CN105701571A - Short-term traffic flow prediction method based on nerve network combination model - Google Patents

Abstract

The invention provides a short-term traffic flow prediction method based on a nerve network combination model. The method is used to construct a counterpropagation nerve network combination prediction model and the short-term traffic flow prediction method is provided based on the model. Aiming at a characteristic of a traffic flow, a fuzzy C mean value clustering algorithm is used to cluster the traffic flow. For bunch generated through clustering, a counterpropagation nerve network prediction model is constructed. According to grade of membership, a weighted sum of prediction model prediction results is calculated and is taken as a final prediction result. In order to increase prediction precision, a taguchi method is used to carry out test designing so as to test influences of different structure parameters on prediction model prediction precision, and an optimum structure parameter is used as an initial structure of the prediction model. By using the method in the invention, the prediction precision of the short-term traffic flow can be effectively increased, an influence of a noise on the prediction precision in training data is reduced and operation time is reasonable.

Description

A kind of short-term traffic flow forecast method based on neural network ensemble model

Technical field

The present invention relates to a kind of short-term traffic flow forecast method, utilize neural network ensemble model to promote the precision of prediction of Short-Term Traffic Flow, reduce the noise impact on precision of prediction in training data, belong to the interleaving techniques application of traffic flow and neutral net。

Background technology

Intelligent transportation system be a kind of energy in real time, accurately, the traffic management system of Effec-tive Function, it has merged the information technology of advanced person, data communication technology, electron controls technology and computer processing technology effectively。The core technology traffic control technology of intelligent transportation system and traffic guidance technology are to solve the traffic congestion in city and improve road network traffic efficiency most effective way, are study hotspots in recent years。Realize the basis of traffic control and traffic guidance be then in real time, short-time traffic flow forecast accurately。

Traffic flow forecasting, according to predicted time length, can be divided into long-term volume forecasting and short term traffic forecasting。Long-term volume forecasting provides the several months even traffic flow forecasting of several years, is mainly used in building the Long-term planning of traffic and transportation system。Short term traffic forecasting focuses on the change in a short time of predicting traffic flow amount, the general forecast flow of following 5 minutes。Short-Term Traffic Flow has non-linear, time variation and the uncertainty of height, this uncertainty may be from environmental factors, such as pavement behavior, Changes in weather etc., it is likely to from emergency situations, such as vehicle accident, big assembly etc., these factors cause Short-Term Traffic Flow Accurate Prediction comparatively difficult。

For the prediction of Short-Term Traffic Flow, researcheres have been proposed for multiple model。The forecast model of early stage, such as exponential smoothing model, time series models and history averaging model mainly based on calculus and mathematical statistics etc., has the problems such as historical data demand is big, precision of prediction is not high。Afterwards, some modern science and technology and method have been introduced in prediction gradually, occur in that some prediction effects model better, such as Kalman filter model, nonparametric Regression Model, gray model for prediction and neural network model etc.。

Artificial neural network is a dynamic system with nonlinearity, and it has powerful nonlinear fitting ability, and thus be accordingly used in forecasting traffic flow has specific advantage。Reverse transmittance nerve network (BP neutral net) model is one of forecast model of being most widely used of forecasting traffic flow aspect, and the reverse transmittance nerve network with a hidden layer can approach any one continuous function with arbitrary accuracy。But, reverse transmittance nerve network forecast model also has many shortcomings, for instance convergence rate slowly, is easily absorbed in local minimum and precision of prediction by its structure influence etc.。Therefore, reverse transmittance nerve network forecast model needs to improve in precision of prediction。In order to promote the predictive ability of neutral net further, it is combined by researcheres with other intelligent method or statistical method, constructs combination forecasting。These built-up patterns often have better precision of prediction relative to single model。

