CN110299011A - A kind of traffic flow forecasting method of the highway arbitrary cross-section based on charge data - Google Patents
A kind of traffic flow forecasting method of the highway arbitrary cross-section based on charge data Download PDFInfo
- Publication number
- CN110299011A CN110299011A CN201910683480.9A CN201910683480A CN110299011A CN 110299011 A CN110299011 A CN 110299011A CN 201910683480 A CN201910683480 A CN 201910683480A CN 110299011 A CN110299011 A CN 110299011A
- Authority
- CN
- China
- Prior art keywords
- section
- vehicle
- time
- car
- traffic flow
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications
- G08G1/0145—Measuring and analyzing of parameters relative to traffic conditions for specific applications for active traffic flow control
Abstract
A kind of traffic flow forecasting method of the highway arbitrary cross-section based on charge data, comprising the following steps: step 1, acquire data, obtain the average overall travel speed of each car on shortest paths;Step 2, various types of vehicles is calculated in each section with stream movement velocity and with stream run duration;Step 3, running time of the vehicle in each section close to truth is estimated;Step 4, the exact time that each car reaches specified cross section place is estimated;Step 5, the magnitude of traffic flow of specified section is obtained;Step 6, specified cross section place traffic flow is predicted based on SAE model.The present invention is no longer needed manually to carry out vehicle flowrate record at highway scene or be recorded from monitor video, and manpower and time is greatly saved.
Description
Technical field
The invention belongs to traffic flow forecasting technical field, in particular to a kind of highway based on charge data is any
The traffic flow forecasting method of section.
Background technique
Forecasting traffic flow, which has become, to be efficiently used the limited road equipment of capacity to avoid traffic congestion and is lined up existing
As, pollution is reduced, is saved the travel time, it is energy saving, promote the important realization rates [1,2] such as intelligent transportation system development.With
The continuous development in the field, the task of forecasting traffic flow is also more and more various, including short-time traffic flow forecast, medium-term and long-term traffic
Stream prediction etc., while the accuracy requirement of forecasting traffic flow is also continuously improved.With the accumulation of highway traffic data, how
It is effective to extract traffic flow data and to select suitable prediction technique to carry out accurately predicting traffic flow most important.
When specifying the forecasting traffic flow of section towards highway, obtains highway and specify traffic flow of section most normal
Method first is that manually carrying out vehicle flowrate record at highway scene or from monitor video being recorded [3].From travelling
Vehicle in obtain traffic information be the another way [4] for predicting the whole volume of traffic, but few drivers are in highway
Road traffic condition is fed back when driving.Wagon detector (VDs)/inductive loop detector, radar and other equipment are commonly installed
On a highway, for detecting traffic condition [5-7], by taking VDs as an example, many information can be collected, such as the passing through of vehicle stops
It is deposited in, running speed, vehicle commander and model, car row length, roadway occupancy etc., passes through these equipment also available traffic flow
Amount.In addition, charging system completely covers the highway of all operations in China, and in charging system, existing electronic charging
(ETC), also there is manual toll collection (MTC).From charge data, the time of entering the station of available each car, the outbound time, enter the station receipts
Take the data such as mouth and outbound charging aperture, running time, operating range, the total travel speed of each car are known that using these data
The information such as the origin and destination of degree and each car.Therefore charging system provides a kind of alternative for obtaining the magnitude of traffic flow, but
The magnitude of traffic flow of specified point on arbitrary cross-section, one of yet unresolved issue are estimated using charge data is, charge
Data and the overall travel speed of single unit vehicle are linked up with, and overall travel speed cannot indicate the traveling speed by vehicle in specified point
Degree, so the specific time that vehicle reaches designated place is still unknown, the magnitude of traffic flow of designated place is also in this way, therefore seldom
There is research to establish and derive model, to obtain the magnitude of traffic flow from the charge record in closed region charging system.Et al. [8]
It proposes and demonstrates the volume of traffic of current road segment and the closely related hypothesis of the volume of traffic of its upstream charge station.Wu et al. [9] root
Condition of road surface is predicted according to the effective information in charge data.Large-scale traffic data collection is improving traffic condition or analysis travelling
Aspect plays an important role [10,11].It, can be based on to it in the highway scenario in view of a large amount of related big datas
The analysis of basic correlation is to study the successful modeling between charge data and traffic flow of section.
In order to solve this problem, many researchers investigated starting point/destination (OD) to and determined more
Information.Although accurate and perfect deductive model may be not present, because vehicle passes through similar geographical feature, including various
Road curvature and road route, these vehicles may have similar driving behavior, therefore propose practice and assume.In addition to space is believed
Breath, time and Changes in weather also will affect car speed.Under this assumption, it may derive from flow velocity degree with similar
Common driving behavior [12] in the different sections of highway of the single unit vehicle of behavior, for example, it is assumed that all vehicles on upward trend in turn
When slow down.These potential correlations help to predict the travel speed in specified section, this is to be worth that further studies to ask
Topic.
