CN108648457A - A kind of method, apparatus and computer readable storage medium of prediction of speed - Google Patents
A kind of method, apparatus and computer readable storage medium of prediction of speed Download PDFInfo
- Publication number
- CN108648457A CN108648457A CN201810690697.8A CN201810690697A CN108648457A CN 108648457 A CN108648457 A CN 108648457A CN 201810690697 A CN201810690697 A CN 201810690697A CN 108648457 A CN108648457 A CN 108648457A
- Authority
- CN
- China
- Prior art keywords
- section
- target road
- road network
- target
- matrix
- 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.)
- Granted
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
- G08G1/0125—Traffic data processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Traffic Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the invention discloses a kind of method, apparatus of prediction of speed and computer readable storage mediums to calculate the initial velocity vector of target road network according to the track data in preset time period；Using the advance trained convolutional neural networks based on road network, process of convolution is carried out to initial velocity vector sum adjoining section matrix, obtains corresponding eigenmatrix；Wherein, in eigenmatrix include the corresponding feature vector in each section；Each feature vector is converted to time series, and using advance trained shot and long term memory network, time series is handled, the corresponding target velocity matrix of target road network is obtained.Since feature vector is the vector obtained in the case where considering road network topology structure situation, influence of the Space Dynamic Evolution feature of mistake to its precision is effectively prevented.So that it is more accurate according to the target velocity vector that this feature vector forecasting goes out, effectively improve the precision of prediction of speed.
Description
Technical field
The present invention relates to urban transportation technical fields, more particularly to the method, apparatus and computer of a kind of prediction of speed
Readable storage medium storing program for executing.
Background technology
With the development of the social economy, the owning amount cumulative year after year of private car, certain metropolitan road reorganization and expansion
Speed cannot be satisfied the requirement of the growth of motor vehicle, thus caused by urban traffic blocking, traffic accident the problems such as have become
Perplex the major issue of urban development and resident trip.For this purpose, intelligent transportation system (Intelligent Transport
Systems, ITS) have become the hot spot of research as a kind of effective solution.And traffic system is someone's participation
, time-varying, huge and complicated system, how the urban highway traffic speed state that the moment varies can be accurately pre-
It surveys, this is one of key problem of ITS.
Each subsystem of ITS, such as adjustment traffic administration control program in time, control traffic congestion；It is sent out for traveler
Cloth trip information provides optimal route selection scheme；Accurate arrival time estimation is carried out, to which recommended user reasonably sets out
Time etc. is required for based on accurate urban transportation prediction of speed.
Traffic speed prediction in recent years is studied extensively by academia and industrial quarters, and correlation technique is broadly divided into Classical forecast
Method and deep learning method.On traditional prediction method, time series analysis is most typical simulated time sequence pattern
Method.Wherein difference ARMA model is predicted by the linear combination to past traffic speed on single section
Traffic speed.Simultaneously in order to describe such as early evening peak of periodical trend, seasonal difference ARMA model is carried
Go out.In addition to this, some traditional machine learning method such as linear regressions and support vector regression are also used for single road
The time sequence model study of section.
In recent years with the continuous development of deep learning, more and more researchers begin to use depth learning technology to carry out
Traffic speed is predicted.Wherein, shot and long term memory network (Long Short-Term Memory, LSTM) is good at capturing longer sequence
Time Dependent, to be used for single channel section traffic speed prediction.
But above method all mainly predicts single section, and the influence in section around is had ignored, city is not considered
The Evolvement of city's traffic speed prediction spatially.In this regard, being good at the convolutional neural networks of capture space relationship
(Convolutional Neural Network, CNN) is used for urban transportation prediction of speed.
During spatial evolution, since the traffic network in city is a topological structure, drilling spatially
Change can be influenced by the topological structure of road network, that is to say, that a section will have a direct impact on the adjoining section of surrounding.And it is traditional
CNN can only learn the unit that closes in matrix, but the other space time velocity matrix of City-level does not ensure that and closes on road
In an adjacent row, the above-mentioned method for carrying out traffic speed prediction to city using CNN does not account for this point to section, therefore can acquire
Some mistake Space Dynamic Evolution feature and influence its precision of prediction.
It is those skilled in the art's urgent problem to be solved as it can be seen that how to promote the precision of prediction of speed.
Invention content
The purpose of the embodiment of the present invention is to provide a kind of method, apparatus and computer readable storage medium of prediction of speed,
The precision of prediction of speed can be promoted.
