An Attention Fusion and Information Recall LSTM Method for Radar Echo Extrapolation
-
摘要: 临近天气预报是气象研究中的热点问题，雷达回波外推技术作为处理临近天气预报的有效手段，具有重要的应用价值。近年来，深度学习技术被应用于处理这一任务，但提高雷达回波外推的预报准确率仍然是一个具有挑战性的问题。在ST-LSTM网络基础上，本文提出一种AFR-LSTM网络，以进一步提高雷达回波外推的预报准确率。首先提出一种注意力融合的时空长短期记忆网络的方法，以关联更多的历史信息，保证信息在传递过程中能够充分关联，减少信息丢失。同时，考虑编码过程中信息丢失问题，在编码器与解码器之间构建信息回忆模块，进一步保存雷达回波预测细节。通过在真实的雷达回波数据集（2019—2021江苏气象雷达数据）上进行消融实验，AFR-LSTM整体效果较好。此外，对该雷达回波数据集进行对比实验，结果表明AFRLSTM在雷达回波预测中评分函数临界成功指数(CSI)值为0.520 9、Heidke Skill Score(HSS)值为0.532 4，并且能较好地保留强回波和位置准确度，优于现有方法，证明了该方法能够获得更准确的预测准确度。Abstract: Nowcasting is a prominent area of research in meteorology, and radar echo extrapolation is an effective technique for generating nowcasts. In recent years, deep learning technology has been applied to this task, but improving the accuracy of radar echo extrapolation forecasting remains a challenge. Based on the ST-LSTM network, this paper proposes an AFR-LSTM network to enhance the accuracy of radar echo extrapolation forecasting. Firstly, an attention fusion method for a spatiotemporal long-short-term memory network is proposed to integrate more historical information, ensuring that information can be fully integrated during the transmission process and reducing information loss. Moreover, we address the issue of information loss in the encoding process by incorporating an information reminiscence module between the encoder and decoder, which helps preserve the details of radar echo prediction. Through ablation experiments conducted on a real radar echo dataset (2019—2021 Jiangsu Meteorological Radar Data), AFRLSTM demonstrates strong overall performance. Comparative experiments on this radar echo dataset also reveal that AFR-LSTM achieves a critical success index (CSI) value of 0.520 9 and a Heidke skill score (HSS) value of 0.532 4 in radar echo prediction, effectively preserving strong echoes and ensuring accurate location prediction. These results outperform existing methods, demonstrating that our proposed method can achieve more accurate image prediction.
-
表 1 不同方法在Moving MNIST数据集上的实验结果(前10帧预测后10帧)
Moving MNIST Method MSE/frame ↓ SSIM/frame ↑ ConvLSTM 103.3 0.707 PredRNN 56.8 0.867 PredRNN++ 46.8 0.898 E3D-LSTM 41.7 0.910 SA-ConvLSTM 43.9 0.913 MotionGRU 34.3 0.928 AFR-LSTM 30.5 0.939 
下载: 导出CSV
表 2 不同方法在雷达数据集上的CSI和HSS评分结果（前10帧预测后10帧）
Reflectivity Threshold CSI HSS 10 20 40 avg 10 20 40 avg ConvLSTM 0.7223 0.4864 0.1601 0.4563 0.6109 0.5283 0.2093 0.4495 PredRNN 0.7421 0.5104 0.1786 0.4770 0.7172 0.5446 0.2338 0.4985 PredRNN++ 0.7484 0.5176 0.1823 0.4828 0.7162 0.5516 0.2589 0.5089 E3D-LSTM 0.7496 0.5195 0.1838 0.4843 0.7259 0.5472 0.2275 0.5002 SA-ConvLSTM 0.7535 0.5198 0.2165 0.4965 0.7286 0.5503 0.2365 0.5051 MotionGRU 0.7565 0.5207 0.2157 0.4976 0.7321 0.5598 0.2832 0.5251 AFR-LSTM 0.7586 0.5306 0.2735 0.5209 0.7326 0.5651 0.2995 0.5324
下载: 导出CSV
表 3 雷达数据集上添加不同模块的CSI、HSS和SSIM评分
Reflectivity Threshold CSI HSS SSIM 10 20 40 avg 10 20 40 avg Baseline 0.7421 0.5104 0.1786 0.4770 0.7172 0.5446 0.2338 0.4985 0.6675 Baseline+SP 0.7527 0.5236 0.2605 0.5123 0.7256 0.5621 0.2795 0.5224 0.6715 Baseline+CH 0.7458 0.5135 0.1915 0.4836 0.7176 0.5498 0.2343 0.5006 0.6948 Baseline+AF 0.7535 0.5283 0.2655 0.5158 0.7286 0.5633 0.2814 0.5244 0.7034 Baseline+Recall 0.7455 0.5112 0.1824 0.4797 0.7181 0.5467 0.2341 0.4996 0.7159 AFR-LSTM 0.7586 0.5306 0.2735 0.5209 0.7326 0.5651 0.2995 0.5324 0.7214
下载: 导出CSV
-
