Multiple-model predictive control framework for multi-input multi-output continuous processes
MetadataShow full item record
This work proposes and develops an approach to transition control of chemical plants based on the development of a state-shared model in a model-predictive control (MPC) framework. Transition control over a large operating space presents a challenging problem, especially for nonlinear multiple-input multiple-output (MIMO) constrained systems. Attractive features of a transition control structure should include rapid and stable closed-loop response. A large number of recent approaches address this issue using multiple fixed and adaptive models, and single or multiple controllers. Regardless, the controller must not only successfully regulate the plant at the initial and final operating points, but also track the reference set point during the transition. In fact, the problem can be considered in two parts - the identification of a model that estimates the plant outputs during the transition and the synthesis of a suitable controller that produces smooth and realistic control action. Satisfactory closed-loop and stable performance of the controller and nonlinear plant is inferred, if the performance of the model in closed-loop with the controller can be guaranteed. By a state-shared model, we mean a linear time-invariant model structure that is a realization of the system and driven by the measured signals - the plant outputs and the manipulated variables. The coefficient matrices in the state-shared model are selected to be a controllable pair by the designer; however, the equation that represents the measured outputs of each model is unique. The description of the measurements is embedded in the coefficient matrix. Any such model necessarily fulfills the requirement that the output of the state-shared model must be an asymptotically-correct estimate of the plant's output, if the process models were selected appropriately. The parameters of the adaptive state-shared models are modified using a stable and convergent adaptive law. By means of adapting the parameters of the equation that represents the measured outputs and switching among fixed and adaptive models, accurate estimates of the plant can be obtained. The use of the state-shared model necessarily relaxes the assumption that the fixed models cover the large operating space. The theoretical underpinnings that permit the development of the state-shared model are stated and proven. Using the state-shared model, the MPC optimal controller or an Hoo robust controller can be designed. Conditions for both controllers to produce stable closed-loop responses for certain classes of systems can be used to establish closed-loop stability in the case of transition control. Critical to the tracking problem is the identification of the transition reference trajectories. Generally, the reference trajectory is determined by experience. Transition control using a state-shared model in either an HQO or MPC framework is demonstrated on several single-input single-output and multiple-input multiple-output continuous, nonlinear processes. Finally, the state-shared model based controller design approach is demonstrated on a plant-wide scale using the widely known Tennessee Eastman (TE) plant.