Controller tuning via constrained Bayesian optimization

Controller parameter tuning is always faced challenges due to the complexity of the dynamic system. Even for professional control practitioners, hand-tuning task is time-consuming and tedious, wasting a lot of time and unable to find the optimal parameters. In this paper, an automatic tuning method using constrained Bayesian optimization is proposed. Firstly, the concept of MPC controller tuning is introduced. Secondly, the constrained Bayesian optimization and how to solve controller tuning problem is proposed. Finally, the proposed method is validated in Shell Heavy Oil Fractionator. Experimental results show that proposed automatic tuning method can achieve good performance with only a small number of evaluations is superior to traditional method.