Sequential Decision-making Based Pid Tuning Methods With Applications In Control Loop Decoupling

PID parameters tuning methods have been widely concerned in theoretical and engineering communities.Relying too much on process models,the mathematical tuning method cannot achieve ideal control performancesin the presence of modeling difficulties or uncertainties of industrial processes.Also,the empirical parameter tuning approaches based on process curves which arenot allowed to produce excessive oscillation in the actual process arerather difficult to be applied in practice.Particularly,the methods above-mentioned cannot deal with the coupling problem of multi control loops where it's difficult to design novel decoupling compensatorsintermsoflimitations of costsand technologies.Practically,PID parameters are usually adjusted according to the operators' experinecs on control performances and operating conditions,performing sequential decision-making steps to separateoperating frequencies of control loopsso as to reduce the coupling effect.Therefore,it is undoubtedly of an academic research significance and engineering application value to study how to emulatethe sequential decision-making steps of operators to implement PID parameter tuning,effectively eliminating the coupling of control loops and realizing intelligent process control operations.The main contribution of this thesis is addressed as follows.1.Considering the difficulty to quantify human operators'experiences of sequential decision-making steps for PID parameter tuning,an empirical modeling method of PID parameter tuning based on sequential decision-making is proposed.Based on the multi-step tuning law of practical PID parameter tuning,the operating frequency of control loop is forced apart,making the coupling effect of control loops weakened,revealing the quantitative relationship between operators' experiences and loop decoupling degrees.2.Facing the problems of high dimensionality and complexity of historical data,this thesis proposes a nonlinear dimensionality reduction method,which uses input training neural networks to retain the valid information of input data,and proves the validity and correctness of this method by using information entropy theories.3.Based on the theory of Equivalent Transfer Function(ETF),the feasibility of the decoupling method of PID parameter tuning loop based on sequential decision-making is proved.The proposed control loop decoupling method is applied to square,non-square systems and the Aspen model of an alcohol distillation process,achieving satisfactory results.The research work is expecting to provide a useful attempt to explore intelligent tuning methods of PID parameters.