Artificial intelligence approaches to model-based control

While of undoubted value for nonlinear identification and control of dynamic systems, neural networks have a number of limitations for practical applications. Thus, in online training, due consideration must be given to the necessity for regularisation with noisy data and to the choice of network architecture. More fundamentally, the nontransparent black-box nature of neural models make it difficult to include a priori system information, and to interpret the final structure meaningfully in terms of physical process characteristics. Neural approaches also fail to exploit the significant body of theoretical results available for conventional model-based control, making it difficult to analyse the closed-loop behaviour in terms of stability and robustness. The aim of this paper is to describe a nonlinear modelling architecture, called the local model network (LMN), which introduces transparency while offering distinct advantages for nonlinear model-based control. Simulation results for a pH neutralisation process are used to illustrate the performance benefits of LMNs for nonlinear dynamic matrix control (DMC) and for nonlinear internal model control (IMC)