Target Point Lightning Safety Risk Early Warning Based on Machine Learning
-
摘要: 收集广东地区1404组包括四个预警类型历史雷暴过程数据样本。结合目标点周围雷电发生的物理特征、雷电灾害的孕灾环境和承灾体特征的7个预报因子,利用四种机器学习算法训练得到面向目标点的雷电安全风险分级预警模型,并开展多指标对各模型进行评价分析,发现无等级模型和四级等级模型中都是随机森林算法的预警准确率最好,分别是95%和73%,而传统的卷积神经网络模型效果不佳。并选取广州塔作为目标点进行模型验证方法可行性,最终得到适应于广东雷暴特征的雷电安全风险预警分级模型。同时,根据本研究过程中可能存在不足提出下一步优化升级思路和方法。Abstract: The present study aimed to develop an accurate lightning risk classification and warning model for target points by using 1404 sets of data from four types of historical thunderstorm processes in Guangdong. Four machine learning algorithms were employed, and seven forecast factors, such as the physical characteristics of lightning occurrence around the target point, the breeding environment of lightning hazard, and the characteristics of the disaster-bearing body, were adopted to conduct multi-index evaluation and analysis of each risk early warning model. The results showed that the random forest algorithm exhibited the highest early warning accuracy in both the no-level model (95%) and the four-level model (73%). In contrast, the traditional convolutional neural network model proved to be ineffective for this purpose. Canton Tower was selected as the target point for model feasibility verification, and a lightning safety risk warning grading model tailored to the characteristics of thunderstorms in Guangdong was obtained. Finally, based on the identified deficiencies in the research process, ideas and methods for future optimization were proposed.
-
Key words:
- lightning /
- lightning safety /
- risk warning /
- machine learning
-
表 1 混淆矩阵评价参数
分类 有雷电安全风险预警 无雷电安全风险预警 20 km内有雷电 真正(True Positive,TP) 假负(False Negative,FN) 20 km内有雷电 假正(False Positive,FP) 真负(True Negative,TN) 表 2 混淆矩阵评价结果
评价参数 CNN RF SVC XGBoost P 153 271 205 264 TN 436 465 465 465 FP 153 23 89 30 FN 5 30 40 24 表 3 四种模型追加指标
评价指标 CNN RF SVC XGBOOST Accuracy 0.546 2 0.738 4 0.565 9 0.721 3 Precision 0.455 7 0.684 0 0.469 1 0.674 6 Recall 0.570 4 0.705 5 0.562 9 0.682 7 log loss 2.6546 1.9380 2.559 1 1.9903 -
[1] 史培军, 应卓蓉. 中国气象灾害对宏观经济增长的影响分析[J]. 北京师范大学学报: 自然科学版, 2016, 52(6): 747-753. [2] 郑国光. 中国气候[M]. 北京: 气象出版社, 2019. [3] 唐懿, 蔡雯悦, 翟建青, 等. 2021年夏季中国气候异常特征及主要气象灾害[J]. 干旱气象, 2022, 40(2): 179-186. [4] COOPER M A, HOLLE R L. Reducing Lightning Injuries Worldwide(Chapter 21)[M]. Berlin : Springer Natural Hazards, 2019. [5] GOMES C. Lightning science, engineering, and economic implications for developing countries[M]. Berlin: Lecture Notes in Electrical Engineering, 2021. [6] YIN Q, LIU H, FAN X, et al. Lightning fatalities in China, 2009-2018[J]. Journal of Agricultural Meteorology, 2021, 77(2): 150-159. [7] MANISH B. SHRIGIRIWAR, RAMESH K. et al. Study of fatalities due to lightning in Nagpur region of Maharashtra[J]. Journal of Indian Academy of Forensic Medicine, 2014, 36(3): 259-262. [8] 殷启元, 郭泽勇, 张义军, 等. 1995—2018年广东地区雷灾伤亡特征[J]. 热带气象学报, 2021, 37(3): 512-520 [9] 马明, 吕伟涛, 张义军, 等. 1997—2006年我国雷电灾情特征[J]. 应用气象学报, 2008, 19(4): 393-400. [10] TOVAR C, ARANGUREN D, LOPEZ J, et al. Lightning risk assessment and thunderstorm warning systems[C]// Intemational Conference of Lighbling Protection, 2014: 1870-1874. [11] WILLIAMS E, MONTANYÀ J. A closer look at lightning reveals needle-like structures[J]. Nature, 2019, 568(7752): 319-320. [12] KUMAR P R, KAMRA A K. Variability of lightning activity in South/Southeast Asia during 1997-98 and 2002-03 El Niño/La Niña events [J]. Atmospheric Research, 2012, 118: 84-102. [13] 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): 11-37. [14] POAKAL D, VEENAJ E, JURKOVI P M, et al. Analysis of orographic influence on hail parameters in NW Croatia[J]. International Journal of Climatology, 2018, 38(3): 5 646-5 658. [15] 易燕明, 杨兆礼, 万齐林, 等. 近50年广东省雷暴、闪电时空变化特征的研究[J]. 热带气象学报, 2006, 22(6): 539-546. [16] 张东, 林芳妮, 何如意, 等. 珠江口以西登陆热带气旋引发粤东暴雨的成因和预报着眼点分析[J]. 热带气象学报, 2019, 35(1): 14-24. [17] 陈绍东, 林奕峰, 区永平. 广州市雷暴日异常的基本气候特征及其与近海海温关系初探[J]. 