Neural network blind equalization optimized by parallel genetic algorithm with partial elitist strategy

Owing to the disadvantage of slow convergence and easy to fall into local minimum of the neural network blind equalization algorithm under high dimensional and non-convex cost function, a parallel genetic algorithm (GA) with partial elitist strategy was proposed to optimize neural network training. According to the neural network topology, individual coding, the control code and the weights were set up to realize the network topology structure and the network weights simultaneously. The individual group was sorted according to the adaptation degree of the optimization iterative process, in order to integrate the advantages of genetic algorithm under the conditions of different genetic operators. Some elite strategies effectively avoid the phenomenon of premature phenomena caused by the optimal individual control in the process of evolution. The simulation results under the nonlinear channel condition show that the method has better convergence performance. © 2016, Editorial Board of Journal on Communications. All right reserved.