Font Size: a A A

Robust Modeling And Predictive Controller Design Based On The Dynamic PLS Method

Posted on:2017-02-14Degree:DoctorType:Dissertation
Country:ChinaCandidate:X JinFull Text:PDF
GTID:1108330485492774Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
The development of modern industry towards complex and large scale, which presents great challenge to conventional first principle models and control methods. With the development of information technology, the automatic acquisition rate of process variables becomes higher. The effective use of these information promotes the development of data-driven method. As a multivariate statistical based data-driven method, partial least square(PLS) has been widely applied to many fields. While, the application of process control is still in the initial stage.In this paper, with the PLS characteristic of noise reduction, dimensionality reduction, eliminating collinearities and self-decoupling, dynamic PLS modeling and model predictive control(MPC) are studied. First, in order to improve the accuracy of the model in the case of contaminated data, a robust dynamic PLS modeling method is proposed. Then, disturbance rejection MPC and offset free control for linear and nonlinear system are proposed. Specific works are as followsTo tackle the sensitivity to outliers in PLS, a new robust dynamic PLS based on outliers detection method is proposed. An improved radial basis function network (RBFN) is used to construct the predictive model of inputs and outputs data, the predictive outputs of RBFN are closed to the true values of the system outputs. Therefore, the errors between the predicitve outputs and the system outputs are white noise, which should be met normal distribution. Hidden markov model method is used to detect the these errors to determine whether the system outputs are outliers. The detected outliers are removed away and replaced by the predictive outputs of RBFN. With this data pre-processing method, a more robust dynamic PLS model is obtained.Due to the principle of dynamic PLS for data compression and iterative modeling, the mismatch between model and actual plant exists, which would be result in incomplete decoupling. In order to suppress the disturbance caused by the coupling. A disturbance rejection generalized predictive control in dynamic PLS framework is proposed. With this method, the multiple single-input single-output(SISO) systems are treated as a special feature of a multi-input multi-output(MIMO) system and the control law is solved. The output tracking error weight of the objective function is reformed as a form which is adjusted in real time according to the deviation between the predicted value and the reference trajectory. The basic principle of weight adjustment is that, the weight of the prediction of each output tracking reference trajectory is consisted of the weighted sum of squared errors of other outputs deviated from their reference trajectory at the same time. Two results of simulation show that the proposed method can reject the disturbance caused by the coupling.State space MPC based on dynamic PLS framework is proposed. With the characteristic of dimensionality reduction and self-decoupling, the mismatch between dynamic PLS model and actual plant is existed. The state feedback is used in the proposed MPC rather than output feedback, which will result in steady state error. In order to solve this problem, two methods are proposed to incorporate the output of the system into the controller. One is to reform the state space model as a velocity form. The other is to introduce a disturbance model into state space model, and apply an observer to estimate the system state. Simulation of Jerome-Ray distillation column model and the reaction process of polyethylene show the effectiveness of the method.An offset free MPC in dynamic PLS framework for nonlinear system is proposed. Frist, dynamic T-S fuzzy model is used in latent variable space to describe nonlinear dynamic characteristic of the system. With this model, an offset free MPC controller is designed. In the controller, disturbance model and state observer are introduced to guaranteed the offset free tracking. Simulation is carried out in the process of PH neutralization titration, The result shows that the proposed method can make the system to track the setpoint without steady-state error.
Keywords/Search Tags:partial least square, outlier detection, model predictive control, disturbance rejection, offset free control, T-S fuzzy mode
PDF Full Text Request
Related items