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 lo⁃cal 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 tra⁃ditional GEP algorithm, as well as the BP and NAR neural network algorithms.
展开▼
