基于BP神经网络的概率径流预测模型  

Probabilistic runoff forecasting model based on BP artificial neural network

作　　者：周娅[1] 郭萍[1,2] 古今今[2] 

机构地区：[1]中国农业大学水利与土木工程学院,北京100083 [2]北京师范大学环境学院生态模拟与城市生态研究所,北京100875

出　　处：《水力发电学报》2014年第2期45-50,共6页Journal of Hydroelectric Engineering

基　　金：国家自然基金面上项目(71071154);水利部公益项目(201001060-02);水利部公益项目(201001061-3)

摘　　要：本文采用多元线性回归模型模拟贝叶斯分析的先验分布和似然函数,并结合反向传播神经网络(Back Propagation Neural Network)建立基于BP神经网络的贝叶斯概率径流预测模型,将模型应用于石羊河出山口六河水系的年径流预测中。为降低BP神经网络的"黑箱"特性对预测精度的影响,在实例应用中结合了区域的水文特性对数据进行预处理,结果表明该方法有效的提高了模型的预测精度;同时相对于确定性水文预测方法而言,贝叶斯概率水文预报定量地、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。his paper presents a Bayesian probabilistic forecasting model based on BP neural network. This model simulates prior distribution and likelihood function with multivariate linear regression. It was used to forecast the annual runoff in the Liuhe river system of the Shiyanghe River basin and its model input data was adjusted according to the hydrological characteristics of the study area to reduce the errors caused by the black box behaviors of BP neural network. The results show that this method improves the model forecasting accuracy. Different from deterministic hydrologic forecasting, Bayesian probabilistic forecasting describes hydrologic forecasting uncertainties using distribution function and hence it is a feasible method that provides more meaningful information for decision making.

关 键 词：水资源 BP神经网络 贝叶斯分析 六河水系径流量 多元回归 

分 类 号：TV124[水利工程—水文学及水资源]