Adaptive Genetic Algorithms for FFN-Based Dynamic Identification

Abstract An adaptive genetic algorithm (AGA) coded in a float number vector which can be potentially used to solve the problem of estimation of the interconnection weights in the large scale neural networks is easily implemented in MATLAB, and a framework of FFN based dynamic system identification using this algorithm is proposed in this paper. In the algorithm, a float number vector representation is used to naturally treat with the multiple parameters in real value and a mechanism of adapting the probability of mutation is introduced in order to avoid the premature convergence. The simulation experiments show that this algorithm outperforms the standard genetic algorithm coded in a binary string (SGA) both in convergent speed and in accuracy. Finally, a FFN-based model of a nonlinear dynamic system is trained by both the SGAs and the AGAs. and the comparative results are given to demonstrate the AGAs’ good performance and efficiency’.