基于EUNITE竞赛数据的中期电力负荷预测

【实例简介】
pdf文档，中文 来源：华北电力大学学报
24 2007 [1 1997 19981 1.2 19991 0.8 840 820 0.6 800 0.4 780 0.2 掩760 720 0 700 0100120 时间/天 680 40 100 120 图3训练过程中的误差变化 时间/天 Fig.3 Errors during training 图1日负荷预测时洲练数据 4 Fig. 1 Training data for day load for ecas ting 35 ←只 5 1.5 05 0-10203040如0的07080901010120 0.8 时间压 图4训练过程中神经元个数变化 Fig 4 Grow th of neurons during train ing 0.2 0 0.95 系0.9 4 0.85 05101520253035 滞后时间/天 如0.8 0.75 图2训练数据的自相关系数 Fig. 2 Autocorrelation coefficients of training dat a (a拟合曲线 0.〔 (3) 10.04 0.0 系 70/1 40 60 [1000000] 时间/天 (b拟合误差 「O1000001 图5训练结果及误差 Fig. 5 Result and error of train in 3- EBF 6, C1994-2012cHinaAcademicJournalElectronicPublishingHouse.Allrightsreservedhttp://www.cnki.net 4 EUNITE 25 820 800 780 760 740 720 26. 26 700 (3) 660 0 5101520253035 时间/天 50/1 [00001 图6日最大负荷预测结果 [01010] 26 ig. 6 Result of for F recasting g 1 SOFNN [716.2,739.7,757.7,781.3],7 Tab 1 Accuracy of training and forecast ing [720.1,738.2,763.7, Method MAPE/ (%O) ME 767.7],7 SOFNN 1.33 44.13 790 1.78 50.04 780 EUNITE 1.95 E770 760 winner 750 报告中ME值没有准确堤供,但可以从报告中估 轵740 计得出约50~60完整的 EUNITE网络竟赛原始数据可 730 从EUNITE网站获得(http://neuron.tuke.sk/compe- 720 tit ion/ index. php) 710 3354 EUNITE MAPE 时间 1.95 [2] 图7周平均最大负荷预测结果 3.2 Fig. 7 Forecasting result of week average max load SOFNN SOFNN WSj 7 2 SOF NN WL WLi deltaWli, delta WLi=WLj-Wli (3 h:× delt al +k× deltal wj, Y h=0.5 8 o1994-2012ChinaAcademicJournalElectronicPublishingHouse.Allrightsreservedhttp:/www.cnki.net 26 2007 820 [1] Chen B J, Chang M W, Lin C J. Load forecasting using 800 support vector machines: a study on eunitE com- 780 tion 2001[J]. IEEE transactions on power systems 760 2004.19(4):1821-1830 [2] Company behind East- Slov akia Power Distrib ution 740 Com pan Worle w ide com petition w ithin the EU 720 NITE network, EUNITE competiton report [RI 700 [3] 680 2004.28 (17):1-11 05101520253035 [4 Leng g, Prasad g, McGinnity T M. An on line algor- 时间/天 rithm for creating self organ izing fuzzy neural networ ks 图8修正后的日最大负荷预测结果 Neural Netw orks, 2004,(17): 1477-1493 [5 Ort iz: Arroyo D, Skov M K, Huynh Q. Accurate Elee- F ig.8 Forecasting results after rev sed tricity Load Forecasting with Artificial Neur al Networks IC. Proceedings of the 2005 International Conference 2 SOFNN on Compu tat io nal Intel ligence for M odel ling, Control and Tah 2 Accuracy of forecast ing Auto mation. and International Conference on I ntelligent Method MAPE/(%) ME Agents, Web Technolo gies and Internet Commerce SOFNN 1.78 50.04 (CIMCAIA WTIC 05). 2005 1.59 41.95 [ 6] Tao X. Input dimens ion reduction for load forecasting EUNITE 1.95 50-60 based on support vector machines [C]. Hong Kong 8 2004 IEEE International Conference of Electric U til y deregulation, res tructur ing and pow er technolog ies MAPE ME 20 [7 Hsu CC. Dynam icall y Optim izing Parameters in Support Vector Regression An A pp licat io n of Electricity Load 4结论 Forecasting [C]. Haw aii: Proceedings of the 3 9th Ilaw aii International Conference on System Sciences 1999 18 Pan kratz A. Forecasting w ith Univariate Box-Jenkins Models [m. John Wiley sons, 1983 SOFNN (1972-),, C1994-2012cHinaAcademicJournalElectronicPublishingHouse.Allrightsreservedhttp://www.cnki.net 【实例截图】
【核心代码】