论文研究-基于非线性修正策略的空气质量预警系统研究.pdf-资料库

weixin_38744270-11795975-4744302542864416962.pdf-第1页.png

第1页 / 共14页

weixin_38744270-11795975-4744302542864416962.pdf-第2页.png

第2页 / 共14页

weixin_38744270-11795975-4744302542864416962.pdf-第3页.png

第3页 / 共14页

weixin_38744270-11795975-4744302542864416962.pdf-第4页.png

第4页 / 共14页

weixin_38744270-11795975-4744302542864416962.pdf-第5页.png

第5页 / 共14页

weixin_38744270-11795975-4744302542864416962.pdf-第6页.png

第6页 / 共14页

weixin_38744270-11795975-4744302542864416962.pdf-第7页.png

第7页 / 共14页

weixin_38744270-11795975-4744302542864416962.pdf-第8页.png

第8页 / 共14页

39 8 2019 8 Systems Engineering — Theory & Practice Vol.39, No.8 Aug., 2019 doi: 10.12011/1000-6788-2018-2470-14 : X831 : A , ( , 116025) , "$%"&’ *",-.. /, 0234Æ56 (isolation forest, iForest) (air quality index, AQI) ;<>, ?@A , D FGHIJGHGH"GHL, MNAPGST (time varying ﬁl- tering based empirical mode decomposition, TVF-EMD)VIJ (modiﬁed butterﬂy optimization algorithm, MBOA);

8 , : 2139 performance of the early warning system; 3) The performance of the developed system is superior to other compared methods, which can provide eﬀective early warning for cities with diﬀerent air pollution level. Keywords air quality early warning; time varying ﬁltering based empirical mode decomposition (TVF- EMD); modiﬁed butterﬂy optimization algorithm (MBOA); outlier robust extreme learning machine (ORELM); nonlinear correction strategy 1 4 , 21 . , , . , , , , , , " . ", $%, !! $, " , "" #$%!’""##&$ [1]. )*+’#*%*($!’&!", ’%$ ($") [2]. ’,( !+ “-"”, %*+" “-"” -, 700 /"#!’. "#, 0)& - , !$ (12 .34, 2# ).(/, "*’’0 1!5( (. (, *12)- ,- PM2.5 +-), 7 .* (air quality index, AQI) ,. AQI 2012 3 .*-,/+5 ( +5(GB3095-2012)), (-8 --8 .PM10 PM2.58 /8 6 + 6 ,/+5", 50 1996 Æ* “0+5” (+5(GB3095-1996)) 1! .* (air pollution index, API) , ,/1./20781, %1/ 1 , " 232241 (/*533. 3 +-4,50, AQI ,/6Æ6/9. ) AQI :-4, 7 (:.0, (##: /-’56/4, 5)6#0;<7; . "#, )&12.*-)= 8%$. =)-), *129/,, /2=6-). 7:6 67$&: 1) 1 =; 2) ;=; 3) 1>=. 1 =9:Æ6 -), 0";<)49*5, -)*5%& [3]. # , Stern [4] .? :@, 15; A, 1 =+- )=?6>. ?;=)?, Æ6, 6?-), >C;")) PM10 @E$+&. , )6 B*F*F*5, GGB*=HH>;"I-), <! 0= [7]. 1D?, AÆ?==@E=*1 >=9:Æ6- ,, @/B;= 2:B**5)C. 7’, A [8] BP Æ?=:/-). Sun [9] H - =)! PM2.5 +:/-). , 31>=A=GC, F "-)=GBH*)-)*&GIB*5 -);, A >=)IB* 5:). /I =-)*, *53D?JH C7B-)=9B -)K6. 7’, Xu [10] / ICEEMDAN-WOA-SVM =)CDILC 6 + +:-), 1.F B=2:- . M [11] Æ 6EFH C7);<:H , )&/ MGWO-SVR B=)I: -), 1.F, 3=50, B="//-), 2-). F$

