基于学习型多策略改进鲸鱼算法的路径规划研究  

Path Planning Based on Reinforcement Learning Multi-Strategy Improved Whale Optimization Algorithm

作　　者：岳凡[1] 艾尔肯·亥木都拉 刘拴 YUE Fan;AIERKEN Haimudul;LIU Shuan(School of Smart Manufacturing Modern Industrial,Xinjiang University,Urumqi 830047,China)

机构地区：[1]新疆大学智能制造现代产业学院,乌鲁木齐830047

出　　处：《组合机床与自动化加工技术》2025年第2期46-51,56,共7页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金资助项目(52265039)。

摘　　要：为解决机器人在路径规划中路径过长与后期寻优停滞的问题,提出了一种学习型多策略改进鲸鱼优化算法(reinforcement learning multi-strategy improvement whale optimization algorithm,RLMIWOA),并在欧式距离的基础上引入了障碍物信息与拐点信息,构建了路径规划适应度函数。首先,引入自适应帐篷映射初始化,使得初始化种群更加均匀;其次,引入了非线性收敛策略平衡算法的开发和探索阶段;然后,通过采用非线性加权因子对最优个体进行扰动,避免了其他个体对最优个体的“盲从”;最后,通过采用强化学习结合ε-精英逐维反向学习策略和动态局部最优逃生策略,提高了算法的收敛效率和跳出局部最优的能力。实验结果表明:RLMIWOA算法可以高效地找到最优路径,在路径搜索方面具有显著的优势。To address the issue of excessive path length and later optimization stagnation in robot path planning,a learning-based multi-strategy improvement whale optimization algorithm(reinforcement learning multi-strategy improvement whale optimization algorithm,RLMIWOA) was proposed.Additionally,obstacle and turning point information were introduced on top of the Euclidean distance,leading to the construction of a path planning fitness function.The population was initially initialized using adaptive tent mapping to achieve a more uniform distribution.Subsequently,the development and exploration stages of a non-linear convergence strategy balancing algorithm were introduced.Further,perturbation of the optimal individual was achieved by employing non-linear weighting factors,thereby preventing other individuals from blindly following the optimal one.Finally,a combination of reinforcement learning with ε-elite per-dimension reverse learning strategy and dynamic local optimal escape strategy was employed to enhance the algorithm′s convergence efficiency and ability to escape local optima.The experimental results indicate that the RLMIWOA algorithm efficiently discovers optimal paths,demonstrating a significant advantage in path searching.

关 键 词：路径规划 强化学习 鲸鱼优化算法 适应度函数 局部最优 

分 类 号：TH166[机械工程—机械制造及自动化] TG659[金属学及工艺—金属切削加工及机床]