Cluster is process set of data objects being divided into subset, and each subset is one bunch so that the object in bunch is similar each other, but dissimilar with the object in other bunch。Clustering method is broadly divided into four classes: division methods, hierarchical method, density based method and based on the method for grid。K average is exactly a kind of typical division methods, it is necessary to the number specified in advance bunch, and bunch division be mutual exclusion。But and not all object all determine and belong to one bunch, it is also possible to belong to several bunches, then can fuzzy set concept for cluster, this fuzzy clustering method that has just been born。Fuzzy c-means is exactly a kind of typical fuzzy clustering method, and each object belongs to different bunches with certain degree of membership, and it is also based on the method divided, it is necessary to the number specified bunch。

Taguchi's method is a kind of quality engineering method of low cost, high benefit, and it emphasizes that the raising of product quality is not by inspection, but by designing。Its thought is mainly in the development and Design stage that product is initial, by selecting design parameter around set desired value, and reduce variation through overtesting bottom line, thus quality is building up in product, make all over products produced have identical, stable quality, greatly reduce cost and loss。In field of quality control, Taguchi's method has been successfully applied to and high-quality product reliable with low-cost design, for instance automobile and consumer electronics product。

Summary of the invention

Technical problem: it is an object of the invention to provide a kind of short-term traffic flow forecast method based on neural network ensemble model, the method solves the problems such as precision of prediction is low, noise is big on the impact of precision of prediction in training data, the time of running is unreasonable of Short-Term Traffic Flow。

Technical scheme: the present invention will build a kind of reverse transmittance nerve network combination forecasting, and on the basis of reverse transmittance nerve network combination forecasting, propose short-term traffic flow forecast algorithm TFBCM, characteristic first against traffic flow, use Fuzzy C-Means Cluster Algorithm that traffic flow is clustered, each bunch that cluster is generated builds a reverse transmittance nerve network forecast model, and asks weighted sum that each forecast model predicts the outcome as finally predicting the outcome according to degree of membership。In order to promote precision of prediction further, adopt Taguchi's method to carry out EXPERIMENTAL DESIGN to test the Different structural parameters impact on forecast model precision of prediction, and use optimum structure parameter as the initiating structure of forecast model。

Short-term traffic flow forecast method based on neural network ensemble model of the present invention, the method builds a kind of neural network ensemble forecast model, utilize the characteristic of traffic flow, with Fuzzy C-Means Cluster Algorithm, traffic flow is divided into a small amount of pattern, and every kind of flow rate mode is set up a neural network prediction model。

Short-term traffic flow forecast method based on neural network ensemble model comprises the following steps:

Step 1) structural parameters of short-term traffic flow forecast are set, specifically comprise the following steps that

Step 1.1) number will be clustered as factors A, make factors A represent the classification of traffic flow, the level of factors A is set, the minimum level of factors A is taken as 3, the highest level is taken as 5, and middle level is taken as 4。Described level represents the consumption or the state in which value that refer to that each factor takes。

Step 1.2) using input number of nodes as factor B, three levels of factor B respectively 1,3,5 are set, these three level represents node three kinds of situations less, general, more successively；

Step 1.3) using node in hidden layer as factor C, factor C is by empirical equationDetermining, described l is node in hidden layer, and arranging n is factor B, and namely n is input number of nodes, and arranging m is output node number, and δ is value constant between 1 to 10。Three levels of factor C are set to 2,3 and 7。

Step 1.4) using learning rate as factor D, factor D value is between 0.01 to 0.1, and its three levels are set to 0.02,0.04 and 0.07。

Step 1.5) use Taguchi's method to test, it is thus achieved that factors A, factor B, factor C, the factor D level when effect value peak, using these levels initial value as structural parameters。In Taguchi's method, useAssessing the effect value of each design factor, described k is the number of times that often group test performs, e_MREIt is average relative error,For the predictive value of traffic flow, the actual value of the traffic flow that θ (T) inputs for user, T is a certain predicted time point, and N is the sum of predicted time point, and p and q is variable。Described Taguchi's method is a kind of quality engineering method, and the method is by selecting design parameter around set desired value, and reduces variation through overtesting bottom line, makes all over products produced have identical, stable quality。