In order to solve the above technical problems, the embodiment of the present invention provides a kind of method of prediction of speed, including：
According to the track data in preset time period, each corresponding initial velocity in section in target road network is calculated；Institute
There is the corresponding initial velocity in the section to constitute the initial velocity vector of the target road network；
Using the advance trained convolutional neural networks based on road network, to the corresponding initial velocity of the target road network to
Amount and adjacent section matrix carry out process of convolution, obtain corresponding eigenmatrix；Wherein, include each in the eigenmatrix
The corresponding feature vector in section；
The feature vector of target road section is converted to time series, and utilizes advance trained shot and long term memory network,
The time series is handled, the corresponding target velocity vector of the target road section is obtained；Wherein, the target road section is
Any one section in all sections that the target road network includes, the corresponding target velocity of all target road sections to
Amount constitutes the target velocity matrix of the target road network.
Optionally, the training process of the convolutional neural networks based on road network and the shot and long term memory network includes：
According to training set track data, each corresponding historical speed in section in the target road network is calculated；All institutes
State the historical speed vector that the corresponding historical speed in section constitutes the target road network；
Using the convolutional neural networks based on road network, section is abutted to the corresponding historical speed vector sum of the target road network
Matrix carries out process of convolution, obtains corresponding history feature matrix；Wherein, include each described in the history feature matrix
The corresponding history feature vector in section；
The history feature vector of target road section is converted to historical time sequence, and utilizes shot and long term memory network,
The historical time sequence is handled, the corresponding predetermined speed vector of the target road section is obtained；All target roads
The corresponding predetermined speed vector of section constitutes predetermined speed matrix of the target road network；
According to predetermined speed matrix and the corresponding actual speed matrix of the target road network, adjustment is described to be based on road
The value of each model parameter of the convolutional neural networks of net and the shot and long term memory network, until each model parameter meets
Preset requirement.
Optionally, the track data according in preset time period, it is corresponding to calculate each section in target road network
Initial velocity includes：
Using following formula, target road section r corresponding initial velocities in cycle time t are calculated
In formula, T indicates to gather by all tracks of the target road section r in cycle time t in the track data；
Traj indicates a track in the track set T；F (traj, t, r) is used to indicate the track by the target road section r
Velocity amplitudes of the traj in cycle time t；R.length indicates the total length of target road section r；StartTime indicates rail
At the beginning of mark traj enters the target road section r in cycle time t；EndTime indicates track traj in the period
The time departure of the target road section r is left in time t.
Optionally, further include：
Using the full articulamentum of convolutional neural networks, the corresponding reference of the target road section is extracted from historical trajectory data
Matrix；
Fusion treatment is carried out to the target velocity matrix and the R-matrix, using obtained fusion results as described in
The newest target velocity matrix of target road network.
Optionally, further include：
Obtain actual speed matrix corresponding with the target velocity matrix；
According to the target velocity matrix and the actual speed matrix obtained after fusion treatment, adjust in network model
The value of each model parameter.
The embodiment of the present invention additionally provides a kind of device of prediction of speed, including computing unit, processing unit and prediction list
Member；
The computing unit, for according to the track data in preset time period, calculating each section phase in target road network
Corresponding initial velocity；The corresponding initial velocity in all sections constitutes the initial velocity vector of the target road network；
The processing unit, for utilizing the advance trained convolutional neural networks based on road network, to the target road
It nets corresponding initial velocity vector sum adjoining section matrix and carries out process of convolution, obtain corresponding eigenmatrix；Wherein, the spy
It includes each corresponding feature vector in the section to levy in matrix；
The predicting unit for the feature vector of target road section to be converted to time series, and is utilized and is trained in advance
Shot and long term memory network, the time series is handled, the target road section corresponding target velocity vector is obtained；Its
In, the target road section is any one section in all sections that the target road network includes, all target roads
The corresponding target velocity vector of section constitutes the target velocity matrix of the target road network.
