基于合作式协同进化算法的神经网络优化

针对一般遗传算法优化神经网络存在的不足，提出合作式协同进化遗传算法实现神经网络结构和权值同步优化方法．首先，结合合作式协同进化遗传算法本身特性和神经网络特点，给出种群分割方法；其次，为了实现结构和权值的同步优化，提出一种新的混合编码方法，并根据该混合编码方法设计新的交叉和变异算子；然后，根据编码结构、代表个体和合作团体之间的关系，提出一种新的结构优化方法；再次，给出进化过程所需代表个体选择、适应度构造方法等．最后，通过双螺旋线问题验证本文算法的有效性．

Optimization of Neural Network Based on Cooperative Co-Evolutionary Algorithm

The optimization of neural network structure and its weights based on cooperative coevolutionary genetic algorithm is proposed for some limitations of conventional genetic algorithm optimizing neural network. Firstly, population partition methodology is presented by combining characteristics of both the cooperative co-evolutionary genetic algorithm and the neural network. Secondly, a novel mixed encoding method is also put forward to perform synchronal optimization of structure and weights of the neural network. Proper crossover and mutation operators were designed based on the encoding method. Next, a novel structure optimization technique is given according to the relationship among encoding, representation and cooperative group. Representative individual selecting and fitness evaluating are given. Finally, the two-spiral problem is used to validate the efficiency of the algorithm presented in this paper.