实数编码多种群遗传算法的改进及应用

The improvement and application of real-coded multiple-population genetic algorithm

本文首先从分析实数编码的GA理论出发,提出实数编码条件下不需要继续沿用二进制编码条件下常用的传统单点、两点及多点交叉算子,基因选择可采用基因位独立(locus independent)的方式进行.据此研制了相应的交叉算子,并给出了一种实现方式,试验计算验证了该算子的有效性.其次,针对地球物理反演问题的复杂性以及遗传算法参数设置对问题的敏感性,将算法性能调整的灵活性作为方法选择的重要标准,采用了多种群遗传算法作为反演方法,设计了两个种群级的改进策略,并验证了其效能.最后,将改进的实数编码多种群遗传算法应用于重力模型反演,完成了梯形变密度体组合模型和2.5度体多边形棱柱组合模型的反演试算,取得了预期效果,并成功地应用于新疆地区实际资料的处理与解释.

This paper first examines the theory of Real-Coded Genetic Algorithm（RCGA） and put forward that locus independent selection strategy should be adopted when performing RCGA crossover operation. According to this strategy a new crossover operator is designed and tested. Secondly, the paper pays great emphasis on the scalability of the GA algorithm when selecting GA strategies, thus the Multiple Population Genetic Algorithm（MPGA） is selected as the framework of the algorithm and two population-level strategies are designed and tested. Combining the new crossover operator and these two strategies together, the paper at last comes to the improved MPGA and it is then applied to the multi-layer gravity inversion of both experimental models and field data of Xinjiang area,China.