|||
复杂系统行为预测的
“机理+辨识”策略
作为对清华大学老师们《电力系统负荷预测研究综述与发展方向的探讨》(电力系统自动化,2004, 28(17): 1-11.)中“类似的这种策略性的升华”的学习和初步回答,我们课题组提出《复杂系统行为预测的“机理+辨识”策略》,作为对“组合预测”策略的细化和发展。
2006-09-29首发在《中国科技论文在线》200609-432,http://www.paper.edu.cn/index.php/default/releasepaper/content/200609-432
后被评为五星级精品论文:精品论文,2007, 2(1): 83-87。见附件,版权归中国科技论文在线。
简单示意图
单一模型®组合预测(1969)®“机理+辨识”预测(2006)
“机理+辨识”策略的六个主要特征
①是“机理+回归+辨识”三阶段预测:机理阶段主要考虑了已知影响因子的作用,回归阶段主要考虑了已知影响因子的未知方式作用,辨识阶段再通过辨识模型对残差进行经验预测。复杂时间序列预测的经典策略是分解成“趋势+季节性+残差”(trend+seasonal+residual)三类成分,再分别预测。但这种分解方法只是从数据到数据,没有利用复杂系统的已知结构等信息。
欢迎您的批评与讨论! 参考文献 [1] 张嗣瀛. 复杂性科学,整体规律与定性研究[J]. 复杂系统与复杂性科学,2005,2(1):71-83 [2] 中华人民共和国国务院. 国家中长期科学和技术发展规划纲要(2006—2020年)[EB/OL]. http://www.gov.cn/ [3] 康重庆,夏清,张伯明. 电力系统负荷预测研究综述与发展方向的探讨[J]. 电力系统自动化,2004,28(17):1-11 [4] Yue YH, Han WX, Zhang WB. Local adding-weight linear regression forecasting method of chaotic series based on degree of incidence[J]. Proceedings of the CSEE,2004, 24(22): 17-20 [5] Jiang CW, Li T. Forecasting method study on chaotic load series with high embedded [J]. Energy Conversion and Management, 2005, 46(5): 667-676 [6] Lelitha V, Laurence RR. A comparison of the performance of artificial[A]. Neural Networks And Support Vector Machines For The Prediction Of Traffic Speed[C]. 2004 IEEE Intelligent Vehicles Symposium, University of Panna, Parma, Italy June 1447,2004: 194-199 [7] Li KP, Gao ZY. Nonlinear dynamics analysis of traffic time series[J]. Modern Physics Letters B, 2004, 18(26): 1-8 [8] Wang DS, He GG. Summary and prospects of the study on traffic chaos[J]. China Civil Engineering Journal, 2003, 36(1): 68-74 [9] Dake Chen, Mark AC Alexey K, et al. Predictability of El Niño over the past 148 years[J]. Nature, 2004, 428 (6984): 733-736 [10] Tziperman, E, Stone, L, Cane, M A, Jarosh, H. El Niño chaos: overlapping of resonances between the seasonal cycle and the Pacific ocean-atmosphere oscillator[J]. Science, 1994, 264(5155): 72-74 [11] 美国National Science Foundation. DDDAS: Dynamic Data Driven Applications Systems[EB/OL]. http://www.nsf.gov/funding [12] Oguchi T, Nijmeijer H. Prediction of chaotic behavior[J]. IEEE Trans. on CAS I: 2005, 52 (11): 2464-2472 [13] 吕金虎, 陆君安, 陈士化. 混沌时间序列分析及其应用[M]. 武汉:武汉大学出版社,2002 [14] Kantz H, Ragwitz M. Phase space reconstruction and nonlinear predictions for stationary and nonstationary Markovian processes[J]. International Journal of Bifurcation and Chaos, 2004, 14(6): 1935-1945 [15] Christopher CS, Alfred W H. Medium-term prediction of chaos[J]. Physical Review Letters, 2006, 96(4): 044101 [16] Garcia SP, Almeida JS. Nearest neighbor embedding with different time delays[J]. Physical Review E, 2005, 71 (3): 037204 [17] Small M, Tse CK. Optimal embedding parameters: a modelling paradigm[J]. Physics D, 2004, 194: 283-296 [18] Kim, HS, Eykholt, R, Salas, JD. Nonlinear Dynamics, Delay Times, and Embedding Windows[J], Physica D, 1999, 127: 48-60 [19] Kugiumtzis D. State Space Reconstruction Parameters in the Analysis of Chaotic Time Series - the Role of the Time Window Length[J]. Physica D, 1996, 95: 13-27 [20] Feeny BF, Lin G. Fractional derivatives applied to phase-space reconstructions[J]. Nonlinear Dynamics, 2004, 38 (1-4): 85-99 [21] Yang SQ, Jia CY. Two practical methods of phase space reconstruction[J]. Acta