基于神经网络的天气发生器  

A Weather Generator Based on Neural Network

作　　者：朱思远 张宸豪 吴邪 耿耀君[1] 杨龙[1] ZHU Si-yuan;ZHANG Cheng-hao;WU Xie;GENG Yao-jun;YANG Long(College of Information Engineering,Northwest A&F University,Xianyang 712100)

机构地区：[1]西北农林科技大学信息工程学院

出　　处：《现代计算机》2019年第18期3-5,23,共4页Modern Computer

基　　金：国家青年基金项目(No.61702422);中央科研基本业务费项目(No.2452017339)

摘　　要：提出一种基于BP神经网络的天气发生器模型,可以根据12个月的月平均天气数据(包括日照强度、最高气温、最低气温、风速、露点温度、降水量),模拟生成365天的逐日天气数据。使用河南省南阳气象站从1990年到2004年的实测数据训练网络模型,并使用该地区2005年到2009年的数据进行模型验证。各天气变量的逐日模拟数据与实测数据在365×5天内的均方根误差,分别为6.310、5.638、5.743、0.951、7.481、9.172;平均绝对百分比误差分别为41.9%、23.6%、7.7%、41.9%、92.9%、208.2%。实验结果表明:除降水量的表现不好之外,该天气发生器模型可以很好地生成接近真实分布的模拟天气数据。Introduces a weather generator model based on BP neural network, which can simulate 365-day daily weather data according to 12-month average weather data (including sunshine intensity, maximum temperature, minimum temperature, wind speed, dew point temperature, precipitation). The network model is trained with the measured data of Nanyang Meteorological Station in Henan Province from 1990 to 2004, and validated with the data of 2005 to 2009. The Root Mean Square Error (RMSE) of daily simulated and measured data for each weather variable in 365×5 days is 6.310, 5.638, 5.743, 0.951, 7.481 and 9.172, respectively;the Average Absolute Percentage Error (MAPE) is 41.9%, 23.6%, 7.7%, 41.9%, 92.9%, 208.2%, respectively. The experimental results show that, except for the poor performance of precipi? tation, the weather generator model can generate simulated weather data close to the real distribution.

关 键 词：BP神经网络 天气发生器 随机模拟 

分 类 号：P4[天文地球—大气科学及气象学] TP183[自动化与计算机技术—控制理论与控制工程]