摘要
针对BP神经网络易陷入局部最优、接近最优解时产生振荡等缺点,利用区间自适应遗传算法自适应移动搜索区间的功能,彻底取代BP神经网络权值阈值反向调整过程。结果表明该方法可使训练误差达到较高的学习精度,有效解决了BP神经网络的困境。将该算法用于大豆油甲酯合成工艺试验中,当训练误差达到2.58×10-8时,神经网络的实际输出值与试验值完全拟合,说明了该方法是一种可靠有效的优化方法。
Since the weights and the thresholds of BP neural networks obtained via existing optimization methods were usually local optimal,these algorithms oscillate upon approaching the optimal solutions.The objective functions of the interval adaptive genetic algorithm were employed to replace the backward adjustment for the weights and thresholds of the BP algorithm.The results revealed that it was able to render training error to higher learning accuracy.Hence,the above difficulties of the BP algorithm could be coped effectively.In addition,the improved algorithm was also applied to optimizing synthesis of soybean oil methyl ester.When the training error was 2.58e-8,the actual output of the neural network was completely fitted to the test value,which showed that the method was a reliable and effective optimization method.
作者
朱会霞
ZHU Hui-xia(School of Management,Liaoning University of Technology,Jinzhou 121001,China)
出处
《辽宁工业大学学报（自然科学版）》
2020年第5期328-331,共4页
Journal of Liaoning University of Technology(Natural Science Edition)
基金
辽宁省教育厅高校科研基金(JQW201715407)。
关键词
区间自适应遗传算法
神经网络
权值阈值
interval adaptive genetic algorithm
neural networks
weights and the thresholds
