LAI Can,WANG Haijiang,LI Jing,et al.The Radar Echo Extrapolation based on ConvLSTM[J].Journal of Chengdu University of Information Technology,2020,35(06):589-593.[doi:10.16836/j.cnki.jcuit.2020.06.001]
基于ConvLSTM的雷达回波外推
- Title:
- The Radar Echo Extrapolation based on ConvLSTM
- 分类号:
- TN957.51+3
- 文献标志码:
- A
- 摘要:
- 为对雷达回波产品进行预测,研究一种基于ConvLSTM的雷达回波外推神经网络模型,采用结构为长短时记忆网络,对其结构做出一些改变后可以使用多维数据作为输入。且采纳单极化SA雷达的反射率数据作为数据集。结果证明:(1)采用相当多或者足够多的数据集,并进行数据预处理,包括数据筛选、排序、拟合、滤波。之后进行标准化,将数据按比例缩放,使其反射率值以0为中心并分布在一个小的区间内。(2)使用卷积层对反射率数据提取特征,并在隐藏层后面接入全连接层和回归层后,可以得到更优化和更准确地预测结果。(3)利用适当的激活函数和一定的正则化方法可以减少模型的过度拟合,提高模型的训练精度。实验结果表明,对18 min内的雷达回波有较好的预测效果,对反射率小于30的回波有较好的预测能力。
- Abstract:
- In order to make predictions for radar echo productions,this paper has researched a radar echo extrapolation neural network model based on ConvLSTM,and the construction adopted by this model is long-short term memory network. I made changes for the structure so multiple dimensions data can be used as input.Single polarization SA radar reflectivity data is regarded as data sets.To achieve ideal prediction effect,the following approaches have been approved effectively:(1)There should be enough or relatively enough data sets,then we preprocess the data sets,including data selection,sort,fitting and filter.Afterwards do standardization, zoom data according to scale,and the reflectivity value is centered on 0 and distributed in a small interval.(2)The convolutional layer is used to extract characteristics of reflectivity data. Placing full connection and regression layers after hidden layers,we can obtain more optimal and accurate prediction results.(3)We take advantage of some activation functions and regularization methods, these can reduce the over-fitting of the model,improve the training accuracy of model.The experimental results have shown that pretty good prediction effect has obtained within 18 minutes,and prediction ability is strong while reflectivity value is small than 30.
参考文献/References:
[1] 俞小鼎,周小刚,王秀明.雷暴与强对流临近天气预报技术进展[J].气象学报,2012,70(3):311-337.
[2] 张蕾,魏鸣,李南,等.改进的光流法在回波外推预报中的应用[J].科学技术与工程,2014,14(32):133-148.
[3] Dixon M,Wiener G.TITAN:Thunderstorm Identification,Tracking,Analysis,and Nowcasting-A Radar-based Methodology[J].Journal of Atmospheric and Oceanic Technology,1993,10(6):785.
[4] Johnson J T,Mackeen P L,Witt A,et al.The Storm Cell Identification and Tracking Algorithm: An Enhanced WSR-88D Algorithm[J].Weather & Forecasting,1998,13(2):263-276.
[5] 乔春贵,郑世林,杨立志,等.质心法雷达回波外推的原理及应用[J].河南气象,2006(3):29-30.
[6] Hilst G R,Russo J A Jr.An objective extrapolation technique for semi-conservative fields with an application to radar patterns[R].Tech Memo No 3,Travelers Weather Research Center,Harford,CT,1960:34.
[7] Austin,G L Bellon A.The use of digital weather radar records for short-term precipitation forecasting[J].Quarterly Journal of the Royal Meteorological Society,1974.
[8] Rinehart R E, Garvey E T.Three-dimensional strom motion detection by conventional weather radar[J].Nature,1978,273:287-289.
[9] LI L.Nowcasting of Motion and Growth of Precipitation with Radar Over Complex Orography[J].j.appl.meteor,1995:34.
[10] 李若楠,张鸿,白志娜,等.气象雷达基数据质量控制及其在回波识别方面的简单应用[J].现代农业科技,2015(5):245-247.
[11] Shanker M,Hu M Y,Hung M S.Effect of data standardization on neural network training[J].Omega,1996,24(4):385-397.
[12] Jozefowicz,Rafal,Zaremba,et al.An Empirical Exploration of Recurrent Network Architectures[C].International Conference on International Conference on Machine Learning.JMLR.org,2015.
[13] Hochreiter S,Schmidhuber J.Long Short-Term Memory[J].Neural computation,1997,9(8):1735-1780.
[14] Pascanu R,Mikolov T,Bengio Y.On the difficulty of training Recurrent Neural Networks[J].2012.
[15] Gers F A,Schmidhuber,Jürgen,et al.Learning to Forget:Continual Prediction with LSTM[J].Neural Computation,2000,12(10):2451-2471.
备注/Memo
收稿日期:2020-07-13 基金项目:国家自然基金民航联合研究基金资助项目(U1733103)