基于多模型预报信息融合的中长期径流预报研究

Study on medium and long-term hydrological forecasting based on data fusion

中长期径流具有非线性时变特点,带有高度的复杂性和不确定性,使用单一算法或模型的预报结果往往不令人满意。因此,本文首先利用GFS降雨预报信息并采用多元线性回归、BP神经网络、季节自回归和新安江模型对浑江桓仁水库流域进行旬径流预报;然后使用自适应联邦滤波算法对四个模型的预报信息进行融合、校正;最后应用结果表明,多模型信息融合能够增强预报的稳定性并提高预报精度,为中长期径流预报方法提供了一定的参考。

This paper presents a concept of fusion information collected from different rainfall-runoff models to produce medium and long-term hydrological forecasts. The objective of data fusion is to enhance forecasting stability and accuracy. We describe a framework of adaptive federated filter algorithm based on the Kalman filtering algorithm, and four models for fusion data collection, i.e. multiple linear regression （MLR） , BP neural network （BP）, seasonal autoregressive （SAR） and Xinanjiang model （XAJ）. The fusion model was validated by simulation and the results suggest that it is a useful model for medium and long-term hydrological forecasting.