Application Of Dynamic Neural Networks To The Model Identification Of Hydraulic System

Perfect results of nonlinear systems identification used traditional methods established on the linear or intrinsically linear system are difficult to get. In this paper, the characteristics of Dynamic Neural Network (DNN) which can approximate any complex nonlinear relationships is studied as following: 1 The NN system identification methods are summarized and are compared with traditional system identification. The history of NN is reviewed and the structure, function and algorithm of Dynamic Elman are analyzed in details. 2 In the paper, the drawback of Standard BP training algorithm of Elman network is analyzed. An improved Elman network with self-feedback gain is acquired. At the same time, by using BP algorithm, the improved Elman network weigh value is adjusted. The proposed method is applied to identification of Hydraulic Automatic Gauge Control (HAGC) of a cold rolling mill. So the validity of the improved Elman network on the high-order nonlinear dynamic system identification is verified. 3 The limitation of the Standard BP algorithms is analyzed, with application to the HAGC of a cold rolling mill, two networks are compared, and one is the improved Elman network with standard BP algorithm, another is the basic Elman network with dynamic BP algorithm. The experimental results indicate that the former can arrive at the same effect of the later. The proposed methods can be used to high-order dynamic system effectively. In this paper, through the identification experiments on a hydraulic system, it shows DNN has a better identified precision and extensive ability. It can well reflect the dynamic characteristics of system. DNN has a great advantage in modeling linear and nonlinear systems.