混沌时间序列的神经网络预测研究

研究了一类特殊非线性系统——混沌系统的预测问题。混沌是一种普遍存在的非线性动力学行为,混沌时间序列难以预测和控制,文章先是通过重构系统状态相空间分析混沌时间序列,然后采用多层前向神经网络对其进行预测。对典型的Lorenz和Mackey-Glass混沌序列预测结果表明,如果训练样本足够多,网络结构简单适当,训练后的网络具有很好的泛化性能,说明神经网络预测方法具有较好的工程实用价值。最后分析神经网络初始权值设置对预测性能的影响,指出改进方向。

Chaotic Time Series Forecasting Based on Neural Networks

The forecast of a special nonlinear system-chaotic system is studied in this paper. The chaotic is a kind of wide spread not line dynamics behavior. Facing the problem of difficulty to estimate and control chaotic time series, the theory of state phase space reconstruction was introduced to analyze the chaotic time series, then the multilayer forward neural network was used to forecast these chaotic time series. The results of predicting the typical Lorenz and Mackey-Glass chaotic time series illustrate that, if there is enough train samples and the neural network has compact structure, the trained network would acquire excellent extensive performance. The effect of the network initial weights was also considered, and an improved way was pointed out in the end.