一种基于深度网络的多环境因素降水量预报模型  

A MODEL FOR PRECIPITATION FORCREASTING OF MUTIPLE ENVIRONMENTAL FACTORS BASED ON DEEP NETWORK

作　　者：张鹏程[1] 张雷[1] 王继民[1] 

机构地区：[1]河海大学计算机学院,江苏南京211100

出　　处：《计算机应用与软件》2017年第9期240-245,261,共7页Computer Applications and Software

基　　金：国家自然科学基金项目(61572171);国家科技支撑计划项目(2015BAB07B01)

摘　　要：为了更好地反映区域降水的变化趋势,开展区域降水量预报显得尤为重要。在流域信息时代存在丰富大数据的情况下,提出一种基于DBN(Deep Belief Nets)深度网络降水量预报模型的新方案。该方案通过模拟大脑神经元的多层结构,并使用反向传播网络对整个网络进行微调。模型使用了与每日降水量息息相关的七种环境因素作为输入向量,未来24小时降水作为输出向量,通过在贵州遵义地区的实验证明了模型的有效性,并与现有方法进行了对比实验,结果表明模型具有更好的预测效果。To better reflect the changing trend of regional precipitation,it is particularly important to develop regional precipitation forecast. Therefore,in the case of abundant information in the era of basin big data,a new method of precipitation forecasting model based on DBN（ Deep Belief Nets） depth network is proposed. The scheme simulated the multilayer structure of the brain neurons and adopted the back propagation network to fine tune the entire network.Besides,the model adopted seven environmental factors that were closely related to the daily precipitation as input vectors,and the next 24 hours precipitation as the output vector. The effectiveness of the proposed model is proved by experiments in Guizhou,Zunyi,and compared with the existing methods. We conclude that the model has better prediction results.

关 键 词：降水量预报 深度学习 多元时间序列 大数据 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]