Model-plant mismatch detection and model update for a run-of-mine ore milling circuit under model predictive control

The performance of a model predictive controller depends on the quality of the plant model that is available. Often parameters in a run-of-mine (ROM) ore milling circuit are uncertain and inaccurate parameter estimation leads to a mismatch between the model and the actual plant. Although model-plant mismatch is inevitable, timely detection of significant mismatch is desirable. Once significant mismatch is detected the model may be partially re-identified in order to prevent deteriorated control performance. This paper presents a simulation study of the detection of mismatch in the parameters of a ROM ore milling circuit model using a partial correlation analysis approach. The location of the mismatch in the MIMO model matrix is correctly detected, and the process model subsequently updated.

► We present a simulation study of the detection of mismatch in the parameters of a ROM ore milling circuit model using a partial correlation analysis approach.
► The location of the mismatch in the MIMO model matrix is correctly detected.
► The process model is updated using a constrained minimization algorithm for partial model re-identification.
► The controller is then redefined using the updated model.