BP神经网络算法的改进及应用  

Improvement and application of BPNN algorithm

作　　者：王美玲[1] 王念平[1] 李晓[1] 

机构地区：[1]解放军信息工程大学电子技术学院,郑州450004

出　　处：《计算机工程与应用》2009年第35期47-48,共2页Computer Engineering and Applications

基　　金：现代通信国家重点实验室基金(No.5143603ZDS0601)

摘　　要：经典的BP算法存在陷入局部极小,算法收敛慢的问题。提出了一种改进的BP神经网络算法,在经典BP算法基础上,引入新的参数以调整经典的神经元转换函数,然后把改进算法应用到实际的教学评估中;利用真实数据的分析结果选取了参数的适当值。结果表明,改进后的算法在收敛速率和误差估计等方面有很好的效果,并实现了对教学效果的合理评价。The main weak point of BP algorithm is that the optimal procedure is easilly trapped into local minimum value and the speed of convergence is very slow.A kind of improvement BP neurel network algorithm is proposed in the paper.The classical non-linear functions is improved by introducing a parameter,after that the improved algorithm is applied into the practical teaching evaluation;Carrided on the comparison with the other values,it indicates that the improved algorithm has the very good effect in the aspects of the speed of convergence and error evaluation.In the end,the logical evaluated teaching effeet is reasonably proved by the selecting appropriate value of the parameter based on the results analysis of the real data.

关 键 词：误差反向传播(BP)算法 非线性函数 教学评价 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]