On Parameterized Controller Based Data-driven Control Method

This dissertation focuses on data driven control methods based on parameterized con-troller. The main work and key contributions are summarized as the following:1. Given a convexly parameterized controller, the necessary and sufficient condition on the existence of robust controller is proposed for a single-input-single-output linear time invariant plant. Based on the analysis, a robust H?controller design method is proposed by applying frequency-domain data and convex optimization. The result is further extended for the plant with multi-model uncertainty or polytopic uncertainty. This method is applicable for both continuous-time and discrete-time system. Moreover, it can directly deal with controller design problem for the plant with time delay or mul-ti-model uncertainty. The correctness and effectiveness are verified by simulation stud-ies.2. Four types of equivalent dynamic linearization structure of the ideal controller are proposed for a class of single-input-single-output discrete-time nonlinear plant with dif-ferent complexity, including the compact form, partial form, full form, and ultra form. Then, the corresponding data-driven model free adaptive control algorithms are de-signed, as well as a uniform framework of controller parameter tuning. Stability anal-yses are developed, and the effectiveness of these methods are verified by simulation studies and experiment on a three-tank level control system. Meanwhile, the advantages of the proposed method are illustrated by the comparisons with other typical data driven control methods.3. To improve the control performance of data-driven model free adaptive control method, the RBF neural network is introduced to directly approximate the parameters of ideal controller. The stability analysis is developed, and the effectiveness and practica-bility are verified by simulations study and experiment on a three-tank level control system. Meanwhile, the applicability of the proposed method for time varying system is also illustrated.4. Based on aforementioned control methods, a pure data driven control scheme is proposed by adapting the idea of lazy learning and constructing a database of input and output data of the system. Two main properties of this scheme are:first, through build-ing the local data model of controlled plant dynamically, the controller parameters are optimized adaptively; second, the burden of building the local data model is reduced by data cleaning of the database and assessment of the data model. In simulation of the control to continuous stirred tank reactor, three types of data driven model free adaptive control schemes are compared. The scope of application is further analyzed and dis-cussed.