摘要
针对现有移动机器人路径优化算法存在的迭代效率低、路径规划距离长等问题,提出一种基于改进Hopfiled神经网络的机器路径优化算法;在世界坐标系内对移动机器人空间运动进行了研究,掌握移动机器人不同时刻的位置信息和移动信息;构建Hopfiled神经网络模型,并采用感知机优化Hopfiled神经网络模型的结构,提升其数据训练能力;同时利用LSTM网络的门控结构替代原网络隐含层的神经元,引入遗忘门、输入门和输出门,提升Hopfiled神经网络的泛化学习能力和样本容纳能力;引入路径评价函数,评价局部区域内的碰撞风险以降低移动机器人之间的碰撞概率;经实验测试得出:改进Hopfiled神经网络模型路径规划均值为104.3 m,耗时均值为122.1 s,随机提取采样点的方差值仅为0.01,显著低于其他的传统路径优化算法。
Aiming at the problems of existing mobile robot path optimization algorithms,such as low iteration efficiency and long path planning distance,a machine path optimization algorithm based on improved Hopfiled neural network is proposed.The space motion of the mobile robot is studied in the world coordinate system,and the position and movement information of the mobile robot at different times are mastered.The Hopfiled neural network model is constructed,and the perceptron is used to optimize the structure of the Hopfiled neural network model to improve the ability of its data training.At the same time,the gating structure of the Long Short-Term Memory(LSTM)network is used to replace the neurons in the hidden layer of the original network,and the forgetting gates,input gates and output gates are introduced to improve the generalization learning and sample holding ability of the Hopfiled neural network.The path evaluation function is introduced to evaluate collision risk in local area to reduce collision probability between mobile robots.Experimental results show that the path planning average value of the improved Hopfiled neural network model is 104.3 m,the average time is 122.1 s,and the variance value of random sampling points is only 0.01,which is significantly lower than other traditional path optimization algorithms.
作者
黄海龙
蔡娟
刘源
HUANG Hailong;CAI Juan;LIU Yuan(College of Information Engineering,Guangzhou Vocational and Technical University of Science and Technology,Guangzhou 510000,China;School of Electronic and Optical Engineering,Nanjing University of Science and Technology,Nanjing 210094,China)
出处
《计算机测量与控制》
2024年第11期204-210,共7页
Computer Measurement &Control