2140 39 [12] -K3D?E@1>=)&/ FEEMD-VMD-DE-ELM B=)LE. O PM2.5 :-), 9/?=-)1.. F;NOF*53D?JH C72 M=-)*, "?I-)1.. , "*,JJH5KPQBG*53D?JH C7)& -)="!(+, /=QR)-)*L&. 6., GGB-)= !F -), ?")&, =-)> NG [13]. -)> NG, -)>S 1 (outlier robust extreme learning machine, ORELM) -)=, ")*52-). >", H N /+- JH C7 —— ERH C7 (modiﬁed butterﬂy optimization algorithm, MBOA) )-) =:H , =-)*. /I =-)*, - )&QRN, / MBOA >B*QR (error nonlinear correction based MBOA, MBOA-ENC) (K)=: QR, U-)1.. ONS 2017 74 10 P9/ 3 T F* —— I C, )7)&- 2*:, YP9/ 5)? 0Q:, I7)&- ) > :2- . 2 6789!$;%<&>?’ ONB*QR(K)&/+- , MA(N: *5-NH N-)NQRN, 6.ON’C: 2.1 @( )*, "13 (time varying ﬁltering based empirical mode decomposition, TVF-EMD) ? Li [15] 2017 +3 (empirical mode decomposition, EMD) EUO. EMD -B3 (ensemble empirical mode decomposition, EEMD) 50, TVF-EMD 6’CHC [16]: 1) >"3PJ"V; 2) 3P/V, "/3S"1F; 3) FCA? 0)W, :TCA6T=*; 4) H*"P9?. TVF-EMD *53D? 6.I’C: (t); -BC Æ6 Hilbert 1T;CIB3S x(t) P"9 A(t) P" ϕ -B P"9 A(t) Q’9 A({tmax}) Q’9 A({tmin}); -B ) A({tmin}) A({tmax}) :X9 β1(t) β2(t), ;CP"A9 a1(t) P" A? a2(t): -BE ) ϕ 1(t) ϕ 2(t): ϕ ({tmin})A2({tmin}) ϕ , 2 a1(t) = β1(t) + β2(t) a2(t) = β2(t) − β1(t) (2) ({tmax})A2({tmax}) :X9, η1(t) η2(t), ;C (1) 2 ; 1(t) = ϕ 2(t) = ϕ η1(t) 1(t) − 2a1(t)a2(t) 2a2 η1(t) 2(t) − 2a1(t)a2(t) 2a2 + + η2(t) 1(t) + 2a1(t)a2(t) , 2a2 η2(t) 2a2 2(t) + 2a1(t)a2(t) ; (3) (4)

8 , : -B ;CQ’R[ ϕ bis(t): 1(t) + ϕ 2(t) bis(t) = ϕ ϕ -B/ L-2 ϕ bis(t), 3P"V; -B0 !3S h(t) = cos[ 2 = η2(t) − η1(t) 4a1(t)a2(t) ; bis(t)dt], ( h(t) 9CSC, :6 B T=X9)3S x(t) : ϕ Y, Y1. m(t); -BÆ ;CR[5\ θ(t), ’. θ(t) ≤ ξ, \U! x(t) IMF, \, V x1(t) = x(t) − m(t), LU :I;I: θ(t) = BLoughlin(t) ϕavg(t) , (6) , BLoughlin(t) 83S Loughlin P"T, ϕavg(t) 3AP" . 2.2 1, RH C7 (butterﬂy optimization algorithm, BOA) ? Arora Singh [17] 2018 +CQH ERC7. R:5;!TB. R2 1W, ! [6. BOA VG+:VW"^H 9. RXW+V f, ?&+")P!, A1Y c\M I Z.* a, F1 f = cI a. !(+O* f(x), RH C7OI’C: -BC 66 n _RGB+ xi; -B H5(+O* f(x) !%_R xi \M Ii; -B ;CR+%_RGÆ9, ^HR g -BE !1Y cZ.* a TW p, 6* r !:QVWXQ’VW; -B ’. r < p, \:QVW, R(QHR, F1 ∗; i + (r2 × g ∗ − xt i) × fi, xt+1 i = xt ∗ , xt i R xi t ]^T3, g _H3, r 0 1 "*, fi R xi XW; -B/ ’. r ≥ p, \:Q’VW, R(::, F1 i + (r2 × xt k R xi, xj, xk t ]^T3; xt+1 i = xt j − xt k) × fi, , xt i, xt j, xt -B0 UC7 X, XX, \Z;I-, \C71Y, H3. ?RH C7=Q’VWN MH G%Z[ , /EC7H * , ON#[+-/-1\TIBRH C7Q’VW1, ERH C7 (modi- ﬁed butterﬂy optimization algorithm, MBOA), RZaZC7 (sine cosine algorithm, SCA)[18], 6 I/-15>], YP!Q’VWR6. , IIIBRH C 75>, aI’C: 2141 (5) (7) (8) (9) -B/ ’. r ≥ p, \6 0 1 "* r4, !’[:Q’VW; -B0 ’. r4 < 0.5, \:RZVW, F1 xt+1 i = xt i + r1 × sin(r2) × |r3 × g ∗ − xt |; i -BÆ ’. r4 ≥ 0.5, \:aZVW, F1 (10) , r1 = a × (1 − t/tmax), a $*, 99 2, t _^T]*, tmax ^T]*, r2 0 2π "*, r3 0 2 "*; xt+1 i = xt i i + r1 × cos(r2) × |r3 × g ∗ − xt |, -B2 UC7 X, XX, \Z;I-, \C71Y, H3.

