遗传算法优化前向神经网络结构和权重矢量

Optimizing Structure and Connection Weights of Feedforward Neural Networks Using Genetic Algorithms

提出了新的遗传算法优化设计前向神经网络的结构和权重矢量。这种新方法的创新在于：二值码串和实值码串的混合编码方法即保留了传统遗传算法的优点，又具有遗传编程和遗传策略的优点；结合遗传算子和ＳｏｌｉｓａｎｄＷｅｔｓ算法生成后代的方法丰富了遗传搜索空间的多样性，加快了遗传算法的收敛速度；对混合编码码串的动态参数编码方法提高了优化精度。

A new genetic algorithm is proposed to optimize the topology and connection weights for neural networks. The mixed encoding schema of binary and real value code not only retains the advantages of traditional genetic method but also gains the advantages of evolutionary programming and evolution strategies. The offspring generation method which combines the genetic operators and Solis and Wets operator diversifys the search space and speeds up the convergence of genetic search. And the dynamic parameter encoding method for the mixed code can obtain more precise connection weights.