Modeling And Predictive Controller Design Based On The Dynamic PLS Method

With the rapid development of modern industry, the scale of industrial production is increasing. The production process and production flow have become more and more complex, which is a major challenge to the traditional mechanism modeling and control strategy. On this condition, data-driven application and theory attract more and more researchers’attention in industry and academia. Due to its advantages, PLS-based data-driven control methods have been widely applied to various fields. However, there are still some specific problems in the industrial process, for example, the nonlinear dynamic characteristics fitting of the PLS, latent variables constraints coupled in PLS control etc. This paper is mainly focusing on strong coupling, constraint processing, nonlinear problems in complex chemical process. Combined with the characteristics of PLS denoising, dimensionality reduction, eliminating collinearity, self decoupling and so on, this paper proposes a multi-loop predictive control strategy based on dynamic PLS modeling method. Technically, we first design linear unconstrained controller and constrainted controller and then extend the algorithm to nonlinear system modeling and control methods and further research on non-linear system online modeling and control problems.The main contents include:(1) Aiming at conventional control strategies in big lag multivariate system, combined with PLS control framework by Kaspar and Ray, Chapter II of this paper proposed a implicit variable dynamic PLS modeling methods and PID control framework. Against the characteristics of strong coupling and difficult modeling of aluminum alloy MIG welding process, the dynamic PLS framework is used in the welding process system modeling and controller design according to its decoupling characteristics. This method transfers the multivariable controller design problem in original space into multi-loop single-variable controller design in latent space and then design PID controller independently in each latent variables control loop. Finally, demonstrate the effectiveness of this method by simulation.(2) Aiming at the constraint of process variables and control variables result from external conditions of process such as pressure, device, environment, and internal characteristics, a higher control requirement is needed for the data-driven based controller design. Consequently, in the Chapter III of the paper, a space projection method (DyPLS) is proposed to deal with constraint, which is combined with iterations method and spatial transform. Constrained model predictive control method is used for model controller design. The simulation shows that this constrained iterative algorithm can achieve the desired result in the case of constraints and pole shift. On this basis, process control is achieved and method of this article is verified on the test model.(3) The nonlinear characteristic is one of the biggest problems in the modeling and control of industrial process. Unfortunately, non-linear characteristic is universal, which is more difficult for nonlinear characteristic modeling of the multi-variable system. Fuzzy model especially the TS model provides an effective way to solve the nonlinear problems. Combined with PLS method with fuzzy method, the multivariable dynamic nonlinear modeling is transferred into the multiple SISO nonlinear dynamic modeling problems. And dynamic fuzzy PLS model is established to describe the process dynamic nonlinear characteristics. Combined with the dynamic fuzzy PLS modeling method, based on PLS control framework proposed by Kaspar and Ray and characteristics of TS fuzzy model, constraint conditions control problem is solved. Meanwhile, fuzzy PFC control is introduced to verify the control performance of fuzzy PFC in pH titration process(4) The process is affected by changes in working conditions, equipment, catalyst, which results in the deviation between original model and current object, having a great impact on the control performance of the model-based controller. What is worse, it leads to instability of the whole system. To cope with this, in Chapter V nonlinear recursive fuzzy PLS model update method, which combines WPLS method with recursive TS modeling, is proposed. The results show that this adaptive algorithm can do better in update the model parameters to achieve the matching between model and process in the case of gain change. On this basis, prediction function controller is introduced to realize the process control. And an experiment of pH titration is simulated to validate this method.