Nonlinear System Identification And Model Predictive Control For Parameter Varying Models

Model predictive control (MPC) algorithm is one of the most successful controllers in process industries such as chemical industry, oil refining, and power plant. Until recently, most industrial applications of MPC have relied on the linear dynamic models. However, the MPC based on linear models often results in poor control performance for the highly nonlinear processes, which limits their applicability especially in process control applications because of the inadequacy of a linear model to predict dynamic behavior of a nonlinear processes.Furthermore, one of the major bottlenecks for the nonlinear model predictive control (NMPC) is the unavailability of an accurate, convenient and reliable mathematical model of the actual physical process. When dealing with dynamic systems with high nonlinearities, the model’s accuracy cannot be guaranteed by maintaining a simple linear relationship. This has led to the development of NMPC in which a more accurate nonlinear model is used for the process prediction and optimization.Recently, parameter varying process identification approaches have been introduced and have attracted great attention from both academia and industry recently. The effectiveness of the parameter varying method has been verified in the field of aerospace (high performance aerospace aircraft, missiles and turbine generator). To find a better model structure to approximate nonlinear processes over a broad operating regime remains an important challenging task for the nonlinear control community.Therefore, focusing on nonlinear system identification and model predictive control for parameter varying models, the specific research contents are as follows:1. To improve the approximation capability of the multi-model LPV model, asymmetric Gaussian weighting functions are introduced and compared with commonly used symmetric Gaussian functions. By this means, locations of operating points can be selected freely. Due to the flexibility of choosing uneven operating points for local linear models over the scheduling variable, the LPV model can be more accurate and effective.2. According to the requirement to requirement of validation, a HVAC (Heating, Ventilation, and Air-conditioning) system is developed in our lab, which includes system design, hardware and software implement. An experiment is performed on the real HVAC to further validate the effectiveness of the proposed approach. The experimental results also show that the asymmetric Gaussian weighing function can obtain improved results for multi-model LPV identification in real industrial processes.3. A novel nonlinear parameter varying (NPV) model structure is introduced for the identification of general nonlinear systems. This model is given in the form of a block-oriented Hammerstein model along with varying parameters, which combines a normalized static artificial neural network with a linear dynamic subsystem. Besides, the identification procedure and test of this specific model structure is given. The effectiveness of this proposed method is validated through simulation studies in the typical MIMO nonlinear control system for the free-radical solution polymerization process of styrene in a jacketed continuous stirred tank reactor (CSTR). The case study demonstrates that the NPV model achieves better performance than the widely used linear parameter varying (LPV) model and Hammerstein-Wiener model.4. A nonlinear MPC design is developed based on the identified NPV model. The control action is computed via a multistep linearization method of nonlinear optimization problem. Meanwhile, the Direct Linear Feedback network is integrated into the design of nonlinear model predictive control. Thence, the computing burden of a nonlinear predictive controller can be greatly reduced. Simulation examples demonstrate the results of model identification and the control performance of nonlinear MPC.