Prophet时序预测模型在电离层TEC异常探测中的应用
|
摘要
提出一种基于Facebook开源的Prophet预测模型进行电离层TEC异常识别的新方法。首先,对比分析了该方法与传统时间序列预测方法(ARIMA模型等)预测电离层TEC建模背景值的精度,以及与经典电离层TEC异常识别方法(滑动四分位法)提取前面对应一致的电离层TEC背景值的精度。结果表明, Prophet预测模型预测建模背景值的精度要明显优于其他方法,且预测的建模精度比ARIMA模型等方法高2.55倍左右,比滑动四分位法高10.74倍左右。同时,在最佳预测建模区间时,其精度值大小比较依次为RMSE_(IQR)=10.5841>RMSE_(ARIMA)=3.2780>RMSE_(Prophet)=0.8469,说明传统探测法预测建模背景值时具有较大的不足。随后,以2017年8月8日九寨沟7.0级地震为例,利用该方法分析了电离层TEC异常扰动情况,并对比验证了该方法的有效性和准确性。实验结果表明:在震前第10 d和第2 d电离层TEC发生较为明显的负异常,第7 d电离层TEC发生较为明显的正异常。对比实验表明, Prophet预测模型的有效性和准确性明显优于滑动四分位法。
This paper proposed a new method for identification of ionospheric TEC anomalies using prophet forecasting model based on Facebook. First, we compared the precision of this new method with the traditional time series forecasting method(Autoregressive Integrated Moving Average, ARIMA models) and the identification method of the classical ionospheric TEC anomalies(Inter Quartile Range, IQR method), in predicting the background values of ionospheric TEC modeling. The results show that the precision of the former is obviously better than the latter two methods: about 2.55 times higher than that of the ARIMA models, and about 10.74 times higher than that of the IQR method. Meanwhile, when the best prediction modeling interval is established, the comparison of precision values is RMSE_(IQR)=10.5841>RMSE_(ARIMA)=3.2780>RMSE_(Prophet)=0.846, indicating that the traditional detection methods have insufficiency in predicting modeling background values. Second, taking the August 8, 2017 Jiuzhaigou earthquake as example, we analyzed its ionosphere TEC anomalies and proved the effectiveness and accuracy of the new method. The results show that obvious ionosphere TEC negative anomalies appeared on the 10 th and 2 nd days before the earthquake, and obvious ionosphere TEC positive anomalies occurred on the 7 th day before the earthquake. In addition, the comparative experiments show that the validity and accuracy of the Prophet forecasting model is significantly better than the IQR method.
This paper proposed a new method for identification of ionospheric TEC anomalies using prophet forecasting model based on Facebook. First, we compared the precision of this new method with the traditional time series forecasting method(Autoregressive Integrated Moving Average, ARIMA models) and the identification method of the classical ionospheric TEC anomalies(Inter Quartile Range, IQR method), in predicting the background values of ionospheric TEC modeling. The results show that the precision of the former is obviously better than the latter two methods: about 2.55 times higher than that of the ARIMA models, and about 10.74 times higher than that of the IQR method. Meanwhile, when the best prediction modeling interval is established, the comparison of precision values is RMSE_(IQR)=10.5841>RMSE_(ARIMA)=3.2780>RMSE_(Prophet)=0.846, indicating that the traditional detection methods have insufficiency in predicting modeling background values. Second, taking the August 8, 2017 Jiuzhaigou earthquake as example, we analyzed its ionosphere TEC anomalies and proved the effectiveness and accuracy of the new method. The results show that obvious ionosphere TEC negative anomalies appeared on the 10 th and 2 nd days before the earthquake, and obvious ionosphere TEC positive anomalies occurred on the 7 th day before the earthquake. In addition, the comparative experiments show that the validity and accuracy of the Prophet forecasting model is significantly better than the IQR method.
引文
[1] Gokhberg M B, Morgounov V A, Yoshino T, et al. Experimental measurement of electromagnetic emissions possibly related to earthquakes in Japan[J]. J Geophys Res Sol Earth, 1982, 87(B9): 7824-7828.
[2] Parrot M. Correlation between GEOS VLF emissions and earthquakes[J]. Ann Geophys, 1985, 3(6): 737-747.
[3] Pulinets S A, Gaivoronska T B, Contreras A L, et al. Correlation analysis technique revealing ionospheric precursors of earthquakes[J]. Natural Hazards & Earth System Sciences, 2004, 4: 697-702.
