通过对传统BP算法的分析，提出了一种改进激励函数的学习方法，并且在神经网络的每一层采用不同的学习速率，以提高训练速度；采用所提出的改进BP算法，训练多层前向神经网络，建立机械手逆运动学模型，仿真结果表明了该算法的有效性；与传统BP算法相比，大大提高了机械手逆运动学的精度。

In this paper, an algorithm in which active function is improved is proposed through analyzing the conventional BP algorithm, and different learning rates are used to increase the learning speed in each layer. The multilayer forward neural networks are used to establish the inverse kinematics models for robot manipulator by this improved BP algorithm. The simulations demonstrate that the proposed method is effective, and improves the inverse kinematics solutions for robot manipulator as compared to the conventional BP algorithm.