Step 2) traffic flow is clustered。User inputs the data set X={x being made up of data object₁,x₂,...x_n, threshold values ε and maximum iteration time maxt；The cluster centre being set with classification composition is V={v₁,v₂,...v_c, u_ijIndicate that jth data object x_jBelong to i-th classification v_iDegree of membership, all u_ijComposition subordinated-degree matrix U, target setting function J be fuzzy set theory square distance and, n is data count, c for cluster number, m be not less than 1 weighting parameters, d_ijFor object x_jWith cluster centre v_iBetween distance, i and j is variable, arranges iterative steps t。Step 2) specifically comprise the following steps that

Step 2.1) c is set as step 1) level of factors A that obtains, user inputs m, is assigned to u with the random number between 0 to 1_ij, all u_ijForm initial subordinated-degree matrix U so that it is meetIterative steps t is set to 0。

Step 2.2) useCalculate the current each element value of cluster centre V, described inIt is u_ijValue when the t time iteration,Represent v_iValue when the t time iteration。

Step 2.3) useCalculate J^t, described J^tIt is the J value when the t time iteration, d_ijFor data object x_jWith classification v_iBetween Euclidean distance, described Euclidean distance is the actual distance in space between two points。

Step 2.4) utilizeSubordinated-degree matrix U is modified by the distance according to object to cluster centre, described inIt is u_ijValue after the t time iteration, d_kjFor data object x_jWith classification v_kBetween Euclidean distance。

Step 2.5) useCalculate J^t+1, described J^t+1It it is J value after the t time iteration。

Step 2.6) judge whether to meet | J^t+1-J^t|≤ε or t is equal to maxt, if it is satisfied, then forward step 3 to), otherwise t=t+1, turn to step 2.2)。

Step 3) use reverse transmittance nerve network to be trained。Described reverse transmittance nerve network, also referred to as BP neutral net, is the multilayer feedforward neural network by Back Propagation Algorithm training。Step 3) specifically comprise the following steps that

Step 3.1) using step 1) the factor B level value that obtains is as network input layer nodes n, factor C level value is as node in hidden layer l, use output layer nodes m, factor D level value is as learning rate λ, and user provides the connection weights ω between input layer, hidden layer and output layer neuron_rsAnd ω_suArranging initial value, user inputs initial hidden layer threshold values a again₀, output layer threshold values b₀, reset maximum iteration time maxt, iterative steps t be set to 0。

Step 3.2) according to user input data set variable X, utilize formulaCalculate hidden layer output H_s, s=1,2 ..., l；

Step 3.3) utilizeCalculate the prediction output O of reverse transmittance nerve network_u, u=1,2 ..., m。

Step 3.4) user inputs desired output Y_u, utilize e_u=Y_u-O_uCalculate forecast error e_u, u=1,2 ..., m。

Step 3.5) utilize ${\mathrm{\ω}}_{rs}={\mathrm{\ω}}_{rs}+{\mathrm{\λH}}_s(1-H_s)x_r\underset{k=1}{\overset{m}{\Σ}}{\mathrm{\ω}}_{ju}{e}_u,$ With ${\mathrm{\ω}}_{\mathrm{su}}={\mathrm{\ω}}_{\mathrm{su}}+{\mathrm{\λH}}_s{e}_u,$ By forecast error e_uBring renewal network into and connect weights ω_rsAnd ω_su, r=1,2 ..., n, s=1,2 ..., l, u=1,2 ..., m。

Step 3.6) utilizeAnd b_u=b_u+e_u, bring forecast error e into_uUpdate network node threshold values a_sAnd b_u, s=1,2 ..., l, u=1,2 ..., m。

Step 3.7) judge whether iteration t has reached maximum iteration time maxt, if reaching, then forward step 4 to), otherwise t=t+1, turn to step 3.2)。