Optionally, it is directed to training for the convolutional neural networks based on road network and the shot and long term memory network
Journey, described device further include adjustment unit；
The computing unit is additionally operable to, according to training set track data, it is corresponding to calculate each section in the target road network
Historical speed；The corresponding historical speed in all sections constitutes the historical speed vector of the target road network；
The processing unit is additionally operable to utilize the convolutional neural networks based on road network, history corresponding to the target road network
Velocity vector and adjacent section matrix carry out process of convolution, obtain corresponding history feature matrix；Wherein, the history feature square
It include the corresponding history feature vector in each section in battle array；
The predicting unit is additionally operable to the history feature vector of target road section being converted to historical time sequence, and profit
With shot and long term memory network, the historical time sequence is handled, obtain the corresponding predetermined speed of the target road section to
Amount；The corresponding predetermined speed vector of all target road sections constitutes predetermined speed matrix of the target road network；
The adjustment unit, for according to predetermined speed matrix and the corresponding actual speed square of the target road network
Battle array adjusts the value of each model parameter of the convolutional neural networks and the shot and long term memory network based on road network, until
Each model parameter meets preset requirement.
Optionally, the computing unit is specifically used for utilizing following formula, and it is right in cycle time t to calculate target road section r
The initial velocity answered
In formula, T indicates to gather by all tracks of the target road section r in cycle time t in the track data；
Traj indicates a track in the track set T；F (traj, t, r) is used to indicate the track by the target road section r
Velocity amplitudes of the traj in cycle time t；R.length indicates the total length of target road section r；StartTime indicates rail
At the beginning of mark traj enters the target road section r in cycle time t；EndTime indicates track traj in the period
The time departure of the target road section r is left in time t.
Optionally, further include extraction unit and integrated unit；
The extraction unit, for the full articulamentum using convolutional neural networks, from historical trajectory data described in extraction
The corresponding R-matrix of target road section；
The integrated unit will be obtained for carrying out fusion treatment to the target velocity matrix and the R-matrix
Fusion results as the newest target velocity matrix of the target road network.
Optionally, further include acquiring unit and adjustment unit；
The acquiring unit, for obtaining actual speed matrix corresponding with the target velocity matrix；
The adjustment unit, for according to the target velocity matrix and the actual speed square obtained after fusion treatment
Battle array adjusts the value of each model parameter in network model.
The embodiment of the present invention additionally provides a kind of device of prediction of speed, including：
Memory, for storing computer program；
Processor, for the step of executing method of the computer program to realize prediction of speed as described above.
The embodiment of the present invention additionally provides a kind of computer readable storage medium, is deposited on the computer readable storage medium
Computer program is contained, the computer program realizes the method for prediction of speed as described above when being executed by processor the step of.
According to the track data in preset time period it can be seen from above-mentioned technical proposal, calculate each in target road network
The corresponding initial velocity in section；The corresponding initial velocity in all sections constitute the initial velocity of the target road network to
Amount；Utilize the advance trained convolutional neural networks based on road network, initial velocity vector sum corresponding to the target road network
Adjacent section matrix carries out process of convolution, obtains corresponding eigenmatrix；Wherein, include each described in the eigenmatrix
The corresponding feature vector in section；This feature vector is converted to time series, and is remembered using trained shot and long term in advance
Recall network, the time series is handled, obtains the corresponding target velocity vector of the target road section；Wherein, target road
Section is any one section in all sections that target road network includes, the corresponding target velocity of all target road sections to
Amount constitutes the target velocity matrix of the target road network.Since feature vector is obtained in the case where considering road network topology structure situation
Vector, effectively prevent influence of the Space Dynamic Evolution feature to its precision of mistake.So that according to this feature vector forecasting
The target velocity vector gone out is more accurate, effectively improves the precision of prediction of speed.
Description of the drawings
In order to illustrate the embodiments of the present invention more clearly, attached drawing needed in the embodiment will be done simply below
It introduces, it should be apparent that, drawings in the following description are only some embodiments of the invention, for ordinary skill people
For member, without creative efforts, other drawings may also be obtained based on these drawings.
Fig. 1 is a kind of flow chart of the method for prediction of speed provided in an embodiment of the present invention；
Fig. 2 a are a kind of structural schematic diagram of simple road network provided in an embodiment of the present invention；
Fig. 2 b be it is provided in an embodiment of the present invention it is a kind of based on each section in road network shown in Fig. 2 a within a preset period of time
Initial velocity schematic diagram；
Fig. 3 is a kind of structural schematic diagram of speed prediction model provided in an embodiment of the present invention；
Fig. 4 is a kind of structural schematic diagram of the device of prediction of speed provided in an embodiment of the present invention；
Fig. 5 is a kind of hardware architecture diagram of the device of prediction of speed provided in an embodiment of the present invention.
