Research On The Application Of Neural Network In The System Identification Of AUV

Up to now, the technology of system identification (SI) is applied widely in many fields, the motion model of the underwater vehicle can be built with the neural network and motion state data can be used directly to recognize the motion characteristic of the underwater vehicle, it can be used as an identifier for the self-adapt controller or a dummy sensor, it can also provide the information for fault diagnosis etc. The purpose of this paper is to apply this technology to an UV produced by HEU (Harbin Engineering University). The results of it will be very useful for the next following research.Firstly, this thesis analyzed the advantages and disadvantages of various neural network structures, with the experimental data of horizontal plane "Z" motion to adjust the parameter set and environment set. This paper tries to find an appropriate neural network for the underwater vehicle on-line identification, form a set of new rules to be used for current networks analyze and parameter adjusting as well as neural network structure comparison and test. Especially, the thesis did lot work on center point problem of RBF network, which can minimize the error. At the sane time, this thesis also discussed the motion parameter estimation of on-line identification, which can be used for the fault-tolerant technique and can dramatic save the cycle index and identification time dramatically.The motion parameter identified by the textual method system identification is fail-safe by validated. The textual method will be very useful for the research of underwater vehicle steering and self-adaptive controlling.