[4] Hayakawa M, Kawate R, Molchanov O A, et al. Results of ultra-low-frequency magnetic field measurements during the Guam Earthquake of 8 August 1993[J]. Geophys Res Lett, 1996, 23(3): 241-244.
[5] Liu J, Chen Y, Chuo Y, et al. Variations of ionospheric total electron content during the Chi-Chi earthquake[J]. Geophys Res Lett, 2001, 28(1): 1383-1386.
[6] Liu J Y, Chen C H, Hattori K. Temporal and spatial precursors in the ionospheric global positioning system (GPS) total electron content observed before the 26 December 2004 M9.3 Sumatra-Andaman Earthquake[J]. J Geophys Res Space Phys, 2012, 115: A09312.
[8] Shen X, Zhang X, Wang L, et al. The earthquake-related disturbances in ionosphere and project of the first China seismo-electromagnetic satellite[J]. Earthquake Science, 2011, 24: 639-650.
[9] Zhang X, Shen X H, Liu J, et al. Ionospheric perturbations of electron density before the Wenchuan Earthquake[J]. International Journal of Remote Sensing, 2010, 31(13): 3559-3569.
[10] 陈鹏. GNSS电离层层析及震前电离层异常研究[J]. 测绘学报, 2013, 42(3): 474-474.
[11] 张学民, 刘静, 赵必强, 等. 玉树地震前的电离层异常现象分析[J]. 空间科学学报, 2014, 34(6): 822-829.
[12] 姚宜斌, 翟长治, 孔建, 等. 2015年尼泊尔地震的震前电离层异常探测[J]. 测绘学报, 2016, 45(4): 385-395.
[13] Yang L, Zhao H, Dong M, et al. Ionospheric Anomaly before Kyushu, Japan Earthquake[J]. Acta Geodaetica et Cartographica Sinica, 2016, 45(S2): 139-146.
[14] 邹斌, 常晓涛, 郭金运, 等. 2014年鲁甸6.5级地震震前TEC异常分析[J]. 测绘科学, 2017, 42(3): 48-54.
[15] Liu J Y, Hattori K, Chen Y I. Application of Total Electron Content Derived from the Global Navigation Satellite System for Detecting Earthquake Precursors[J]. Pre-Earthquake Processes: A Multidisciplinary Approach to Earthquake Prediction Studies, Geophysical Monograph 234, 2018, 17: 305-317.
[16] He L, Heki K. Three-dimensional distribution of ionospheric anomalies prior to three large earthquakes in Chile[J]. Geophys Res Lett, 2016, 43(14): 7287-7293.
[17] Heki K, Enomoto Y. MW dependence of the preseismic ionospheric electron enhancements[J]. J Geophys Res Space Phys, 2015, 120(8): 7006-7020.
[18] Heki K. Ionospheric electron enhancement preceding the 2011 Tohoku-Oki earthquake[J]. Geophys Res Lett, 2011, 38(17): 136-147.
[19] 张小红, 任晓东, 吴风波, 等. 震前电离层TEC异常探测新方法[J]. 地球物理学报, 2013, 56(2): 441-449.
[20] 张小红, 任晓东, 吴风波, 等. 自回归移动平均模型的电离层总电子含量短期预报[J]. 测绘学报, 2014, 43(2): 118-124.
[21] Tulunay E, Senalp E T, Radicella S M, et al. Forecasting total electron content maps by neural network technique[J]. Radio Science, 2006, 41(4): RS4016.
[22] Oyeyemi E O, Poole A W V, Mckinnell L A. On the global short-term forecasting of the ionospheric critical frequency foF2 up to 5 hours in advance using neural networks[J]. Radio Science, 2005, 40(6): RS6012.
[23] Francis N M, Brown A G, Cannon P S, et al. Prediction of the hourly ionospheric parameter f(O)F(2) using a novel nonlinear interpolation technique to cope with missing data points[J]. Journal of Geophysical Research Space Physics, 2001, 106(A12): 30077-30083.
[24] Huang Z, Yuan H. Ionospheric single-station TEC short-term forecast using RBF neural network[J]. Radio Science, 2014, 49(4): 283-292.
[25] Xie S F, Chen J, Huang L K, et al. Ionospheric TEC Prediction Based on Holt-Winters Models[J]. Journal of Geodesy & Geodynamics, 2017, 37: 72-76.
[26] Sean J. Taylor & Benjamin Letham. Forecasting at Scale[J]. The American Statistician, 2018, 72(1): 37-45.
