder is constructedto extract private features from process scenarios and locate similarscenarios from historical DWPC tasks. Based on the located scenarios,the multistep-ahead disturbance and its uncertainty are directly predictedthrough multioutput Gaussian process regression. Finally, the obtaineddisturbance results are incorporated into the MPC framework, whichcontinuously enhances MPC performance in DWPC tasks. Two case studiesdemonstrate the effectiveness of the proposed MPC.

Automated summary

This study presents a self-evolving offset-free model predictive control algorithm for dynamic working-point change tasks in industrial processes. The algorithm reduces the impact of model-plant mismatch and improves the dynamic performance of MPC by using historical scenarios similar to the current operational scenario for disturbance prediction.