Step 4) carry out short-term traffic flow forecast。User inputs predicted time point T successively, from step 3) the prediction output of the reverse transmittance nerve network that obtains, take c predictive value O₁,O₂,...,O_c；According to step 3) predicted time point is for the degree of membership of each bunch in subordinated-degree matrix U, by degree of membership as weights, asks the weighted sum of each bunch corresponding reverse transmittance nerve network predictive value as the predictive value O at predicted time point T_T, namelyDescribed T is predicted time point, T=1,2 ..., n。

Beneficial effect: this invention address that short-term traffic flow forecast problem, each forecast model predicts the outcome use fuzzy set theory summation as final predictive value, relax the noise impact on precision of prediction in training data to a certain extent, there is following beneficial effect:

(1) present invention uses Taguchi's method to select suitable structural parameters, and arranges the structure of forecast model according to these parameters, improves precision of prediction further。

(2) present invention is when artificial specified structure parameter, precision of prediction has just been significantly higher than traditional reverse transmittance nerve network algorithm and CITFF algorithm, and compare genetic algorithm, use Taguchi's method choice structure parameter, it is possible to reduce more forecast error in the short period of time。

(3) present invention is higher for the precision of prediction of Short-Term Traffic Flow, and the time of calculating is reasonable。

Accompanying drawing explanation

Fig. 1 is short-term traffic flow forecast solution framework figure。

Detailed description of the invention

The present invention is directed to crossroad short-term traffic flow forecast, in being embodied as, consider the multiformity of traffic flow distribution form, model training adequacy and precision of prediction, reduce stochastic generation or artificial these parameter values of appointment。Below that the present invention is for a more detailed description。

First the present invention builds a kind of reverse transmittance nerve network combination forecasting in being embodied as, and this model is combined by reverse transmittance nerve network forecast model, fuzzy C-means clustering model and structural parameters preference pattern。

Reverse transmittance nerve network is a kind of multilayer feedforward neural network, and it is mainly characterized by signal propagated forward, error back propagation。In propagated forward, input signal successively processes from input layer through hidden layer, until output layer。The neuron state of each layer only affects next layer of neuron state。If output layer can not get desired output, then proceed to back propagation, adjust network weight and threshold values according to forecast error, so that reverse transmittance nerve network prediction output constantly approaches desired output。The present invention adopts the reverse transmittance nerve network of single hidden layer to carry out the prediction of Short-Term Traffic Flow in being embodied as。

Fuzzy C-means clustering provides the approach how data point of hyperspace is grouped into certain number of group, first the method randomly selects some cluster centres, all data points are all endowed the fuzzy membership that cluster centre is certain, then pass through alternative manner and constantly revise cluster centre, in iterative process with all data points of minimization to the distance of each cluster centre be subordinate to the weighted sum of angle value for optimization aim。The present invention adopts fuzzy clustering method that traffic flow is divided in being embodied as, and so each time point just can belong to different pattern with certain degree of membership。Owing to cluster accuracy has little influence on the precision of forecast model, admittedly the FCM Algorithms adopting amount of calculation relatively small clusters, with the whole efficiency of boosting algorithm。

The selection of prediction model parameters is most important for its precision of prediction, should by verifying the different parameters impact on precision of forecasting model。In EXPERIMENTAL DESIGN, the factor represents the independent variable in test, is the relevant factor and the condition that affect test index, and level represents consumption or the state in which that each factor takes。Current test design method mainly has 4 kinds: trial and error pricing, Graph One factor test method(s), total divisor test method(s) and Taguchi's method。Taguchi's method considers the advantage taking into account a Graph One factor test method(s) and total divisor test method(s), from all possible combination, select that there is typicality, representational combination, test combinations is made to be uniformly distributed in global scope, good overall view can be reflected, wish that again test combinations is few as much as possible, then use orthogonal table。The present invention adopts Taguchi's method to design and test in being embodied as, and obtains the effect value of each design factor according to result of the test, selects the Reasonable Parameters of combination forecasting so that it is precision of prediction promotes further。

