本文已被：浏览 85次 下载 40次
投稿时间：2025-07-01 修订日期：2026-04-17
投稿时间：2025-07-01 修订日期：2026-04-17
中文摘要: 本文利用1961—2022年中国西北地区东部155个国家级气象观测站降水量以及全球大气、海温、海冰等历史数据资料，基于长短期记忆网络模型（LSTM）深度学习算法，融合了时间卷积网络（TCN）和嵌入卷积注意力机制（CBAM）算法，建立了中国西北地区东部夏季降水预测模型（CBAM-TCN-LSTM），并开展了预测技巧检验，与多种机器学习算法预测能力进行对比。结果表明：多种深度学习算法融合的降水预测模型预测技巧优于单一深度学习算法模型，独立样本检验阶段（2018—2022年）中国西北地区东部夏季降水预测趋势异常综合评分（PS）在60～80，均值达73.81，空间距平相关系数（ACC）除2020年为负值外，其余各年份均为正值，平均值为0.14，有明显提升。对比了目前7种主流的机器学习算法，CBAM-TCN-LSTM模型在PS、距平符号一致率（PC）、ACC、均方根误差（RMSE）和平均绝对误差（MAE）评价指标均优于其他模型。在2023年汛期预测业务中，CBAM-TCN-LSTM模型得到成功应用，准确预测了中国西北地区东部大部地区夏季降水偏少的特征，PS为81。通过采用具有强大时序预测能力的LSTM模型，融合TCN和嵌入CBAM建立的降水预测模型，可以为区域降水预测提供重要参考依据，具有较好的推广应用前景。
中文关键词: 深度学习，融合算法，中国西北地区东部，汛期降水，客观预测
Abstract:Based on the precipitation data of 155 meteorological stations in the eastern part of Northwest China from 1961 to 2022, as well as historical datasets including global atmosphere data, sea surface temperature (SST), and sea ice data, this study integrates the temporal convolutional network (TCN) module and the convolutional block attention module (CBAM) into the long short-term memory (LSTM) deep learning algorithm. A climate intelligent prediction model, CBAM-TCN-LSTM, based on the fusion of deep learning algorithms is thereby established for summer precipitation in the eastern part of Northwest China. The predictive performance of the model is verified, and its predictive capability is compared with that of other deep learning algorithms. The results show that the intelligent prediction model based on the fusion of multiple deep learning algorithms outperforms single-algorithm deep learning models. During the independent sample validation period (2018-2022), the prediction score (PS) for summer precipitation prediction ranged from 60 to 80, with an average value reaching 73.81. The anomaly correlation coefficient (ACC) was positive for all years except 2020, with an average value of 0.14, representing a significant improvement over other models. Compared with current seven mainstream models of machine learning algorithms, the CBAM-TCN-LSTM model presents superior performance across all five evaluation metrics: PS, percentage correct (PC), ACC, mean absolute error (MAE), and root mean square error (RMSE). In addition, the CBAM-TCN-LSTM model was successfully applied in the 2023 flood season prediction operation. It accurately predicted the characteristics of below-normal summer precipitation in most areas of the eastern part of Northwest China, with a PS as high as 81. By building a precipitation prediction model that combines the TCN module and the CBAM module on the basis of the LSTM model which has strong time-series predictive capability, this study provides a scientific basis and technical support for regional precipitation prediction, and the model holds a good prospect for popularization and application.
文章编号： 中图分类号： 文献标志码：
基金项目:国家自然科学基金联合基金项目（U22A20577）、宁夏重点研发计划（2022BEG02020）、宁夏青年拔尖人才工程、宁夏智能数字预报技术与应用创新团队项目、中国气象局创新发展专项（CXFZ2021J024、CXFZ2025Q022）和宁夏自然科学基金项目（2026AAC030558）共同资助
引用文本：
朱晓炜,刘嘉华,张肃诏,刘立波,杨建玲,耿硕飚,王璠,2026.基于多种深度学习算法融合的中国西北地区东部夏季降水客观预测方法研究[J].气象,52(10):1191-1201.
ZHU Xiaowei,LIU Jiahua,ZHANG Suzhao,LIU Libo,YANG Jianling,GENG Shuobiao,WANG Fan,2026.Objective Prediction Method for Summer Precipitation in the Eastern Part of Northwest China Based on Fusion of Multiple Deep Learning Algorithms[J].Meteor Mon,52(10):1191-1201.
朱晓炜,刘嘉华,张肃诏,刘立波,杨建玲,耿硕飚,王璠,2026.基于多种深度学习算法融合的中国西北地区东部夏季降水客观预测方法研究[J].气象,52(10):1191-1201.
ZHU Xiaowei,LIU Jiahua,ZHANG Suzhao,LIU Libo,YANG Jianling,GENG Shuobiao,WANG Fan,2026.Objective Prediction Method for Summer Precipitation in the Eastern Part of Northwest China Based on Fusion of Multiple Deep Learning Algorithms[J].Meteor Mon,52(10):1191-1201.
