论文部分内容阅读
基于全局搜索的进化算法和一种局部搜索算法——结构化的非线性参数优化方法(SNPOM),提出两种混合的优化算法来估计RBF神经网络中的参数:1)初始化一定数目的种群作为SNPOM的初始值得到其适应值,通过选择、交叉和替换策略来更新种群;2)采用进化算法运行一定的代数,从最终群体中选取一些个体进一步用SNPOM来优化.这两种混合优化算法的本质是用进化算法为SNPOM搜寻最优初始值,以得到全局最优解.仿真实验结果表明,该混合算法比单独使用进化算法或SNPOM更优,且优于其他一些算法.
Based on the global search evolutionary algorithm and a local search algorithm - Structured Nonlinear Parameter Optimization (SNPOM), two hybrid optimization algorithms are proposed to estimate the parameters in the RBF neural network: 1) Initialize a certain number of populations as The initial value of SNPOM gets its fitness value, and the population is updated by selection, crossover and replacement strategies. 2) The evolutionary algorithm is used to run some algebra, and some individuals from the final population are further optimized by SNPOM. The essence is to use the evolutionary algorithm to search the optimal initial value of SNPOM to obtain the global optimal solution.The simulation results show that the hybrid algorithm is better than the other algorithms alone or using SNPOM algorithm.