论文部分内容阅读
设计间接编码方案表达网络结构,利用二维细胞自动机中的元细胞表示网络连接存在与否,对元细胞的坐标和值分别演化,实现对网络结构的生长和剪枝.应用二进制量子化粒子群算法,采用特定的适应度函数,生成和进化元细胞的坐标.设计元细胞的邻域演化规则,演化元细胞的值.利用浮点量子化粒子群算法训练当前网络,逐步得到最终稳定网络结构及其参数.实验结果表明,当应用于不同规模的网络结构设计时,算法复杂度变化不大,并且具备稳定的收敛性能.
Design indirect coding scheme to express the network structure, the use of two-dimensional cellular automata in the presence or absence of network connectivity, cell coordinates and values of the evolution of the network structure to achieve the growth and pruning. Application of binary quantization particles Group algorithm to generate and evolve the coordinates of metamerism cells by using a specific fitness function, and to design the neighborhood evolution rules of meta-cells and the value of evolved meta-cells. By using the floating-point quantum particle swarm optimization algorithm to train the current network, the final stable network Structure and its parameters.Experimental results show that when applied to the design of network structures of different scales, the complexity of the algorithm does not change much and has stable convergence performance.