设为首页
加入收藏
A Particle Swarm Optimization-neural Network Ensemble Prediction Model for Persistent Freezing Rain and Snow Storm in Southern China
-
摘要: 利用逐日气温和降水量数据、NCEP/NCAR再分析资料以及预报场资料，通过分析提取我国南方区域持续性低温雨雪过程及其预报因子，使用粒子群-神经网络方法建立非线性的统计集合预报模型 (PSONN-EPM)，对我国南方区域持续性低温雨雪过程进行预报试验。结果表明：以过程的冷湿程度及影响范围为标准，将低温雨雪过程分为一般过程和严重过程，并建立不同的预报模型效果较好。通过10 d独立样本预报试验看，基于粒子群-神经网络方法建立的集合预报模型比基于逐步回归方法建立的预报模型的预报平均相对误差小，对严重过程预报能力高于对一般过程预报，且这种非线性统计集合建模方法在建模过程中不需要调整神经网络参数，在实际预报业务中值得尝试。Abstract: Based on daily minimum temperature, maximum temperature and precipitation data of 756 stations in China, National Center for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) reanalysis data during 1951-2013 and NCEP 24 h forecast data, a nonlinear statistical ensemble prediction model based on the particle swarm optimization-neural network (PSONN-EPM) is developed for predicting and verifying the regional persistent freezing rain and snow storm process in southern China by analyzing and extracting significant predictors. Results show that model performance can be effectively improved when dividing low-temperature processes into the general process and severe process which are constructed based on cold extents, humidity and influence ranges of the freezing rain and snow storm processes. In 10-day independent forecast test, the average relative errors for the general process and the severe process are 2.04 and 0.6 using stepwise regression equation forecast method, while those are 1.33 and 0.30 by using PSONN-EPM technique. It means forecast errors are reduced by 0.71 and 0.3 as compared with the stepwise regression method. In addition, the predication result for the severe freezing rain and snow storm process is better than that for the general process. The PSONN-EPM integrates predictions of multiple ensemble members, thus the prediction accuracy and stability are higher than those of the traditional linear regression method. Furthermore, such method does not contain any tunable parameters, and is applicable for practical operational weather prediction.
-
图 1 冷湿指数与850 hPa温度相关分布
(阴影表示达到0.001显著性水平)
Fig. 1 Correlation of PT value to 850 hPa temperature
(the shaded denotes passing the test of 0.001 level)
图 2 一般过程 (a) 和严重过程 (b) 粒子群-神经网络集合预报模型拟合预报
Fig. 2 Fitting values of general (a) and severe (b) processes by the particle swarm optimization-neural network ensemble prediction model
表 1 1951—2013研究区域内年持续性极端低温雨雪事件
Table 1 Cold rain and snow events in the study area during 1951-2013
开始日期 结束日期 持续日数/d 影响站数 过程日最大
冷湿指数1954-12-07 1954-12-15 9 3 13.05 1954-12-26 1955-01-10 16 30 70.42 1956-01-06 1956-01-12 7 6 20.18 1956-01-20 1956-01-26 7 4 9.08 1957-01-12 1957-01-16 5 4 20.86 1957-02-04 1957-02-16 13 10 18.77 1958-01-15 1958-01-19 5 4 8.83 1958-01-29 1958-02-04 7 5 12.91 1960-01-23 1960-01-28 6 3 5.68 1961-01-11 1961-01-16 6 3 7.43 1962-01-15 1962-01-29 15 3 8.34 1964-01-23 1964-02-04 13 8 22.9 1964-02-15 1964-02-28 14 46 112.27 1966-12-25 1967-01-12 19 11 17.74 1967-02-10 1967-02-15 6 4 6.56 1968-02-01 1968-02-10 10 11 21.06 