Identification of an unmanned helicopter system using optimised neural network structure

A method for offline system identification to model the attitude dynamics of an unmanned helicopter using neural network techniques has been developed. The processed test data obtained from various test flights were used to train a neural network ARX (autoregressive structure with extra inputs) using Levenberg-Marquardt computation. An optimised network structure was obtained using a combination of cross validation, correlation analysis, Lipschitz criterion and weights regularisation methods to ensure good generalisation ability of the model. Satisfactory correlation analysis results were achieved which indicated that the offline model using the proposed methods contain all information about the dynamics of the system. The proposed methods would allow users to design a better prediction model with the aid of the correlation analysis from estimation theory. The results show that the proposed method is effective in modelling the coupled UAS helicopter dynamics with acceptable accuracy.