Design And Application Of Generalized Predictive Controller Based On SVR And Lazy Learning

Model prediction control is a new class of computer control algorithms and developed rapidly in recent years because of its multi-step forecasting, rolling optimization and feedback correction control strategy, and this method has better control effect, suitable for industrial process of which the mathematical model is hard to construct or pretty complex,therefore, it increasingly draws the attention of the engineering both at home and abroad and has been successfully applied in machinery manufacturing, power electronics, metallurgy industry and chemical process control systems. Predictive control algorithm is an advanced process control and thus has drawn more and more attention. After several years of rapid development, it has made significant achievements in both theoretical research and practical.During the development process of predictive control, predictive control algorithm for linear time-invariant systems research has become mature gradually, and has been successfully applied to the actual industry.However, for complex nonlinear systems, predictive control is still not mature enough yet.Therefore,how to make predictive control be used in complex nonlinear systems has become the key focus of the academia.From the engineering point, people want the controlled model to be as simple as possible, but it could still able to maintain good robustness,to meet the needs of real-time control.Moreover,the control method is relatively simple and easy to implement.Support vector regression,which is based on statistical learning theory,is a new modeling and system identification tools. It can overcome the shortcomings of the local minima in neural network control, the poor generalization ability and the structure which needs to be identified, it has a good generalization performance, with advantages of being able to approximate the complex nonlinear function as well as good generalization performance.So it is concerned by the control community in recent years.Predictive control method is a kind of model-based control, and thus support vector machine system identification methods can be applied to predictive control method. Focusing on the characteristics of the control method, the support vector regression and lazy learning algorithm are added to a generalized predictive control which is based on the CARMA model.Finally, the application of this method to a highly non-linear binding of typical control system and the simulation results show the effectiveness and feasibility of the proposed controller.