1969-01-11 1969-01-17 7 17 26.24 1969-01-28 1969-02-09 13 23 80.65 1969-02-14 1969-02-28 15 4 6.19 1971-01-15 1971-02-05 12 16 63.73 1972-12-29 1973-01-06 9 6 13.25 1972-02-03 1972-02-11 9 51 138.71 1974-01-23 1974-02-12 21 29 65.51 1975-12-08 1975-12-15 8 22 73.48 1976-12-26 1977-01-17 23 18 36.67 1977-01-26 1977-02-04 10 21 64.39 1980-01-29 1980-02-13 16 38 79.62 1981-01-25 1981-01-31 7 9 17.59 1982-02-06 1982-02-15 10 7 37.71 1983-12-22 1984-01-02 12 7 51.72 1983-01-08 1983-01-23 16 12 31.13 1984-12-18 1984-12-31 14 13 31.27 1984-01-16 1984-02-11 27 28 50.74 1989-01-11 1989-01-16 6 9 35.09 1989-01-29 1989-02-09 12 3 6.87 1990-01-30 1990-02-04 6 7 11.65 1991-12-25 1991-12-31 7 4 24.31 1993-01-13 1993-01-24 12 14 23.36 1996-02-17 1996-02-26 10 32 81.31 1998-01-18 1998-01-25 8 3 13.11 2000-01-27 2000-02-05 10 6 25.30 2004-02-03 2004-02-08 6 3 7.63 2008-01-13 2008-02-15 34 71 264.24 2010-02-16 2010-02-20 5 3 10.47 2011-01-02 2011-02-01 31 24 118.86 2012-01-21 2012-01-27 7 6 14.64 2013-01-02 2013-01-13 12 5 7.28 
下载: 导出CSV
表 2 一般过程建模所选预报因子 (F=3)
Table 2 Predictors used in general process forecasting models (F=3)
因子序号 因子名称 相关系数 X3 850 hPa江南区域气温 -0.37 X5 非洲北部上空850 hPa与700 hPa温度差 -0.37 X8 200 hPa印度半岛西北部与青藏高原区域高度差 0.30 X9 500 hPa北太平洋北部与南部的高度差 0.25 X12 700 hPa菲律宾北部区域的湿度 -0.26 X16 500 hPa印度半岛与内蒙古区域的水平风速差 0.34 X17 700 hPa印度半岛西北部区域水平风速 0.27 X18 850 hPa太平洋夏威夷和库克群岛区域水平风速差 0.34 X21 850 hPa长江中上游与越南北部区域垂直风速差 0.34 X22 500 hPa孟加拉湾北部与蒙古区域的风速差 0.40
下载: 导出CSV
表 3 严重过程建模所选预报因子 (F=3)
Table 3 Predictors used in severe process forecasting models (F=3)
因子序号 因子名称 相关系数 X1 850 hPa与700 hPa江南区域温度差 -0.49 X6 850 hPa贝加尔湖区域温度 0.35 X9 500 hPa贝加尔湖到我国东北区域高度 0.42 X12 700 hPa长江中下游区域湿度 0.43 X13 700 hPa孟加拉湾区域上空湿度 0.43 X14 850 hPa孟加拉湾到越南北部上空湿度 0.58 X15 850 hPa赤道索马里上空湿度 -0.62 X19 850 hPa半太平洋北部区域与南部区域水平风速差 0.36 X23 700 hPa鄂霍次克海与蒙古高原区域垂直风速差 0.41 X25 850 hPa江南与东海区域垂直风速差 0.58
下载: 导出CSV
表 4 两种过程的粒子群-神经网络预报模型独立样本预报效果
Table 4 Statistics of predicted values of independent samples from two different processes using the particle swarm optimization-neural network ensemble prediction model
一般过程 (F=3) 严重过程 (F=3) 实况值 预报值 误差 相对误差/% 实况值 预报值 误差 相对误差/% 6.39 5.44 -0.95 15 24.06 32.70 8.64 36 5.00 4.82 -0.18 4 20.41 29.46 9.05 44 7.28 8.52 1.24 17 26.09 36.19 10.10 39 5.25 11.39 6.14 117 21.28 22.94 1.66 8 4.82 28.45 23.63 490 65.69 41.38 -24.31 37 5.80 8.50 2.70 47 47.37 61.42 14.05 30 3.81 6.39 2.58 68 118.86 49.22 -69.64 59 2.49 6.68 4.19 168 62.95 60.33 -2.62 4 1.08 4.29 3.21 297 27.82 27.83 0.01 1 1.35 2.84 1.49 111 15.86 22.14 6.28 40
下载: 导出CSV
表 5 两种过程逐步回归方程独立样本预报结果
Table 5 Statistics of predicted values of independent samples from two different processes using stepwise regression equation
一般过程 (F=3) 严重过程 (F=3) 实况值 预报值 误差 相对误差/% 实况值 预报值 误差 相对误差/% 6.39 6.84 0.45 7 24.06 45.09 21.03 87 5.00 5.98 0.98 20 20.41 31.55 11.14 55 7.28 8.49 1.21 17 26.09 38.47 12.38 47 5.25 11.43 6.18 118 21.28 9.50 -11.78 55 4.82 16.49 11.67 242 65.69 57.08 -8.61 13 5.80 11.35 5.55 96 47.37 63.77 16.40 35 3.81 10.19 6.38 167 118.86 52.22 -66.64 56 2.49 10.80 8.31 334 62.95 56.09 -6.86 11 1.08 8.10 7.02 650 27.82 -0.32 -28.14 101 1.35 6.55 5.20 385 15.86 -6.04 -21.90 138
