Disturbance Observation Method Based On Kalman Filter Applied In Model Predictive Control

Since been developed in the70s of last century, model predictive control (MPC)with its lower request to the mathematical model, excellent control performance andconvenience online calculation, has become a very important sub-discipline with richtheory foundation and practical application and has attracted much attention from theindustrial control field. However, the issue of the unmeasurable disturbance, which iswidespread in industrial processes, is difficult to be dealt with and requires further study.In most implementations of MPC technology, this problem is solved by incorporating aconstant output disturbance into the process model, but this method can not achieveoffset-free control if unmeasured disturbances enter the process elsewhere. Based on theKalman filter theory, by using a more general disturbance model which is superior tothe constant output disturbance model, the problem of unmeasurable disturbance can besolved, and the necessary conditions for offset-free model predictive control are alsoaddressed.Dynamic control algorithm (DMC) is the most extensively applied algorithms ofmodel predictive control. Due to the deviation between the predicted values and actualmeasured values is assumed to be invariant, DMC can not effectively estimatedisturbance in advance and lead to the limited capacity of DMC in dealing withunmeasurable disturbance. In order to improve the control performance of DMC, twonew algorithms with feed-forward compensation are obtained based on Kalman filter,thereby the capacity of suppress unmeasurable disturbance is significantly improved.By utilizing disturbance model, the unmeasurable disturbance vectors areaugmented as the states of control system, and Kalman filter is used to estimateunmeasurable disturbance and its effect on the system output. With the disturbanceestimated, unmeasured disturbance can be suppressed of feed-forward compensationcontrol strategy. Since the linearization theory is introduced to design Kalman filter andpredictive controller, the improved algorithms require less on-line calculation work,satisfy the real-time control demand and eliminate affect of unmeasurable disturbance tothe output of the system. Simulation results show that this method can significantlyimprove the control performance of DMC and perfectly achieve offset-free controlunder unmeasurable disturbance.