In order to avoid the population premature into local minimum,a new averaging method based on a random initial population was introduced into the genetic algorithm.The initial population is stochastically divided into several sub populations to form niches,with the purpose to maintain the population diversity,make the individuals in a sub population not display prematurity phenomenon,and improve the convergence speed of the algorithm as well.The adaptive technique is employed to control the crossover and mutation probability,therefore the algorithm can find the optimal solution quickly.Simulation results show that,compared with traditional RBF neural network optimized by genetic algorithm,the new algorithm is characterized by less iterations,higher precision,and greatly improved convergence speed.