2017 2nd Inernaonal Conference on Sofware, Mulmeda and Communcaon Engneerng (SMCE 2017) ISBN: 978-1-60595-458-5 Predcon of Ol Demand Based on Tme Seres Decomposon Mehod Nan MA * and Yong LIU College of Informaon Engneerng, Henan Unversy of Scence and Technology, Luoyang 471000, Chna *Correspondng auhor Keywords: Tme-Seres decomposon, Large ol depo, Ol demand, Predcon algorhm. Absrac. The forecas of ol demand n large ol depo s he man bass for he allocaon of ol and dsrbuon. The accurae and effcen real-me forecas of ol demand s an mporan guaranee for he effecve managemen and raonal use of ol. In hs paper, he amoun of ol n and ou of large ol depos s absraced no me seres model, and he man facors nfluencng he me seres are exraced by me seres decomposon mehod. The calculaon process and he changng rules of hese facors are suded, and he forecasng model of ol demand of large ol depo s consruced. The example shows ha he me seres decomposon mehod can realze he effecve forecas of large ol depo demand. Inroducon The forecas of ol demand n large ol depo s he man bass for he ol o managemen and dsrbuon. The accurae and effcen real-me forecas of ol demand s an mporan guaranee for he effecve managemen and raonal use of ol. In hs paper, he amoun of ol n and ou of large ol depos s absraced no me seres model, and he man facors nfluencng he me seres are exraced by me seres decomposon mehod. The calculaon process and he changng rules of hese facors are suded, and he forecasng model of ol demand of large ol depo s consruced. The example shows ha he me seres decomposon mehod can realze he effecve forecas of large ol depo demand. FM Chang (2013) combned wh me seres decomposon mehod and AI machne learnng mehod, hrough he analyss of hsorcal daa analyss, forecas a Tawan readmll pars manufacurng company orders quany, o acheve cos savngs purposes. Xu F e al. (2016) hrough he analyss of a large number of moble daa raffc rends and he developmen rend of random changes, he use of me seres decomposon mehod, accurae denfcaon and predcon of cellular daa raffc. Prema V e al. (2017) usng me seres decomposon mehod o acheve he wnd speed predcon. In hs paper, we compare and analyze he dfferen seasonal wnd speeds n one year and fnd ha when he wnd season comes, he wnd speed s unsable and he predcon error s hgher, and he wnd speed predcon error s reduced by 6.7%. Rojo J e al. (2017) usng he exponenal smoohng mehod o predc he concenraon of arborne pollen. In hs paper he changes of pollen concenraon s an mporan ndex o change he plan lfe cycle, hrough me seres modelng he varaon rend of he pollen concenraon n he ar, analyss of he nfluence of pollen concenraon season, predc fuure pollen concenraon n he ar, n order o acheve he agrculural crop harves forecas. Aue A e al. (2013) analyzes he advanages and dsadvanages of me seres decomposon mehod, based on he me seres decomposon mehod, a mulple nnovaon algorhm s nroduced o descrbe he concenraon of parcles n he ar, and he ar polluon ndex s predced. Ths paper presens a me seres predcon mehod based on decomposon, accordng o hsorcal ol mpor and expor daa of large ol depo, he change rule of ol ngress s analyzed by forecasng model. And as a bass for predcs he demand of large ol depos and provdes relable bass for he managemen of large ol depos. 484
