径流序列的相空间重构神经网络预测模型

Neural network forecasting model for phase space reconstruction of runoff series

在水文水资源领域中引入混沌理论,将相空间重构理论与神经网络理论相结合,提出了径流时间序列预测模型.通过相空间重构,把一维径流时间序列拓展为多维序列,而多维序列可挖掘更为丰富的信息,有利于神经网络的训练.研究表明,利用神经网络建模可以较好地解决非线性问题,使预测更符合实际.以汉江石泉水库逐月平均入库径流序列为例,建立了径流时间序列相空间重构与神经网络耦合预测模型,计算结果表明,模型有较高的预测精度.

The chaos theory was introduced into hydrology and water resources field, and a runoff time-series foreeasting model was proposed by combination of the theories of phase space reconstruction with neural network. One-dimensional runoff time-series was developed to multi-dimensional runoff time-series by reconstruction of phase space. As the multidimensional runoff time-series involves the ergodie information, it can provide abundant information, which is favorable for ANN training. The modeling with ANN can effectively solve the nonlinear problems, and make the result of forecasting in good accordance with the reality. As a ease study of the monthly average runoff into the Shiquan Reservoir of the Hanjiang River, a phase space reconstruction and neural network combined runoff time-series forecasting model was established, and the resuh shows that the model is of high precision of forecasting.