FREEWAYTRAFFICFLOWMODELINGBASEDONRECURRENTNEURALNETWORANDWAVELETTRANSFORM基于遞歸神經(jīng)網(wǎng)絡(luò)和小波變換的高速公路交通流建模6Conclusions^吉論Thehighlynonlinearanddynamiccharacteristicsofthemacroscopictrafficflowrequireamodelingapproach,whichcandealwiththecomplexnonlinearrelationshipsamongthespeed,flowanddensity.宏觀交通流的高度非線性和動態(tài)特性，需要一個建模方法，它可以處理的速度，流量和密度之間的復(fù)雜的非線性關(guān)系。Atthesametime,effectivemeasureshavetobetakentoeliminatetrafficnoiseanddisturbance.Inthispaper,amethodofwaveletdenoisingandtrafficflowmodelingbyanElmanrecurrentneuralnetworkispresented.同時，必須采取有效措施來消除交通噪聲和干擾。在本文中，一種小波去噪和交通流的Elman神經(jīng)網(wǎng)絡(luò)建模這工作了。Simulationresultsshowthatthewavelettransformcaneffectivelyeliminatethenoisesignal,andthattheElmannetworkcanaccuratelydescribetherealbehavioroffreewaytrafficflow.仿真結(jié)果表明，小波變換可以有效地消除信號中的噪聲，和Elman神經(jīng)網(wǎng)絡(luò)能夠準確地描述高速公路交通流的真實行為。Thismethodisofgreatimportancetorealizeon-linemodelingandcontroloffreewaytrafficflow.此方法對實現(xiàn)高速公路交通流的在線建模與控制具有重要意義。FuzzySelf-AdaptivePIDControllerforFreewayRampMetering高速公路匝道控制的模糊自適應(yīng)控制器Abstract:Aimingatthenonlinearandtime-varyingcharacteristicsoffreewaytrafficsystem,afuzzyself-adaptivePIDcontrollerisdesignedandappliedtofreewayrampmeteringinthispaper.摘要：針對高速公路交通系統(tǒng)的非線性、時變特性，設(shè)計了一種模糊自適應(yīng)控制器，并將其應(yīng)用于高速公路匝道控制系統(tǒng)。Atrafficflowmodeltodescribethefreewayflowprocessisfirstlybuilt.Basedonthemodelandinconjunctionwithnonlinearfeedbacktheory,afuzzy-PIDrampcontrolleristhendesigned.TherampmeteringrateisdeterminedbythePIDcontrollerwhoseparametersaretunedbyfuzzylogicaccordingtothedensitytrackingerroranderrorvariation.首先建立了一個交通流模型來描述高速公路的流程。基于該模型，并結(jié)合非線性反饋理論，一個模糊的控制器，控制器，然后設(shè)計。斜坡根據(jù)密度跟蹤誤差和誤差變化情況，通過模糊邏輯對控制器參數(shù)進行調(diào)整，確定了測量速度。Gaussandtrianglecurvesareusedforthemembershipfunctionsofthefuzzyvariables.Therulebaseincluding49fuzzyrulesisalsoestablished.Finally,thecontrolsystemissimulatedinMATLABsoftware.Theresultsshowthatthiscontrollerdesignedhasfastresponse,gooddynamicandsteady-stateperformance.Itcanachieveadesiredtrafficdensityalongthemainlineofafreeway,andcanmakevehiclestravelmoreefficientlyandsafely.Thisapproachisquiteeffectivetofreewayrampmetering.高斯和三角曲線用于模糊變量的隸屬函數(shù)。還建立了包括49個模糊規(guī)則的規(guī)則庫。最后，對控制系統(tǒng)進行了仿真研究的軟件。結(jié)果表明，該控制器具有響應(yīng)快、動態(tài)性能和穩(wěn)態(tài)性能。它可以實現(xiàn)所需的交通密度，高速公路主線，并能使車輛行駛更安全、更安全。這種方法是相當(dāng)有效的高速公路匝道控制。2.FreewayTrafficFlowModel高速公路交通流模型Consideramultiplelane(入)freewaysectionwithasingleentryramp,asshowninfigure1.考慮多車道(入)具有單一入口匝道路段，如圖1所示。Assumethatattimeslicek,vehiclesflowintoagivensectionatarateofqu(k)vehiclesperhourperlanefromitsupperboundaryandr(k)vehiclesperhourfromtheentryramp.假設(shè)在時間片上，車輛進入一個給定的區(qū)段在一個區(qū)段(克)每車道從它的上邊界和(鉀)車輛每小時從入口匝道的速度。Theydischargeatarateofq(k)vehiclesperhourperlaneatitslowerboundary.他們以每小時的速度在其較低的邊界處排放一個問(克)的車輛。Bytheconservationprinciple,thenumberofvehiclesinthisfreewaysegmentattime按養(yǎng)護原則，在這條高速公路的車輛數(shù)量在時間k+1,N(k+1)wouldbeN(k1)=N(k)t[qu(k)-q(k)r(k)](1)Figure1AfreewaysectionN(k1)=N(k)t[qu(k)-q(k)r(k)](1)Definetrafficdensityas,whereislengthoftheroadsegment.Equation(1)cannowbewrittenintermsofdensity定義交通密度,哪里是公路段的長度。方程(1)現(xiàn)在可以用密度來書寫：(k1)=：(k)(t/：x)[qu(k)-q(k)r(k)/]⑵Empiricalevidencesuggeststhatarelationshipbetweenflowandtrafficdensityexists.Thisrelationship,denotedasisgenerallyreferredtoasthefundamentaldiagramoftrafficflow.Manyformsoffundamentaldiagramshavebeenproposed,allofwhichsharesomecommonfeatures:經(jīng)驗證據(jù)表明，流量和交通密度之間的關(guān)系存在。