On the traffic flow forecasting method that highway specifies section, classical time series models, Kalman filtering mould
Type, Markov model and support vector machines (SVM) model have been used for the field.In these models, time series models,
Such as rolling average autoregression model (ARIMA) is integrated, lay particular emphasis on the time mode for extracting historical data.By summarizing dependency number
Rule between, Markov model can determine the road condition of future time.In conjunction with linear state equations, Kalman's filter
Wave pattern can obtain the optimal estimation [13-15] of road condition.Based on SVM method, Soviet Union et al. [16] discovery analysis model exists
It is better than prediction technique in terms of accuracy, and more efficient.It is most of dependent on introducing historical data in these conventional methods
Feature assume following volume of traffic.To deep neural network (DNN) model, Recognition with Recurrent Neural Network based on deep learning
(RNN) it is automatic to arrive shot and long term memory unit (LSTM) model based on RNN, door control unit model (GRU) and stack again for model
Encoder (SAE) model, has formd a variety of forecasting traffic flow means.Wherein, deep learning method is non-thread as one group of depth
Property topological model, can replace traditional linear method to carry out in the irregular traffic data that real world the Caspian Sea is measured
Precision of prediction can be improved by extracting the feature hidden in data in modeling.
As one group of non-linear topological model of depth, deep learning method can substitute classical linear method.By mentioning
The feature being hidden in data is taken, successfully the irregular data of real world is modeled to improve precision of prediction [17,18].
For example, hole et al. [19] are extracted four features of trip with subway communications and transportation data to predict trip requirements.Yi, Jung and
Bae et al. [20] predicts that real-time traffic amount, accuracy rate reach using deep neural network (DNN) model with 5 minutes intervals
99%, but data scale very little.Recognition with Recurrent Neural Network (RNN) [21] is widely used in another method of traffic volume forecast, it
The feature being hidden in data can be saved, the predicting traffic flow amount in terms of time and space.Shot and long term memory network (LSTM)
It is the structure based on RNN model refinement, there is input gate, out gate and forgetting door [22], these doors and storage unit constitutes LSTM
Model, it can learn the long-term dependence [23] between input data.Gating cycle unit (GRU) model is LSTM model
Variant, be considered performing better than [24] than LSTM in traffic volume forecast field.All these deep structures and multilayer nerve net
Network has successful performance in terms of the internal feature for extracting data, and the potential feature and mode of historical data are mined out,
The accuracy of prediction can be greatly improved.The successful application of deep learning model first is that stack autocoder (SAE) model,
The model has the pre- geodesic structure of deep layer, and has good performance [25,26] on forecasting traffic flow.
It is main at present using manually carrying out vehicle flowrate record at highway scene or from monitoring when obtaining traffic flow data
Video carries out record and extracts data from wagon detector (VDs), but in practical applications, there is also many for these methods
In place of shortcomings and deficiencies.Firstly, manual record vehicle flowrate can expend a large amount of manpower and time, and artificial statistical vehicle flowrate by
To several factors influence, such as monitor video is unintelligible, monitoring device covering not comprehensively, monitoring device damage, personnel can not be each
Period is in working condition etc., can all lead to data statistics omission or mistake of statistics, causes the missing of data and statistics accurate
Rate decline.Secondly, there is also many problems for other Road Detection equipment.By taking VDs as an example, it can capture the traffic of fixed position
Flow, travel speed and occupancy, but as VDs equipment uses, it also will appear some problems.
First: in view of needing a large amount of VDs equipment on highway, the purchase of these equipment and installation cost very it is high simultaneously
And it the time needed for being regularly maintained and regularly updating the VDs of damage and is short of hands.
Second: VDs detection accuracy is sometimes not high enough, especially when vehicle flowrate is excessively intensive, will lead to vehicle flowrate
It counts missing rate and error rate increases.
Third: since VDs is placed on fixed position, to report the traffic condition of these specific positions, rather than it is any
The traffic condition of position, that is to say, that traffic condition can only be captured by being mounted on the VDs of specific cross section place, not had
In the case that VDs and VDs damage is installed, it is difficult to identify the traffic condition of its corresponding position, correspondingly also resulting in can not obtain
The magnitude of traffic flow of corresponding position.
Summary of the invention
The forecasting traffic flow side of the purpose of the present invention is to provide a kind of highway arbitrary cross-section based on charge data
Method, to solve the above problems.
To achieve the above object, the invention adopts the following technical scheme:
A kind of traffic flow forecasting method of the highway arbitrary cross-section based on charge data, comprising the following steps:
Step 1, data are acquired, obtain the average overall travel speed of each car on shortest paths: the receipts original from each
Time of entering the station, outbound time, charge station's entrance and the charge station's outlet information of each car are obtained in expense data record, according to above-mentioned
Data obtain the average overall travel speed of each car on shortest paths;
Step 2, vehicle classification combines step 1 to calculate the average overall travel speed of each vehicle, calculates all kinds of after classification
Vehicle is in each section with stream movement velocity and with stream run duration;
Step 3, it according to the most short driving path of each car, determines which section vehicle has passed through, and is obtained pair by step 2
Answer estimating with stream run duration then in conjunction with the actual travel time that each car is exported from charge station entrance to charge station for section
Vehicle is calculated in each section close to the running time of truth;
Step 4, it calculates each car and reaches the running time that specified section is spent from charge station's entrance, then basis
Corresponding running time estimates the exact time that each car reaches specified cross section place;
Step 5, the time that specified section is reached according to the calculated each car of step 4, in required time interval cohesion
All records are closed to obtain the magnitude of traffic flow of specified section;
Step 6, specified cross section place traffic flow is predicted based on SAE model.
