基于优化RBF神经网络的并联机器人位姿检测

Pose detection of parallel robot based on optimized RBFNN

针对采用遗传算法(GA)学习训练RBF神经网络进行并联机器人位姿检测存在的早熟收敛问题,提出一种交替使用GA和Levenberg-Marquart(LM)算法的混合学习算法,首先运用仅需预先给定基函数宽度的最近邻聚类算法,学习训练得出基函数的中心,然后运用所提出的混合学习算法学习网络,确定RBF神经网络的参数,并最终实现对机器人的位姿检测。仿真结果表明,所设计算法提高了网络学习能力,并最终提高了机器人位姿检测精度。

In order to overcome the insufficiency of premature convergence when using RBF Neural Network（RBFNN）,which is optimized by Genetic algorithm（GA） to detect the pose of parallel robot.This paper proposes a new hybrid study algorithm,which uses the GA and Levenberg-Marquart（LM） to optimize the neural network＇s parameters in turn,improve the network＇s learning capability,and improve the detection accuracy of pose of parallel robot.First,using the nearest neighbor-clustering algorithm,which only needs to present radius of basis function,obtains the center vector of basis function.Then using the above hybrid study algorithm to learn and optimize the neural network,fix the RBF neural network＇s parameters,and realize the pose detection of the robot finally.The simulation result shows that the designed algorithm improves the learning ability of the network,and increases the accuracy of pose detection.