【摘要】 提出了一种适用于神经网络控制器在线学习的ＢＰ算法，它用共扼梯度因子λ确定学习的方向，通过奖惩系数γ调整学习步长η，从而具有高速收敛性、快速跟踪能力及较强的鲁棒性。最后通过一个４层前馈ＮＮ网络对标准控制序列进行学习仿真，证实了该算法的优越性更多还原

【Abstract】 A fast varied step and conjugate gradient based BP algorithm (VSCGBP) is presented in this paper. The VSCGBP is a novel on line BP learning algorithm, which uses a conjugate gradient factor λ to determine the learning direction and uses a bonus penalty coeffient γ to adjust the learning step η . This approach guarantees stability, convergence and robustness of the algorithm. The VSCGBP is compared with the conventional BP algorithm by training a four layer neural network with a standard control series in this paper. Simulation results show that the proposed algorithm is outperformed.更多还原