Further, in step 1, according to charge station's entrance and charge station's outlet information, Dijkstra shortest path is utilized
Algorithm calculates the shortest path and corresponding operating range of every record, and the corresponding running time of each car can going out by each car
Time of standing subtracts the time of entering the station and is calculated, then the average overall travel speed of each car on shortest paths can calculate,
The calculation formula are as follows:
Wherein T " is the outbound time of vehicle j, and T ' is the inbound time of vehicle j, and L (j) is row of the vehicle j in shortest path
Distance is sailed,It is the average overall travel speed of vehicle j.
Further, in step 2, the classification method of vehicle: the type of vehicle of each charge station record be according to axle into
Row classification, car and twin shaft truck are divided into small vehicle, and bus and three, four axis trucies are medium sized vehicle, five axis
Above vehicle is oversize vehicle.
Further, in step 2, the average overall travel speed of each vehicle is calculated by step 1, then by above-mentioned vehicle point
Class standard calculates all types of vehicles in each section with flow velocity degree and with stream time, calculation formula are as follows:
WhereinIt is the average overall travel speed of vehicle j, mkIt is the quantity in the charging data record in i-th section, k table
Show the type of vehicle, LiIt is the length in i-th section,Each car i-th section with flow velocity degree, with stream when
Between.
Further, in step 3, running time of the vehicle in each section close to truth, calculation formula are estimated are as follows:
Wherein ti,kIt is estimation running time of each car in i-th of section, k indicates the type of vehicle, and T is from each vehicle
The running time exported from charge station entrance to charge station, n is exported from charge station entrance to charge station from each vehicle
Section quantity;Based on the charge data monthly obtained, monthly timing more new variablesAnd ti,k。
Further, in step 4, the running time in each section on its most short driving path can be calculated by step 3, so
Vehicle can be calculated afterwards, and specified section the time it takes, calculation formula are reached from charge station's entrance are as follows:
Wherein Δ tAIt is the running time from charge station's entrance to specified section A, x is from charge station's entrance to specified section
Section number, txIt is the estimation running time in x-th section of section, LDisThe section for being the upstream charge station in x-th of section and specifying
The distance between, LxIt is the length in x-th of section;
Then the time that each car reaches specified section, calculation formula are estimated are as follows:
TA=T '+Δ tA (6)
Wherein TAIt is the time that single unit vehicle reaches specified section, T ' is the time that each car enters charging aperture.
Further, in step 5, time interval is 5a minutes, and a is positive integer.
Compared with prior art, the present invention has following technical effect:
The present invention extracts the highway arbitrary cross-section location history volume of traffic by proposing from large area charge data
Algorithm introduces SAE model to predict the magnitude of traffic flow, and using mean absolute error (MAE) come to model prediction result
It is assessed, realizes from charge data the traffic flow forecasting method for predicting that highway specifies cross section place, have
Following advantage,
First: due to no longer needing manually to carry out vehicle flowrate record at highway scene or be recorded from monitor video, greatly
Manpower and time are saved greatly.
Second: it is external dry that the visual field is unintelligible, monitoring device covering is not complete, detecting devices damage or detection accuracy are not high enough etc.
The factor of disturbing is greatly decreased, and the integrality of data increases substantially.
Third: charge data is automatically generated by computer, statistics omit or mistake of statistics rate it is almost nil, data it is accurate
Rate greatly improves.
4th: since charging system is established on the highway that the whole nation has opened operation comprehensively, so not needing to pacify again
Fill other additional magnitude of traffic flow detection devices, be greatly saved the manpowers of all kinds of detection device installation and maintenances, financial resources and when
Between cost.
Further, using the specified section historical traffic flow data obtained from charge data, by phase
Section forecasting traffic flow is answered to verify, it is ensured that the correctness of algorithm.
Further, by obtaining the magnitude of traffic flow from charge data, resource can not only be saved for a long time, including manpower,
Time, money etc., additionally it is possible to which the application field for expanding charge data allows charge data to play the effect made the best use of everything.
Further, by introduce with flow velocity degree, with the stream time as each section actual travel speed and running time
Reference value considerably reduces influence of the different situations of different sections of highway to vehicle actual travel situation.
Further, by combine each car each section with stream time and each car from charge station entrance to charge station
The actual travel time of outlet calculates vehicle in the running time in each section, it is ensured that the accuracy of calculated result.
Further, enter the time of charge station in conjunction with vehicle and vehicle reaches specified section from charge station's entrance and spent
The running time of expense obtains the time that each car reaches specified section, it is ensured that the correctness of estimation result.
Further, by using to the preferable SAE model of forecasting traffic flow adaptability, traffic fluxion can deeply be extracted
According to feature, improve the accuracy rate of prediction result.
Detailed description of the invention
Fig. 1 is that VDs equipment detects traffic condition schematic diagram.
Fig. 2 is the process schematic for obtaining the running time in each section.
Fig. 3 is the principle and structure chart of autocoder.
Fig. 4 is the process schematic of the specified link traffic flow of present invention prediction.
Fig. 5 is the description figure from charge data.
Fig. 6 is for vehicle in each section with stream time diagram.