Tme-seres Decomposon Mehod The me seres decomposon mehod s o check, comple and analyze he hsorcal values obaned by he me seres accordng o he order of me. Accordng o he me seres, he developmen process, drecon and rend of he hsorcal values are refleced and exended o fnd he hsorcal values. Tme changes and changes n he law o nfer a perod of me may reach he developmen scale, level, speed and proporon of real me predcon mehod. Based on he hsorcal ol resource allocaon of a large ol depo, hs paper esablshes he me seres of ol ngress and ouflow, realzes he forecas of ol demand by me seres decomposon mehod, and provdes effecve daa guaranee for he demand managemen of large ol depo. Forecasng Model Dependng on he operaon of a large ol depo, he facors ha affec ol demand and reserves are long-erm rend demand, seasonal change demand, cyclcal demand and emporary (rregular) demand. The me seres can be expressed as Eq. 1. y = f(t, S, C, I ). (1) In he formula, T ndcaes a long-erm rend facor, S ndcang a seasonal change, C ndcang a perodc change, I ndcang an rregular change. Tme seres decomposon commonly used models are addve models and mulplcaon models. The addve model s: Y = T + S + C + I. (2) The mulplcaon model s: Y = T S C I. (3) Due o he neracon beween he four facors of he ol daa of he ol depo suded n hs paper, he daa of hs paper are suded by he mulplcave model. Forecasng Process and seps Sep 1. Calculae long-erm rends. The me seres of he orgnal daa s 12 monhs movng average. The resulng daa s no seasonal, and he randomness facor s small or even no. Because he randomness around he nermedae value of he flucuaons, he number of 12 sum afer he average, posve and negave flucuaons o a ceran exen offse each oher, so ha here s no randomness. Smlarly, he second monh of he fuel daa o he hreenh monh of he fuel daa added evenly, does no conan seasonal and random facors, can be obaned whou seasonal and random facors of a ceran average number of rows M. n Y 1 M (4) n Where M s he movng average, n s he movng average acual nerval, and Y s he acual value n he n-segmen me. In order o elmnae he lag of he daa, n he prevous movng average on he bass of a movng average, ha s, he cenralzaon process. Use he movng average mehod o remove seasonal facors and rregular change values, ge he movng average M. M T C. (5) Ths value M conans long-erm rends and cyclcal changes. Ths se of long-erm rend values s lnearly fed o remove he cyclcal changes and oban long-erm rend values T. Sep 2. Calculae season ndex. Eq. 5 conans only long-erm rend and cycle varaon. The rend s removed accordng o Eq. 3 he resuls conan only seasonal and oulers. 485
Y T S C I S I M T C. (6) In order o elmnae he rregular change value, he SI value of he same monh s averaged, he seasonal facor s obaned, and he seasonal facor s normalzed, and he adjused seasonal ndex was S. 12 S 1 12 Sep 3. Calculaed cycle varaon facor. Accordng o he long-erm rend and cycle varaon rend conaned n Eq. 3, and he resul of he long-erm rend T, whch can be obaned from sep 1, he cycle varaon facor can be calculaed. C M T. (7) Sep 4. Calculaed predcve value. Accordng o he above analyss, he predced rend value and he correspondng seasonal ndex mulplcaon, he forecas value can be calculaed. y T S (=1,2,,12). (8) Applcaon Example Analyss In hs paper, we hrough he hsorcal daa of a large ol depo from June 2009 o June 2015 for sascal analyss, me seres model s esablshed and he daa and predc he acual hsory from July 2015 o July 2016 ol values are compared o verfy he feasbly of hs algorhm. Sep 1. Calculae long-erm rends. The average monhly movng average of hsorcal ol daa s shown n Table 1. Table 1. The army monhly fuel consumpon me seres decomposon able from June 2009 o December 2015. Tme Monh Order Value Once T C Twce T C S I(%) T C(%) Y Jun.2009 1 226 1934.05 Jy.2009 2 118 1947.1 Aug.2009 3 154 1960.15 Sep.2009 4 121 1973.2 Oc.2009 5 347 1986.25 Nov.2009 6 504 1999.3 Dec.2009 7 385 481 535.55 71.88 2012.35 26.61 Jan.2010 