2142 2.3 3, 39 @1> (extreme learning machine, ELM) ?-\0]1\9Z [19] 2006 , +3bK[Æ?, 613-);C]TO=#HC. \ O*K1Æ?50, 6?=-)N"1>N [20,21]. βiG(ai, bi, xj) = yj, j = 1, 2,··· , N, (11) , ai b]KSC i , bi i b]KSC9, G b]KEO*, βi b ]K i E K"Q29, L b]KE*, N ]-TO*. i=1 ELM -)=F1 fL = L / Fa, (0R (11) H(a1,··· , aL, b1,··· , bL, x1,··· , xN ) = Hβ = Y , ⎡ ⎢⎣ G(a1, b1, x1) ... G(a1, b1, xN ) β = [β1,··· , βL]T, Y = [y1,··· , yN ]T. ⎤ ⎥⎦ , ··· G(aL, bL, x1) ... ··· G(aL, bL, xN ) ... , H b]K ]c. 9]c<0R (15) (16) ;C: Hβ − Y = HH +Y − Y = minβHβ − Y , β = H +Y . , H + b]K ]c H Moore-Penrose 9^]c. , ELM =)]-J9$_1, KÆ6#7 =-)*. /bJ 9)=-)*&, Zhang Luo [22] /JQ@1> (outlier robust extreme learning machine, ORELM) =. ORELM =(+O*1YH , F1: , k R\ H*, 6]> 9^*"^ef, 5=6?=: N, e N ]TO-)>. 0R (17) 69_g (augmented Lagrange multiplier, ALM) =3, ^T6F1: (12) (13) (14) (15) (16) (17) (18) β2 minβe1 + s.t. e = Y − Hβ. 1 k 2 ⎧⎪⎨ ⎪⎩ βt+1 = arg minβLμ(et, β, λt), et+1 = arg mineLμ(e, βt+1, λt), λt+1 = λt + μ(Y − Hβt+1 − et+1). Y − et + λt 1 μ Y − Hβt+1 + λt −1 H T H + 2 kμI H T , μ . ⎧⎪⎪⎨ ⎪⎪⎩ βt+1 = et+1 = shrink , λt t ]^T_g2, μ = 2N/||Y ||1 ‘H*, βt+1 et+1 <0R (19) : μ , (19) 2.4 HI, ][-)=#=!-)>, /I =-)*, :-), QRI=-)1., I -)=*. >-)QR(KPQ >QR,LC7, =-)*. "*>QR,_= -)*, HH:6>-)9I=-)935(KU-)1., /35(K eG_QR(K= F N. "#, /b =-)*, ON /B*Q R(K, AC 3 I:

8 , : 2143 -BC 6>""‘2 ! Actual(t) ""‘2 t " 89, F orecast(t) t " -)9, \= t " -)>: Error(t) = Actual(t) − F orecast(t). (20) -B )&-)=, :>-) H5>""‘2F, ! Error(t − d) t − d -)>, f[·] )&-)=, )>‘2 :-), \>‘2 t " -)9: EF orecast(t) = f[Error(t − 1), Error(t − 2),··· , Error(t − d)]. (21) -B :B*QR, U-)9 PQ+2QR(K)=-)9:QR, =-)*. ON /+ MBOA >B*QR (error nonlinear correction based MBOA, MBOA-ENC) (K, <)&-)= F [·] U-)9: F F orecast(t) = F [EF orecast(t), F orecast(t)], (22) , F [·] B*O*, ? MBOA-ORELM =*5:]!, [ EF orecast(t) F orecast(t), F F orecast(t). 2.5 TVF-EMD-ORELM-MBOA-ENC 5K N7;6.67, ON /B*QR(K- —— TVF-EMD-ORELM- MBOA-ENC, M<8ac"- : GI-)acB*QRac. ac: GI-)ac Oac(+)& TVF-EMD-ORELM B-)=, " AQI GI-). GI-)ac, :6)&*5-N: AQI *5-, Æ6 TVF-EMD *53D?(IB AQI *53 G IMF . X, )d IMF Æ6-)N)& ORELM =:-), % -)9)d-)1.:-U-)9. 7 =Uh*, Oac-67P Q257. #, /7 =-)1.*, 5=6/6/9, (7)& TVF-EMD- ORELM =: 100 ], 9-)1.A9OacU-)1., ac- (/F! -)M. ac-: B*QRac Oac(+)ac-)1.:QR, B*QR(K, A8g=: >- )=>QR=. , >-)=, )>‘2:-), >‘2-)9; >QR =, (>-)9ac-)9:-U-)9. ac‘, ac--)= : 100 ], 9A9U-)1., - 2*. 3 LM?’ ONP9I TF*:, 7)&- 2*. 3.1 NO@( /7)&- 2*, ONP9I AQI *5:,. I, a #, cL, e1 B *Q&, kd?e; , f #, .49N , >"b( L +, ?f6Rg, SS, fg, ) :,6L; , e0 #, L&, L, Ac)& . mF7;, F; 6 >C6#F, " , F; 3 ,):

