A Comparison Study between Inferred State-Space and Neural Network Based System Identifications Using Adaptive Genetic Algorithm for Unmanned Helicopter Model

In this paper, system identifications of an unmanned aerial vehicle (UAV) based on inferred state space and multiple neural networks were presented. In this work an optimization approach was used to conclude an inferred state space and the multiple neural networks system identifications based on the genetic algorithms separately. The UAV is a multi-input multi-output (MIMO) nonlinear system. Models for such MIMO system are expected to be adaptive to dynamic behavior and robust to environmental variations. This task of accurate modeling has been achieved with multi-neural network
architecture in the most recent years. The presented work is focusing on an inferred state space based system identification which is a new approach seldom used, but it is also easier and more stable compared with the multi-network based system identification during the modeling of dynamic behavior of nonlinear systems. In other words the number of inputs used in the genetic algorithm to obtain an inferred state space is almost one third of the number of inputs needed to develop the multi-layer recurrent neural network architecture to simulate the required dynamic behavior of a real model. The neural network models are based on the autoregressive technique with linear and nonlinear networks. The simulation results presented in this paper show the superiority of the inferred state space model compared with the autoregressive technique based multi-neural network.