Specific embodiment
Below in conjunction with attached drawing, the present invention is further described:
Please refer to Fig. 1 to Fig. 6, a kind of traffic flow forecasting method of the highway arbitrary cross-section based on charge data, packet
Include following steps:
Step 1, data are acquired, obtain the average overall travel speed of each car on shortest paths: the receipts original from each
Time of entering the station, outbound time, charge station's entrance and the charge station's outlet information of each car are obtained in expense data record, according to above-mentioned
Data obtain the average overall travel speed of each car on shortest paths;
Step 2, vehicle classification combines step 1 to calculate the average overall travel speed of each vehicle, calculates all kinds of after classification
Vehicle is in each section with stream movement velocity and with stream run duration;
Step 3, it according to the most short driving path of each car, determines which section vehicle has passed through, and is obtained pair by step 2
Answer estimating with stream run duration then in conjunction with the actual travel time that each car is exported from charge station entrance to charge station for section
Vehicle is calculated in each section close to the running time of truth;
Step 4, it calculates each car and reaches the running time that specified section is spent from charge station's entrance, then basis
Corresponding running time estimates the exact time that each car reaches specified cross section place;
Step 5, the time that specified section is reached according to the calculated each car of step 4, in required time interval cohesion
All records are closed to obtain the magnitude of traffic flow of specified section;
Step 6, specified cross section place traffic flow is predicted based on SAE model.
In step 1, according to charge station's entrance and charge station's outlet information, calculated using Dijkstra shortest path first every
The shortest path and corresponding operating range of item record, the corresponding running time of each car can be subtracted by the outbound time of each car
Time of entering the station is calculated, then the average overall travel speed of each car on shortest paths can calculate, the calculation formula
Are as follows:
Wherein T " is the outbound time of vehicle j, and T ' is the inbound time of vehicle j, and L (j) is row of the vehicle j in shortest path
Distance is sailed,It is the average overall travel speed of vehicle j.
In step 2, the classification method of vehicle: the type of vehicle of each charge station's record is classified according to axle,
Car and twin shaft truck are divided into small vehicle, and bus and three, four axis trucies are medium sized vehicles, vehicles more than five axis
For oversize vehicle.
In step 2, the average overall travel speed of each vehicle is calculated by step 1, then by above-mentioned vehicle classification standard meter
All types of vehicles are calculated in each section with flow velocity degree and with stream time, calculation formula are as follows:
WhereinIt is the average overall travel speed of vehicle j, mkIt is the quantity in the charging data record in i-th section, k table
Show the type of vehicle, LiIt is the length in i-th section,Each car i-th section with flow velocity degree, with stream when
Between.
In step 3, running time of the vehicle in each section close to truth, calculation formula are estimated are as follows:
Wherein ti,kIt is estimation running time of each car in i-th of section, k indicates the type of vehicle, and T is from each vehicle
The running time exported from charge station entrance to charge station, n is exported from charge station entrance to charge station from each vehicle
Section quantity;Based on the charge data monthly obtained, monthly timing more new variablesAnd ti,k。
In step 4, the running time in each section on its most short driving path can be calculated by step 3, then
It calculates vehicle and reaches specified section the time it takes, calculation formula from charge station's entrance are as follows:
Wherein Δ tAIt is the running time from charge station's entrance to specified section A, x is from charge station's entrance to specified section
Section number, txIt is the estimation running time in x-th section of section, LDisThe section for being the upstream charge station in x-th of section and specifying
The distance between, LxIt is the length in x-th of section;
Then the time that each car reaches specified section, calculation formula are estimated are as follows:
TA=T '+Δ tA (6)
Wherein TAIt is the time that single unit vehicle reaches specified section, T ' is the time that each car enters charging aperture.
In step 5, time interval is 5a minutes, and a is positive integer.
Based on SAE model in specified section predicting traffic flow amount method, including,
Autocoder, the fine processing of SAE model and data, traffic flow forecasting and Performance Index;
The principle and structure of single autocoder are as shown in Fig. 3, and each autocoder has three-decker, and
It needs to rebuild input layer;First layer is input layer, and the last layer is to rebuild layer, both there is identical unit number;Hidden layer is used
In extracting data characteristics by one group of data of input { x1 (l), x2 (l) ..., xn (l)) }, wherein xi (l) ∈ RD indicates l
The unit of layer;In the Nonlinear Processing process of autocoder, coded treatment refers to that the feature of input data can be in hidden layer
Middle obtain simultaneously is expressed as a (xi (l)), and decoding process, which is autocoder, to be decoded a (xi (l)) and to be redeveloped into xi (l) ' defeated
Out, calculation formula are as follows:
A (x)=f (W1x+b1) (7)
X '=g (W2a(x)+b2) (8)
Wherein W1And W2It is encoder matrix and decoding matrix respectively, they are the weight matrix of each autocoder;B1 and
B2 is coding and decoding bias vector;F (x) and g (x) is activation primitive used in neural network;In the method, encoded
Journey and decoding process are all using the linear unit function max (0, x) of rectification;
In addition, reconstruction error is the major parameter for assessing performance, it is defined as model variable, is expressed as θ, calculation formula
Are as follows:
In general, the hidden layer unit number of non-linear self-encoding encoder may be more than input layer unit number, this results in self-editing
Code device may learn identity function, or input data is simply only copied as output, mention so as to cause from model
The feature taken becomes useless;We use random inactivation (dropout) method after cataloged procedure, pass through random erasure one
Unit, the unit temporarily connected together with it are also deleted, and obtain the network of one more " thin ", different dropout values also can shadow
