| Intelligent Transportation Systems (ITS) is a major subject in the transportationfield under study at present in the world. It came into being under the backgroundof contemporary science and technology full development. ITS is considered as amost effective measure to settle urban traffic jam, improve driving safety, raiseoperational efficiency , as well as reduce air pollution. ITS will become the modeand developing trend of the modernized transportation system in 21st century .It isan important sign for transportation entering information age.Traffic information is not only the prerequisite and main content of ITS but alsoa key factor for carrying out study smoothly. Prediction of dynamic trafficinformation is an important part of ITS. Proper estimation of network travel time isa main parameter of real-time dynamic prediction; it is also a crucial step duringthe researching of traffic flow guidance and study of theory.While analyzing road networks, the most basic parameter, which can express theoptimum model of system, is traffic volume (traffic flow rate) passing the routesegment or node, Secondly is impedance of each route segment. We usually taketravel time as impedance for it may be representative of physical attribute of roadsegment characteristics. For subscribers who care for traffic flow guidance systemor others, journey time is a most direct parameter. Travel time prediction is a keypart for realizing ITS since it aims at using road rationally, but it is also a maindifficult work because of its characters of real-time and dynamic.Transportation system is a non-linear system whose structure and inner variablesare complicated .It has much input and export parameters. Adopting somefactors and parameters or individual model can only reflect part of system, so weintroduce scientific integration of many variables with data obtained by differentdetectors or effective compound of many models, which can remarkably improveprediction accuracy and estimation of simulation. Recently, integrated non-linearsystem technology, data fusion technology are active day by day. We select thedynamic compound prediction to resolve the prediction of real-time dynamictraffic information in this paper. This paper includes five parts as following:1. A new processing method for traffic investigation data is brought forward.2. A compound model for real time traffic flow prediction is given. First we put forward several sub models, then through optimum weightingintegration, we get the following models: x? (t) =ωj (t)x? (t) +ωj (t)x? (t) +ωj (t)x? +ωj (t)x? (t) +ωj (t)x? (t) (1) j 1 j1 2 j2 3 j3 4 j4 5 j5In the equation, x? (t), x? (t), x? (t), x? (t),x? (t) are predicted values at time t j1 j2 j3 j4 j5respectively gotten by (2),(3),(4),(5),(6); Wj(t) = ωj (t), ωj (t), ωj (t), ωj (t), ωj (t) ( ) T is a column vector of 1 2 3 4 5weighting coefficient for each sub models. An optimum model decides it.Followings are sub models: Sub model 1: x? (j (t) =a0 + a1 t + a2 t2 k) (2) 1 k k k In equation (2) , t stands for time, a0 , a1 , a2 does for regression k k k coefficients Sub model 2: x? (j (t) = b0 + b1 x(j (t ?1) + b2 x(j (t ? 2) k) k) k) (3) 2 k k kIn above equation, b0 , b1 , b2 are regression coefficients; k k kx j(t ?1) and x j(t ? 2) denote traffic flow for route segment j at time t-1and t-2respectively.Sub model 3: x? (j (t) = c0 + c1 x j (t ?1) + c2 x j (t ? 2) + c3 x j1?1(t ?1) k) 3 k k ?1 k ?1 k + c4 xj1?1(t ? 2) + c5 xj2?1(t ?1)+ c6 xj2?1(t ? 2) (4) k k kc0 , c1 , c2 , c3 , c4 , c5 , c6 are coefficients. Route segment k k k k k k k j1 ?1 ,j2 ?1are neighbors of j,t-1,t-2 denote one or two moments prior to time t . Sub model 4(superposing sub model with trending nuchal) Traffic flow of route segment j at time t is related to t and have some trend, sowe give the hypothesis: x? (j (t) = d0 ed1 + x? (j (t) k) kt k) (5) 4 k 2d0 , d1 are coefficients. k k Sub model 5(liner superposition sub model)For traffic flow of segment j at time t is related to t, so x? (j (t) = h0 + h1 t + x? (j (t) k) k) (6) 5 k k 2 h0 , h1 are superposing coefficients. k k3. Road network real-time dynamic travel time compound prediction model andapplication method 1) The following is a Road network real-time dynamic travel time compoundprediction model and application method whose traffic flow submits to steadyPoisson Distributions. And the trip includes signalized intersection. Given the following assumption: the jth road segment is marked j(0 ≤j ≤n );the vehicle travel time over the jthroad segment during T interval is marked Tj (r)(t) ; the queue time at the followingintersection of the jth road segment is Tj (q) (t) ; the time traveling over thefollowing intersection of the jth road segment is marked as Tj (t) . (c) In order to calculate the average travel time(Tj(t) ) for the jth road segment,the equation is as the following: Tj(t) = Tj (t) + Tj (t) + Tj (t) + Tj (t) (r) (q) (c) (e) (7)where,Tj (t) = Tij =(Lj ? L(ij ) vi . (r) (r) q) The average queue time for a single lane isHere: L ——the length of the jth road segment. j L(ij ——the average queue length in the downstream intersection of the jth q)road segment during i time interval. vi ——the average queue length for all vehicles in the network during iinterval. T(ij ——the average travel time for the jth road segment during i interval. r) To improve prediction accuracy, we add an adjusted item into multi variablestatistic regression method. We put forward a common equation and work outprediction results. We can get the exact value of invariable parameter through uniting. And then,about given data of volume, road segment length, split and so on. We can get theoperational time of reciprocal road. So the method have virtue of easily gatheringdata, adjusting abduction interval, fast speed of calculation, high precision and soon. It improves the practicability of running time forecasting model, realize thefunction of real time dynamic abduction about flow. It is easy to extend and apply. 2) Road network real time dynamic operational time forecast application... |
PDF Full Text Request