Font Size: a A A

Data Driven Modelling And Model Free Adaptive Control With The Applications To Complex Industrial Systems

Posted on:2018-07-27Degree:DoctorType:Dissertation
Country:ChinaCandidate:S D LiuFull Text:PDF
GTID:1318330542491101Subject:Traffic Information Engineering & Control
Abstract/Summary:PDF Full Text Request
This thesis mainly investigates the theoretical studies concerning the data-driven modeling and data driven model free adaptive control,and the applications in different practical problems.Specifically,the contributions of this work are summarized as follows.1.An enhanced genetic back propagation neural network with link switches(EGA-BPNN-LS)is proposed to address a data-driven modeling problem for the gasification processes inside UGI gasifiers.The model of the syngas temperature plays a dominant role in the UGI gasification process.However,the relationship of the inputs and the outputs of the model is difficult to be established via the first principles due to the practical complexity of the gasification process.EGA-BPNN-LS,incorporates a neural network with link switches(NN-LS),an enhanced genetic algorithm(EGA)and the Levenberg-Marquardt(LM)algorithm.It can not only learn the relationships between the control inputs and the system outputs from the historical data with the help of an optimized network structure through a combination of EGA and NN-LS,but also makes use of the network's gradient information via the LM algorithm.EGA-BPNN-LS is applied to model the UGI gasification process by using a set of data collected from the Ruixing Chemical Industry Group Co.,Ltd,Shandong,China.Thus,the effectiveness of EGA-BPNN-LS is verified.2.A modified lazy learning approach incorporated with relevance vector machine(MLL-RVM)is proposed for online modeling of the syngas temperature during the UGI gasification process.The model of the syngas temperature is a typical multi-input and multi-output(MIMO)nonlinear system,and its accurate mathematical model is difficult to be estibalished due to the complexity of the gasification process.In order to address this issue,the proposed MLL-RVM,appling the novel relevance vector machine(RVM)into the lazy learning control method,is used to modeling the real-time local RVM for the syngas temperature.Further,the effectiveness of MLL-RVM is verified by a series of experiments based on the data collected from Ruixing Chemical Industry Group Co.,Ltd.3.Two novel data-driven control methods,named compact-form-dynamic-linearization-based model free adaptive predictive control approach combined with the lazy learning(LL-CFDL-MFAPC),and patial-form-dynamic-linearization-based model free adaptive predictive control approach combined with the lazy learning(LL-CFDL-MFAPC),are proposed for a class of discrete-time single-input and single-output(SISO)nonlinear systems.The essential idea of these two methods is that by using the advantage of the database real time query of lazy learning method,both the online and the offline input and output data can be simultaneously utilized to uptate the parameters of the model free adaptive predictive controllers.Moreover,by using different dynamic linearization techniques,including the CFDL and PFDL techniques,the controller designing processes of these two methods can be implemented only using the process I/O data rather than the plant model.Meanwhile,the controllers of these two methods have strong robustness because the prediction mechanism participates.Besides,the stability and convergence of LL-CFDL-MFAPC,as well as the the the stability of LL-PFDL-MFAPC,are guaranteed by theoretical analysis under several reasonable assumptions.Further,LL-CFDL-MFAPC is applied to address the oxygen concentration control problem during the oxygen-enriched system in UGI gasification process,and LL-PFDL-MFAPC is applied in a practical three-tank water level control system.4.A novel double-successive-projection based model free adaptive control method(DSP-MFAC)is proposed.The significance of DSP-MFAC is that it provides a new proof method and new research tool for the model free adaptive control.By using DSP-MFAC,the structures of the controller and its parameters estimator are symmetric similar,which makes the analysis on the control performance more easily.Moreover,compared with conventional model free adaptive control under the framework of the contracting mapping principle,DSP-MFAC is more flexable to extended,since different controller structures and corresponding time varying parameters estimation methods can be designed when the defined norms in DSP-MFAC are different.Meanwhile,the stability of DSP-MFAC is guaranteed by rigorous mathematical analysis.Further,numerical simulation results show that DSP-MFAC is effective and applicable.
Keywords/Search Tags:Data-driven modeling, data-driven control, fixed-bed intermittent gasification process, neural networks, genetic algorithm, relevance vector machine, successive projection, lazy learning, three-tank water level system
PDF Full Text Request
Related items