Specific implementation mode
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation describes, it is clear that described embodiments are only a part of the embodiments of the present invention, rather than whole embodiments.Based on this
Embodiment in invention, without making creative work, what is obtained is every other by those of ordinary skill in the art
Embodiment belongs to the scope of the present invention.
In order to enable those skilled in the art to better understand the solution of the present invention, with reference to the accompanying drawings and detailed description
The present invention is described in further detail.
Next, the method that a kind of prediction of speed that the embodiment of the present invention is provided is discussed in detail.Fig. 1 is that the present invention is implemented
A kind of flow chart of the method for prediction of speed that example provides, this method include：
S101：According to the track data in preset time period, each corresponding initial speed in section in target road network is calculated
Degree；The corresponding initial velocity in all sections constitutes the initial velocity vector of target road network.
With the fast development of global positioning system, wireless communication and development of Mobile Internet technology, track data increasingly holds
It easily obtains, a large amount of track data can map out the traffic speed situation in entire city.In embodiments of the present invention, Ke Yili
The velocity amplitude in each section is calculated with track data.
Target road network can be need to carry out it link speed predicti section set namely target road network in include
Multiple sections.
In embodiments of the present invention, cycle time can be preset.For example, cycle time could be provided as 20 minutes.
Often pass through the track data of target road network of a cycle time acquisition.
In practical applications, the speed in each section is predicted in needing to target road network in the next cycle time
When, can obtain with the track data of the cycle time tight adjacent preceding several cycle times be used as with reference to value namely it is default when
Between section may include one or more cycle times.
In embodiments of the present invention, the specific value of preset time period is not limited.For subsequently introduce conveniently, with
Track data in 3 cycle times is used as with reference to being worth, to predict the speed in each section in target road network in the next cycle time
It is unfolded to introduce for angle value.Wherein, this 3 cycle times are to be pressed for time adjacent preceding 3 week with next cycle to be predicted
Time phase.
By taking preset time period includes 3 cycle times as an example, it is assumed that cycle time is 20 minutes, to predict 10 o'clock sharp of the morning
, then can be 20 minutes with 9 points to the velocity amplitude of 10: 20 partial objectives for road network of the morning in 9 o'clock sharp to the morning in the morning, 20 minutes at 9 points in mornings are supreme
40 minutes 9 points of noons and 40 minutes at 9 points in mornings to the track data in this 3 cycle times of 10 o'clock sharp of the morning are used as with reference to being worth.
The calculation of the initial velocity in each section is similar in target road network in each cycle time, with a cycle time
For, in the concrete realization, following formula can be utilized, target road section r corresponding initial velocities in cycle time t are calculated
In formula, T indicates to gather by all tracks of target road section r in cycle time t in track data；Traj is indicated
A track in the set T of track；F (traj, t, r) is used to indicate the track traj by target road section r in cycle time t
Velocity amplitude；R.length indicates the total length of target road section r；StartTime indicates track traj in cycle time t
At the beginning of the interior r into target road section；EndTime indicates that track traj leaves target road section r in cycle time t
Time departure.
The calculating process of the initial velocity in reference object section can calculate the entire target road network in cycle time t
All sections on initial velocity, here use initial velocity vector XtIt indicates,
Wherein, | E | indicate the quantity in section in target road network.
It include 7 in road network shown in Fig. 2 a referring to the structural schematic diagram of one shown in Fig. 2 a simple road network
Section is followed successively by r0-r6, it is assumed that preset time period is 20 minutes, and the road network in the morning such as scheme by 8 points to 8 points 20 minutes initial velocities
Shown in 2b, the unit of the corresponding initial velocity in each section is km/h, correspondingly, the corresponding initial velocity vector X of the road networkt=
[40,38.6,23.5,11.6,21.7,35.7,25.2]。
S102：Utilize the advance trained convolutional neural networks based on road network, initial velocity corresponding to target road network
Vector sum abuts section matrix and carries out process of convolution, obtains corresponding eigenmatrix.
Wherein, in eigenmatrix include the corresponding feature vector in each section.
Since traditional CNN can not be directly applied in the road network data with topological structure in this way, in the present invention
In embodiment, the topological structure of road network is embedded into convolution using convolutional layer (Look-up Convolutional, LC) is searched
In operation, the convolutional neural networks for introducing LC layers can be referred to as the convolutional neural networks based on road network.
In the concrete realization, an adjacent section matrix M is defined first to go to record the adjoining section in all sections.