Ring the function of model;
SAE model is used using each autocoder as separate unit, and they are stacked to create deep layer
Network, the model with deep learning method, which usually has, to have three layers above, and the number of nodes in the quantity of layer and every layer will affect mould
The prediction result of type;SAE model is a kind of structure for successively stacking autocoder, and each layer is all an autocoder,
For coding and decoding data;Input layer passes data to first hidden layer, and then hidden layer is extracted by encoding operation
Feature simultaneously passes them to second hidden layer, while the output of reconstruction is deleted from network;Second and later hide
Layer executes identical operation, until reaching the last one hidden layer;Meanwhile every layer is all calculated using greedy layering unsupervised learning
Method carries out pre-training, with the weight of optimization layer;When pre-training process is completed, using the output of the last one hidden layer as defeated
Enter, and the parameter of model is finely adjusted by backpropagation (BP) algorithm;
In order to assess the prediction error of SAE model and at the end of the study compared with other prediction models, common property
Energy index is mean absolute error (MAE), average relative error (MRE) and root-mean-square error (RMSE) for the pre- of assessment models
Error between measured value and actual value;Since MAE value can reflect the actual conditions of prediction error, this research selects MAE
Value assesses the results of different models;
Wherein xiIt is real data, xi' it is prediction data;
It is whole to realize that process is attached as shown in figure 4, first from each receipts in the big regional scope around particular link section
Take station and collects history charge record;The receipts of the large area around are all recorded in by most of vehicular traffic of cross section place
Take in data;Then the traffic on the intersection in the traffic information and road on road is estimated using magnitude of traffic flow model
Amount;Finally, our advantages using the SAE model based on historical data, specified section is predicted at various time intervals
The magnitude of traffic flow.
Embodiment:
The traffic flow forecasting method that specified section is predicted based on charge data of unification described in the invention, such as attached drawing 4
It is shown, including five major parts, it is that acquisition, the calculating vehicle of charge data reach the time of specified section, polymerization specifies and breaks respectively
The assessment of face historical traffic flows, the forecasting traffic flow based on SAE model, prediction result, each section are specific as follows.
1, the acquisition of charge data.
By 2018, Shannxi Expressway total kilometrage was up to 5386 kilometers.We have collected Chinese Shan in this research
The original charge data of all freeway toll stations of Xi Sheng, and Shaanxi freeway net is selected to receive as a closed region
Charge system.This research selects respective stretch from Xi'an in the high speed of city, and as shown in Fig. 5, specified section is from western high-new
Charge station is to Chang'an charge station, middle position, section, and the distance of cross section place to upstream charge station is 2 kilometers, the receipts for record of charging
Collecting the time is from January, 2018 in April, 2018.Because the volume of traffic of the section is sufficiently large, the basic demand of research is met.
In addition, the section is located at the center in city, the magnitude of traffic flow also has certain Tide Characteristics, it is ensured that comes from charge data
Obtain the reliability of traffic flow data.
2, the time that vehicle reaches specified section is calculated.
In charge record from January, 2018 of selection in April, 2018, the magnitude of traffic flow is exported using from charge data
Algorithm obtains the time that each car reaches specified section, as shown in Fig. 6, in order to illustrate with flow velocity degree and with the stream time and per each
The correlation of the estimation running time in section, has randomly selected small-sized, and medium-sized and large-scale three classes vehicle is from three bridges around city high speed
Charge data sample of the charge station to Chang'an charge station.The running time of three classes vehicle is 7:30 to 8:00 in morning in morning, driving
Route also includes western high-new charge station to the specified cross section in Chang'an charge station section.
As shown in Fig. 6 (a), the length in each section is differed from 2.4 kms to 7.0 kms on entire travel path.Fig. 6
(b) average overall travel speed of three samples on entire travel path is shown, it can be seen that in practical situations, using each
Section with flow velocity degree rather than ensemble average travel speed is more accurate.Fig. 6 (c) show each section vehicle with stream when
Between, the running time of large-scale wagon flow is longer than the running time of other two kinds of vehicles, and the running time of middle-size and small-size wagon flow is fairly close, this
Reflect that the vehicle of both types is possible similar in travel speed in the real world.Fig. 6 (d) is shown from entrance charge station
To the estimation running time in each section of outlet charge station, it is concluded that going out the running time of estimation and the length in section
It is positively correlated.By taking small vehicle sample as an example, estimation time of shortest path section G4-G5 is 89 seconds, when the estimation of longest section G2-G3
Between be 243 seconds, do not consider the abnormal conditions such as traffic congestion, it is as a result consistent with actual conditions.It is above results showed that with vehicle
Type and each section reach the accuracy of the time calculated result of specified section with flow velocity degree, with the vehicle that the stream time is foundation.
3, it polymerize specified section historical traffic flows.
The time that specified section is reached according to all types of vehicles that flow exported algorithm obtains, by the traffic flow of specified section
Amount data aggregate gets up, i.e., the volume of traffic is taken together and summed by certain time interval.Later by the magnitude of traffic flow of polymerization
Data preparation and selects the traffic flow data of first trimester as training set, the 4th month data at the form of data set
As test set.Both guaranteed to separate training set data and test set data in this way, and had in turn ensured training set and test set data
Time span long enough, greatly reduce prediction result and the probability of random error occur.
4, the forecasting traffic flow based on SAE model.
