The Research Of Non-parameters Model Adaptive Control Method

With the rapid development of current industrial field, the productiveprocess is getting more and more complicated, so it is difficult to control,forecast and evaluation by traditional model parameter control method. All theprocess data of enterprise are fully utilized, and it is important to developdata-driven control theory and method which has important theoretical andpractical significance.This paper selects Model Free Adaptive Control method to be the objectof study, which belongs to the field of data-driven control. In the view ofcompensate the high order controller output, extend the first order pan-modelof the traditional MFAC to second order pan-model, then the Second-OrderPan-model MFAC is re-derived.The original MFAC method only put forward to the concept of FullFormat Dynamic Linearization (FFDL), but not give the detail derive processof FFDL method. Because of the complexity of expression of FFDL, theproof of convergence has always been a problem in this field. Based on firstorder FFDL-MFAC method is detail derived in this paper (In order to differentfrom the traditional MFAC, the proposed method is named Improved Non-Parameters Model Adaptive Control (INPMAC)). Then the stability andconvergence of the proposed algorithm are rigorously proved. Under someappropriate assumptions, the outputs of system and controller are convergent,and the system output can ultimately track the given constant reference.In the process of real industrial environment and simulation research, it isfound that the choice of MFAC controller parameter has great effects oncontrol results. However, for the present situation of the rarely research ofMFAC parameter tuning, the parameter tuning algorithm of second-orderMFAC controller is proposed in this paper, which is basis of traditionalperformance index combine with gradient descent method. In the realindustrial environment, the random noise has serious impact on system.However, with asymmetric probability density function of Non-Gaussianrandom noise is the most common interference in the real environment. In thiscase, for the problem of traditional performance index should not be representsystem information, a parameter tuning method based on Minimum Entropyoptimization is proposed, in which the feature of entropy is used to describeaccurately for uncertainty, so as to realize the parameteroptimization in real industrial environment.At last, based on operating requirements and the analysis of thecharacteristics of a65t/h nature circulation boiler, a complete control schemefor a boiler should be designed. Specifically, due to the typical nonlinear andcoupled characteristics of the outlet pressure of a boiler, MFAC method is used to maintain it at the appropriate values. The automatic control system isrealized using the DCS of SIEMENS SIMATIC PCS7Box. Detailedcomparisons between MFAC and the traditional PID control demonstrate thepracticability and robustness of the MFAC control scheme.