An improved state-space model structure and a corresponding predictive functional control design with improved control performance

Abstract

Conventional state-space model predictive control requires a state estimator/observer to access the state information for feedback controller design. Its drawbacks are the numerical convergence stability of the observer and closed-loop control performance deterioration with activated plant input/output constraints. The recent direct use of measured input and output variables to formulate a non-minimal state-space (NMSS) model overcomes these problems, but the subsequent controller is too sensitive to model mismatch. In this article, an improved structure of NMSS model that incorporates the output-tracking error is first formulated and then a subsequent predictive functional control design is proposed. The proposed controller is tested on both model match and model mismatch cases for comparison with previous controllers. Results show that control performance is improved. In addition, a linear programming method for constraints dealing and a closed form of transfer function representation of the control system are provided for further insight into the proposed method. IntroductionModel predictive control (MPC) has been a very active research area in both control theory and control engineering since it first appeared in the 1970s. The main focus is on the three most general MPC design methods: finite impulse response (FIR) or step response models, transfer function models and state-space models. The earlier FIR model-based MPC represented by dynamic matrix control (DMC) (Cutler and Ramaker 1980) and quadratic DMC (Garcia and Morshedi 1986) can only deal with stable processes and the design model order is usually high. The representative article of transfer function model-based MPC can be seen in generalised predictive control (GPC) by Clarke, Mohtadi, and Tuffs (1987), which can deal with both stable and unstable processes. The main disadvantage is that it is considered to be less effective for multi-variable processes and real-time application. The third category is MPC based on state-space models, and it has attracted much interest from researchers since it can effectively use the states' information for state feedback, constraints consideration and even for providing important information for the controlled process. A number of state-space model-based predictive control methods have been developed

Design of state space linear quadratic tracking control using GA optimization for batch processes with partial actuator failure

“…Example 1. This section gives a non-minimum phase systems with time delays and the state space model is given by [24]: Follow the proposed design, the state variable is chosen as…”

Design and implementation of an improved linear quadratic regulation control for oxygen content in a coke furnace

“…However, the state observer based SSMPC also suffers some drawbacks such as numerical difficulty, which leads to the research on non-minimal SSMPC that does not require the state observers because the NMSS model directly incorporates the measured process inputs, outputs and their past values into a state variable. Till now, both theoretical and application results have been witnessed [29][30][31][32][33]. MPC on industrial coke unit was also witnessed in recent years.…”

Improved infinite horizon LQ tracking control for injection molding process against partial actuator failures

“…Recently, model predictive control (MPC) has also been proposed to improve control performance [22][23][24]. However, there is still space for further design methods to improve control performance under model/plant mismatch and partial actuator failure [25].…”