The present invention is in being embodied as, the structural parameters of Clustering Model and forecast model are determined initially with Taguchi's method, then use Fuzzy C-Means Cluster Algorithm that traffic flow is clustered, and according to bunch division training data generated, train a reverse transmittance nerve network forecast model corresponding to each bunch。During prediction, with all of forecast model, predicted time point is predicted, and according to the predicted time point degree of membership for each forecast model, obtains the weighted sum of each predictive value, as final predictive value。This model framework is as shown in Figure 1。

The present invention is in being embodied as, and on the basis of reverse transmittance nerve network combination forecasting, carries out short-term traffic flow forecast。Considering the multiformity of traffic flow distribution form, flow dimension is substantially divided by the present invention, uses different forecast models to describe different flow rate modes, and the optimum number divided then is determined with test。In order to avoid the problem that pattern quantity is too much brought, the present invention in being embodied as general constrained clustering number less than 5。Due to the uncertainty of Short-Term Traffic Flow, the present invention uses degree of membership to represent that each time point belongs to the probability of certain pattern, and seeks, according to degree of membership, the weighted sum that each forecast model predicts the outcome, as final predictive value。For promoting the precision of prediction further, the present invention adopts Taguchi's method design experiment, defines the Rational structure parameter of Clustering Model and forecast model with less test number (TN), uses the structure of these parameter initialization models, it is possible to make to predict the outcome more accurate。

1, the choosing of structural parameters

The structure of reverse transmittance nerve network and initial weights thereof and threshold values are on its precision, convergence rate and whether can be absorbed in local minimum and all can have certain impact。The present invention considers choosing of reverse transmittance nerve network structural parameters in being embodied as。

The present invention is in being embodied as, and design factor A directly affects the output of Clustering Model, and design factor B, C and D determine the structure of neural network prediction model。The selection of each design factor and level thereof describes as follows:

1) design factor A: cluster number determines FCM Algorithms to be needed to gather traffic flow for several classes, also just determines and needs to set up several forecast model。This parameter is relatively big on the impact of precision of prediction, need to judge its optimum value by test。According to general knowledge, traffic flow at least can be divided three classes by its size, i.e. peak, low ebb and common discharge, so its floor level is defined as 3。Considering the problem existing for efficiency and CITFF algorithm, cluster number is unsuitable too much, so highest level is defined as 5, the 2nd level then takes intermediate value 4。

2) design factor B: input number of nodes determines the number of the historical traffic flows information being supplied to reverse transmittance nerve network forecast model。If input number of nodes is very few, then it is not provided that enough information is to forecast model, causes that precision of prediction declines；And if input number of nodes is too much, some noise informations can be introduced again, also can impact prediction precision。The three of input number of nodes level set are 1,3 and 5 by the present invention in being embodied as, and cover less, the general and more three kinds of situations of input node respectively。

3) design factor C: node in hidden layer affects the learning capacity of neutral net, if very few, then cannot produce enough connection weight number of combinations to meet the study of some samples；If too much, then the generalization ability of study network later is deteriorated。For the determination of node in hidden layer, determined by below equation:Described l is node in hidden layer, and arranging n is factor B, and namely n is input number of nodes, and arranging m is output node number, and δ is value constant between 1 to 10。Owing to input number of nodes is set to 1,3 and 5, and output node number is fixed as 1, according to this formula, the three of node in hidden layer levels can be respectively set to 2,3 and 7。

4) design factor D:BP algorithm learns by reverse propagated error and in the way of revising weights threshold values, and the size of learning rate all has considerable influence for convergence rate and training result。If learning rate is too little, study can be made excessively slow；If learning rate is too big, then may result in vibration or disperse。Rule of thumb, the value of learning rate should be comparatively suitable between 0.01 to 0.1, therefore the three of learning rate levels can be set to 0.02,0.04 and 0.07。

