基于轮换寻优的遗传算法在神经网络中的应用

基于最优控制中的轮换寻优思想 ,对遗传算法进行了改进。综合采用交叉编码方法和多参数级联之点映射编码方法对寻优参数进行编码 ,为了避免遗传算法中经常出现的过早收敛现象的发生 ,把近亲回避交叉策略和最优保留方法应用到遗传算法中 ,对神经网络的权值和阈值进行了分组轮换寻优 ,成功地完成了对多层前馈神经网络的训练 ,并与常规的BP算法和常规的遗传算法进行了比较。仿真结果表明 ,改进算法的效果比常规的BP算法和常规的遗传算法要好。这种寻优方法把传统的寻优方法和遗传算法结合起来 ,为全局寻优方法提出了一种途径 ,但更通用的全局寻优方法还有待进一步研究

Application of Genetic Algorithm Based on Alternating Optimization to Neural N etwork

In this paper the genetic algorithms is improved based on alternating optimization method. Crossing coding methods and multiple parameter cascading methods are used into the coding of parameters. In order to avoid the premature convergence, the relatives-avoidance cross method and elitist preserve method are applied, the weight value of BP network is optimized in group, and the training of MBP is completed successfully. The improved method is compared with the common BP algorithms and genetic algorithms. The simulation result shows that the new method is better than the common BP algorithms and genetic algorithms. The method proposed in this paper is combination of the conventional optimum method and genetic algorithms, which presented a new approach of the whole optimum, but the more universal optimum method still should be studied .