基于小波-BP神经网络的贝叶斯概率组合预测模型及其应用

Wavelet-BP Neural Network Bayesian Probabilistic Combination Forecasting Model and Its Application

采用一元线性回归模型模拟贝叶斯分析的先验分布和似然函数,建立了基于小波-BP神经网络的贝叶斯概率组合预测模型,在结合区域的水文特性对数据进行预处理的基础上,将其应用于老挝Namngum水库的月径流量预测中,结果表明,该方法较单一BP神经网络模型和小波-BP神经网络模型而言,有效的提高了月径流量的预测精度;同时相对于确定性水文预报方法而言,基于小波-BP神经网络的贝叶斯概率组合预测模型定量地、以分布函数形式描述水文预报的不确定度,能向用户提供更多、更全面的信息,为决策提供更有价值的技术支持。

This paper presents a Wavelet-BP neural network Bayesian probabilistic combination forecasting model by simulating prior distribution and likelihood function with a linear regression .Its model input data is adjusted and it is used to forecast the monthly runoff in the Namngum Reservoir .The results show that this method improves the accuracy of Wavelet-BP neural network model forecasting .Different from deterministic hydrologic forecasting ,Bayesian probabilistic forecasting describes hydrologic forecasting uncertainties by using distribution function and hence it is a feasible way to provide more meaningful information for decision-making .