为了解决脉冲整形实验中经常碰到的遗传算法收敛速度慢,早熟等问题,我们对传统的遗传算法进行了几点改进,例如:将两个个体间的欧几里得距离作为判断是否进行交叉操作的判据之一,而不再仅仅依靠个体的适应度值(fitness),这样能有效地保持种群的基因多样性,提高交叉算子的效率;第二,引入多个交叉算子共同作用于种群.由于算子的组合效应,共同作用产生的子代适应度值要优于任何一个算子单独作用时产生的子代适应度值.因而可以产生更大的探索范围,防止算法收敛在某个局部最优解;第三,为了提高收敛速度,我们提出一种新的插值方式:非线性插值,即依据频谱的强度大小决定插值点的密度.我们初步将此改进算法应用到飞秒整形光路输出光的相位补偿实验中,得到了比较令人满意的结果.
In order to solve the problems of slow convergence and prematurity of genetic algorithms commonly encountered in pulse shaping experiments, we proposed some improvements of traditional genetic algorithm, for example: 1. The Euclidean distance is used as the criteria of performing cross-over, instead of depending on individual fitness only. 2. Multiple cross-over operators are used to operate on the population together. Because of the cooperative effect, the fitness values from multiple cross-over operators are better than that from any individual operator. Hence a larger searching space is generated, which prevents the algorithm to converge to a local optimal solution. 3. In order to achieve faster convergence, we propose a new interpolation method, nonlinear interpolation, in which the interpolated number density depends on the intensity of laser spectrum. The improved algorithm is tested in the optimization of the output spectrum phase from a femtosecond laser pulse shaping setup.