Multiloop Internal Model Controller Design Based On PLS Model

Nowadays, multivariable control is still one of the most important areas that attracts a lot of researchers, where there are many interesting but difficult problems unsolved, such as decoupling control for complex multivariable systems, nonlinear multivariable control systems etc. Due to the sophisticated mechanism inherent in the process as well as the existence of random noise and unknown disturbance, it poses an extreme difficulty to establish rigorous model for high-dimensional systems based on first-principles. And such limitation has greatly compromised the practical application of optimization and control algorithm which used the first-principles model. With the rapid development of information and computer technology, the data-driven methods and theories has achieved tremendous breakthrough in various areas, like chemical process, biochemistry, economics et.al, which arouse significant attention in both academic and industrial field. As a multivariable regression technique, although the Partial Least Squares(PLS) has gained its popularity and success in process modeling, fault diagnosis and soft sensor, little research work has been done to integrate PLS into control structure. To address strong decouple and nonlinearity in multivariable system, we propose three novel control schemes which design the multi-loop internal model controller (IMC), adaptive internal model controller and nonlinear internal model controller in the PLS subspace respectively by fully taking advantages of PLS's merits in dimension reduction, self-decoupling structure and collinear elimination. The main topics of this thesis involve:(1) Proposing a multi-loop IMC scheme based on a dynamic PLS model. With reference to the poor performance that traditional controllers yield for multivariable systems with large time delay, we come up with a new IMC design method by incorporating the IMC algorithm into the PLS control scheme that proposed by Kaspar/Ray. To guarantee the system stability in the presence of uncertainties, robustness assessment for the proposed control scheme is addressed and the stability bounds are provided as well. Unlike the traditional decoupling MIMO systems, the proposed control strategy automatically decomposes the MIMO systems into a multi-loop control system in the reduced subspace. By projecting the measurable disturbance into the reduced subspace, a multi-loop feed-forward control is applied to achieve better performance for disturbance rejection. Simulation results for two representatives of distillation column are used to further demonstrate this new strategy outperforms conventional control schemes in servo behavior and disturbance rejection.(2) Developing a new multi-loop adaptive internal model control algorithm within the PLS control scheme to cope with the model mismatch problem. Process plants often encounter perturbations caused by environment changes, operation mode shift or aging equipments, which enlarge the structure gap between the plants and model, and even deteriorate the stability of the whole system with time going by. In this paper, we apply the recursive least square(RLS) algorithm to eliminate the model errors in each PLS subspace where the adaptive IMCs are built on, and provide a proof of the parameter convergence. With perfect online updating, the IMC control system regains stability as well as good control performance. Two benchmarks are applied to verify the effectiveness of the proposed multi-loop adaptive IMC algorithm.(3) Nonlinearity inherently exists in nearly every practical process, which forces the development of nonlinear control schemes. The traditional approaches are always restricted to many complex mathematical models as well as rigid assumptions, which significantly halt their feasibilities. Compared to these methods, we provide a novel dynamic nonlinear PLS model which connects the ARX and neural network(NN) structure in a series fashion in each PLS subspace , and a nonlinear IMC algorithm is then designed thereon. Similar to the PLS based IMC mentioned above, the proposed nonlinear IMC scheme provides the advantage of decomposing multivariable nonlinear system into several univariate systems in the latent space, and cascaded mode of ARX-NN facilitates the optimization and stability analysis of the whole system. The simulation results of pH neutralization process have demonstrated the feasibility of the algorithm.