Abstract— Model predictive control (MPC) has become the leading form of advanced multivariable control in the chemical process industry. The objective of this work is to introduce a multiple model adaptive control strategy for multivariable MPC. The method of approach is to design multiple linear MPC controllers. This strategy maintains performance of multiple linear MPC controllers over a wide range of operating levels. One important contribution is that the strategy combines several multiple linear MPC controllers, each with their own linear state space model describing process dynamics at a specific level of operation. One of the linear MPC controller output is selected as multiple model adaptive controller’s output based on the current value of the measured process variable. The tuning parameters for the linear MPC controller are obtained using Genetic Algorithm (GA). The capabilities of the multiple model adaptive strategy for MPC controller are investigated on Two Tank Conical Interacting System (TTCIS) through computer simulation. Keywords : Model predictive controller, Adaptive control, Multiple model, Genetic algorithm I. INTRODUCTION he control of liquid level in tanks and flow between the tanks is a basic problem in process industries. The process industries require the liquids to be pumped, stored in tanks and then pumped to another tank. Many times the liquid will be processed by chemical or mixing treatment in the tanks, but always the level of the fluid in the tanks must be controlled. A level that is too high may upset reaction equilibria, cause damage to equipment, or result in spillage of valuable or hazardous material. If the level is too low, it may have bad consequences for the sequential operations. Hence, control of liquid level is an important and common task in process industries. Conical tanks find wide applications in process industries, namely hydrometallurgical industries, food processing industries, concrete mixing industries, sewage water treatment industries and wastewater treatment industries. Their shape contributes to better drainage of solid mixtures, slurries and viscous liquids at the bottom of the tank. So control of conical tank presents a challenging problem due to its non-linearity and constantly changing cross section. Hence, the conical tank process is taken up for study here. Computer process control, beginning in the 1960’s, initially used traditional linear control algorithms. However, the highly non linear characteristics of many process caused problems. Consequently, in the 1970’s, self regulating controllers [1,2] were developed to enhance explicit and implicit model-based controllers and controller tuning. The process industries, however, found few successes with early, hard-to-tune adaptive techniques [3]. Meanwhile, moving horizon and linear programming methods were being revived in the non-adaptive MPC. Industrial successes with MPCs renewed academic interest in these methods and new formulations of MPCs emerged [4]. These multivariable controllers were based on easily understood process models, could incorporate constraints, and were relatively easy to tune. In addition, their performance seemed less sensitive to varying time delays, one of the major limitations of the early adaptive controllers. To improve the robustness of the adaptive controllers, some researchers began to employ extended-horizon strategies [5]. Likewise, predictive controllers were improved by incorporating adaptive techniques [6]. Over the past decade, MPC has established itself in industry as an important form of advanced control [7] due to its advantages over traditional controllers [8],[9]. MPC displays improved performance because the process model allows current computations to consider future dynamic events. For example, this provides benefit when controlling processes with large dead times or non minimum phase behavior. MPC allows for the incorporation of hard and soft constraints directly in the objective function. In addition, the algorithm provides a convenient architecture for handling multivariable control due to the superposition of linear models within the controller. Anandanatarajan et al. [10] have designed the globally linearized controller for a first order non-linear system with dead time for a conical tank level process based on simulation. Madhubala et al. [11] have discussed the performance of the genetic algorithm based fuzzy controller for a conical tank. They examined that the fuzzy based control is better for compensating the set point and load changes than the PI controller. N.S. Bhuvaneswari et al. [12] proposed a Neuro based Model reference Adaptive Control for conical tank level process. In our work, we investigate the TTCIS with the focus of deriving a multi model adaptive strategy for model predictive controller for closed-loop A Multiple Model Adaptive Control Strategy for Model Predictive controller for Interacting Non Linear Systems V.R.RAVI, Dept. of Electronics & Instrumentation Engg., Velammal Engineering College, Chennai, India. vrravi_2007@yahoo.co.in T.THYAGARAJAN Dept. of Instrumentation Engineering, M.I.T Campus, Anna University Chennai, Chennai, India thyagu_vel@yahoo.co.in M.MONIKA DARSHINI Dept. of Electronics & Instrumentation Engg., Velammal Engineering College, Chennai, India. monikadarshini@yahoo.com T