Physica Sinica, 2002, 51(11): 2452-2458 [22] Li HC,Zhang JS. Local Prediction of Chaotic Time Series Based on Support Vector Machine[J]. Chinese Physics Letters, 2005, 22(11): 2776 – 2779 [23] Ren R, Xu J, Zhu SH.Prediction of chaotic time sequence using least squares support vector domain[J]. Acta Physica Sinica, 2006, 55 (2): 555-563 [24] Gan JC, Xiao XC. Nonlinear adaptive multi-step prediction of chaotic time series based on points in the neighborthood[J]. Acta Physica Sinica, 2003, 52(12): 2995-3001 [25] Ma JH,Chen YS,Xin BG. Study on prediction methods for dynamics systems of nonlinear chaotic time series[J]. Applied Mathematics and Mechanics, 2004, 25(6): 605- 611 [26] Li HC, Zhang JS,Xiao XC. Neural Volterra filter for chaotic time series prediction[J]. Chinese Physics, 2005, 14(11): 2181-2188 [27] Cui WZ, Zhu CC, Bao WX, Liu JH. Chaotic time series prediction using mean-field theory for support vector machine[J]. Chinese Physics, 2005, 14(5): 922-919 [28] 杨正瓴,张广涛,陈红新,林孔元. 短期负荷预测“负荷趋势加混沌”法的参数优化[J]. 电网技术,2005,29(4):27 – 30, 44 [29] Mark A Cane. The evolution of El Niño, past and future[J]. Earth and Planetary Science Letters, 2005, 230 (3-4): 227-240 [30] Dake Chen, M A Cane, A Kaplan, S E Zebiak, D Huang, Predictability of El Niño in the past 148 years[J]. Nature, 428, 733-736, 2004. [31] Michael E Mann, Mark A Cane, Stephen E Zebiak, Amy ClementT. Volcanic and Solar Forcing of the Tropical Pacific over the Past 1000 Years[J]. Journal Of Climate, 2005, 18(3): 447-456 [32] J Lean, G Rottman, J Harder, G Kopp. SORCE contributions to new understanding of global change and solar variability[J]. Solar Physics, 2005, 230(1-2): 27-53 [33] J Hansen, M Sato, R Ruedy, et al. Efficacy of climate forcings[J]. Journal Of Geophysical Research-Atospheres, 2005, 110 (D18): Art. No. D18104 [34] Ping Chang, Yue Fang, R. Saravanan, Link Ji, Howard Seidel1. The cause of the fragile relationship between the Pacific El Niño and the Atlantic Niño [J]. Nature, 2006, 443(7109): 324-328 [35] P. Foukal, C. Fröhlich, H. Spruit and T. M. L. Wigley. Variations in solar luminosity and their effect on the Earth's climate[J]. Nature, 2006, 443(7108): 161-166 [36] 韩延本,赵娟,李志安. 由地球自转的年际变化预测El Niño事件[J]. 科学通报,2001, 46(22): 1858-1861 [37] 郑大伟,丁晓利,周永宏,陈永奇,李志林,廖新浩. El Niño事件的前兆现象在日长和海平面观测中的反映[J]. 科学通报,2000, 45(23):2572-2576 [38] 韩延本,李志安,赵娟. 天文学与自然灾害的相关研究[J]. 北京师范大学学报:自然科学版,2000, 36(4): 555-557 [39] 任振球. 全球变化研究的新思维[J]. 地学前缘,2002, 9(1): 27-33 [40] 苏旸. 气候变化的天文理论[J]. 地球物理学进展,2000,15(3): 102-111 [41] 谢炯光,曾琮,纪忠萍. 中国近30 年来气象统计预报进展[J]. 气象科技, 2003, 31(2): 67 –83 [42] F. Atger. The Skill of Ensemble Prediction Systems [J]. Monthly Weather Review, 1999, 127(9): 1941–1953
②模型评价:在复杂系统预测中,建议对预测中采用的多个模型的表现(预测结果)进行评价,如“平均误差”(代表预测的系统误差)、“平均绝对值误差”、“最大误差”、“重要数据的预测误差(如对最大值、最小值、特定数值等)”等进行统计考核,以确定该模型在多模型合成中的地位和作用。
③预测结果的灵活合成:根据对系统将来行为的预测目的,根据各模型的预测表现,由控制误差的关键量,采用灵活的多套预测值的合成,以期在人们最感兴趣的未来行为预测中得到最优效果。这是对组合预测、ensebmle预测(因散预测,也可译为集合预测)技术的进一步发展。
④概率化预测:由于采用了灵活的多模型预测,可以用一定的方法把这些预测结果的概率统计性质,用概率的方法表示,使得对预测结果的风险进行更准确和科学的评估。
⑤非平稳数据的平稳化技术。通过“机理+辨识”策略,以及差分等技术,实现非平稳数据的平稳化,提高辨识预测的准确率。
⑥预测准确率上限和可预测性研究:在“机理+辨识”预测策略的第一阶段,对系统的预测准确率上限和可预测性进行研究。
Archiver|手机版|科学网 ( 京ICP备07017567号-12 )
GMT+8, 2024-10-31 09:45
Powered by ScienceNet.cn
Copyright © 2007- 中国科学报社