Operating Region Partition And Model-Plant Mismatch Correction For MLD-MPC Systems

The multi-model approach has received more and more attention because of its ability to handle those highly nonlinear plants with wide operating ranges and large set-point changes.The multi-model approach has two key steps:the multi-model decomposition and the multi-model combination.In the multi-model combination,MLD model can integrate linear sub-models under a united framework to capture the whole system dynamics,and MLD-based MPC has become an ideal candidate for dealing with nonlinear system control.We focus on two preconditions for MLD-MPC to get great control performance.First,it is essential to complete the multi-model decomposition,that is,to establish the linear model bank and obtain the operating region partition.Second,correction strategy is necessary to deal with the model-plant mismatch.As for the above two aspects,based on the existing researches of multi-model decomposition,a new operating region decomposition method for the nonlinear system with unmeasurable state variables is proposed in this dissertation,and then a new and effective output-based correction strategy is initiated for tackling the model-plant mismatch.The main contributions of this dissertation are as follows:(1)The multi-model set is established for nonlinear systems,and an MLD model is used to integrate linear sub-models.Then a global predictive controller(MLD-MPC controller)is designed.The control effect of MLD-MPC is compared with that of classical soft-switching/hard-switching PID.The simulation results show the effectiveness of MLD-MPC in the control of nonlinear systems.(2)A multi-model operating region partition method based on state estimation is proposed for nonlinear systems with unmeasured state variables.Firstly,the influence of closed-loop feedback on system dynamics is considered,and operating region partition is combined with system control.Then,due to unmeasured state variables,the moving horizon estimation is introduced and integrated into the operating region partition.Finally,the operating region partition is transformed into a bilevel programming problem.(3)A new and effective output-based correction strategy is initiated for MLD-MPC systems with the model-plant mismatch,which is intuitive and easy to carry out online.In the proposed strategy,correction vectors are introduced,and meanwhile the sub-models switching influence is taken into account by introducing logic variables into the correction.Then the correction term containing correction vectors and logic variables is transformed into linear inequalities that can be incorporated in the MLD model.Simulation results show that when the proposed output correction is applied,the static output error caused by mismatch can be eliminated and satisfactory overall control performance can be obtained.