计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (22): 142-147.DOI: 10.3778/j.issn.1002-8331.2005-0357

• 模式识别与人工智能 • 上一篇    下一篇

基于改进量子免疫算法的神经网络集成

曹林,王之腾,陈亮,李洪顺,高申,张自立   

  1. 1.解放军陆军工程大学,南京 210007
    2.军事交通学院 汽车士官学校,安徽 蚌埠 233011
    3.中国人民解放军 31697部队
    4.中国人民解放军 65370部队
  • 出版日期:2020-11-15 发布日期:2020-11-13

Neural Network Ensemble Based on Improved Quantum Immune Algorithm

CAO Lin, WANG Zhiteng, CHEN Liang, LI Hongshun, GAO Shen, ZHANG Zili   

  1. 1.Army Engineering University of PLA, Nanjing 210007, China
    2.Automobile NCO Academy, Army Military Transportation University, Bengbu, Anhui 233011, China
    3.Unit 31697 of PLA, China
    4.Unit 65370 of PLA, China
  • Online:2020-11-15 Published:2020-11-13

摘要:

针对量子免疫算法在神经网络集成结论生成时存在精英损失和过早收敛的问题,提出了改进量子免疫算法。改进算法在免疫选择时采用精英策略保留最优个体,提升了收敛效率,并引入反转策略增加个体多样性,加强了全局搜索能力。仿真实验结果表明,改进量子免疫算法是集成结论优化的有效方法,泛化性能明显优于简单平均、推广集成等传统方法。

关键词: 精英策略, 反转策略, 量子免疫算法, 神经网络集成

Abstract:

Aiming at the problems of elite loss and premature convergence when neural network ensemble generates its conclusion, an improved quantum immune algorithm is proposed. It keeps the best individuals by elitism strategy to improve efficiency and adds the diversity of the individuals by contrarian strategy to strengthen global searching ability. The simulation results show that the improved quantum immune algorithm is a valid optimization method of ensemble conclusions with better generalization ability than traditional methods such as basic ensemble method and generalized ensemble method.

Key words: elitism strategy, contrarian strategy, quantum immune algorithm, neural network ensemble