基于BVMD-Attention-GRU的中长期干旱预测
Mid-Long Term Drought Prediction Based on BVMD-Attention-GRU
谢颂扬 1马廷淮2
作者信息
- 1. 南京信息工程大学计算机学院, 江苏 南京 210044
- 2. 南京信息工程大学计算机学院, 江苏 南京 210044;江苏海洋大学计算机工程学院, 江苏 连云港 222005
- 折叠
摘要
干旱是一种由长期缺水导致的现象,及早发现干旱现象并预测其程度,对于科学防旱抗旱至关重要.为此,提出一种基于变分模态分解算法(VMD)和融合注意力机制(Attention)的门控循环单元(GRU)的干旱指数预测方法.首先使用蝴蝶优化算法(BOA)对 VMD进行参数寻优,将标准化降水蒸散发指数(SPEI)数据分解为一组波动性较小的子序列;然后将注意力机制引入 GRU模型,对各子序列进行预测;最后将各子序列预测结果加和得到SPEI预测值.使用BVMD-Attention-GRU模型对乌鲁木齐市SPEI进行预见期为 6 个月的中长期预测,并构建 GRU、VMD-GRU、BVMD-GRU 模型进行对比试验.试验结果表明,BVMD-Attention-GRU模型具有更高的预测精度,适用于中长期干旱预测.
Abstract
Drought is a phenomenon caused by long-term water shortage.Early detection and prediction of drought are crucial for scientific drought prevention and control.To address this,a drought index prediction method based on var-iational modal decomposition algorithm(VMD)and gated recurrent unit(GRU)with attention mechanism is proposed.Firstly,the butterfly optimization algorithm(BOA)is used to optimize the parameters of VMD,which decomposes the standardized precipitation evapotranspiration index(SPEI)data into a set of subsequences with smaller volatility.Then,attention mechanism is introduced into the GRU model to predict each subsequence.Finally,the predicted values of each subsequence are summed up to obtain the SPEI prediction value.The BVMD-Attention-GRU model is applied to predict the SPEI in Urumqi city with a mid-long term forecast period of 6 months,and comparative experiments are conducted u-sing GRU,VMD-GRU,and BVMD-GRU models.The experimental results show that the BVMD-Attention-GRU model has higher prediction accuracy and is suitable for mid-long term drought prediction.
关键词
干旱预测/蝴蝶优化算法/变分模态分解/注意力机制/门控循环单元Key words
drought prediction/butterfly optimization algorithm/variational modal decomposition/attention mecha-nism/gated recurrent unit引用本文复制引用
基金项目
国家重点研发计划(2021YFE0104400)
出版年
2024
NETL
NSTL
维普
万方数据