8 883 590 651.41 135.62 2025.4 32.16 Feb.2010 9 783 713 780.41 100.37 2038.45 38.28 Mar.2010 10 536 848 910.43 58.90 2051.5 44.38 Apr.2010 11 759 973 1032.72 73.54 2064.55 50.02 May.2010 12 953 1093 1160.62 82.15 2077.6 55.86 Jun.2010 13 1539 1228 1275.53 120.67 2090.65 61.01 Jy.2010 14 1585 1323 1342.17 118.10 2103.7 63.80 Jy.2014 62 3766 2382 2346.04 160.52 2730.1 85.93 Aug.2014 63 869 2310 2365.27 36.72 2743.15 86.22 Sep.2014 64 1090 2421 2441.69 44.65 2756.2 88.59 Oc.2014 65 833 2463 2511.95 33.16 2769.25 90.71 Nov.2014 66 1754 2561 2590.59 67.69 2782.3 93.11 Dec.2014 67 5277 2620 2620.12 201.42 2795.35 93.73 Jan.2015 68 2012 2808.4 Feb.2015 69 2599 2821.45 Mar.2015 70 3495 2834.5 Apr.2015 71 4464 2847.55 486
May.2015 72 3944 2860.6 Jun.2015 73 1837 2873.65 Table 1 s he hsorcal daa of ol depo, he fourh column for a movng average afer he resuls obaned, he ffh column for he average of he resuls obaned afer he second move. Accordng o he secondary movng average, drawng no a scaer plo, lnear fng, as shown n Fgure 1, excludng he cyclcal facors C. Fgure 1. The monhly rend of ol consumpon rend map. As can be seen from Fgure 2, he rend of he whole mage s on he rse. Takng he me as he ndependen varable, and he movng average seres as he dependen varable on he long-erm rend T lnear fng, wh 2010a fed o he lnear regresson Eq. 9: T a b a 13.05 b 1921 ( 1,2,...,73 ). (9) Column 7 of Table 1 s he long-erm rend forecas calculaed from he curve fng equaon. Sep 2. Calculae season ndex. The values of he "S I rao" column n Table 1 are aggregaed n he same column accordng o he daa of he same monh n each year, and he monhly average value of S,and he average values were sandardzed, elmnaes he numercal rregular varaons I represen only seasonal. Draw he S I scaer plo as shown n Fgure. 2. Fgure 2. S I scaer plo. The seasonal ndex S n Table 2 represens he average of S I for each year of he hsorcal year, he number of whch does no nclude he rregular change facor, seasonal ndex S wll add 1185.42576%, no equal o he sum of he seasonal ndex rao of welve n 1200% monhs, need o be adjused for seasonal ndex. The 1185.42576% and 1200% dvson, 1.012515072 for he adjusmen coeffcen, and he seasonal ndex of each monh mulpled by S s adjused S. Table 2. The seasonal ndex. Monh Seasonal Index Modfed Seasonal Index 1 98.74751529 99.98334755 2 98.12258371 99.35059491 487
3 98.08385358 99.31138007 4 99.78868493 101.0375475 5 101.105361 102.3707018 6 100.6148067 101.8740082 7 101.4281395 102.69752 8 99.76722005 101.015814 9 98.04507165 99.27211278 10 96.49628094 97.70393884 11 95.64127328 96.8382307 12 97.32675231 98.54480362 1185.167543 1200 Sep 3. Calculaed cycle varaon facor. The second movng average s dvded by he correspondng long-erm rend forecas and he perod change s gven n column 8 of Table 1. Sep 4. Calculaed predcve value. Through he above calculaon resuls, we can predc he amoun of ol n July 2015. Table 2 n July adjused season ndex S = 102.69752, he forecas of ol 7 demand for July 2015 s avalable from he rend equaon: y74 T 74S7 2886.7 (102.69752) 2964.57. (10) From 2015 o July n July, he acual fuel consumpon compared wh he forecas daa, he error rao s as shown n Table 3: Table 3. The error beween predced and acual values. Tme Value Predced Value Percen error Jy.2015 3523 2964.57 Jy.2015 Aug.2015 2491 2911.88 Aug.2015 Sep.2015 2618 2904.83 Sep.2015 Oc.2015 3249 2918.98 Oc.2015 Nov.2015 2553 2923.24 Nov.2015 Dec.2015 2427 2920.30 Dec.2015 Jan.2016 3410 2965.23 Jan.2016 Feb.2016 2672 2938.55 Feb.2016 Mar.2016 3348 2962.66 Mar.2016 Apr.2016 2786 3020.19 Apr.2016 May.2016 2728 3059.35 May.2016 Jun.2016 3489 3061.56 Jun.2016 Jy.2016 3230 3073.76 Jy.2016 100.00% The error rao beween predced value and acual value 50.00% 0.00% -50.00% -100.00% Fgure 3. The error rao beween predced value and acual value. As can be seen from Table 3 and Fg. 3 ha he percenage of error beween he predced value and he acual value s 4.83%~20.33%, and he closer he me s, he more accurae he predcon value 488