With 4 kinds of design factors, every kind of design factor level number is 3, therefore the orthogonal table described in use table 1 carrys out design experiment。The present invention adopts following average relative error e in being embodied as_MRERepresent the error of prediction,DescribedFor the predictive value of traffic flow, the actual value of the traffic flow that θ (T) inputs for user, T is a certain predicted time point, and N is the sum of predicted time point。The error of test meets the little characteristic of prestige, the smaller the better, in being embodied as, adopts in Taguchi's method and hopes that little characteristic η is to assess the effect value of each design factor, namelyDescribed k is the number of times that in orthogonal table, often group test performs, and p and q is variable。

In being embodied as, by performing test, calculate the η often organizing test, the η of test corresponding to each level of certain factor is added and averages, the effect value of each level of this factor can be drawn。Choose each factor and there is the level best of breed as structural parameters initial value of the highest effect value。

2, the cluster of traffic flow

The present invention adopts Fuzzy C-Means Cluster Algorithm to carry out the cluster of traffic flow in being embodied as, and cluster numbers c is pre-determined by Taguchi's method design experiment。Data-oriented collection X={x₁,x₂,...x_n, fuzzy C-means clustering seeks to be divided into X c class, and c cluster centre is V={v₁,v₂,...v_c}。In fuzzy clustering, it is not as hard clustering method and like that each object is strictly divided into a certain apoplexy due to endogenous wind, but obtain each object degree of membership corresponding to all classes。If using u_ijRepresent that jth object belongs to the degree of membership of the i-th class, then u_ij∈ [0,1], and have:Namely each object is 1 for the degree of membership sum of all classes。The target of fuzzy C-means clustering still for minimizing inter-object distance and maximizing between class distance, so its object function can be set to fuzzy set theory square distance and, namelyDescribed n be object sum, c for cluster number, m be not less than 1 weighting parameters, d_ijFor object x_jWith cluster centre v_iBetween distance, here adopt Euclidean distance calculate。

Fuzzy C-means clustering is by iteration adjustment subordinated-degree matrix U and cluster centre V so that object function is minimum, specifically comprises the following steps that

(2.1) selected cluster numbers c and weighting parameters m, and initialize subordinated-degree matrix U with the random number between 0 to 1⁰So that it is meetConstraints。Iterative steps t is set to 0。

(2.2) useCalculate the current each element value of cluster centre V。

(2.3) useCalculate J^t, described J^tIt is the J value when the t time iteration, d_ijFor data object x_jWith classification v_iBetween Euclidean distance, described Euclidean distance is the actual distance in space between two points。

(2.4) utilizeSubordinated-degree matrix U is modified by the distance according to object to cluster centre, described inIt is u_ijValue after the t time iteration, d_kjFor data object x_jWith classification v_kBetween Euclidean distance。

(2.5) useCalculate J^t+1, described J^t+1It it is J value after the t time iteration。

(2.6) judge whether to meet | J^t+1-J^t|≤ε or t is equal to maxt, if it is satisfied, continue following operation, and otherwise t=t+1, turn to (2.2)。

According to maximum subjection principle, each object should belong to the class that the degree of membership of its correspondence is maximum。

3, the training of reverse transmittance nerve network

Traffic flow data, through cluster, defines c bunch, corresponding to c pattern of traffic flow。In order to the corresponding forecast model of each Model Establishment, the training set training these models first should be marked off。Temporally tie up and the data in former training set are divided, for each bunch, all training datas identical with object in this bunch for time dimension are divided into a sub-training set, form c sub-training set so altogether, with every sub-training set one reverse transmittance nerve network of training。The training process of reverse transmittance nerve network includes following step:

(3.1) netinit。The network input layer nodes n that obtains before employing, node in hidden layer l, output layer nodes m, learning rate λ, given input layer, connection weights ω between hidden layer and output layer neuron_rsAnd ω_su, more given hidden layer threshold values a₀, output layer threshold values b₀, reset maximum iteration time maxt, iterative steps t be set to 0。

(3.2) hidden layer output calculates。According to input variable X, calculate hidden layer outputS=1,2 ..., l。

