Research On Multivariable Identification And Its Application

There exist some linear correlations between input and the noises of theout put and less information available in closed loop system, which leads manycla ssical identification methods unable to be used for closed-loop controlsystem directly. In this paper, anovel method NPSO applying to any inputsign al is presented to solve those problems. Besides, we introduce a method toest imate the initials of the unknown parameters. On this basis, some effectivestra tegies are proposed to solve the problems of the identification for theunknown high-order system and multivariable system.First of all, we introduce the background and development about mymajor as well as the basic knowledge of least squares method. An improvedmethod is brought in to estimate the parameters of Box-Jenkins model.Besides, we solve the identification problem of complex high-order systemvia building its characteristic model.Secondly, a method based on sub models and sub-sub models ispresented for multivariable identification in open loop. After MIMO system mo del is decomposed into many sub models, we build the Sub-Sub modelwith auxiliary output variable.It is a good direction to utilize existingalgorithm based on SM and SSM to estimate the multivariable parameters.Thirdly, a novel frequency response estimation method is proposed to beapp lied to both open and closed loop identification. PSO is utilized to solvethe problem of the choice of a damping factor in the frequency responseide ntification. In addition, the method is improved that the input signal is nolonger confined to the step response test signal, for any continuous signal canmaintain output convergence and be second-order derivative in the s-domain,it can be applied to the method.Last but not the least, considering the safety and economic factors, wenee d a method to get an unbiased and accurate model in closed loop. A novelopt imal method combined the global search capability of PSO and exactlyloc al optimization of NLJ is proposed to estimate the model from closed-loopstep response. There is an equivalent conversion from the closed loopidentification to open loop identification via adapting some suitable strategies.With the initials provided by frequency response estimation method, NPSOcan get rid of blindness and inefficiency caused by stochastic search. Besides,NPSO is applied for continuous-time MIMO system in closed-loop.