基于混沌天牛群算法优化的神经网络分类模型  

Neural Network Model for Classification Based on Chaotic Beetle Swarm Algorithm

作　　者：王丽 陈基漓[1,2] 谢晓兰 徐荣安[1] WANG Li;CHEN Ji-li;XIE Xiao-lan;XU Rong-an(College of Information Science and Engineering, Guilin University of Technology, Guilin 541004, China;Guangxi Key Laboratory of Embedded Technology and Intelligent Systems, Guilin 541004, China)

机构地区：[1]桂林理工大学信息科学与工程学院,桂林514004 [2]广西嵌入式技术与智能系统重点实验室,桂林514004

出　　处：《科学技术与工程》2022年第12期4854-4863,共10页Science Technology and Engineering

基　　金：国家自然科学基金(61762031);广西科技重大专项(桂科AA19046004);广西重点研发项目(桂科AB18126006)。

摘　　要：针对传统反向传播(back-propagation,BP)神经网络受初始权阈值影响大且易陷入局部极值,标准天牛须搜索算法局部搜索能力差、寻优精度低等问题,提出一种自适应步长因子的混沌天牛群算法用于优化BP神经网络分类模型。通过增加天牛种群,引入自适应步长更新策略优化天牛须搜索算法的局部搜索能力,使其跳出局部最优,提高算法的计算精度;利用Logistic混沌映射产生新个体,替换性能较差的个体,增强全局搜索效果。为了改善BP神经网络对非均衡数据集中少数类的分类效果,采用SMOTE算法处理非均衡数据集。将改进的天牛须搜索算法用于优化BP神经网络中的初始权值和阈值,建立改进的天牛须搜索及反向传播神经网络(improved beetle antennae search and back propagation neural network,IBAS-BPNN)分类模型,提高BP神经网络分类模型的准确率。为验证分类模型的性能,将改进的BP神经网络分类模型与其他6种典型的分类算法进行比较。实验结果表明:IBAS-BPNN分类模型的平均分类正确率高于其他算法。改进的混沌天牛群算法泛化能力强,鲁棒性好,具有一定的优越性。Aiming at the problems of traditional back-propagation(BP)neural network that are greatly affected by the initial weight threshold and easily fall into local extremes,and the standard beetle antennae search algorithm has poor local search ability and low optimization accuracy.An adaptive step factor chaotic beetle swarm algorithm was proposed to optimize the BP neural network classification model.The local search ability of the beetle antennae search algorithm was optimized by increasing the beetle population and introducing an adaptive step size update strategy,so that the algorithm jumped out of the local optimum and improved the calculation accuracy of the algorithm.Logistic chaotic mapping was used to generate new individuals,and individuals with poor performance were replaced,which enhanced the global search effect.In order to improve the classification effect of the BP neural network on the minority classes in the unbalanced data set,the SMOTE algorithm was used to process the unbalanced data set.The improved beetle antennae search and back propagation neural network(IBAS-BPNN)classification model was established to optimize the initial weight and threshold in the BP neural network to improve the accuracy of the BP neural network classification model.In order to verify the performance of the classification model,the improved BP neural network classification model was compared with other six typical classification algorithms.The experimental results show that the average classification accuracy of the IBAS-BPNN classification model is higher than other algorithms.The improved chaotic beetle swarm algorithm has strong generalization ability,good robustness,and certain advantages.

关 键 词：天牛须搜索(BAS)算法 反向传播(BP)神经网络 Logistic混沌 SMOTE算法 分类 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]