For a section r, adjacent set S can be usedrIndicate the adjoining section in the section, specifically：
Sr=r, r' ∈ E | r'.s=r.e or r'.e=r.s }；
Wherein, r.s indicate section r starting point, r.e indicate section r terminal, r'.s=r.e indicate using the terminal of r as
All sections of starting point, r'.e=r.s indicate all sections using the starting point of r as terminal.Therefore SrHave recorded section r itself
And all sections using the terminal of r as starting point and all sections using the starting point of r as terminal.
For all sections, we can obtain a maximum adjacent quantity, be indicated by A, to which we can construct
One A × | E | matrix M.The i-th row of M have recorded section riAll of its neighbor section, that is,In section.
For the schematic diagram of road network shown in Fig. 2 a, third section r3Adjoining section collection be combined into S=0,2,3,
6 }, wherein 0 indicates section r0, 2 indicate section r2, 3 indicate section r3, 6 indicate section r6.Thus M [:, 3]=[0,2,3,6]T,
M[:, 3] and indicate section r3All of its neighbor section matrix.
Since the adjoining section in each section is not quite similar, in embodiments of the present invention, the mark in section itself can be used
Information such as id fills the value lacked until A.
After the topological structure for expressing road network with M, needs to embed it in convolution operation and be drilled to extract Spatial distributions
The feature of change.
For arbitrary 1 layer, by the output X of M and preceding layerl-1As input, to obtain the output X of current layerl, wherein l
Indicate LC layers of number.
Wherein, first layer input can from the speed of the velocity vector of current period time and preceding p cycle time to
Amount composition, i.e. X0=[Xt,Xt-1,…,Xt-p]T。
Wherein, the value of p can be set according to actual demand, as long as ensureing when the value and model training of p every
The number for the cycle time that training data includes matches.Assuming that in model training, every training data is foundation
The speed of preceding 3 cycle time target road network predicts the speed of the 4th cycle time target road network, then is executing prediction of speed
When, the corresponding velocity vector of current period time is subtracted, the value of p is 3-1=2.
As traditional convolution operation, in order to obtain more diversified feature, it can be rolled up using multiple convolution kernels
Product.
By taking k-th of convolution kernel as an example, k-th of eigenmatrix can be obtained after searching convolution operation, formula is as follows：
Wherein, k-th eigenmatrix is really by | E | the feature vector in a section forms, and the feature in i-th of section to
Amount can be obtained by following formula,
In this formula, for indicating that look-up is operated, it is tieed up according to the numerical value in Q as in P second L (P, Q)
The index of degree returns to a vector or matrix.
Specifically L (P, Q)=P [:, Q], therefore in this formula, L (M, i)=M [:, i] and it indicates to obtain section i's
All of its neighbor section, L (Xl-1,M[:, i]) it is exactly to go to obtain subcharacter matrix again by all of its neighbor section, we use hereIt indicates.
Next, being exactly to carry out convolution to the submatrix by traditional convolution operation, * indicates convolution operation, Wl,kIt is one
The convolution kernel of a h × A, h are usually arranged as 1 or 2, go scanning submatrix can be obtained with convolution kernelIt can be to
Lower formula indicates：
Wherein, relu is to correct linear activation primitive, takes input value and 0 maximum value, specially relu (x)=max (0,
x)。
In addition, gradient disappears in order to prevent and fast-training procedures can be followed by embodiments of the present invention in each LC
A upper batch standardization layer (batch normalization, BN).
Wherein, the BN layers of layer for commonly accelerating network processes speed for convolutional neural networks in the prior art, to its work
Principle repeats no more.
S103：Feature vector is converted to time series, and using advance trained shot and long term memory network, to the time
Sequence is handled, and the corresponding target velocity vector of target road section is obtained.
Wherein, target road section is any one section in all sections that target road network includes, all target road sections
Corresponding target velocity vector may be constructed the target velocity matrix of target road network.
After by multiple LC layers, we can obtain the Spatial distributions characteristic pattern on entire road network, it is assumed that we are most
The output that LC layers of the latter is
Wherein, knFor the convolution kernel number used in the last one LC layers.