[1] ZOU H, WU S, SHAN J, et al. A method of radar echo extrapolation based on TREC and Barnes filter[J]. Journal of Atmospheric and Oceanic Technology, 2019, 36(9): 1 713-1 727. [2] DEL MORAL A, RIGO T, LLASAT M C. A radar-based centroid tracking algorithm for severe weather surveillance: Identifying split/merge processes in convective systems[J]. Atmospheric Research, 2018, 213: 110-120. [3] WOO W, WONG W. Operational application of optical flow techniques to radar-based rainfall nowcasting[J]. Atmosphere, 2017, 8(3): 48. [4] 冯树民, 刘浩, 李来成. 雨雪天气下轨道交通客流预测模型[J]. 哈尔滨工业大学学报, 2022, 54(9): 1-6. [5] YIN X, WU G, WEI J, et al. Deep learning on traffic prediction: Methods, analysis and future directions[J]. IEEE Transactions on Intelligent Transportation Systems, 2021, 23(6): 4 927-4 943. [6] 王博文, 王景升, 王统一, 等. 基于长短时记忆网络的Encoder-Decoder多步交通流预测模型[J]. 重庆大学学报, 2021, 44(11): 71-80. [7] LV Y, DUAN Y, KANG W, et al. Traffic flow prediction with big data: a deep learning approach[J]. IEEE Transactions on Intelligent Transportation Systems, 2014, 16(2): 865-873. [8] OPREA S, MARTINEZ-GONZALEZ P, GARCIA-GARCIA A, et al. A review on deep learning techniques for video prediction[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 44(6): 2 806-2 826. [9] 莫晨, 邵洁. 基于自注意力的多模态LSTM的动作预测[J]. 计算机工程与设计, 2022, 43(4): 1 083-1 088. [10] 吴哲夫, 张令威, 刘光宇, 刘光灿. 基于空间自适应卷积LSTM的视频预测[J]. 计算机应用与软件, 2020, 37(09): 62-67, 110. [11] HSIEH J T, LIU B, HUANG D A, et al. Learning to decompose and disentangle representations for video prediction[J]. Advances in Neural Information Processing Systems, 2018, 31. [12] 黄兴友, 马玉蓉, 胡苏蔓. 基于深度学习的天气雷达回波序列外推及效果分析[J]. 气象学报, 2021, 79(5): 817-827. [13] 王式太, 张定红, 殷敏, 等. 基于LSTM模型的降雨短临预报[J]. 无线电工程, 2021, 51(11): 1 278-1 283. [14] 方巍, 庞林, 张飞鸿, 等. 对抗型长短期记忆网络的雷达回波外推算法[J]. 中国图象图形学报, 2021, 26(5): 1 067-1 080. [15] 郭瀚阳, 陈明轩, 韩雷, 等. 基于深度学习的强对流高分辨率临近预报试验[J]. 气象学报, 2019, 77(4): 715-727. [16] HE W, XIONG T, WANG H, et al. Radar echo spatiotemporal sequence prediction using an improved ConvGRU deep learning model[J]. Atmosphere, 2022, 13(1): 88. [17] LIN Z, LI M, ZHENG Z, et al. Self-attention convlstm for spatiotemporal prediction[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2020, 34(7): 11 531-11 538. [18] WU H, YAO Z, WANG J, et al. MotionRNN: A flexible model for video prediction with spacetime-varying motions[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 15 435-15 444. [19] SHI X, CHEN Z, WANG H, et al. Convolutional LSTM network: A machine learning approach for precipitation nowcasting[J]. Advances in Neural Information Processing Systems, 2015, 28. [20] WANG Y, LONG M, WANG J, et al. PredRNN: Recurrent neural networks for predictive learning using spatiotemporal LSTMs[J]Advances in Neural Information Processing Systems, 2017, 30: 879-888. [21] WANG Y, GAO Z, LONG M, et al. Predrnn++: Towards a resolution of the deep-in-time dilemma in spatiotemporal predictive learning[C]// International Conference on Machine Learning. PMLR, 2018: 5 123-5 132. [22] TRAN Q K, SONG S. Multi-channel weather radar echo extrapolation with convolutional recurrent neural networks[J]. Remote Sensing, 2019, 11(19): 2 303. [23] WANG Y, JIANG L, YANG M H, et al. Eidetic 3D LSTM: A model for video prediction and beyond[C]//International conference on learning representations. 2018. [24] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[J]. Advances in Neural Information Processing Systems, 2017, 30. [25] SRIVASTAVA N, MANSIMOV E, SALAKHUDINOV R. Unsupervised learning of video representations using lstms[C]//International conference on machine learning. PMLR, 2015: 843-852. [26] HORÉ A, ZIOU D. Image Quality Metrics: PSNR vs. SSIM[C]//2010 20th International Conference on Pattern Recognition. 2010: 2 366-2 369. -
点击查看大图
图(7) / 表(3)
计量
- 文章访问数: 320
- HTML全文浏览量: 23
- PDF下载量: 370
- 被引次数: 0
粤公网安备 4401069904700003号