下载: 导出CSV
表 6 一般过程不同F值逐步回归方法和神经网络方法独立样本预报误差 (单位：%)
Table 6 Statistics of predicted errors of independent samples from general processes using stepwise regression method and neural network method with different F values (unit:%)
F=2 F=3 F=4 逐步回归 神经网络 逐步回归 神经网络 逐步回归 神经网络 37 20 7 15 0 18 48 25 20 4 21 18 24 9 17 17 26 14 127 91 118 117 125 98 250 323 242 490 246 723 116 51 96 47 89 29 184 76 167 68 148 41 390 285 334 168 291 132 678 386 650 297 573 241 532 279 385 111 328 151
下载: 导出CSV
表 7 严重过程不同F值逐步回归方法和神经网络方法独立样本预报误差 (单位：%)
Table 7 Statistics of predicted errors of independent samples from severe processes using stepwise regression method and neural network method with different F values (unit:%)
F=2 F=3 F=4 逐步回归 神经网络 逐步回归 神经网络 逐步回归 神经网络 53 20 87 36 107 63 6 92 55 44 114 46 56 25 47 39 89 44 86 26 55 8 20 30 2 31 13 37 3 33 40 30 35 30 29 28 47 54 56 59 51 61 0 29 11 4 10 14 109 15 101 1 57 7 188 59 138 40 50 67
下载: 导出CSV
表 8 全样本不同F值粒子群-神经网络集合预报方法独立样本预报结果
Table 8 Statistics of predicted values of independent samples from all the samples using the particle swarm optimization-neural network ensemble prediction model
实况值 F=2 F=3 F=4 预报值 误差 相对误差/% 预报值 误差 相对误差/% 预报值 误差 相对误差/% 6.39 21.35 14.96 234 19.45 13.06 204 26.04 19.65 308 5.00 10.75 5.75 115 10.87 5.87 117 17.86 12.86 257 7.28 14.99 7.71 106 14.21 6.93 95 19.06 11.78 162 5.25 15.28 10.03 191 17.59 12.34 235 20.72 15.47 295 4.82 16.44 11.62 241 14.72 9.90 205 21.08 16.26 337 5.80 13.24 7.44 128 11.39 5.59 96 22.28 16.48 284 3.81 3.89 0.08 2 6.83 3.02 79 7.80 3.99 105 2.49 2.88 0.39 16 3.92 1.43 57 0.84 -1.65 66 1.08 1.11 0.03 3 1.80 0.72 67 7.29 6.21 575 1.35 0.74 -0.61 45 1.68 0.33 25 3.09 1.74 129
下载: 导出CSV
-
[1] 陶诗言, 卫捷.2008年1月中国南方严重冰雪灾害过程分析.气候与环境研究, 2008, 13(4):337-351. http://www.cnki.com.cn/Article/CJFDTOTAL-QHYH200804001.htm [2] 丁一汇, 王遵娅, 宋亚芳, 等.中国南方2008年1月罕见低温雨雪冰冻灾害发生的原因及其与气候变暖的关系.气象学报, 2008, 66(5):808-825. doi: 10.11676/qxxb2008.074 [3] 李崇银, 杨辉, 顾薇.中国南方雨雪冰冻异常天气原因的分析.气候与环境研究, 2008, 13(2):113-122. http://www.cnki.com.cn/Article/CJFDTOTAL-QHYH200802000.htm [4] 赵思雄, 孙建华.2008年初南方雨雪冰冻天气的环流场与多尺度特征.气候与环境研究, 2008, 13(4):351-367. http://www.cnki.com.cn/Article/CJFDTOTAL-QHYH200804002.htm [5] 彭艳, 王钊, 刘安麟, 等.2008年1月中国南部低温雨雪冰冻天气特征及其与东亚大气环流异常探讨.大气科学学报, 2010, 33(5):634-640. http://www.cnki.com.cn/Article/CJFDTOTAL-NJQX201005016.htm [6] 钱维宏, 张宗婕.南方持续低温冻雨事件预测的前期信号.地球物理学报, 2012, 55(5):1501-1512. doi: 10.6038/j.issn.0001-5733.2012.05.007 [7] 张宗婕, 钱维宏.中国冬半年区域持续性低温事件的前期信号.大气科学, 2012, 36(6):1269-1279. doi: 10.3878/j.issn.1006-9895.2012.11227 [8] 金龙, 吴建生, 林开平, 等.基于遗传算法的神经网络短期气候预测模型.高原气象, 2000, 24(6):981-987. http://cpfd.cnki.com.cn/Article/CPFDTOTAL-ZGQX200312005034.htm [9] Jin L, Yao C, Huang X Y.A nonlinear artificial intelligence ensemble prediction model for typhoon intensity.Mon Wea Rev, 2008, 136:4541-4554. doi: 10.1175/2008MWR2269.1 [10] Yao C, Jin L, Zhao H S.Ensemble prediction olmon-soon index with a genetic neural network model.Acta Meteor Sinica, 2009, 23(6):701-712. https://www.researchgate.net/profile/Yuehui_Chen/publication/222432560_Flexible_neural_trees_ensemble_for_stock_index_modeling/links/0fcfd50a70809e1d25000000.pdf?inViewer=true&disableCoverPage=true&origin=publication_detail [11] 陆虹, 金龙, 缪启龙, 等.