基于灰狼优化算法的长短期记忆网络在时间序列预测中的应用  

Long short term memory networks based on grey wolf optimizer in time series forecasting

作　　者：王书芹[1,2] 华钢[1] 郝国生[2] 谢春丽[2] 刘海强[1] WANG Shuqin;HUA Gang;HAO Guosheng;XIE Chunli;LIU Haiqiang(School of Information and Control Engineering,China University Xuzhou,Jiangsu 221116, China;School of Computer Science and Technology,Jiangsu Normol University, Xuzhou,Jiangsu 221116,China)

机构地区：[1]中国矿业大学信息控制学院,江苏徐州221116 [2]江苏师范大学智慧教育学院,江苏徐州221116

出　　处：《中国科技论文》2017年第20期2309-2314,共6页China Sciencepaper

基　　金：国家自然科学基金资助项目(51574232;61502212;61673196)

摘　　要：提出了1种基于灰狼优化算法的长短期记忆(long short term memory,LSTM)模型。结合灰狼优化算法全局收敛的优点,将其应用于长短期记忆网络中参数的优化,克服了传统的长短期记忆网络所采用的随时间反向传播(back propagation through time,BPTT)算法易于收敛于局部最优的缺点。将所提出的模型应用于时间序列预测,实验结果表明,其性能优于基于BPTT的LSTM。A longshort termmemorymodelbasedonGreyWolfOptimizer isproposed. Thebackpropagationthrough time for parameters updating in LSTM may converge to local optima.. Since grey wolf optimizer presents the advantages of global conver-gence, in this paper, it is used for optimizing the parameters of the LSTM. Proposed model is applied for time series forecasting and the simulation illustrates that it outperforms BPTT-based LSTM.

关 键 词：时间序列预测 长短期记忆 灰狼优化算法 

分 类 号：O211.61[理学—概率论与数理统计] TP18[理学—数学]