Font Size: a A A

Subspace Methods For System Identification And Predictive Control Design

Posted on:2008-05-31Degree:DoctorType:Dissertation
Country:ChinaCandidate:H YangFull Text:PDF
GTID:1118360215476818Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
It is urgent to find appropriate identification methods for the morden industrial processes. Some important problems remain unsolved for the complicated multi-variable systems. Being a young but promising field, subspace methods have attracted many attentions in the identification and control fields. However, there are still many open problems to be solved. In this paper, subspace identification researches with predictive control design are considered. The main parts including the implementation of new algorithms, the property analysis and the predictive controller design.A new recursive subspace identification algorithm is proposed based on the frame of forgetting factor. This method overcomes the problem of the computation burden and satisfies the requests of online update. Therefore, the subspace methods are extended from offline identification to online applications. In order to make the new recursive method convergence fast, the RLS-like forgetting factor is introduced to construct the new data Hankel matrices. In the framework of subspace methods, the gradient type subspace tracking method is used to implement the update of the state subspace, which results the unbiased estimate. Different from traditional offline method, the system matrices are recursively computed using RLS method. The convergence analysis of the proposed method is given and finally, the efficiency of this method is illustrated with a simulation example.A new closed-loop identification method is proposed to implement consistency estimation of closed-loop subspace identification. The effect of the closed-loop modeling is analyzed carefully and taken account into the identification process. In order to eliminate the effect of the correlation between noise and input under feedback, the input-output error sequences are augmented into input subspace in the construction of subspace projection. The consistency of parameter estimator is proven in theory, which implements asymptotic consistent estimation of closed-loop multivariable systems. Simulation example is given to test the efficiency of this method.The new subspace methods are proposed for two kinds of nonlinear systems, including bilinear system and PWL system. The new problems and challenges appearing in the nonlinear applicationes are solved, which makes subspace identification more applicable to the modeling and control of industrial process. With regards to bilinear system, a major drawback is that the subspace identification methods induce enormous dimension of the data matrices, which grows exponentially with the increase of the model order. A computationally efficient subspace identification procedure for bilinear systems is proposed to provide a solution to tackle the computational difficulties. The identification of a class of piecewise linear (PWL) systems composed of state-space sub-models is also considered. The identification of the PWL model includes two steps. First, obtain the polyhedral partition of the regression space based on a new fuzzy G-K clustering technique. In each cluster, the nonconsecutive parts of the input and output data that correspond to one of the local linear sub-models can be used to obtain the system matrices. Different transitions are used to force the models to the same state space basis. Finally, we verify the applicability subspace identification method through numerical examples.Data-driven predictive control is implemented based on the previous researches, such as nonlinear identification and closed-loop subspace identification. The correlation of the model and controller design reduces the number of the design parameters. In the existence of system noise,a new robust predictive controller is proposed for LTV multivariable system. Through subspace identification, uncertainty model is used for better description of the controlled system. The new predictive controller is designed with the combination of robust control strategy. The system modeling and design of robust controller are included in a framework, which increases robustness of the control system. Being a subclass of nonlinear systems, bilinear system is useful to approximate a class of nonlinear systems and implement predictive control in many circumstances. Therefore, a nonlinear predictive control is implemented by exploiting the structural properties of the identified bilinear subspace predictor model. The open-loop optimization problem of MPC that is nonlinear in nature is solved with series quadratic programming without any approximations. These improvements in system modeling and optimization solver make the bilinear subspace MPC approach more applicable to real industry processes. The proposed control approaches are illustrated with a simulation of a nonlinear continuously stirred tank reactor (CSTR) system.
Keywords/Search Tags:Subspace identification, predictive control, online recursive identification, closed-loop identification, nonlinear identification, consistency, convergency
PDF Full Text Request
Related items