After putting training set and test set data in order, we carry out the prediction of traffic flow based on SAE model.But
If the structure of model is different, precision of prediction can also change.In order to obtain excellent optimal prediction result, what this research considered
Influence factor is as follows,
First: using identical Construction of A Model, the time interval difference of training set data polymerization may result in different
Prediction result.Therefore, we select data aggregate time interval be respectively come within 5 minutes, 15 minutes, 30 minutes and 60 minutes into
Row training and test.
Second: different random inactivation (dropout) values may also can generate different precision of predictions.To many traffic flows
It predicts for network, it is optimal that dropout value, which is 0.5,.But pass through many experiments, the best dropout value of this research is
0.2。
Third: the number of unit in the quantity of hidden layer and every layer also affects the prediction essence of model to a certain extent
Degree.The MRE value of the different structure of SAE model, MAE value and RMSE value as shown in appendix 1, the value of each structure be six times experiment after
The average value of acquisition.A series of experiments show 2 layers can be 15 minutes interval in optimum structures, number of unit be [300,
300].It was spaced similarly for 5 minutes, the traffic flow forecasting at 30 minutes intervals and 60 minutes intervals, optimal number of unit
Distribution is respectively [300,400,300], [400,400,400], [400,400,400,400].
TABLE 1
MRE VALUE, MAE VALUE AND RMSE VALUE OF DIFFERENT STRUCTURES OF THE SAE
MODEL
SAE model is adjusted by predefined parameter later, then will be trained in training set data input model, training set number
According to being assessed, prediction result is obtained.
5, the assessment of prediction result.
The prediction result of SAE model is compared with other deep learning model prediction results, including LSTM, DNN, GRU
With RNN model.After a series of tests, we obtain the MAE values of each model prediction result to carry out assessment models performance, such as
Shown in subordinate list 2, SAE network was shown in interval than certain network better performances 30 minutes intervals and 60 minutes, but at 5 points
Prediction accuracy in clock and 15 minutes intervals is slightly worse than other models, but accuracy is very close.With the increasing of time interval
Add, the accuracy of RNN model prediction result declines rapidly compared with SAE model, this is because gradient disappearance is difficult RNN model
Long-term forecasting traffic flow in progress.But SAE model can pass through coding and decoding journey with the continuous variation of the magnitude of traffic flow
Sequence rebuilds the traffic flow data of input, shows in long-term traffic flow forecasting good.Therefore SAE model and other models
It is more preferable compared to practical function.
TABLE 2
PERFORMANCE COMPARISON OF THE MLAE FOR SAE, THE DNN,
THE GRU, THE RNN AND THE LSTM
The present invention is drawn by extracting the highway arbitrary cross-section location history volume of traffic from large area charge data
Enter SAE model to predict the magnitude of traffic flow, has investigated a kind of computation model of low cost to replace expensive detection to set
It is standby, cost, time and manpower are both saved, and realize the high-precision forecast of the magnitude of traffic flow.
Claims (7)
1. a kind of traffic flow forecasting method of the highway arbitrary cross-section based on charge data, which is characterized in that including following
Step:
Step 1, data are acquired, obtain the average overall travel speed of each car on shortest paths: the charge number original from each
According to time of entering the station, outbound time, charge station's entrance and the charge station's outlet information for obtaining each car in record, according to above-mentioned data
Obtain the average overall travel speed of each car on shortest paths;
Step 2, vehicle classification combines step 1 to calculate the average overall travel speed of each vehicle, calculates various types of vehicles after classification
In each section with stream movement velocity and with stream run duration;
Step 3, it according to the most short driving path of each car, determines which section vehicle has passed through, and corresponding road is obtained by step 2
Section is estimated with stream run duration then in conjunction with the actual travel time that each car is exported from charge station entrance to charge station
Vehicle is in each section close to the running time of truth;
Step 4, it calculates each car and reaches the running time that specified section is spent from charge station's entrance, then according to correspondence
Running time estimate the exact time that each car reaches specified cross section place;
Step 5, the time that specified section is reached according to the calculated each car of step 4, institute is polymerize in required time interval
There is record to obtain the magnitude of traffic flow of specified section;
Step 6, specified cross section place traffic flow is predicted based on SAE model.
2. a kind of traffic flow forecasting method of highway arbitrary cross-section based on charge data according to claim 1,
It is characterized in that, according to charge station's entrance and charge station's outlet information, utilizing Dijkstra shortest path first meter in step 1
The shortest path and corresponding operating range of every record are calculated, the corresponding running time of each car can be by the outbound time of each car
Subtracting the time of entering the station is calculated, then the average overall travel speed of each car on shortest paths can calculate, the calculating
Formula are as follows:
Wherein T " is the outbound time of vehicle j, and T ' is the inbound time of vehicle j, L (j) be vehicle j shortest path traveling away from
From,It is the average overall travel speed of vehicle j.
3. a kind of traffic flow forecasting method of highway arbitrary cross-section based on charge data according to claim 1,
It is characterized in that, the classification method of vehicle: the type of vehicle of each charge station's record is classified according to axle in step 2
, car and twin shaft truck are divided into small vehicle, and bus and three, four axis trucies are medium sized vehicles, more than five axis
Vehicle is oversize vehicle.
