Abstract: The Multi Artificial Fish Warm Algorithm（MAFWA） based on the work principle of basic Artificial Fish Warm（AFW） is presented for extracting the classification rules of continuous variable. The code of artificial fish is designed in terms of the characteristics of extracting the classification rules. Then the fitness function to evaluate the quality of the regular rule is established and some formulas to calculate some key parameters for its application in extracting classification rules are defined. Meanwhile, in order to avoid the MAFWA falling in the local optima, the crossover operator and mutation operator of the AFW are designed based on the mutation and crossover idea of Genetic Algorithm（GA）. Then the MAFWA with Cross and Mutation（MAFWA_CM） is proposed. At last, the algorithm is tested on the Iris and Wine data sets. The experimental results show: （1）the algorithm can extract the classification rules with high precision in a short time. （2）When referred to the efficiency of the convergence and the precision of the rule, the MAFWA_CM is superior to the single AFW and is closer to the multi particle swarm algorithm.

Key words: Multi Artificial Fish Warm Algorithm（MAFWA）, Cross and Mutation（CM）, classification rule

摘要： 在基本人工鱼群算法的基础之上构建了用于解决连续变量空间分类规则提取的多群体人工鱼群算法，根据分类规则提取问题的特性设计了人工鱼的编码规则，并在此编码基础上定义了进行规则评价的适应值函数以及相关状态更新公式。为克服人工鱼群算法易陷入局部最优解的缺陷，引入了遗传算法中的交叉变异思想，设计了基于人工鱼的交叉及变异算子，提出了利用多种群交叉变异人工鱼群算法生成分类规则的算法思想。利用Iris和Wine数据集作为测试数据，结果表明：（1）该算法能够快速生成精度较高的分类规则；（2）在收敛效率及规则精度上全面优于基本多群体人工鱼群算法，并达到了多群体微粒群算法的性能水平。

关键词: 多群体人工鱼群, 交叉变异, 分类规则