«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

34

期数:

2024年05期

页码:

117-125

栏目:

人工智能

出版日期:

2024-05-10

文章信息/Info

Title:

Improved Snake Optimizer of Multi-strategy

文章编号:

1673-629X(2024)05-0117-09

作者:

权浩迪; 刘勇国*; 傅翀; 朱嘉静; 张云; 兰刚; 李巧勤

电子科技大学 信息与软件工程学院,四川 成都 610054

Author(s):

QUAN Hao-di; LIU Yong-guo*; FU Chong; ZHU Jia-jing; ZHANG Yun; LAN Gang; LI Qiao-qin

School of Information & Software Engineering,University of Electronic Science & Technology of China,Chengdu 610054,China

关键词:

蛇优化算法; 启发式算法; 优化问题; 多策略改进; 神经网络

Keywords:

snake optimizer; heuristic algorithm; optimization problem; multi-strategy improvement; neural network

分类号:

TP181

DOI:

10.20165/j.cnki.ISSN1673-629X.2024.0049

摘要:

为改进蛇优化算法(Snake Optimizer,SO)在探索方式、变量计算、空间搜索方式和种群更新方式等方面存在的不足,提出了一种多策略改进的蛇优化算法(Improved Snake Optimizer,ISO)。 首先,提出探索寻优策略,根据个体相对于优势个体的位置更新自身的位置,使种群在前期快速收敛到最优解附近。 其次,优化变量计算方式,将 SO 算法中的指数运算改进为多项式运算,提高 SO 的时间效率。 同时引入动态调整搜索空间的机制,随种群进化迭代次数的增加逐步扩展搜索范围以提高寻优能力。 最后,引入优势进化策略,淘汰适应度较差的个体并结合优势个体的基因产生新个体,快速提高种群优势基因比例以增加收敛速度。 对不同基准测试函数进行寻优实验,分别与经典 SO 算法和 5 种启发式算法进行对比,结果表明 ISO 具有较强的寻优能力。 为进一步验证所提算法的高效性和实用性,将 ISO 应用于全连接神经网络的优化问题,结果表明基于 ISO 优化的神经网络具有更优的分类效果。

Abstract:

An Improved Snake Optimizer (ISO) of multi-strategy is proposed to address the limitations of the Snake Optimizer (SO) in exploration strategy, variable computation, space searching and population updating. Firstly, an optimized exploration strategy is proposed,where individuals update their positions based on their relative positions to the best individual. This allows the population to quickly converge to the vicinity of the optimal solution in the early stage. Secondly,variable computation is optimized by replacing the exponential operations in the SO with polynomial operations to improve the time efficiency of SO.Additionally, we introduce the dynamic adjustment mechanism of the search space, gradually expanding the search range with the increase in population evolution iterations to enhance the optimization capability. Finally,an advantage evolution strategy is introduced,which eliminates individuals with lower fitness and combines the genes of dominant individuals to generate new individuals. This strategy accelerates convergence by rapidly increasing the proportion of dominant genes in the population.Experiments were conducted on different benchmark test functions,comparing ISO with the classical SO and five heuristic algorithms. The results demonstrate that ISO exhibits strong optimization capabilities. To further validate the efficiency and practicality of the proposed algorithm,ISO is applied to the optimization problem of fully connected neural networks. The results show that neural networks optimized based on ISO achieve superior classification perform-ance.

常用功能

更新日期/Last Update: 2024-05-10