4. Advanced Control Performance Assessment

This chapter deals with extensions of the MV benchmark that need substantially more information about the plant than just the time delay. An extension of the MV benchmark is the approach of generalised MV (GMV) benchmarking, minimising a weighted sum of the control error and control effort. More general but rigorous extensions are the linear-quadratic Gaussian (LQG) benchmark and the model-predictive control (MPC) assessment. These benchmarks are useful when more information on controller performance, such as how much can the output variance be reduced without significantly affecting the controller output variance is needed, or for cases where actuator wear is a concern. This chapter provides an overview of these advanced methods. It is particularly shown how to use routine operating data to distinguish between poor performance due to plant–model mismatch and that due to improper tuning of the MPC controller. Moreover, performance measures that estimate potential benefit from re-identification of the process model or re-tuning of the controller are introduced. This is essential in MPC monitoring, as a process model is a substantial component of the MPC controller.