[27] Taylor S J, Letham B. prophet: automatic forecasting procedure. R package version 0.3, 2017.
[28] Zhao N Z, Liu Y, Jennifer K Vanos et al. Day-of-week and seasonal patterns of PM2.5 concentrations over the United States: Time-series analyses using the Prophet procedure[J]. Atmospheric Environment, 2018, 192: 116-127.
[29] 赵海山, 杨力, 徐世依. 太阳活动高低年电离层TEC变化特性分析[J]. 导航定位学报, 2017, 5(1): 24-30.
[30] Harvey A C, Shephard N. Structural time series models[C]//Maddala G, Rao C, Vinod H (Eds.), Handbook of Statistics[M]. Vol 11. Amsterdam: Elsevier, 1993.
[31] Byrd R H, Lu P, Nocedal J, et al. A Limited Memory Algorithm for Bound Constrained Optimization[J]. SIAM Journal on Scientific Computing, 1995, 16(5): 1190-1208.
[32] 刘权明. 基于Prophet的CPI指数预测[J]. 中国管理信息化, 2018, 21(13): 122-123.
[33] Brockwell P J, Davis R A. Introduction to Time Series and Forecasting[M]. Third ed. Springer Texts in Statistics, 2016.
[34] Niu M, Wang Y, Sun S, et al. A novel hybrid decomposition-and-ensemble model based on CEEMD and GWO for short-term PM 2.5, concentration forecasting[J]. Atmospheric Environment, 2016, 134: 168-180.
[35] Wang P, Zhang G, Zhang H, et al. novel hybrid-Garch model based on Arima and SVM for PM2.5 concentrations forecasting[J]. Atmospheric Pollution Research, 2017, 8(5): 850-860.
[36] Mini K G, Kuriakose Somy, Sathianandan T V. Building-model CPUE series for the fishery along northeast coast of India: A comparison between the Holt-Winters, ARIMA and NNAR models[J]. Journal of the Marine Biological Association of India, 2015, 57(2): 75-82.
[2] Parrot M. Correlation between GEOS VLF emissions and earthquakes[J]. Ann Geophys, 1985, 3(6): 737-747.
[3] Pulinets S A, Gaivoronska T B, Contreras A L, et al. Correlation analysis technique revealing ionospheric precursors of earthquakes[J]. Natural Hazards & Earth System Sciences, 2004, 4: 697-702.
[4] Hayakawa M, Kawate R, Molchanov O A, et al. Results of ultra-low-frequency magnetic field measurements during the Guam Earthquake of 8 August 1993[J]. Geophys Res Lett, 1996, 23(3): 241-244.
[5] Liu J, Chen Y, Chuo Y, et al. Variations of ionospheric total electron content during the Chi-Chi earthquake[J]. Geophys Res Lett, 2001, 28(1): 1383-1386.
[6] Liu J Y, Chen C H, Hattori K. Temporal and spatial precursors in the ionospheric global positioning system (GPS) total electron content observed before the 26 December 2004 M9.3 Sumatra-Andaman Earthquake[J]. J Geophys Res Space Phys, 2012, 115: A09312.
[8] Shen X, Zhang X, Wang L, et al. The earthquake-related disturbances in ionosphere and project of the first China seismo-electromagnetic satellite[J]. Earthquake Science, 2011, 24: 639-650.
[9] Zhang X, Shen X H, Liu J, et al. Ionospheric perturbations of electron density before the Wenchuan Earthquake[J]. International Journal of Remote Sensing, 2010, 31(13): 3559-3569.
[10] 陈鹏. GNSS电离层层析及震前电离层异常研究[J]. 测绘学报, 2013, 42(3): 474-474.
[11] 张学民, 刘静, 赵必强, 等. 玉树地震前的电离层异常现象分析[J]. 空间科学学报, 2014, 34(6): 822-829.
[12] 姚宜斌, 翟长治, 孔建, 等. 2015年尼泊尔地震的震前电离层异常探测[J]. 测绘学报, 2016, 45(4): 385-395.
[13] Yang L, Zhao H, Dong M, et al. Ionospheric Anomaly before Kyushu, Japan Earthquake[J]. Acta Geodaetica et Cartographica Sinica, 2016, 45(S2): 139-146.
[14] 邹斌, 常晓涛, 郭金运, 等. 2014年鲁甸6.5级地震震前TEC异常分析[J]. 测绘科学, 2017, 42(3): 48-54.