Before handling feature vector, need to deform the feature vector corresponding to each section
For a time seriesWherein each feature vector can be obtained by lower formula,
Using the corresponding time series in each section as the input of LSTM.Hiding sequence can be iteratively obtained using LSTM
[h0,…,ht,…,hp]。
Wherein,
Based on last hidden state, we can remove the traffic speed of z cycle time after prediction, tool with z unit
Body calculates as follows：
yi=[yi,t+1,yi,t+2,…,yi,t+z]；
Finally connect the prediction output composition target velocity matrix Y in all sectionsST,
YST=[y0,y1,…,y|E|-1]T。
In embodiments of the present invention, using based on road network convolutional neural networks and shot and long term memory network to target road network
The prediction for carrying out speed, realizes the combination to space-time characterisation.And LC is introduced in the convolutional neural networks based on road network
Layer, to which the topological structure of road network to be embedded into convolution operation so that the velocity amplitude predicted is more accurate.
Utilizing the prediction of convolutional neural networks and shot and long term memory network to target road network progress speed based on road network
Before, need to adjust each model parameter in convolutional neural networks and shot and long term memory network based on road network, wherein model is joined
Number is used parameter when executing corresponding function.
Illustratively, can regard the combination of convolutional neural networks and shot and long term memory network based on road network as one
A network model.In embodiments of the present invention, the corresponding history whithin a period of time of each section in target road network can be utilized
Track data is trained the network model, so as to adjust corresponding model parameter.
Specifically, each corresponding historical speed in section in target road network can be calculated according to training set track data；
The corresponding historical speed in all sections constitutes the historical speed vector of target road network.Then the convolutional Neural net based on road network is utilized
Network carries out process of convolution to target road network corresponding historical speed vector sum adjoining section matrix, obtains corresponding history feature
Matrix；Wherein, in history feature matrix include the corresponding history feature vector in each section；History feature vector is turned
It is melted into historical time sequence, and utilizes shot and long term memory network, historical time sequence is handled, obtains target road section correspondence
Predetermined speed vector；The corresponding historical speed vector of all target road sections constitutes predetermined speed matrix of target road network.
Using convolutional neural networks and shot and long term memory network based on road network, training set track data is handled,
So that it is determined that it is similar to go out the step of course of work of predetermined speed matrix of target road network is with above-mentioned S101-S103, herein no longer
It repeats.
In embodiments of the present invention, one day, one week or one month track number before current time can be chosen
According to as training corpus.
It, can be using the track data of target road network in the time before current time as training language for one day
Material can then get 72 track datas as training corpus in conjunction with assuming that cycle time is 20 minutes in above-mentioned introduction.
With in the prior art by historical data adjust model parameter in the way of it is similar, can be with when being trained to model
Training corpus is divided into training set track data and verification collection track data.Using training set track data, network mould is adjusted
The model parameter of type.It is verified according to verification collection track data to whether each model parameter in network model reaches requirement.
Since the values for actual speed in each section in training corpus is known quantity, in embodiments of the present invention, it can incite somebody to action
According to the actual speed matrix of predetermined speed matrix and target road network, convolutional neural networks and shot and long term note based on road network are adjusted
The value of each model parameter of network is recalled, until each model parameter meets preset requirement.
When adjusting model parameter, the velocity amplitude predicted and values for actual speed can be subjected to mean square error calculating, so
The value of model parameter is adjusted by back-propagation algorithm afterwards.
When verifying model parameter, track data can be collected according to verification, calculate predetermined speed matrix and actual speed square
The mean square error of battle array, and using the mean square error as judging whether the model parameter of network model meets the foundation of preset requirement.
When the square mean error amount is compared with last square mean error amount, mean square error becomes hour, then illustrates network model
Model parameter has the space further adjusted, then can further be adjusted to model parameter according to training set track data
It is whole.
When the square mean error amount is compared with last square mean error amount, when mean square error is constant, then illustrate network model
Model parameter has been adjusted to optimal, can terminate the training to network model at this time.
When the square mean error amount is compared with last square mean error amount, when mean square error becomes larger, then illustrate last adjustment
Model parameter afterwards have reached it is optimal, then can using the model parameter of last time adjustment as the model parameter of network model,
And training of the end to network model.
According to the track data in preset time period it can be seen from above-mentioned technical proposal, calculate each in target road network
The corresponding initial velocity in section；The corresponding initial velocity in all sections constitutes the initial velocity vector of target road network；Using pre-
First trained convolutional neural networks based on road network, to the corresponding initial velocity vector sum of target road network abut section matrix into
Row process of convolution obtains corresponding eigenmatrix；Wherein, include the corresponding feature in each section in eigenmatrix to
Amount；This feature vector is converted to time series, and using advance trained shot and long term memory network, time series is carried out
Processing obtains the corresponding target velocity vector of target road section；Wherein, target road section is in all sections that target road network includes
Any one section, the corresponding target velocity vector of all target road sections constitutes the target velocity matrix of target road network.Due to
Feature vector is the vector obtained in the case where considering road network topology structure situation, and the Space Dynamic Evolution for effectively preventing mistake is special
Levy the influence to its precision.So that it is more accurate according to the target velocity vector that this feature vector forecasting goes out, effectively improve speed
Spend the precision of prediction.
