基于差分进化生物地理学优化的多层感知器训练方法  

Multi-layer perceptron using hybrid differential evolution and biogeography-based optimization

作　　者：王娟[1,2] 吴宪祥[2] 曹艳玲[2] 

机构地区：[1]西安翻译学院工程技术学院,西安710105 [2]西安电子科技大学空间科学与技术学院ICIE研究所,西安710071

出　　处：《计算机应用研究》2017年第3期693-696,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61105066);中央高校基本科研业务费专项资金资助项目(JB141305)

摘　　要：针对生物地理学优化训练多层感知器存在的早熟收敛以及初始化灵敏等问题,提出一种基于差分进化生物地理学优化的多层感知器训练方法。将生物地理学优化(biogeography-based optimization,BBO)与差分进化(differential evolution,DE)算法相结合,形成改进的混合DE_BBO算法;采用改进的DE_BBO来训练多层感知器(multi-layer perceptron,MLP),并应用于虹膜、乳腺癌、输血、钞票验证四类数据分类。与BBO、PSO、GA、ACO、ES、PBIL六种主流启发式算法的实验结果进行比较表明,DE_BBO_MLP算法在分类精度和收敛速度等方面优于已有方法。The problems of premature convergence and initialization-sensitive are often experiencing when train the multi-layer perceptron using the biogeography-based optimization. This paper proposed a novel multi-layer perceptron training method using hybrid differential evolution and biogeography-based optimization. This paper introduced the differential evolution to the biogeography-based optimization to construct the hybrid DE_BBO algorithm and then used the hybrid DE_BBO algorithm for training MLPs. In order to investigate the efficiencies of DE_ BBO in training MLPs,this paper employed four classification datasets,including the Iris dataset,the breast cancer dataset,the blood transfusion datasets and the banknote authentication dataset. Comparing with six well-known heuristic algorithms,including BBO,PSO,GA,ACO,ES,and PBIL in a statistically significant way,the experimental results show that training MLPs using hybrid DE_BBO is significantly better than the current heuristic learning algorithms in terms of convergence speed and convergence accuracy.

关 键 词：生物地理学优化 差分进化 多层感知器 数据分类 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]