A Data-driven Performance Assessment Approach for MPC Using Improved Distance Similarity Factor

To keep the whole control system running well, a controller in Model Predictive Control (MPC) system plays an important role. Data-driven performance assessment approach can detect the poor performance of the controller in time and avoid the crash of the whole system. This paper proposes a method based on improved distance similarity factor in order to improve the accuracy of performance assessment. In this factor, Bhattacharyya distance is used for detecting the similarity of the real-time I/O data and historical I/O data. It considers both the mean absolute difference and the variance so as to enlarge the fluctuation change of the system I/O data and to improve the accuracy of performance assessment. A simulation on Wood-Berry distillation model is made to verify the effectiveness of this method.