基于驱动分析的LSTM干旱预测模型研究  

Study on Drought Prediction Model of LSTM Based on Drive Analysis

作　　者：李艳玲 巩雅杰 LI Yan-ling;GONG Ya-jie(School of Mathematics and Statistics,North China University of Water Resources and Electric Power,Zhengzhou 450046,China)

机构地区：[1]华北水利水电大学数学与统计学院,河南郑州450046

出　　处：《数学的实践与认识》2022年第5期92-102,共11页Mathematics in Practice and Theory

基　　金：国家自然科学基金(51679189);河南省科技攻关计划(212102310306)。

摘　　要：干旱是世界上影响面最广、造成损失最大的自然现象之一.论文利用1961-2020年黄河流域河南段的逐月气象数据,计算不同时间尺度的标准化降水蒸散指数(SPEI),在Copula熵的基础上根据Hampel准则选择干旱驱动因子,构建多变量长短时记忆(LSTM)神经网络预测模型.结果表明:以驱动分析选择出的水汽压、湿度、温度及降水量作为输入变量集的多变量LSTM模型预测精度较高;预测精度随着SPEI时间尺度的增大而提高,尤其对长期干旱有较好的预测效果;黄河流域河南段的东部地区有发生轻、中度干旱的风险,为相关部门制定防旱措施提供依据.Drought is one of the most widespread and most damaging natural phenomena in the world.This paper calculated the Standardized Precipitation Evapotranspiration Index(SPEI) on diverse time scales using the monthly meteorological data of the Henan section of the Yellow River Basin during 1961-2020.This paper selected the driving factors of drought according to the Hampel criterion based on Copula entropy,constructed the prediction model of multivariate long short-term memory(LSTM) neural network.The results show that the multivariate LSTM model with water vapor pressure,humidity,temperature and precipitation selected by driving analysis as input variables has a higher predictive ability.The prediction accuracy of the model improves with the increase of SPEI time scale,especially for longterm drought.There is a risk of mild to moderate drought in the eastern part of the Yellow River Basin,which provides a basis for relevant departments to formulate drought prevention measures.

关 键 词：LSTM Copula熵 驱动因子 干旱预测 SPEI 

分 类 号：P426.616[天文地球—大气科学及气象学]