Convergence Analysis of Online Decoupling Based on Neural Network

Aimed at a novel online decoupling algorithm of the neural network, which makes the cross-correlation function as the target function and the weights are trained by the hybrid genetic algorithms based on real (floating)-coded, its convergence is analyzed. Firstly, the uniform random step sequence is selected to excite adequately the MIMO system. Secondly, it is discussed that cross-correlation function has some effect on the convergence of algorithm. Finally, the convergence of the Hooke-Jeeves pattern search and FGA is discussed. The theoretical analysis and simulation results indicate that the algorithm is convergent and efficient.