In practical applications, when carrying out the prediction of speed of road network other than the changing pattern of spatial evolution and time series,
The cyclically-varying rule and traffic scene of each section speed also have having a certain impact for the prediction of section velocity amplitude.
In the embodiment of the present invention, in order to further enhance the accuracy of link speed predicti, the complete of convolutional neural networks can be utilized to connect
Layer is connect, the corresponding R-matrix of target road section is extracted from historical trajectory data；Target velocity vector sum R-matrix is carried out
Fusion treatment, using obtained fusion results as the newest target velocity vector of target road section.
Wherein, R-matrix be may include the velocity vector extracted based on traffic speed cyclically-varying rule and is based on
The velocity vector extracted under the influence of different traffic scene factors.
On some sections of target road network, interval will present similitude to traffic speed at the same time daily, weekly
It is spaced at the same time and also will present out certain tendency.In order to extract this partial information, in embodiments of the present invention may be used
To use a full articulamentum come d days before extracting average speed and another full articulamentum used to go to extract first w weeks and
Gesture information finally obtains velocity vector Y in conjunction with themP。
Wherein, the specific value of d and w can be determined according to the cycle variation law of each section speed in target road network.
Traffic context data may include having festivals or holidays, weather conditions and some metadata for example, weekend, non-weekend, star
Phase is several, which hour, peak, non-peak etc..It can go to extract these letters using a full articulamentum in embodiments of the present invention
Breath, is then gone low latitude Feature Mapping obtaining velocity vector Y to higher-dimension again with another full articulamentumC。
It in the concrete realization, can be by obtain two speed YPAnd YCIt is added, obtains R-matrix YE, then using ginseng
Matrix number come merge R-matrix and target velocity vector, the fusion results that will be obtainedAs newest target velocity vector, tool
Body formula is as follows,
Wherein, WSTFor YSTCorresponding model parameter, WEFor YECorresponding model parameter.
It in embodiments of the present invention, can be by convolutional neural networks, shot and long term memory network and convolution based on road network
The combination of the full articulamentum of neural network is as a complete network model, structural schematic diagram such as Fig. 3 institutes of the network model
Show, the convolutional neural networks based on road network may include have it is LC layers multiple, in order to prevent gradient disappear and accelerate training process,
Each LC layers can connect one BN layers, and BN layers belong to common technology in the prior art, are repeated no more to its concrete operating principle.
After multiple LC layers and BN layers of processing, the corresponding eigenmatrix of target road network can be obtained, will include in this feature matrix
Each feature vector be transformed into time series after, each time series can be handled using shot and long term memory network,
Obtain the target velocity matrix Y corresponding to target road networkST.Pass through full articulamentum FCS1 can extract velocity vector YP；Pass through
Full articulamentum FCS2 can extract velocity vector Yc；To YPAnd YcIt is summed to obtain R-matrix YE.In the Fusion of Fig. 3
Include corresponding model parameter WSTAnd WE, using hyperbolic functions tanh to YSTAnd YEFusion treatment is carried out, final mesh is obtained
Mark rate matrices
According to network model shown in Fig. 3, the velocity amplitude in each section in target road network can be predicted.It is instructed in model
Practice the stage, actual speed matrix Y corresponding with target velocity matrix can be obtainedt；According to the target obtained after fusion treatment
Rate matricesWith actual speed matrix Yt, adjust the value of each model parameter in network model.Specifically, loss can be passed through
Function (Loss) calculates target velocity matrixWith actual speed matrix YtMean square error, i.e.,Foundation should
Mean square error updates each model parameter in network model by back-propagation algorithm.
