機(jī)器學(xué)習(xí)在ENSO預(yù)測(cè)會(huì)商中的應(yīng)用

作者：李晨彤

單位：國(guó)家海洋環(huán)境預(yù)報(bào)中心, 北京 100081

關(guān)鍵詞：ENSO 可解釋機(jī)器學(xué)習(xí) 多模式 智能會(huì)商

分類號(hào)：P456

出版年·卷·期（頁(yè)碼）：2022·39·第一期（91-103）

摘要：

基于多模式集合預(yù)報(bào)的思想，利用可解釋機(jī)器學(xué)習(xí)方法——決策樹算法建立了多模式ENSO預(yù)測(cè)結(jié)果智能會(huì)商系統(tǒng)。分別使用基于Boosting的GBDT、XGBoost、lightGBM和基于Bagging的RF 4種決策樹模型方法，結(jié)合隨機(jī)搜索交叉驗(yàn)證、網(wǎng)格搜索交叉驗(yàn)證兩種超參數(shù)調(diào)整方法對(duì)決策樹模型的超參數(shù)進(jìn)行優(yōu)化調(diào)整，根據(jù)不同超前預(yù)報(bào)時(shí)效分別建立多模式ENSO預(yù)測(cè)結(jié)果智能會(huì)商系統(tǒng)，對(duì)多模式預(yù)測(cè)結(jié)果進(jìn)行集合訂正，并給出各模式預(yù)測(cè)結(jié)果在智能會(huì)商系統(tǒng)中的特征重要性。該智能會(huì)商系統(tǒng)模擬了ENSO預(yù)測(cè)會(huì)商過程，實(shí)現(xiàn)了讀取各模式預(yù)測(cè)結(jié)果、訓(xùn)練模型、給出預(yù)測(cè)結(jié)論及預(yù)測(cè)依據(jù)、預(yù)測(cè)結(jié)果可視化等流程的自動(dòng)化，同時(shí)實(shí)現(xiàn)了智能調(diào)參的功能。

Bansed on the concept of multi-model ensemble forecasting, this study establishes an intelligent consultation system for multi-model intelligent consultation system of ENSO prediction using the interpretable machine learning method named decision tree algorithm. The hyper parameters of four decision tree models of GBDT based on Boosting, XGBoost, lightGBM and Random Forest (RF) based on Bagging are optimized and adjusted by using two hyper parameter adjustment methods of random search cross-validation and grid search cross-validation. The intelligent consultation system of multi-model ENSO prediction results is established according to different prediction leading time, which makes integrated correction on the multi-model prediction results and provides the feature importance of the prediction result of each model in the intelligent consultation system. The intelligent consultation system simulates the consultation process of ENSO prediction, which realizes the automation of the processes of reading the prediction results of each model, training the model, giving the prediction conclusion and prediction basis and the visualization of the prediction results, and realizes the function of intelligent parameter tuning. The intelligent consultation system collectively revises the multi-modal ENSO prediction results. The results show that machine learning also has some advantages in the multi-modal result consultation, which provide a reference for consultations of ENSO prediction in the future.

