Modelling engineering systems using analytical and neural techniques: Hybridization

From real input/output data, different control-oriented models of a quadrotor unmanned aerial vehicle (UAV) are obtained by applying different identification methods. Parametric techniques, neural networks, neuro-fuzzy inference systems, and the hybridization of some of them are applied. The identified models are analyzed and compared in the time and frequency domains. We conclude that the hybridization of analytical and intelligent techniques is a good choice to model of complex systems while keeping a good balance between accuracy and computational cost. In addition, off-line trained neural networks and adaptive networks with on-line learning are analyzed, and their advantages and disadvantages regarding modelling are presented. The influence of the partition of the training and validation dataset on the model error is also discussed.