个体最优共享GEP算法及其气象降水数据预测建模  

Individual optimal sharing GEP algorithm and its application in forecast modeling of meteorological precipitation

作　　者：彭昱忠[1,2] 元昌安[1] 李洁[3] 许明涛[1] 陈冰廉[1] 

机构地区：[1]广西师范学院计算机与信息工程学院,广西南宁530021 [2]广西师范学院北部湾环境演变与资源利用教育部重点实验室,广西南宁530001 [3]广西科技师范学院数计系,广西柳州545004

出　　处：《智能系统学报》2016年第3期401-409,共9页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金项目(61562008;41575051);广西科学研究与技术开发计划项目(1598019-1);广西高校科学技术研究重点项目(ZD2014083)

摘　　要：针对基因表达式编程算法存在进化后期收敛慢且容易陷入局部最优而降低其数据建模的性能问题,和降水量因受诸多自然因素相互影响而难以准确地建模与预测的问题,提出了一种改进的基因表达式编程算法。该算法具有染色体最优状态记忆功能,在进化过程中可以按条件学习自身的历史经验知识,以加强局部搜索能力和促进收敛,同时尽量控制个体的趋同化而保持种群的多样性。3组不同区域和不同类型的真实降水数据集的实验验证了其可以改善传统GEP算法后期收敛慢的问题,寻优能力更强,降水数据拟合和预测效果均显著优于传统GEP算法、BP神经网络和NAR神经网络等算法。Gene expression programming （ GEP ） is characterized by slow convergence and ease of falling into a local optimum in the later stages of its evolution. Many methods are difficult to model and use to accurately forecast precipitation because of the simultaneous influence of many natural factors. In this paper, we propose an improved GEP algorithm, which has an optimal state memory function, can learn from historical experience in the process of evolution to strengthen the local search ability, and can thus promote convergence and, at the same time, control the convergence of individuals and maintain the diversity of the population. The experimental results of three groups from different regions and different actual precipitation data sets show that the proposed algorithm can improve the slow convergence problem of the traditional GEP algorithm and has better search ability. Experimental results also show that the proposed algorithm＇s ability to fit and forecast precipitation data is significantly better than that of traditional GEP algorithm, as well as the BP and NAR neural network algorithms.

关 键 词：基因表达式编程 经验共享 时间序列 气象建模 降水预测 演化计算 演化建模 

分 类 号：P457.6[天文地球—大气科学及气象学] P409