将TD方法同神经网络相结合进行时间序列实时建模预测  

COMBINING THE METHODS OF TEMPORAL DIFFERENCES WITH NEURAL NETWORK FOR REAL TIME MODELLING AND PREDICTION OF TIME SERIES

作　　者：杨璐[1] 洪家荣[1] 黄梯云[1] 

机构地区：[1]哈尔滨工业大学计算机科学与工程系,哈尔滨工业大学管理学院

出　　处：《计算机学报》1996年第9期695-700,共6页Chinese Journal of Computers

摘　　要：本文首先探讨了基于神经网络的时间序列预测模型的建立机制，然后提出可将基于神经网络的时序直接多步预测模型的实时建模问题看成是延时加强学习问题，从而可将ＴＤ法与ＢＰ法相结合用于解决实时建模预测问题．本文对太阳黑子问题和外汇汇率问题进行了实时建模和预测，其结果表明，本文提出的实时建模预测方法是可行的．This paper discusses the construction method of time series prediction models by neural networks. This paper proposes that the problem of real time modeling of time series based on neural networks can be treated as delayed reinforcement learning, so can be solved by combining the temporal differences methods with BP algorithm. The modelling and prediction of the sunspot series and the currency exchange rate that the method is feasible.

关 键 词：神经网络 时间序列 预测 TD方法 线性回归分析 

分 类 号：O212.1[理学—概率论与数理统计] TP18[理学—数学]