參考文獻(xiàn)：

[1] 李崇銀, 穆穆, 周廣慶, 等. ENSO機(jī)理及其預(yù)測(cè)研究[J]. 大氣科學(xué), 2008, 32(4):761-781. Li C Y, Mu M, Zhou G Q, et al. Mechanism and prediction studies of the ENSO[J]. Chinese Journal of Atmospheric Sciences, 2008, 32(4):761-781. [2] 丑紀(jì)范. 天氣和氣候的可預(yù)報(bào)性[J]. 氣象科技進(jìn)展, 2011, 1(2):11-14. Chou J F. Predictability of weather and climate[J]. Advances in Meteorological Science and Technology, 2011, 1(2):11-14. [3] 陶祖鈺, 趙翠光, 陳敏. 談?wù)劷y(tǒng)計(jì)預(yù)報(bào)的必要性[J]. 氣象科技進(jìn)展, 2016, 6(1):6-13. Tao Z Y, Zhao C G, Chen M. The necessity of statistical forecasts[J]. Advances in Meteorological Science and Technology, 2016, 6(1):6-13. [4] 許小峰. 從物理模型到智能分析——降低天氣預(yù)報(bào)不確定性的新探索[J]. 氣象, 2018, 44(3):341-350. Xu X F. From physical model to intelligent analysis:a new exploration to reduce the uncertainty of weather forecast[J]. Meteorological Monthly, 2018, 44(3):341-350. [5] Jin E K, Kinter Ⅲ J L, Wang B, et al. Current status of ENSO prediction skill in coupled ocean-atmosphere models[J]. Climate Dynamics, 2008, 31(6):647-664. [6] 任福民, 袁媛, 孫丞虎, 等. 近30年ENSO研究進(jìn)展回顧[J]. 氣象科技進(jìn)展, 2012, 2(3):17-24. Ren F M, Yuan Y, Sun C H, et al. Review of progress of ENSO studies in the past three decades[J]. Advances in Meteorological Science and Technology, 2012, 2(3):17-24. [7] Clarke A J. El Niño physics and El Niño predictability[J]. Annual Review of Marine Science, 2014, 6:79-99. [8] 陳大可, 連濤. 厄爾尼諾-南方濤動(dòng)研究新進(jìn)展[J]. 科學(xué)通報(bào), 2020, 65(35):4001-4003. Chen D K, Lian T. Frontier of El Nino-southern oscillation research[J]. Chinese Science Bulletin, 2020, 65(35):4001-4003. [9] Barnston A G, Tippett M K, L'Heureux M L, et al. Skill of realtime seasonal ENSO model predictions during 2002-11-is our capability increasing?[J]. Bulletin of the American Meteorological Society, 2012, 93(5):631-651. [10] 陳靜, 陳德輝, 顏宏. 集合數(shù)值預(yù)報(bào)發(fā)展與研究進(jìn)展[J]. 應(yīng)用氣象學(xué)報(bào), 2002, 13(4):497-507. Chen J, Chen D H, Yan H. A brief review on the development of ensemble prediction system[J]. Journal of Applied Meteorological Science, 2002, 13(4):497-507. [11] Mylne K R, Evans R E, Clark R T. Multi-model multi-analysis ensembles in quasi-operational medium-range forecasting[J]. Quarterly Journal of the Royal Meteorological Society, 2002, 128(579):361-384. [12] 鄭飛. ENSO集合預(yù)報(bào)研究[D]. 北京:中國(guó)科學(xué)院研究生院(大氣物理研究所), 2007. Zheng F. Research on ENSO ensemble prediction[D]. Beijing:Institute of Atmospheric Physics, Chinese Academy of Sciences, 2007. [13] Yan X Q, Tang Y M. An analysis of multi-model ensembles for seasonal climate predictions[J]. Quarterly Journal of the Royal Meteorological Society, 2013, 139(674):1179-1198. [14] 王琳. ENSO的多模式集合預(yù)報(bào)研究[D]. 成都:成都信息工程大學(xué), 2017. Wang L. Study on ENSO multi-model ensemble predictions with the statistical correction method[D]. Chengdu:Chengdu University of Information Technology, 2017. [15] 郭煒豪, 溫文, 王曉春, 等. NINO3.4指數(shù)的多模式集合預(yù)報(bào)方法[J]. 熱帶氣象學(xué)報(bào), 2019, 35(2):262-267. Guo W H, Wen W, Wang X C, et al. A multimodel ensemble method for NINO3.4 index forecast[J]. Journal of Tropical Meteorology, 2019, 35(2):262-267. [16] 王琳. 基于多模式集合的季節(jié)-年際氣候預(yù)測(cè)方法研究[D]. 武漢:中國(guó)地質(zhì)大學(xué), 2020. Wang L. Study on seasonal to interannual climate prediction with multi-model ensemble method[D]. Wuhan:China University of Geosciences, 2020. [17] 陳溢豪, 張?zhí)N斐, 周倩, 等. 集合預(yù)報(bào)及其在季節(jié)尺度氣候預(yù)測(cè)中的應(yīng)用[J]. 海洋預(yù)報(bào), 2020, 37(6):102-111. Chen Y H, Zhang Y F, Zhou Q, et al. Ensemble forecasting and its application in seasonal climate forecast[J]. Marine Forecasts, 2020, 37(6):102-111. [18] 焦李成, 楊淑媛, 劉芳, 等. 神經(jīng)網(wǎng)絡(luò)七十年:回顧與展望[J]. 