基于递归网络的混沌时间序列预测

为了能对时问序列充分建模，从混沌的慨念入手，将混沌与神经网络相结合，利用人工神经网络的拟合特性，提出了递归网络的混沌时间序列预测方法。给出了递归神经网络预测的基本理论、数学模型、及具体步骤，并通过由杜芬方程所产生的混沌时间序列对该神经网络进行了模拟实验。仿真结果表明，该方法远好于前馈网络的预测效果，其预测误差在10^-15的数量级上。

Chaotic Time Series Prediction Based on Recursive Networks

In order to sufficiently model time series, we began with the conception of chaos, used the fitting ability of neural network, and proposed the method based on recurrent neural network, we introduced the basic theory, math model and predict step of the method. We simulated the method using chaotic time series produced by Duffing equation. Recurrent network is a network with feedback, it can keep information in the net. Though it has more complicated structure and algorithm, it shows more powerful fitting and predicting ability. The results shows that the predicted error is on the magnitude order of 10^-15.