Abstract:

The small-scale unmanned helicopter (SUH) is a typical nonlinear dynamical system whose identification experiments are difficult to implement due to its inherent instability. The data sampled is thus small and non-informative, causing a loss to the estimation performance of conventional blackbox identification methods. In this article, a new method based on support vector machine (SVM) is proposed to identify the flight dynamics of SUHs at hover. Priori knowledge is used to simplify the dynamics and determine the non-linear regression form of the model. The mapping relationship is therefore constructed from the primal input space to the highdimensional feature space, with the corresponding kernel function determined. Least aquare SVMs (LS-SVMs) are applied to compute the SVM models. The corresponding non-linear parametric models are obtained consequently, which express the internal couplings explicitly while showing high prediction accuracy, and are suitable for non-linear controller design.

Key words: small-scale unmanned helicopter, support vector machines, least square methods, kernel function, flight dynamics, identification