計(jì)算機(jī)學(xué)報(bào), 2016, 39(8):1697-1717. Jiao L C, Yang S Y, Liu F, et al. Seventy years beyond neural networks:retrospect and prospect[J]. Chinese Journal of Computers, 2016, 39(8):1697-1717. [19] Reichstein M, Camps-Valls G, Stevens B, et al. Deep learning and process understanding for data-driven Earth system science[J]. Nature, 2019, 566(7743):195-204. [20] Sebastian S, Gabriele M. Weather and climate forecasting with neural networks:using general circulation models (GCMs) with different complexity as a study ground[J]. Geoscientific Model Development, 2019, 12(7):2797-2809. [21] Ruti P M, Tarasova O, Keller J H, et al. Advancing research for seamless Earth system prediction[J]. Bulletin of the American Meteorological Society, 2020, 101(1):E23-E35. [22] Wheeling K. Machine learning improves weather and climate models[J]. Eos, 2020, 101, doi:10. 1029/2020EO142422. [23] De Vos M G, Hazeleger W, Bari D, et al. Open weather and climate science in the digital era[J]. Geoscience Communication, 2020, 3(2):191-201. [24] Bauer P, Dueben P D, Hoefler T, et al. The digital revolution of Earth-system science[J]. Nature Computational Science, 2021, 1(2):104-113. [25] 賀圣平, 王會(huì)軍, 李華, 等. 機(jī)器學(xué)習(xí)的原理及其在氣候預(yù)測(cè)中的潛在應(yīng)用[J]. 大氣科學(xué)學(xué)報(bào), 2021, 44(1):26-38. He S P, Wang H J, Li H, et al. Machine learning and its potential application to climate prediction[J]. Transactions of Atmospheric Sciences, 2021, 44(1):26-38. [26] Ham Y G, Kim J H, Luo J J. Deep learning for multi-year ENSO forecasts[J]. Nature, 2019, 573(7775):568-572. [27] Zhang Q, Wang H, Dong J Y, et al. Prediction of sea surface temperature using long short-term memory[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(10):1745-1749. [28] Aguilar-Martinez S, Hsieh W W. Forecasts of Tropical Pacific Sea Surface temperatures by neural networks and support vector regression[J]. International Journal of Oceanography, 2009, 2009:167239. [29] Feng Q Y, Vasile R, Segond M, et al. ClimateLearn:a machinelearning approach for climate prediction using network measures[J]. Geoentific Model Development Discussions, 2016, doi:10. 5194/gmd-2015-273. [30] Nooteboom P D, Feng Q Y, López C, et al. Using network theory and machine learning to predict El Niño[J]. Earth System Dynamics Discussions, 2018, doi:10. 5194/esd-2018-13. [31] 許柏寧, 姜金榮, 郝卉群, 等. 一種基于區(qū)域海表面溫度異常預(yù)測(cè)的ENSO預(yù)報(bào)深度學(xué)習(xí)模型[J]. 科研信息化技術(shù)與應(yīng)用, 2017, 8(6):65-76. Xu B N, Jiang J R, Hao H Q, et al. A deep learning model of ENSO prediction based on regional sea surface temperature anomaly prediction[J]. e-Science Technology & Application, 2017, 8(6):65-76. [32] 何丹丹, 姜金榮, 郝卉群, 等. 基于深度學(xué)習(xí)的ENSO預(yù)報(bào)方法研究[J]. 科研信息化技術(shù)與應(yīng)用, 2019, 10(1):38-47. He D D, Jiang J R, Hao H Q, et al. Research on ENSO forecasting method based on deep learning[J]. e-science Technology & Application, 2019, 10(1):38-47. [33] 蔣國(guó)榮, 張韌, 沙文鈺. 用EOF展開和人工神經(jīng)網(wǎng)絡(luò)方法預(yù)測(cè)ENSO的研究[J]. 海洋預(yù)報(bào), 2001, 18(3):1-11. Jiang G R, Zhang R, Sha W Y, et al. The study of forecasting ENSO by using EOF appsoach and neural network method[J]. Marine Forecasts, 2001, 18(3):1-11. [34] Dijkstra H A, Petersik P, Hernández-García E, et al. The application of machine learning techniques to improve El Niño prediction skill[J]. Frontiers in Physics, 2019, 7:153. [35] Breiman L. Random forests[J]. Machine Learning, 2001, 45(1):5- 32. [36] Quinlan R J. Induction of decision trees[J]. Machine Learning, 1986, 1(1):81-106. [37] Dietterich T G. An experimental comparison of three methods for constructing ensembles of decision trees:bagging, boosting, and randomization[J]. Machine Learning, 2000, 40(2):139-157. [38] Friedman J H. Greedy function approximation:a gradient boosting machine[J]. The Annals of Statistics, 2001, 29(5):1189-1232. [39] Chen T Q, Guestrin C. XGBoost:a scalable tree boosting system[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco, California, USA:ACM, 2016. [40] Ke G L, Meng Q, Finley T, et al. LightGBM:a highly efficient gradient boosting decision tree[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, CA, USA:ACM, 2017:3149-3157. [41] Liaw A, Wiener M. Classification and Regression by random Forest[J]. R News, 2002, 2(3):18-22. [42] Ren H L, Zheng F, Luo J J, et al. A review of research on tropical Air-Sea interaction, ENSO dynamics, and ENSO prediction in China[J]. Journal of Meteorological Research, 2020, 34(1):43-62.

服務(wù)與反饋：

【文章下載】【發(fā)表評(píng)論】【查看評(píng)論】【加入收藏】