4. a kind of traffic flow forecasting method of highway arbitrary cross-section based on charge data according to claim 1,
It is characterized in that, calculating the average overall travel speed of each vehicle by step 1, then by above-mentioned vehicle classification standard in step 2
All types of vehicles are calculated in each section with flow velocity degree and with stream time, calculation formula are as follows:
WhereinIt is the average overall travel speed of vehicle j, mkIt is the quantity in the charging data record in i-th section, k indicates vehicle
Type, LiIt is the length in i-th section,Each car i-th section with flow velocity degree, with the stream time.
5. a kind of traffic flow forecasting method of highway arbitrary cross-section based on charge data according to claim 1,
It is characterized in that, estimating running time of the vehicle in each section close to truth, calculation formula in step 3 are as follows:
Wherein tI, kEstimation running time of each car in i-th of section, k indicates the type of vehicle, T be from each vehicle from
The running time that charge station's entrance is exported to charge station, n are from the road that each vehicle is exported from charge station entrance to charge station
Segment number;Based on the charge data monthly obtained, monthly timing more new variablesAnd tI, k。
6. a kind of traffic flow forecasting method of highway arbitrary cross-section based on charge data according to claim 1,
It is characterized in that, can calculate the running time in each section on its most short driving path by step 3 in step 4, then can
Specified section the time it takes, calculation formula are reached from charge station's entrance to calculate vehicle are as follows:
Wherein Δ tAIt is the running time from charge station's entrance to specified section A, x is the road from charge station's entrance to specified section
Number of segment, txIt is the estimation running time in x-th section of section, LDisIt is between the upstream charge station in x-th of section and specified section
Distance, LxIt is the length in x-th of section;
Then the time that each car reaches specified section, calculation formula are estimated are as follows:
TA=T '+Δ tA (6)
Wherein TAIt is the time that single unit vehicle reaches specified section, T ' is the time that each car enters charging aperture.
7. a kind of traffic flow forecasting method of highway arbitrary cross-section based on charge data according to claim 1,
It is characterized in that, time interval is 5a minutes, and a is positive integer in step 5.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910683480.9A CN110299011A (en) | 2019-07-26 | 2019-07-26 | A kind of traffic flow forecasting method of the highway arbitrary cross-section based on charge data |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910683480.9A CN110299011A (en) | 2019-07-26 | 2019-07-26 | A kind of traffic flow forecasting method of the highway arbitrary cross-section based on charge data |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110299011A true CN110299011A (en) | 2019-10-01 |
Family
ID=68031887
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910683480.9A Pending CN110299011A (en) | 2019-07-26 | 2019-07-26 | A kind of traffic flow forecasting method of the highway arbitrary cross-section based on charge data |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110299011A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111126713A (en) * | 2019-12-31 | 2020-05-08 | 方正国际软件(北京)有限公司 | Space-time hot spot prediction method and device based on bayonet data and controller |
CN111260122A (en) * | 2020-01-13 | 2020-06-09 | 重庆首讯科技股份有限公司 | Method and device for predicting traffic flow on expressway |
CN111292533A (en) * | 2020-02-11 | 2020-06-16 | 北京交通大学 | Method for estimating flow of arbitrary section of highway at any time period based on multi-source data |
CN111724589A (en) * | 2020-06-03 | 2020-09-29 | 重庆大学 | Multi-source data-based highway section flow estimation method |
CN111882858A (en) * | 2020-06-01 | 2020-11-03 | 重庆大学 | Method for predicting queuing length of expressway abnormal events based on multi-source data |
CN112233426A (en) * | 2020-11-23 | 2021-01-15 | 重庆市市政设计研究院有限公司 | Bridge real-time early warning method device and system based on RFID and multi-source data |
CN113053114A (en) * | 2021-03-12 | 2021-06-29 | 广西计算中心有限责任公司 | Dynamic prediction and inspection method for exit station and exit time of illegal vehicles on highway |
CN113345252A (en) * | 2021-06-08 | 2021-09-03 | 重庆大学 | Short-time prediction method and device for lower-channel flow of toll station |
CN113525354A (en) * | 2021-07-28 | 2021-10-22 | �田�浩 | Detour path planning and control system for parking navigation |
CN114999162A (en) * | 2022-08-02 | 2022-09-02 | 北京交研智慧科技有限公司 | Road traffic flow obtaining method and device |
CN115376308A (en) * | 2022-05-26 | 2022-11-22 | 南京工程学院 | Method for predicting automobile running time |
CN116504076A (en) * | 2023-06-19 | 2023-07-28 | 贵州宏信达高新科技有限责任公司 | Expressway traffic flow prediction method based on ETC portal data |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102800198B (en) * | 2012-08-15 | 2015-05-13 | 重庆大学 | Measuring and calculating method for traffic flow of section of expressway |
CN106327864A (en) * | 2015-07-10 | 2017-01-11 | 北京大学 | Traffic flow estimation method based on network charging data of highway |
CN107527501A (en) * | 2017-06-05 | 2017-12-29 | 交通运输部公路科学研究所 | The building method of travel time data and the method for predicting the motorway journeys time between a kind of highway station |
-
2019
- 2019-07-26 CN CN201910683480.9A patent/CN110299011A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102800198B (en) * | 2012-08-15 | 2015-05-13 | 重庆大学 | Measuring and calculating method for traffic flow of section of expressway |
