%0 Journal Article
%T 基于深度学习的短时强降水天气识别<br>Recognition of Short-Time Heavy Rainfall Based on Deep Learning
%A 路志英
%A 任一墨
%A 孙晓磊
%A 贾惠珍
%J 天津大学学报（自然科学与工程技术版）
%D 2018
%R 10.11784/tdxbz201703106
%X 气象预报人员面临的问题之一是如何准确有效地识别短时强降水天气．短时强降水是一种主要由强对流天气形成的气象灾害，产生原因与空气湿度、大气中的水分以及温湿等物理量参数有关，由此提出基于物理量参数和深度学习模型DBNs的短时强降水天气识别模型．首先，利用SMOTE算法人工合成短时强降水少数类(相对于非短时强降水天气类)样本，调整原始数据集不均衡分布问题；然后通过含有高斯玻耳兹曼机的深度学习模型对地面大气监测站逐小时加密的观测量，以及常用于天气预报分析的物理量等低层特征构造出抽象的高层特征，发现数据特征内在关系；最后实现了DBNs短时强降水的自动识别模型．结果表明，该方法能够较为准确地识别短时强降水，对于短时强降水的命中率、误警率和临界成功指数，都有着较好的表现．<br>One of the key studies for meteorological practitioners is how to recognize and predict short-time heavy rainfall accurately and effectively. The short-time heavy rainfall is a severe meteorological disaster that is mainly caused by strong convective weather, which is related to such physical parameters as air humidity, moisture in the atmosphere, temperature and humidity. In this paper, a recognition model of the short-time heavy rainfall based on physical parameters and deep learning model DBNs is constructed. Firstly, SMOTE algorithm is used to synthesize a few samples of the short-time heavy rainfall, which is much less than normal weather, to adjust the distribution of the original data set. Secondly, a deep learning model with a Gaussian Boltzmann machine is constructed based on the observed data from automatic monitoring stations on a local ground and the physical quantities commonly used in weather forecast analysis. Finally, the automatic recognition model of short-term heavy rainfall is obtained. Through the analysis of the experimental results, the model can accurately recognize the short-time heavy rainfall, and have a good performance in the POD, FAR and CSI of short-time heavy rainfall recognition
%U http://journals.tju.edu.cn/zrb/oa/darticle.aspx?type=view&id=201802001