(3.3) output layer output calculates。Export according to hidden layer, calculate the prediction output of reverse transmittance nerve network $O_u=\underset{s=1}{\overset{l}{\mathrm{\Σ}}}{H}_s{\mathrm{\ω}}_{\mathrm{su}}-b_u,$ U=1,2 ..., m

(3.4) Error Calculation。User inputs desired output Y_u, calculate forecast error e_u=Y_u-O_u, u=1,2 ..., m。

(3.5) right value update。By forecast error e_uBring renewal network into and connect weights ω_rsAnd ω_su, namelyω_su=ω_su+λH_se_u, r=1,2 ..., n, s=1,2 ..., l, u=1,2 ..., m。

(3.6) threshold values updates。According to forecast error e_u, update network node threshold values a_sAnd b_u, namelyb_u=b_u+e_u, described s=1,2 ..., l, u=1,2 ..., m。

(3.7) judging whether iteration t has reached maximum iteration time maxt, if reaching, then forwarding step 4 to), otherwise t=t+1, turn to (3.2)。

It is embodied as the present invention is above-mentioned, Taguchi's method is used to determine suitable structural parameters, such as, the orthogonal table utilizing table 1 carrys out design experiment, often group test performs 5 times, obtain the effect value of 3 each corresponding levels of four kinds of design factors, choose each factor and there is the level best of breed as structural parameters initial value of the highest effect value, design factor and level result of the test thereof in Table 2。In ensuing enforcement, choose the traffic flow (any one sky) of some day in training set, it is used for Fuzzy C-Means Cluster Algorithm to cluster, form many sub-training sets altogether, with every sub-training set one reverse transmittance nerve network of training, it is corresponding in turn to different flow rate modes。It is being embodied as final stage, is exporting result, subordinated-degree matrix U according to the prediction of reverse transmittance nerve network, calculate the predictive value on particular prediction time point。

Table 1 is orthogonal table example,

Table 1

Table 2 is design factor example。

Table 2

Claims (4)

1. the short-term traffic flow forecast method based on neural network ensemble model, it is characterised in that the step that the method comprises is:

Step 1) structural parameters of short-term traffic flow forecast are set,

Step 2) traffic flow is clustered, user inputs the data set X={x being made up of data object₁,x₂,...x_n, threshold values ε and maximum iteration time maxt；The cluster centre being set with classification composition is V={v₁,v₂,...v_c, u_ijIndicate that jth data object x_jBelong to i-th classification v_iDegree of membership, all u_ijComposition subordinated-degree matrix U, target setting function J be fuzzy set theory square distance and, n is data count, c for cluster number, m be not less than 1 weighting parameters, d_ijFor object x_jWith cluster centre v_iBetween distance, i and j is variable, arranges iterative steps t；

Step 3) use reverse transmittance nerve network to be trained, described reverse transmittance nerve network, also referred to as BP neutral net, is the multilayer feedforward neural network by Back Propagation Algorithm training；

Step 4) carry out short-term traffic flow forecast, user inputs predicted time point T successively, from step 3) the prediction output of the reverse transmittance nerve network that obtains, take c predictive value O₁,O₂,...,O_c；According to step 3) predicted time point is for the degree of membership of each bunch in subordinated-degree matrix U, by degree of membership as weights, asks the weighted sum of each bunch corresponding reverse transmittance nerve network predictive value as the predictive value O at predicted time point T_T, namelyDescribed T is predicted time point, T=1,2 ..., n。

2. a kind of short-term traffic flow forecast method based on neural network ensemble model according to claim 1, it is characterised in that described step 1) structural parameters that arrange short-term traffic flow forecast specifically comprise the following steps that

Step 1.1) number will be clustered as factors A, make factors A represent the classification of traffic flow, the level of factors A is set, the minimum level of factors A is taken as 3, the highest level is taken as 5, and middle level is taken as 4, and described level represents the consumption or the state in which value that refer to that each factor takes；

