摘要： 在特征选择中，特征子集的优化结果影响分类识别的正确率。简单遗传算法存在早熟收敛和局部搜索能力弱的缺陷。在进化过程中，保持合适的个体选择压力，抑制种群多样性的快速下降，是提高遗传算法性能的关键。该文提出一种新的自适应约束惩罚措施，应用相关联赛选择和相关家庭竞争算子对基本遗传算法进行改进，并用于通信信号调制特征选择。仿真结果表明，该算法的收敛性和稳定性均有显著提高。

关键词: 遗传算法, 特征选择, 种群多样性, 选择压力, 约束惩罚

Abstract: It is the optimization effect of feature subset that always affects the accuracy of classification. Simple Genetic Algorithm(SGA) has the drawbacks of premature convergence and poor local searching ability. The crucial problem for improving SGA’s performance lies on how to solve the conflict between the population diversity and the individual selection pressure. This paper proposes an Improved Genetic Algorithm(IGA) in which adaptive constraint penalty is proposed while two operators, the Correlative Tournament Selection(CTS) and the Correlative Family Selection(CFS) are employed. Simulation results show that the performance of the new algorithm is much better than that of SGA when both are applied to feature selection in automatic modulation recognition.

Key words: genetic algorithm, feature selection, population diversity, selection pressure, constraint penalty

中图分类号: 

• TP301