针对常规BP算法收敛速度慢和易陷入局部极小的问题,文章提出了一种新的BP算法———SASSFBP算法。该算法根据训练中最近两个梯度的符号及其相对大小来动态地改变权步因子,提高了神经网络的收敛速度,并同时结合模拟退火算法来避免陷入局部极小。仿真实验结果表明:SASSFBP算法在收敛速度与运算精度,以及避免陷入局部极小的能力等方面均明显优于常规的BP算法。 Aiming at the problem of slow convergence and easy falling into local minimum for conventional BP algorithm, a new BP algorithm called SASSFBP is proposed. The algorithm dynamically changes the weight step factor according to the sign and its relative size of the two latest gradient in training, and improves the convergence speed of the neural network, meanwhile, it avoids falling into a local minimum by combining the simulated annealing algorithm. The simulation results show that the SASSFBP algorithm outperforms the conventional BP algorithm in terms of convergence speed and precision, as well as the ability to avoid falling into local minima.