Step 1.2) using input number of nodes as factor B, three levels of factor B respectively 1,3,5 are set, these three level represents node three kinds of situations less, general, more successively；

Step 1.3) using node in hidden layer as factor C, factor C is by empirical equationDetermining, described l is node in hidden layer, and arranging n is factor B, and namely n is input number of nodes, and arranging m is output node number, and δ is value constant between 1 to 10, and three levels of factor C are set to 2,3 and 7；

Step 1.4) using learning rate as factor D, factor D value is between 0.01 to 0.1, and its three levels are set to 0.02,0.04 and 0.07；

Step 1.5) use Taguchi's method to test, it is thus achieved that factors A, factor B, factor C, the factor D level when effect value peak, using these levels initial value as structural parameters。In Taguchi's method, useAssessing the effect value of each design factor, described k is the number of times that often group test performs, e_MREIt is average relative error, For the predictive value of traffic flow, the actual value of the traffic flow that θ (T) inputs for user, T is a certain predicted time point, and N is the sum of predicted time point, and p and q is variable；Described Taguchi's method is a kind of quality engineering method, and the method is by selecting design parameter around set desired value, and reduces variation through overtesting bottom line, makes all over products produced have identical, stable quality。

3. a kind of short-term traffic flow forecast method based on neural network ensemble model according to claim 1, it is characterised in that described step 2) specifically comprise the following steps that

Step 2.1) c is set as step 1) level of factors A that obtains, user inputs m, is assigned to u with the random number between 0 to 1_ij, all u_ijForm initial subordinated-degree matrix U so that it is meetIterative steps t is set to 0；

Step 2.2) useCalculate the current each element value of cluster centre V, described inIt is u_ijValue when the t time iteration,Represent v_iValue when the t time iteration；

Step 2.3) useCalculating Jt, described Jt is the J value when the t time iteration, d_ijFor data object x_jWith classification v_iBetween Euclidean distance, described Euclidean distance is the actual distance in space between two points；

Step 2.4) utilizeSubordinated-degree matrix U is modified by the distance according to object to cluster centre, described inIt is u_ijValue after the t time iteration, d_kjFor data object x_jWith classification v_kBetween Euclidean distance；

Step 2.5) useCalculate J^t+1, described J^t+1It it is J value after the t time iteration；

Step 2.6) judge whether to meet | J^t+1-J^t|≤ε or t is equal to maxt, if it is satisfied, then forward step 3 to), otherwise t=t+1, turn to step 2.2)。

4. a kind of short-term traffic flow forecast method based on neural network ensemble model according to claim 1, it is characterised in that described step 3) specifically comprise the following steps that

Step 3.1) using step 1) the factor B level value that obtains is as network input layer nodes n, factor C level value is as node in hidden layer l, use output layer nodes m, factor D level value is as learning rate λ, and user provides the connection weights ω between input layer, hidden layer and output layer neuron_rsAnd ω_suArranging initial value, user inputs initial hidden layer threshold values a again₀, output layer threshold values b₀, reset maximum iteration time maxt, iterative steps t be set to 0；

Step 3.2) according to user input data set variable X, utilize formulaCalculate hidden layer output H_s, s=1,2 ..., l；

Step 3.3) utilizeCalculate the prediction output O of reverse transmittance nerve network_u, u=1,2 ..., m；

Step 3.4) user inputs desired output Y_u, utilize e_u=Y_u-O_uCalculate forecast error e_u, u=1,2 ..., m；

Step 3.5) utilizeWithBy forecast error e_uBring renewal network into and connect weights ω_rsAnd ω_su, r=1,2 ..., n, s=1,2 ..., l, u=1,2 ..., m；

Step 3.6) utilizeAnd b_u=b_u+e_u, bring forecast error e into_uUpdate network node threshold values a_sAnd b_u, s=1,2 ..., l, u=1,2 ..., m；

Step 3.7) judge whether iteration t has reached maximum iteration time maxt, if reaching, then forward step 4 to), otherwise t=t+1, turn to step 3.2)。