基于相空间重构和GA-BP网络的径流预测

将相空间重构理论与神经网络以及遗传算法相结合提出了径流时间序列预测模型,通过相空间重构将一维径流时间序列拓展为多维序列,挖掘了更为丰富的信息,反映出系统的非线性特征,有利于神经网络建模和训练。研究表明,基于相空间重构理论的遗传算法和BP神经网络组合模型可较好地解决径流预测。以深圳宝安铁岗水库月径流为例,采用小波消噪对数据预处理,利用遗传算法训练BP神经网络,计算结果表明模型具有较高的预测精度。

Runoff Forecasting Based on Phase Space Reconstruction and GA-BP Neural Network

The runoff time series forecasting model is proposed in this paper by combining the theory of phase space reconstruction,Neural Network and genetic algorithm.One-dimensional runoff time series is expanded to multi-dimensional runoff time-series by reconstruction of phase space,which embodies abundant information to reflect nonlinear characteristics of the system and is useful for ANN training.The study shows that the proposed combination model can solve runoff forecasting problem effectively.It takes the monthly runoff of the Tiegang Reservoir in Baoan,Shenzhen as a case study,which the wavelet filtered noise is used to pre-process data and the BP neural network is trained by GA.The result demonstrates the model has high precision for runoff forecasting.