摘要
为研究建立网格化气象预报神经网络模型,对人工智能网格化气象预报可能涉及的气象数据特征进行了讨论,提出采用“多时段、多近邻模式”进行人工智能网格化气象预报的方法;以雷电临近预警预报的神经网络模型研究为例,对该方法进行实例检验:选取闪电、雷达组合反射率、液态水含量、回波顶高等参数,经处理后得到1296个输入维度的神经网络模型。该模型以福建省2016年、2017年历史闪电数据进行训练,测试集准确率最终维持在95%左右;使用福建省2018年5~8月的数据对模型的准确率进行了检验,相比常规预报算法准确率提高了13.9%。结果表明,应用“多时段、多近邻模式”处理网格化气象预报大类别量输出的思路是可行的。
In order to study and establish a neural network model of gridded weather forecasting,this paper discusses the characteristics of meteorological data that may be involved and proposes a method using multi-period and multi-neighborhood model to deal with artificial intelligence gridded weather forecast.Taking the neural network model of lightning warning as an example to test this method.The neural network model has 1296 input dimensions after processing four parameters(the lightning location data,radar combined reflectivity,liquid water content and echo top height).Based on the historical lightning data of Fujian province in 2016 and 2017,the accuracy of the test set is finally maintained at about 95%.The accuracy of the model is tested using lightning data from May to August 2018 in Fujian province,which is higher by 13.9%compared to the accuracy of conventional forecasting algorithms.The results show that it is feasible to apply multi-period and multi-neighbor mode to deal with the output of large categories of gridded weather forecast.
作者
刘冰
张烨方
吴生灿
朱彪
LIU Bing;ZHANG Yefang;WU Shengcan;ZHU Biao(Fujian Meteorological Disaster Prevention Technology Center,Fuzhou 350008,China;Nanping Meteorological Bureau,Nanping 353000,China)
出处
《智能计算机与应用》
2022年第2期187-190,共4页
Intelligent Computer and Applications
基金
福建省科技计划项目(2019Y0063)
灾害天气国家重点实验室开放课题(2021LASW-B07)
关键词
神经网络
网格化气象预报
人工智能
雷电预警
neural network
grid weather forecast
artificial intelligence
thunder warning