热带气象学报, 2004, 20(1). 106-112. [18] 马瑞阳, 郑栋, 姚雯, 等. 雷暴云特征数据集及我国雷暴活动特征[J]. 应用气象学报, 2021, 32(3): 358-369. [19] 张恒进, 郑永光. 基于逐时观测的1971—2010年中国大陆雷暴气候特征[J]. 气象学报, 2022, 80(1): 54-66. [20] 张义军, 周秀骥. 雷电研究的回顾和进展[J]. 应用气象学报, 2006, 17(6): 829-834. [21] LAI J, LIU Y, DU J, et al. Lightning Detection Technology and Application[C]//2019 International Conference on Meteorology Observations (ICMO), 2019: 1-5. [22] FERRO M A D S, YAMASAKI J, PIMENTEL D R M, et al. Lightning risk warnings based on atmospheric electric field measurements in Brazil[J]. Journal of Aerospace Technology and Management, 2011, 3: 301-310. [23] LÓPEZ, E. PÉREZ, J. HERRERA, D. Aranguren and L. Porras. Thunderstorm warning alarms methodology using electric field mills and lightning location networks in mountainous regions[C]//2012 International Conference on Lightning Protection (ICLP), 2012. [24] MENG Q, YAO W, XU L. Development of lightning nowcasting and warning technique and its application[J]. Adv Meteor, 2019, 2019: 1-9. [25] ZENG Q, WANG Z, GUO F, et al. The application of lightning forecasting based on surface electrostatic field observations and radar data [J]. Journal of Electrostatics, 2013, 71(1): 6-13. [26] LYNN B, YAIR Y. Prediction of lightning flash density with the WRF model[J]. Advances in Geosciences, 2010, 23: 11-16. [27] XIANG L, XIANG J, GUAN J, et al. A Novel reference-based and gradient-guided deep learning model for daily precipitation downscaling [J]. Atmosphere, 2022, 13(4): 511. [28] LIN T, LI Q, GENG Y, JIANG L, et al. Attention-based dual-source spatiotemporal neural network for lightning forecast[J]. IEEE ACCESS 2019, 7: 158 296-158 307. [29] LIU W, WANG Y, ZHONG D, et al. ConvLSTM Network-based rainfall nowcasting method with combined reflectance and radar-retrieved wind field as inputs[J]. Atmosphere 2022, 13(3): 411. [30] GENG Y, LI Q, LIN T, et al. LightNet: A dual spatiotemporal encoder network model for lightning prediction[J]. Association for Computing Machiner, 2019: 2 439-2 447. [31] 杨璐, 南刚强, 陈明轩, 等. 基于三种机器学习方法的降水相态高分辨率格点预报模型的构建及对比分析[J]. 气象学报, 2021, 79(6): 1 022-1 034. [32] 孙健, 曹卓, 李恒, 等. 人工智能技术在数值天气预报中的应用[J]. 应用气象学报, 2021, 32(1): 1-11. [33] 刘海知, 徐辉, 包红军, 等. 机器学习分类算法在降雨型滑坡预报中的应用[J]. 应用气象学报, 2022, 33(3): 282-292. [34] 赵旭寰, 王振会, 肖稳安, 等. 神经网络在雷暴预报中的应用初步研究[J]. 热带气象学报2009, 25(3): 357-360. [35] SHI T, HU D, REN X, et al. Investigation on the lightning location and warning system using artificial intelligence[J]. Journal of Sensors, 2021: 6108223. [36] WANG G, KIM W H, KIL G S, et al. An intelligent lightning warning system based on electromagnetic field and neural network[J]. Energies, 2019, 12(7): 1 275. [37] BAO R, ZHANG Y, MA B J, et al. An artificial neural network for lightning prediction based on atmospheric electric field observations[J]. Remote Sensing, 2022, 14(17): 4 131. [38] 王志斌, 肖艳姣, 王珏, 等. 基于卷积神经网和SVM雷电监测预警[J]. 自然灾害学报, 2022, 31(1): 219-225. [39] 周康辉, 郑永光, 王婷波. 利用深度学习融合NWP和多源观测数据的闪电落区短时预报方法[J]. 气象学报2021, 79(1): 1-14. [40] ZHOU K, ZHENG Y, DONG W, et al. A deep learning network for cloud-to-ground lightning nowcasting with multisource data[J]. J Atmos Ocean Tech 2020, 37(5): 927-942. [41] LECUN Y, BOSER B, DENKER J S, et al. Backpropagation applied to handwritten zip code recognition[J]. Neural Comput, 1989: 541-551. [42] LECUN Y, BOTTOU L. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2 278-2 324. [43] MALDONADO S, LÓPEZ J, VAIRETTI C. Time-weighted fuzzy support vector machines for classification in changing environments[J]. Inform Sciences, 2021, 559: 97-110. [44] HAKALA K, KAEWPHAN S, BJORNE J, et al. Neural network and random forest models in protein function prediction[J]. IEEE/ACM transactions on computational biology and bioinformatics, 2022: 1 772-1 781. [45] SHENDRYK Y, GORROD E. Leveraging airborne lidar data and gradient boosting for mapping the density of different sized trees[J]. IEEE J-STARS, 2021: 1 572-1 579.