CN106327864A (en) * | 2015-07-10 | 2017-01-11 | 北京大学 | Traffic flow estimation method based on network charging data of highway |
CN107527501A (en) * | 2017-06-05 | 2017-12-29 | 交通运输部公路科学研究所 | The building method of travel time data and the method for predicting the motorway journeys time between a kind of highway station |
Non-Patent Citations (2)
Title |
---|
YINLI JIN等: "SAE network: a deep learning method for traffic flow prediction", 《IEEE》 * |
靳引利等: "基于OD的高速公路断面交通流量推算方法", 《交通信息与安全》 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111126713B (en) * | 2019-12-31 | 2023-05-09 | 方正国际软件(北京)有限公司 | Space-time hot spot prediction method and device based on bayonet data and controller |
CN111126713A (en) * | 2019-12-31 | 2020-05-08 | 方正国际软件(北京)有限公司 | Space-time hot spot prediction method and device based on bayonet data and controller |
CN111260122A (en) * | 2020-01-13 | 2020-06-09 | 重庆首讯科技股份有限公司 | Method and device for predicting traffic flow on expressway |
CN111292533A (en) * | 2020-02-11 | 2020-06-16 | 北京交通大学 | Method for estimating flow of arbitrary section of highway at any time period based on multi-source data |
CN111882858A (en) * | 2020-06-01 | 2020-11-03 | 重庆大学 | Method for predicting queuing length of expressway abnormal events based on multi-source data |
CN111882858B (en) * | 2020-06-01 | 2022-05-20 | 重庆大学 | Multi-source data-based method for predicting queuing length of highway abnormal event |
CN111724589A (en) * | 2020-06-03 | 2020-09-29 | 重庆大学 | Multi-source data-based highway section flow estimation method |
CN112233426A (en) * | 2020-11-23 | 2021-01-15 | 重庆市市政设计研究院有限公司 | Bridge real-time early warning method device and system based on RFID and multi-source data |
CN112233426B (en) * | 2020-11-23 | 2024-02-02 | 重庆市市政设计研究院有限公司 | Bridge real-time early warning method device and system based on RFID and multi-source data |
CN113053114A (en) * | 2021-03-12 | 2021-06-29 | 广西计算中心有限责任公司 | Dynamic prediction and inspection method for exit station and exit time of illegal vehicles on highway |
CN113345252A (en) * | 2021-06-08 | 2021-09-03 | 重庆大学 | Short-time prediction method and device for lower-channel flow of toll station |
CN113345252B (en) * | 2021-06-08 | 2022-07-22 | 重庆大学 | Short-time prediction method and device for lower-path flow of toll station |
CN113525354A (en) * | 2021-07-28 | 2021-10-22 | �田�浩 | Detour path planning and control system for parking navigation |
CN115376308A (en) * | 2022-05-26 | 2022-11-22 | 南京工程学院 | Method for predicting automobile running time |
CN114999162A (en) * | 2022-08-02 | 2022-09-02 | 北京交研智慧科技有限公司 | Road traffic flow obtaining method and device |
CN116504076A (en) * | 2023-06-19 | 2023-07-28 | 贵州宏信达高新科技有限责任公司 | Expressway traffic flow prediction method based on ETC portal data |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110299011A (en) | A kind of traffic flow forecasting method of the highway arbitrary cross-section based on charge data | |
CN108197739B (en) | Urban rail transit passenger flow prediction method | |
CN112700072B (en) | Traffic condition prediction method, electronic device, and storage medium | |
Wang et al. | Fine-grained traffic flow prediction of various vehicle types via fusion of multisource data and deep learning approaches | |
CN104778834B (en) | Urban road traffic jam judging method based on vehicle GPS data | |
Fang et al. | FTPG: A fine-grained traffic prediction method with graph attention network using big trace data | |
CN106710215B (en) | Bottleneck upstream lane grade traffic status prediction system and implementation method | |
CN104658252A (en) | Method for evaluating traffic operational conditions of highway based on multisource data fusion | |
CN111063204B (en) | Expressway vehicle speed prediction model training method based on toll station flow | |
CN103871246A (en) | Short-term traffic flow forecasting method based on road network space relation constraint Lasso | |
CN102592447A (en) | Method for judging road traffic state of regional road network based on fuzzy c means (FCM) | |
CN105006147A (en) | Road segment travel time deducing method based on road space-time incidence relation | |
CN109272745A (en) | A kind of track of vehicle prediction technique based on deep neural network | |
CN111583628B (en) | Road network heavy truck traffic flow prediction method based on data quality control | |
CN109118787A (en) | A kind of car speed prediction technique based on deep neural network | |
CN113537626B (en) | Method for predicting neural network combined time sequence by aggregating information difference | |
Wang et al. | Forecasting traffic volume at a designated cross-section location on a freeway from large-regional toll collection data | |
CN111581538A (en) | Highway charging data-based high-speed traffic flow state inference method | |
CN108364464A (en) | A kind of public transit vehicle hourage modeling method based on probabilistic model | |
CN116052435B (en) | Urban road congestion influence range definition and road influence calculation method | |
CN107180534A (en) | The express highway section average speed method of estimation of support vector regression fusion | |
CN109493449A (en) | A kind of lorry loading method for estimating state based on lorry GPS track data and high speed transaction data | |
CN108830414B (en) | Load prediction method for commercial charging area of electric automobile | |
CN111524358B (en) | Regional radiation sexual communication flow prediction method | |
CN111784027A (en) | Urban range electric vehicle charging demand prediction method considering geographic information |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191001 |