基于BP神经网络的混沌时间序列预测方法及应用研究

神经网络在时间序列的预测中得到广泛的应用,但神经网络模型的输入层神经元个数的选取仍然没有一个明确的解析式来表达.为解决这个问题,在非线性动力系统中,根据混沌理论重构相空间,通过最大Lyapunov指数判定时间序列是否存在混沌现象,存在则通过G-P算法计算出混沌吸引子的关联维数,进而获得相空间的嵌入维数作为神经网络的神经元个数.通过上述方法对铝现有价格进行建模,验证该方法对时间序列的短期预测有较好的精度,在此基础上,对未来一段时间铝价格进行预测.

The Research of Prediction Methods and Application of Chaotic Time Series Based on BPNN

Neural network is widely used in time series forecasting, but still not a clear analytical expression for how to choose the number of the input layer neurons of the neural net- work model. To solve this problem, the paper reconstructs phase space, and determines the series is chaotic time series or not by Lyapunor index in the nonlinear dynamical systems. The correlation dimension of chaotic attractor is calculated through the G-P algorithm if the series is chaotic time series. Then embedding dimension of phase-space is obtained as the number of neurons in neural networks. Testing aluminum price data proves above-men- tioned method has good accuracy for short-term forecasting of time series. Then the paper forecasts aluminum price in the future by the method.