BP神经网络算法的一种改进

为了减小标准BP算法中迭代次数并提高其收敛速度，提出了将负梯度下降法与DFP变尺度算法相结合进行权值修正的方法，在误差寻优初期，首先采用标准BP算法进行迭代，每迭代一次的工作量较小、所需存贮量较少，且对初始点的要求不高。当寻优过程开始接近最优时，更改寻优算法，即使用DFP变尺度算法。最后，通过MATLAl3实现。结果表明改进后的BP算法减少了迭代次数，提高了寻优的收敛速度。

Improvement on BP algorithm in artificial neural network

To reduce the iterative time and speed constringency in conventional BP algorithm,a improved weighted algorithm,which combines the method of negative gradient descent with DFP variable scale algorithm,is presented in the paper.At the initial optimization stages,iteration method based on conventional BP algorithm is adopted for the reason of the advantage in lesser iterative workload and memory cell as well as the low demand on initial point.Then,the DFP variable scale algorithm is adopted during the optimization process near optimum point.The MATLAB simulation shows the improved weighted algorithm is of more efficien than conventional BP algorithm.