基于改进Elman神经网络的光功率预测方法  

Optical power prediction method based on improved Elman neural network

作　　者：陈晓娟[1] 徐梦[2] 赵亮[3] 刘柏良 钱欣[5] 

机构地区：[1]长春理工大学电子与信息工程学院,长春130022 [2]东北电力大学信息工程学院,吉林省吉林市132012 [3]吉林省电力有限公司,长春130021 [4]国网江苏省电力公司经济技术研究院,南京210008 [5]国网江苏省电力公司南京供电公司,南京210019

出　　处：《光通信技术》2016年第8期12-14,共3页Optical Communication Technology

基　　金：国家自然科学基金(61271115)资助

摘　　要：提出了一种改进的Elman神经网络的光功率趋势预测方法。增加了输出层与承接层之间的反馈环节,引入了附加动量和变学习率算法,建立了基于改进Elman神经网络的预测模型。实验表明,BP和传统Elman神经网络相比,改进后的Elman神经网络具有良好的动态性能,训练速率快,实现了对光纤线路状态趋势的预测。In this paper, an optical power trend prediction method based on the improved Elman neural net- work is proposed. The feedback link being added between output layer and context layer, by which a predic- tion model based on modified Elman neural network is established with the additional momentum and vari- able learning speed algorithm.The simulation result shows that the modified model has better dynamic perfor- mance and faster training speed compared with the BP and traditional Elman neural network, which also shows the prediction of the trend of optical fiber line state is realizable.

关 键 词：光功率预测 ELMAN神经网络 反馈环节 训练误差 

分 类 号：TN914.332[电子电信—通信与信息系统]