影响广西热带气旋年频数的神经网络预测模型.南京气象学院学报, 2003, 26(1):56-62. http://www.cnki.com.cn/Article/CJFDTOTAL-NJQX200301007.htm [12] 吴振玲, 潘璇, 董昊, 等.天津市多模式气温集成预报方法.应用气象学报, 2014, 25(3):293-301. doi: 10.11898/1001-7313.20140305 [13] 张振华, 苗春生, 曾智华, 等.一种人工神经网络云分类方法的改进与应用.应用气象学报, 2012, 23(3):355-363. http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=20120312&flag=1 [14] 闵晶晶, 孙景荣, 刘还珠, 等.一种改进的BP算法及在降水预报中的应用.应用气象学报, 2010, 21(1):55-62. doi: 10.11898/1001-7313.20100107 [15] 黄小刚, 费建芳, 陈佩燕.利用神经网络方法建立热带气旋强度预报模型.应用气象学报, 2009, 20(6):699-705. doi: 10.11898/1001-7313.20090607 [16] Kennedy J, Eberhart R C.Particle Swarm Optimization//Pro IEEE International Conference on Neural Networks.1995. [17] Wu J S, Jin L.Study on the meteorological prediction model using the learning algorithm of neural ensemble based on PSO algorithms.Journal of Tropical Meteorology, 2009, 15(1):83-88. http://en.cnki.com.cn/Article_en/CJFDTOTAL-RDQX200806013.htm [18] Machado T R, Lopes H S.A Hybrid Particle Swarm Optimization Model for the Traveling Salesman Problem//Adaptive and Natural Computing Algorithms.New York:Springer, 2005:255-258. [19] Jin L, Zhu J S, Huang Y, et al.A nonlinear statistical ensemble model for short-range rainfall prediction.Theoretical and Applied Climatology, 2015, 119(3-4):791-807. doi: 10.1007/s00704-014-1161-8 [20] Jin L, Huang Y, Zhao H S.Ensemble Prediction of Monthly Mean Rainfall with a Particle Swarm Optimization-Neural Network Model.13th IEEE International Conference on Information Reuse and Integration, 2012:287-294. [21] 龚志强, 王晓娟, 崔冬林, 等.区域性极端低温事件的识别及其变化特征.应用气象学报, 2012, 23(2):195-204. doi: 10.11898/1001-7313.20120208 [22] 韩荣青, 陈丽娟, 李维京, 等.2-5月我国低温连阴雨和南方冷害时空特征.应用气象学报, 2009, 20(3):312-320. doi: 10.11898/1001-7313.20090307 [23] 郑维忠, 倪允琪.热带和中纬太平洋海温异常对东北夏季低温冷害影响的诊断分析研究.应用气象学报, 1999, 10(4):394-401. http://qikan.camscma.cn/jams/ch/reader/view_abstract.aspx?file_no=199904105&flag=1 [24] Zhang H, Qin J, Li Y.Climatic background of cold and wet winter in southern China: Part Ⅰ observational analysis.Climate Dynamics, 2011, 37(11-12):2335-2354. doi: 10.1007/s00382-011-1022-4 [25] Qian X, Miao Q L, Zhai P M, et al.Cold-wet spells in mainland China during 1951-2011.Nat Hazards, 2014, 70:975-994. doi: 10.1007/s11069-013-0856-y [26] 王绍武.中国冷冬的气候特征.气候变化研究进展, 2008, 4(2):68-72. http://www.cnki.com.cn/Article/CJFDTOTAL-QHBH200802005.htm [27] 王东海, 柳崇健, 刘英, 等.2008年1月中国南方低温雨雪冰冻天气特征及其天气动力学成因的初步分析.气象学报, 2008, 66(3):405-422. doi: 10.11676/qxxb2008.038 [28] 杨贵名, 孔期, 毛冬艳, 等.2008年初"低温雨雪冰冻"灾害天气的持续性原因分析.气象学报, 2008, 66(5):638-648. http://www.cnki.com.cn/Article/CJFDTOTAL-QXXB200805016.htm [29] 纪忠萍, 高晓容, 谷德军, 等.广东低温阴雨的低频振荡及环流特征.应用气象学报, 2013, 24(1):32-42. doi: 10.11898/1001-7313.20130104 -
计量
- 摘要浏览量: 3331
- HTML全文浏览量: 1200
- PDF下载量: 625
- 被引次数: 0