[15] Liu J Y, Hattori K, Chen Y I. Application of Total Electron Content Derived from the Global Navigation Satellite System for Detecting Earthquake Precursors[J]. Pre-Earthquake Processes: A Multidisciplinary Approach to Earthquake Prediction Studies, Geophysical Monograph 234, 2018, 17: 305-317.
[16] He L, Heki K. Three-dimensional distribution of ionospheric anomalies prior to three large earthquakes in Chile[J]. Geophys Res Lett, 2016, 43(14): 7287-7293.
[17] Heki K, Enomoto Y. MW dependence of the preseismic ionospheric electron enhancements[J]. J Geophys Res Space Phys, 2015, 120(8): 7006-7020.
[18] Heki K. Ionospheric electron enhancement preceding the 2011 Tohoku-Oki earthquake[J]. Geophys Res Lett, 2011, 38(17): 136-147.
[19] 张小红, 任晓东, 吴风波, 等. 震前电离层TEC异常探测新方法[J]. 地球物理学报, 2013, 56(2): 441-449.
[20] 张小红, 任晓东, 吴风波, 等. 自回归移动平均模型的电离层总电子含量短期预报[J]. 测绘学报, 2014, 43(2): 118-124.
[21] Tulunay E, Senalp E T, Radicella S M, et al. Forecasting total electron content maps by neural network technique[J]. Radio Science, 2006, 41(4): RS4016.
[22] Oyeyemi E O, Poole A W V, Mckinnell L A. On the global short-term forecasting of the ionospheric critical frequency foF2 up to 5 hours in advance using neural networks[J]. Radio Science, 2005, 40(6): RS6012.
[23] Francis N M, Brown A G, Cannon P S, et al. Prediction of the hourly ionospheric parameter f(O)F(2) using a novel nonlinear interpolation technique to cope with missing data points[J]. Journal of Geophysical Research Space Physics, 2001, 106(A12): 30077-30083.
[24] Huang Z, Yuan H. Ionospheric single-station TEC short-term forecast using RBF neural network[J]. Radio Science, 2014, 49(4): 283-292.
[25] Xie S F, Chen J, Huang L K, et al. Ionospheric TEC Prediction Based on Holt-Winters Models[J]. Journal of Geodesy & Geodynamics, 2017, 37: 72-76.
[26] Sean J. Taylor & Benjamin Letham. Forecasting at Scale[J]. The American Statistician, 2018, 72(1): 37-45.
[27] Taylor S J, Letham B. prophet: automatic forecasting procedure. R package version 0.3, 2017.
[28] Zhao N Z, Liu Y, Jennifer K Vanos et al. Day-of-week and seasonal patterns of PM2.5 concentrations over the United States: Time-series analyses using the Prophet procedure[J]. Atmospheric Environment, 2018, 192: 116-127.
[29] 赵海山, 杨力, 徐世依. 太阳活动高低年电离层TEC变化特性分析[J]. 导航定位学报, 2017, 5(1): 24-30.
[30] Harvey A C, Shephard N. Structural time series models[C]//Maddala G, Rao C, Vinod H (Eds.), Handbook of Statistics[M]. Vol 11. Amsterdam: Elsevier, 1993.
[31] Byrd R H, Lu P, Nocedal J, et al. A Limited Memory Algorithm for Bound Constrained Optimization[J]. SIAM Journal on Scientific Computing, 1995, 16(5): 1190-1208.
[32] 刘权明. 基于Prophet的CPI指数预测[J]. 中国管理信息化, 2018, 21(13): 122-123.
[33] Brockwell P J, Davis R A. Introduction to Time Series and Forecasting[M]. Third ed. Springer Texts in Statistics, 2016.
[34] Niu M, Wang Y, Sun S, et al. A novel hybrid decomposition-and-ensemble model based on CEEMD and GWO for short-term PM 2.5, concentration forecasting[J]. Atmospheric Environment, 2016, 134: 168-180.
[35] Wang P, Zhang G, Zhang H, et al. novel hybrid-Garch model based on Arima and SVM for PM2.5 concentrations forecasting[J]. Atmospheric Pollution Research, 2017, 8(5): 850-860.
[36] Mini K G, Kuriakose Somy, Sathianandan T V. Building-model CPUE series for the fishery along northeast coast of India: A comparison between the Holt-Winters, ARIMA and NNAR models[J]. Journal of the Marine Biological Association of India, 2015, 57(2): 75-82.