长江上游月降水人工神经网络预测模型

ARTIFICIAL NEURAL NETWORK MODELS FOR FORECASTING MONTHLY PRECIPITATION IN THE UPPER YANGTZE RIVER

长江上游月降水量预测对于三峡库区及整个长江流域水资源管理具有重要意义。根据长江上游不同气候区降水差异,选取玉树、九龙和宜宾3个代表性气象站点近60a的月降水量数据,运用反向传播神经网络、径向基函数神经网络、广义回归神经网络和多元线性回归法,确定降水时滞和降水月份,建立月降水预测模型,来预测未来一个月的降水量,并采用均方误差和判定系数来验证和对比各种模型的模拟效果。结果显示:人工神经网络模型总体上优于多元线性回归,特别是反向传播神经网络的模拟结果各站表现较好,在确定合理的输入变量和网络结构后,可以尝试作为长江上游各站月降水预测模型。

Monthly precipitation forecast for the upper Yangtze River is very essential to the water resources management for the entire Yangtze River basin.Three typical meteorological stations were selected respectively in three different climatic zones.All the selected stations contained nearly 60 years of monthly precipitation records in the upper Yangtze River.This paper estimated the month of precipitation and precipitation time delay parameter,and established monthly precipitation forecasting model using back-propagation neural network,radial basis function neural network,generalized regression neural network and multiple linear regression method respectively,to predict the precipitation of coming month.Then,the mean square error and coefficient of determination were used to verify the simulation accuracy of various models and the model simulation results.The results show that artificial neural network prediction model is superior to multiple linear regression in general.Especially,the performance of the back-propagation neural network is better than the others.It can be determined as an effective monthly precipitation methods for the upper Yangtze River after determining reasonable input variables and network structure.