基于粒子群优化神经网络的水下链式机器人直航阻力预报  

Direct Route Drag Prediction of Chain-structured Underwater Vehicle Based on Neural Network Optimized by Particle Swarm Optimization

作　　者：康帅 俞建成[1,2] 张进[1,2] 金乾隆 胡峰[1,2] KANG Shuai;YU Jiancheng;ZHANG Jin;JIN Qianlong;HU Feng(State Key Laboratory of Robotics,Shenyang Institute of Automation,Chinese Academy of Sciences,Shenyang 110016;Institutes for Robotics and Intelligent Manufacturing,Chinese Academy of Sciences,Shenyang 110016;China University of Chinese Academy of Sciences,Beijing 100049)

机构地区：[1]中国科学院沈阳自动化研究所机器人学国家重点实验室,沈阳110016 [2]中国科学院机器人与智能制造创新研究院,沈阳110016 [3]中国科学院大学,北京100049

出　　处：《机械工程学报》2019年第21期29-39,共11页Journal of Mechanical Engineering

基　　金：国家自然科学基金(U1709202);机器人学国家重点实验室自主课题(2019-Z13)资助项目

摘　　要：多单体水下机器人串联组成的水下链式机器人具有航行效率高、稳定性能好、搭载能力强等优势,对其直航阻力的精确预报可实现更有效的运动控制和更合理的动力编组。针对由于水下链式机器人各单体间耦合关系复杂及使用计算流体力学分析阻力耗时较长导致无法快速准确进行阻力预报问题,开展了水下链式机器人直航阻力预报研究。利用计算流体力学分析获得大量输入量(单体数量、航速和单体间间距)与输出量(直航阻力)样本数据,使用BP神经网络建立输入量与输出量模型关系,并通过粒子群算法优化神经网络的初始权值和偏差以改善BP神经网络易陷入局部极值点和过拟合等问题。由大量测试样本的预报结果可知:基于粒子群优化的BP神经网络算法比传统BP神经网络算法预报结果更准确,在给定不同速度和间距测试中均方误差分别降低了2.04×10–5和7.40×10–6;在5单体水下链式机器人以0.25 m/s2的加速度做匀加速运动过程中,基于粒子群优化的BP神经网络模型预报结果的平均相对误差为0.42%,精度较高。试验结果说明所提方法是可行且有效的。The chain-structured underwater vehicle is composed of several autonomous underwater vehicles in series. It has advantages in terms of navigation efficiency, stability, and carrying capacity. The motion can be better controlled and the power can be better organized if its direct route drag can be predicted accurately. Aiming at the problem that the drag prediction of chain-structured underwater vehicle cannot be carried out quickly and accurately because of the complex coupling relationship among the units and the long time-consuming of computational fluid dynamics(CFD) analysis, the research on the direct route drag prediction carried out. A large number of input(number of units, velocity and spacing between units) and output(direct route drag) sample data are obtained by using CFD. BP neural network is used to establish the relationship between input and output. Particle swarm optimization is used to optimize the initial weights and biases of the neural network to improve the problem that BP neural network is easy to fall into local extreme points and over-fitting. The prediction results of a large number of test samples show that the BP neural network algorithm based on particle swarm optimization is more accurate than the traditional BP neural network algorithm, and the mean square error is reduced by 2.04×10–5 and 7.4×10–6 respectively in the tests of given different velocities and spacing. The average relative error of BP neural network model optimized by particle swarm optimization is 0.42% during the uniform acceleration of the 5-unit chain-structured underwater vehicle with an acceleration of 0.25 m/s2. The accuracy of prediction results is high. The experimental results show that the proposed method is feasible and effective.

关 键 词：水下链式机器人 神经网络 粒子群算法 阻力预报 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]