基于BP神经网络的改进算法研究

BP算法使用优化算法中的梯度下降法，梯度下降法的不足，使BP算法收敛速度慢，计算量比较大，且收敛速度与初始权的选择有关；学习时，无法保证可以得到最小值。BP的改进算法提出权值更新的快速收敛方法，使用MatLab对改进算法进行仿真，结果表明改进算法具有高效性和有效性。

Research on the Improved Algorithm Based on BP Neural Network

The BP neural algorithm exploits the decreasing gradient algorithm, the one depriving from the optimized algorithm, that causes a low con- vergence rate and an intensive computation to the BP neural algorithm. It is proved that the convergence rate is related to the choice of the initial right and that, when learned, the minimum cannot be guaranteed. In light of the updated BP algorithm, puts forward a method for rapid convergence rate and tries a simulation of the updated algorithm through MatLab. The results show that the improved algorithm is both highly efficient and effective.