Nonlinear System Identification using Genetic Algorithm Based Recurrent Neural Networks

In this paper, a new genetic algorithm (GA) is developed to optimize the architecture of a recurrent artificial neural network (RANN) with multiple hidden layers. A new direct matrix mapping encoding (DMME) method is proposed to efficiently and effectively represent the architecture of a neural network. A modified back-propagation (BP) algorithm is utilized to tune the weights and other parameters of RANNs. The RANN optimized by this algorithm has been applied to the identification of nonlinear dynamic systems with unknown nonlinearities. Three types of RANN-based nonlinear models are proposed to describe the behavior of nonlinear systems. The effectiveness of these models and identification algorithms are extensively verified in the identification of several complex nonlinear systems such as "smart" actuator preceded by hysteresis, and friction-plague harmonic drive