【作者】 彭斯俊； 杜伟伟；

【Author】 Siju Peng Weiwei Du (School of Sciences, Wuhan University of Technology, Hubei, Wuhan, 430070)

【机构】 武汉理工大学理学院；

【摘要】 针对误差反向传播神经网络模型(BP模型)学习收敛速度慢、容易陷入局部极小点等缺点,本文对BP网络模型进行了改进。对原始数据采用非线性的归一化函数,使其避免取得S型函数的极值点,提出一种更加有效的学习率的改进算法, 提高了网络的收敛速度;采用一种新的权值及阈值初始化方法,以避免训练时误差陷入局部极小的可能性;最后,本文以教学质量的评估为例,将改进算法与传统的BP算法进行比较,验证了该算法的优越性。更多还原

【Abstract】 In accordance with the shortcoming that BP neural network exists, such as the rate of studying is slow, often getting into local least values and so on, this paper has carried on the improvement. We use nonlinear function to normalize the original data to avoid it obtain the pole of sigmoid function, and put forward a more valid method of study rate to improve the speed of network training. We use a new method of weight and bias value initialization to avoid the error getting into the local minimum value when network is training. At last, this paper takes the valuation of the teaching quantity as an example to compare the improved BP learning algorithm with the traditional BP learning algorithm. We can see that the improved BP learning algorithm introduced in this paper is superior to the traditional BP learning algorithm.更多还原