2144 39 6 1 79:Q;RS<TUV>; hi n n dgn d o 84 e 87 102 97.2927 111.0876 121.3858 388 483 499 22 25 21 j 3111.0826 4885.3111 4695.7992 ql dghÆj 6 2 @RB N N i=1 |Fi − Ai| N i=1(Fi − Ai)2 N ghhrq IA = 1 −N | × 100% i=1 | Ai−Fi ×N i=1(|Fi − ¯A| + |Ai − ¯A|)2 i=1(Fi − Ai)2/ i=1(Fi − Ai)2)/( ggj dghÆjmn M AP E = 1 N M AE = 1 N RM SE = iho N jfk T IC = i=1 A2 i + 1 N Ai 1 N 1 N 1 N N i=1 F 2 i ) - ,6LÆ6/9. #, ’.7)&- 3 6 >FL A"2- , \B)&- 2*. ON7:6 AQI *54*5, <)"" 2015 7 1 2017 6 30 , 731 *5, ’ 1 71. #, dC*55;*;.+9’F 1 71. ON(%CT O*5]-)-8’6ac AQI GI-), , P9 2015 7 1 2016 12 31 ; 550 *5]TO, P9 2017 1 1 2017 6 30 ; 181 *5 )TO. , B*QRac, (ac)TO]TO)TO8’, , P 9 2017 1 1 2017 5 31 ; 151 *5ac-]TO, P9 2017 6 1 2017 6 30 ; 30 *5)TO. 3.2 5KCD /)7)&- :",/, ONP9/ 5 +.+:,/, AAi)> (MAE)A6H> (RMSE)Ai)>p0 (MAPE)kD * (TIC) ;D$*.* (IA). d,/.+;C0R’F 2 71. , Ai i " AQI K9, Fi i " AQI -)9, ¯A AQI A9, N 7)TOf. , MAERMSE MAPE 6,/-), *9\F-); TIC 6,/ -)N, 99^l 0 1, 2 0 \F -)NM; : Nb,/=-)*L .+, ? IA 3=’-)N a;70, ON:6 IA ) >= : N:,/. mF7;, ON(S-) -)N: N 3 j)7)&: 1,/. 3.3 NOE 400 500 600 700 100 200 300 400 200 I Q A I Q A 300 400 0 0 500 100 200 300 /7)&- 2*, ON; /8:)0,, 6.’C. 3.3.1 F3 q, .**5#=JC, *-)=)J90?_1, #7 =-)*. /ON*5 = JC, :6+2T$)C7 —— & II (isolation forest, iForest)[23,24] )IB .**5:k, )1.’ 1 71, 100 0 0 500 400 300 200 100 0 0 I Q A 100 200 300 400 500 600 700 100 200 300 400 500 600 700 Z 1 [Æ AQI SIJKNP

8 , : 2145 " 3 C=!*JC. /bJC)=-)*&, ON:6)JT OC6?M: NJQ@1>)&-)=:-). 3.3.2 NOC :/t)0, (acGI-)1. ELMRELMORELMMBOA-ORELM EMD-ORELM :)0, ON7PQ*53D?ERH C7JQ@1> H*, ON B)(K2*. F 3 1/F;d= 3 -)1.. ?F 3 ,/.+, TVF-EMD-ORELM =,/.+ 3 =, GF ELMRELMORELMMBOA-ORELM EMD-ORELM =50M=6/=-)-) N: N. X, <)0 >="-)1., [=H*, 6.:&)0: 1) 3-)=, ORELM =6=-)1., p ORELM =)JTOC: N/=, !F =-)*; 2) ORELM =50, MBOA-ORELM =-)1./=, p ERH C72 =-)*; 3) <) ORELMEMD-ORELM TVF-EMD- ORELM 3 =:)0, " 3 =-)1.]1=, /*53D?B )(K2*, "F TVF-EMD *53D?%H EMD *53D?, /F =-)*. 6 3 ]^_-Q‘ a

资料库

论文研究-基于非线性修正策略的空气质量预警系统研究.pdf

相关推荐

开发技术

热门标签

最新资料