Abstract: Due to the limitation of the fixed mathematical structure, it can not fit all the data; as a result, some experimental results may be discarded in constructing fertilization model, which results in a waste of some data. To solve these problems, the algorithm of fertilization construction model based on co-evolution theory was proposed. The algorithm divided the whole construction model problem into two sub-problems: model structure construction and model parameter optimization. In addition, the two sub-problems were abstracted as co-evolution among various groups. The processes of using genetic programming to construct the model structure and genetic algorithm to optimize the model parameters were carried out in collaboration. The results show that the algorithm not only generates the dynamic model automatically based on the historical experimental data but also has higher accuracy.

Key words: co-evolution, Genetic Programming (GP), Genetic Algorithm (GA), fertilization model, fitness function

摘要： 采用数理统计方法进行施肥模型构造，由于受到固定的数学结构的限制，导致有一些实验结果因不能被模型拟合而被舍弃，造成了一些数据的浪费。针对这些问题，提出了基于协同进化理论的施肥模型构建算法，将模型构建问题分解为模型结构构建与模型参数优化两个子问题，并将这两个子问题抽象成多种群间协同进化。使用遗传规划算法进行模型结构构建，使用遗传算法对模型参数进行优化，两个过程协同进行。实验结果表明，该算法能够在历史实验数据的基础上自动生成动态模型，同时具有较好的准确度。

关键词: 协同进化, 遗传规划, 遗传算法, 施肥模型, 适应度函数