s. The forecasng algorhm of hs subjec has good effec and praccal sgnfcance for he predcon and analyss of ol demand. Conclusons In hs paper, based on he long-erm rend and seasonal varaon of ol use n large ol depo, a forecasng model of ol demand for large ol depo based on me seres decomposon s proposed. The model dvdes ou he seasonal facors n he hsorcal fuelng daa by me seres, and uses he quadrac movng average mehod o move he average rend of he ol use. By modfyng he seasonal facors, he forecas of he ol demand s realzed. The applcaon example proves he feasbly and effecveness of he model, whch provdes a good reference for he large ol depo managemen forecas. References [1] Chang, F. Mchael. Forecasng Order Quany for Treadmll Par by Comparson of Tme Seres Forecas Technologes and Arfcal Inellgence Mehods. Advances n Susanable and Compeve Manufacurng Sysems. Sprnger Inernaonal Publshng, 2013:695-704. [2] Xu, Fengl, e al. "Bg Daa Drven Moble Traffc Undersandng and Forecasng: A Tme Seres Approach." IEEE Transacons on Servces Compung 9.5(2016):1-1. [3] Prema, V., and K. U. Rao. "Tme seres decomposon model for accurae wnd speed forecas." Renewables: Wnd, Waer, and Solar 2.1(2015):1-11. [4] Rojo, Jesús, e al. "Modelng pollen me seres usng seasonal-rend decomposon procedure based on LOESS smoohng." Inernaonal Journal of Bomeeorology 61.2(2017):335. [5] Aue, Alexander, D. D. Nornho, and S. Hörmann. "On he predcon of funconal me seres." Transplanaon Proceedngs 10.3(2013):589-91. [6] Corez, Paulo, and J. P. Donae. "Global and decomposon evoluonary suppor vecor machne approaches for me seres forecasng." Neural Compung and Applcaons 25.5(2014):1053-1062. [7] Zhang, Langl, and B. Pan. Characerscs Analyss on Speed Tme Seres wh Emprcal Mode Decomposon as Vehcle Drvng Towards an Inersecon. Informaon Technology and Inellgen Transporaon Sysems. Sprnger Inernaonal Publshng, 2017. [8] Domenech, Josep, e al. "A axonomy of web predcon algorhms." Exper Sysems wh Applcaons 39.9(2012):8496-8502. [9] Km, Chul, S. H. Nam, and I. Joe. A Log Regresson Seasonaly Based Approach for Tme Seres Decomposon Predcon n Sysem Resources. Advances n Compuer Scence and Ubquous Compung. Sprnger Sngapore, 2015. [10] Zhou, Feng, and P. I. De-Chang. "Predcon Algorhm for Seasonal Saelle Parameers Based on Tme Seres Decomposon." Compuer Scence (2016). [11] Taylor, James W., and R. D. Snyder. "Forecasng nraday me seres wh mulple seasonal cycles usng parsmonous seasonal exponenal smoohng." Omega 40.6(2012):748-757. [12] Klepsch, Johannes, and C. Klüppelberg. "An Innovaons Algorhm for he predcon of funconal lnear processes." (2016). [13] Coussemen, Krsof, S. Lessmann, and G. Versraeen. "A comparave analyss of daa preparaon algorhms for cusomer churn predcon: A case sudy n he elecommuncaon ndusry." Decson Suppor Sysems (2016). 489
[14] L, Meng, e al. "Chaos Tme Seres Predcon Based on Membrane Opmzaon Algorhms." Scenfc World Journal 2015.2(2015):589093. [15] Aman, Sama, C. Chelms, and V. K. Prasanna. "Influence-Drven Model for Tme Seres Predcon from Paral Observaons." Lyrnx (2015). [16] Ak, R, O. Fnk, and E. Zo. "Two Machne Learnng Approaches for Shor-Term Wnd Speed Tme-Seres Predcon." IEEE Transacons on Neural Neworks & Learnng Sysems 27.8(2015):1734-1747. 490