這種關(guān)系，表示為通常被稱為交通流基本圖。許多形式基本圖已經(jīng)被提出，所有這些都有一些共同的特點：(1)flowiszerowhendensityiszero;(2)thereisamaximumdensity(oftenreferredtoasjamdensity)thatcorrespondstobumper-to-bumpertraffic(atwhichflowisalsozero);and(3)thereexistsatrafficdensityatwhichflowismaximal(oftenreferredtoascriticaldensity).(1)密度為零時的流量為零；(2)有一個最大密度(通常稱為果醬密度)，對應(yīng)于保險杠到保險杠的流量(在流量也是零)；O)存在流量最大的交通密度(通常稱為臨界密度)。Whentrafficdensityisbelowcriticaldensity,flowincreaseswithdensity;andwhentrafficdensityisabovecriticaldensity,flowdecreaseswithdensity.交通密度低于臨界密度時，流量隨密度增大而增大；當(dāng)交通密度超過臨界密度時，流量隨密度的增加而減小。Equation(2)andthefundamentaldiagramconstituteadiscreteformofthewellknowntrafficflowmodelcalledLWRmodel[4].方程(2)和基本圖構(gòu)成的離散的眾所周知的交通流模型的lwr模型[4]形式。TheLWRmodelprovidesasimpleandtractableframeworkfortrafficsimulationandcontrolstudies.lwr模型提供了一個簡單的框架，交通仿真與控制研究。Itgivesgoodqualitativeresultsindescribingsuchtrafficphenomenaasshockwavesattrafficbottlenecks,progressionofatraffichump,andstartingandstoppingtrafficatintersections.它提供了良好的定性結(jié)果，在交通瓶頸，交通駝峰的進展，并停止交通在路口的交通瓶頸，描述這樣的交通現(xiàn)象。TheLWRmodelisusedinthispaperasthesystemmodelfordevelopingtraffic-responsiverampcontroller.lwr模型作為本文發(fā)展交通響應(yīng)匝道控制器的系統(tǒng)模型。Substituteinto(2),anddefine,weobtainequation(3).替代成(2),并定義，我們得到方程(3)。■(k1)=：(k)一二f[：(k)]：qu(k)：r(x)/■(3)Withinitialcondition,equation(3)givesacompletedescriptionofthefreewayflowprocess.初始條件，方程(3)給出了高速公路交通流過程的完整描述。FuzzySelf-AdaptivePIDControllerforFreewayRampMetering高速公路匝道控制的模糊自適應(yīng)控制器Theoptimizationprocedureswithgeneticalgorithmareasfollows.遺傳算法的優(yōu)化程序如下(1)Determinationofoptimizedparametersandencodedmode.優(yōu)化參數(shù)和編碼方式的確定TheparametersneededtobeoptimizedareproportionalcoefficientKpandintegralcoefficientKi,andtheencodedmodeisrealnumbercode.待優(yōu)化參數(shù)比例系數(shù)KP和積分系數(shù)Ki和編碼方式是實數(shù)編碼。(2)Generationofinitialpopulationrandomly隨機初始種群的生成TheinitialpopulationisthegenecodesofPIparameters.初始種群是參數(shù)的基因編碼。Evaluationoffitnessfunction健身功能評價Fitnessfunctionisthelinkerbetweentheproblemandthegeneticalgorithm.EachchromosomewillbetranslatedintothePIparameterstocontroltheobjective.TheperformanceofeachcandidatePIcontrolisevaluatedthroughthevalueofitsfitness.健身功能是連接問題和遺傳算法的連接程序。每一個染色體將被翻譯成一個參數(shù)來控制目標。每個候選的性能控制是通過其健身價值評估。Aperformanceindexisaquantitativemeasureofsystemperformance.Forfreewayrampmetering,agoodPIcontrolshouldminimizetheerrorbetweenthedesiredtrafficdensityandtheactualtrafficdensitytozeroquickly,soaperformanceindexisdefinedas(8)性能指標是衡量系統(tǒng)性能的一個定量指標。對于高速公路匝道控制，一個良好的控制應(yīng)盡量減少所需的交通密度和實際交通量之間的誤差密度迅速降至零，所以性能指標定義為(8)wheree(k)isthedensitytrackingerror.Thebiggerthesumsofsquare-errorofaPIcontrol,thesmallerthefitnessfunction.Therefore,thefitnessfunctionisassigned,toeachcandidatePIcontrol.其中電子密度跟蹤誤差。一個有點控制的平方誤差的總和越大，越小的適應(yīng)度函數(shù)。因此，健身功能是分配的，每個候選PI控制ROLSelectionoperator選擇算子Besidestheroulette-wheelselection,weadoptaspecialselectionoperator,elitismoperator[8],whichsidestepsselection,crossover,andmutationandsimplyplacesthefittestPIcontrolfromthepreviousgenerationintothenextgenerationwithoutmodification.ElitismoperatorcanensurethatthebestPIcontroldon'thavetheprobabilityofmissing.除了輪盤賭選擇，我們采用了一種特殊的選擇算子、精英算子[8],它回避了選擇、交叉、變異和簡單的地方，優(yōu)勝劣汰的PI控制的上一代沒有修改的下一代。精英算子可以確保最佳的PI控制沒有丟失的概率。Crossoveroperator交叉算子Crossoveroperatorproducesnewindividualsincombiningtheinformationcontainedintwoparents.Thisisdonebycombiningthevaluesoftwoparents.Amethodcalleddiscreterec