摘要
本文针对BP神经网络培训过程中的局部最优解问题提出了一种新的培训方法——GDR-SA法,分别对1个具有13个决策变量的网络和1个具有108个决策变量的催化裂化分馏塔优化操作建模网络进行了培训。结果表明这一方法可以有效地克服局部最优解给网络培训所带来的困扰。
A local optimal solution often occurs in the training process of artificial neural network. To cope with this situation, a new training method of BP net is proposed. The method has been applied to training a net with 13 decision variables and another net for modeling optimizing operation of FCC main fractionator with 108 decision variables. The results show that the method can overcome successfully the problems of local optimal
solution and speed up training process.
出处
《化工学报》
EI
CAS
CSCD
北大核心
1994年第5期573-579,共7页
CIESC Journal
关键词
神经网络
局部最优解
分馏塔
BP
artificial neural network, local optimal solution, decision variable, fractionator
