西安电子科技大学学报 ›› 2019, Vol. 46 ›› Issue (5): 120-127.doi: 10.19665/j.issn1001-2400.2019.05.017

• • 上一篇    下一篇

Tent混沌和变邻域局部搜索优化的GSA

娄奥1,姚敏立1,贾维敏2,袁丁1   

  1. 1. 火箭军工程大学 作战保障学院, 陕西 西安 710025
    2. 火箭军工程大学 核工程学院, 陕西 西安 710025
  • 收稿日期:2019-05-15 出版日期:2019-10-20 发布日期:2019-10-30
  • 作者简介:娄 奥(1995—),男,火箭军工程大学硕士研究生,E-mail:la5310@qq.com.
  • 基金资助:
    国家自然科学基金(61179004);国家自然科学基金(61179005)

Tent chaos and variable neighborhood local search optimized GSA

LOU Ao1,YAO Minli1,JIA Weimin2,YUAN Ding1   

  1. 1. School of Military Operational Support, Rocket Force Univ. of Engineering, Xi’an 710025, China
    2. School of Nuclear Engineering, Rocket Force Univ. of Engineering, Xi’an 710025, China
  • Received:2019-05-15 Online:2019-10-20 Published:2019-10-30

摘要:

针对引力搜索算法存在的易陷入局部最优、精度有待提高等问题,提出一种Tent混沌和变邻域局部搜索优化的引力搜索算法。首先改进Tent混沌,利用其遍历均匀性、随机性初始化种群,增强算法的全局搜索能力;然后改进粒子速度和引力系数公式,加快算法的收敛速度;最后设计一种基于莱维飞行的变邻域局部搜索策略,引导种群脱离局部最优,提高寻优精度。仿真结果显示,新算法能有效地抑制局部最优,相较其他测试算法有更好的寻优精度和稳定性。利用新算法优化径向基函数神经网络,对非线性系统的辨识结果证明,改进后的径向基函数神经网络比标准径向基函数神经网络和反向传播神经网络具备更好的模型逼近能力和泛化水平。

关键词: 引力搜索算法, 混沌, 局部搜索, 神经网络, 系统辨识

Abstract:

An improved gravitational search algorithm (GSA) optimized by Tent chaos and Variable neighborhood Local search (TVL-GSA) is proposed to overcome the problem of easily falling into local optimum and defect of improving accuracy. First, tent chaos is improved to initialize the population and enhance the global search ability of the algorithm by using its ergodic uniformity and randomness; second, the particle speed and gravity coefficient formulas are improved to accelerate the convergence speed; third, a variable neighborhood local search strategy based on Levy flight is designed to guide the population to escape from local optimum and improve search accuracy. Simulation results show that the new algorithm can effectively inhibit the local optimum and has a better optimization accuracy and stability than other test algorithms. The new algorithm is used to optimize the radial basis function neural network (RBFNN). The identification results of the nonlinear system show that the improved RBFNN has a better model approximation ability and generalization level than the standard RBFNN and back propagation neural networks (BPNN).

Key words: gravitational search algorithm, chaotic, local search, neural network, system identification

中图分类号: 

  • TP18