基于Attention-BiLSTM混合模型的月尺度降水量预测  

Monthly Precipitation Prediction Based on Attention-BiLSTM Model

作　　者：成玉祥 肖丽英[1] 王萍根 刘祥周 章晨晖 CHENG Yuxiang;XIAO Liying;WANG Pinggen;LIU Xiangzhou;ZHANG Chenhui(School of Hydraulic and Ecological Engineering,Nanchang Institute of Technology,Nanchang 330099,China)

机构地区：[1]南昌工程学院水利与生态工程学院,江西南昌330099

出　　处：《人民珠江》2024年第6期73-81,共9页Pearl River

基　　金：国家自然科学基金项目(52069014、52069015);2021年江西省学位与研究生教育教学改革研究项目(JXYJG-2021-210);江西省科技厅项目(20212BDH81002);南昌工程学院大学生创新创业计划项目(2021011、2022066)。

摘　　要：降水受到多种气象因素的影响,从而导致降水预测精度不高。针对这个问题,在考虑影响降水的多个气象因素基础上,通过Attention机制赋予各种气象因素不同的权重,结合双向长短期记忆神经网络(BiLSTM),提出了改进的Attention-BiLSTM混合模型去实现月尺度降水量的预测。以江西省南昌气象站为例,将1989—2018年的逐月降水量与逐月气象因素(气温、蒸发量、气压等)观测资料作为模型输入数据,通过Attention机制识别出各种气象因素的权重,从而提高BiLSTM模型对降水量的预测性能。结果表明:Attention-BiLSTM混合模型可有效地提高降水量预测的精度;通过Attention机制的修正,显著地改善了原有的BiLSTM模型降水量预测值偏低的问题。Precipitation is affected by various meteorological factors,leading to low prediction accuracy.To solve this problem,multiple meteorological factors affecting precipitation were considered,and the attention mechanism was used to assign different weights to various meteorological factors.Combined with the bidirectional long short-term memory neural network(BiLSTM),an improved attention-BiLSTM model was proposed to predict monthly precipitation.By taking the Nanchang Meteorological Station in Jiangxi Province as an example,the observation data of monthly precipitation and meteorological factors(temperature,evaporation,pressure,etc.)from 1989 to 2018 were used as input data for the model.The attention mechanism identified the weights of various meteorological factors to improve the prediction performance of the BiLSTM model for precipitation.The results show that the attention-BiLSTM model can effectively improve the accuracy of precipitation prediction.Through the correction of the attention mechanism,the low precipitation prediction values by the original BiLSTM model are significantly improved.

关 键 词：月尺度降水 气象因子 Attention机制 BiLSTM 预测性能 

分 类 号：TV213.9[水利工程—水文学及水资源] P457.6[天文地球—大气科学及气象学]