Research On Multiple Models Adaptive Control Of Nonlinear System

With the development of productivity and control theory, the controlled system is often highly nonlinear. Traditional adaptive controller is extensively studied in time-invariant system with unknown parameters or time-variant system with slow drifting parameters. While the system has abrupt changes in parameters, the settling time is long, the overshoot of the system is big and the performance is not good.For this problem, the multi-model adaptive control of complex nonlinear systems is introduced in this paper from the following aspects.1) To deal with affine nonlinear systems, the adaptive dynamic programming (ADP) algorithm is combined with the idea of multiple-model control. The proof of convergence and stability are given. In ADP algorithm, neural networks are designed to to approximate the performance function and control strategy, and the control inputs are restricted in a bounded range. However, the controller cannot guarantee the dynamic quality of the system, especially the transient response process, and the overshoot is often large. According to the idea of multiple models control, ladder-shaped reference trajectory is designed. The system states track the final value of reference trajectory in the whole control process step by step. Simulation results are shown and analyzed, demonstrating the feasibility and validity of the proposed optimal tracking control scheme.2) Nonlinear discrete time system is modeled and by back propagation (BP) neural networks used to approximate the dynamic character. The weights of neural networks are updated by using a dead-zone algorithm. For the situation that jumping change parameters exist, multi-model is built by multiple neural networks with different weights to cover the uncertainty of parameters. Different kind of combination of fixed model, adaptive model will be used for robust multiple models adaptive control. At every sample time, a performance index function will be used to choose the optimal model and the corresponding controller. The proof of stability and convergence of MMAC are given, and the significant efficacy of the proposed methods is tested by simulation.3) The multi-model control method is applied to the actual blast furnace burden distribution system. A large amount of burden surface data from radars are classified, and the multiple models set of burden surface is built. When the expected burden surface is given, multiple control strategies are designed based on multiple burden surfaces of the model set, and multiple burden distribution matrices are obtained. In every charging distribution period, the real time burden surface data will be matched with the model set based on switching mechanism, and the corresponding charge distribution matrices will be selected for charge distribution until the expected burden surface is produced. The proposed control strategy is applied to a2500m3blast furnace in an Iron and Steel Plant, the economic and technical indexes of blast furnace have been improved greatly.The main innovation points in this paper are as follows:1) A multiple setpoints tracking control strategy based on ADP is proposed. A non-quadratic performance functional is introduced to tackle the control constraints by consistently restricting the control inputs in a bounded range. Control inputs will not abrupt changes due to different set point According to the idea of multiple-model control, system states track the final value of the reference trajectory step by step, ensuring the stability of the system and significantly reducing the overshoot and response-time.2) BP neural networks will be used to cover the uncertainty of the parameters of the system. Considering the unmodeling dynamics of the system, a robust adaptive controller based on the BP neural network is proposed. Three kinds of combinations of adaptive model and fixed model are used to make the multi-model set, and a switching law is suitably defined to make the decision of the best model. The main contribution is the rigorously proof of convergence and stability of the system.3) In the actual blast furnace burden distribution control system, the transcendental multiple models set of burden surface, corresponding burden distribution matrices and switching mechanism are built. Feedback mechanism is formed from the observed burden surface data of radars, and closed-loop control is realized. This method helped to overcome disadvantages of open-loop control of traditional blast furnace burden distribution, and the accuracy and efficiency of burden distribution is greatly improved.