Fig. 4 be a kind of structural schematic diagram of the device of prediction of speed provided in an embodiment of the present invention, including computing unit 41,
Processing unit 42 and predicting unit 43；
Computing unit 41, for according to the track data in preset time period, it is opposite to calculate each section in target road network
The initial velocity answered；The corresponding initial velocity in all sections constitutes the initial velocity vector of target road network；
Processing unit 42, for using the advance trained convolutional neural networks based on road network, being corresponded to target road network
Initial velocity vector sum adjoining section matrix carry out process of convolution, obtain corresponding eigenmatrix；Wherein, it is wrapped in eigenmatrix
The corresponding feature vector in each section is included；
Predicting unit 43 for feature vector to be converted to time series, and is remembered using trained shot and long term in advance
Network handles time series, obtains the corresponding target velocity vector of target road section；Wherein, target road section is target road
Any one section in all sections that net includes, the corresponding target velocity vector of all target road sections constitute target road network
Target velocity matrix.
Optionally, it is directed to the training process of convolutional neural networks and shot and long term memory network based on road network, device is also
Including adjustment unit；
Computing unit is additionally operable to, according to training set track data, calculate each corresponding history speed in section in target road network
Degree；The corresponding historical speed in all sections constitutes the historical speed vector of target road network；
Processing unit is additionally operable to utilize the convolutional neural networks based on road network, to target road network corresponding historical speed vector
Process of convolution is carried out with adjacent section matrix, obtains corresponding history feature matrix；Wherein, include every in history feature matrix
A corresponding history feature vector in section；
Predicting unit is additionally operable to history feature vector being converted to historical time sequence, and utilizes shot and long term memory network,
Historical time sequence is handled, the corresponding predetermined speed vector of target road section is obtained；The corresponding history of all target road sections
Velocity vector constitutes predetermined speed matrix of target road network；
Adjustment unit is adjusted for the actual speed matrix according to predetermined speed matrix and target road network based on road network
The value of each model parameter of convolutional neural networks and shot and long term memory network, until each model parameter meets preset requirement.
Optionally, computing unit is specifically used for utilizing following formula, and it is corresponding in cycle time t to calculate target road section r
Initial velocity
In formula, T indicates to gather by all tracks of target road section r in cycle time t in track data；Traj is indicated
A track in the set T of track；F (traj, t, r) is used to indicate the track traj by target road section r in cycle time t
Velocity amplitude；R.length indicates the total length of target road section r；StartTime indicates track traj in cycle time t
At the beginning of the interior r into target road section；EndTime indicates that track traj leaves target road section r in cycle time t
Time departure.
Optionally, further include extraction unit and integrated unit；
Extraction unit extracts target road section for the full articulamentum using convolutional neural networks from historical trajectory data
Corresponding R-matrix；
Integrated unit makees obtained fusion results for carrying out fusion treatment to target velocity vector sum R-matrix
For the newest target velocity vector of target road section.
Optionally, further include acquiring unit and adjustment unit；
Acquiring unit, for obtaining actual speed matrix corresponding with target velocity matrix；
Adjustment unit, for according to the target velocity matrix and actual speed matrix obtained after fusion treatment, adjusting network
The value of each model parameter in model.
The explanation of feature may refer to the related description of embodiment corresponding to Fig. 1 in embodiment corresponding to Fig. 4, here no longer
It repeats one by one.
According to the track data in preset time period it can be seen from above-mentioned technical proposal, calculate each in target road network
The corresponding initial velocity in section；The corresponding initial velocity in all sections constitutes the initial velocity vector of target road network；Using pre-
First trained convolutional neural networks based on road network, to the corresponding initial velocity vector sum of target road network abut section matrix into
Row process of convolution obtains corresponding eigenmatrix；Wherein, include the corresponding feature in each section in eigenmatrix to
Amount；This feature vector is converted to time series, and using advance trained shot and long term memory network, time series is carried out
Processing obtains the corresponding target velocity vector of target road section；Wherein, target road section is in all sections that target road network includes
Any one section, the corresponding target velocity vector of all target road sections constitutes the target velocity matrix of target road network.Due to
Feature vector is the vector obtained in the case where considering road network topology structure situation, and the Space Dynamic Evolution for effectively preventing mistake is special
Levy the influence to its precision.So that it is more accurate according to the target velocity vector that this feature vector forecasting goes out, effectively improve speed
Spend the precision of prediction.
Fig. 5 is a kind of structural schematic diagram of the device 50 of prediction of speed provided in an embodiment of the present invention, including：
Memory 51, for storing computer program；
Processor 52, for the step of executing computer program to realize the method such as above-mentioned prediction of speed.
The embodiment of the present invention additionally provides a kind of computer readable storage medium, is stored on computer readable storage medium
Computer program, when computer program is executed by processor the step of the realization such as method of above-mentioned prediction of speed.
It is provided for the embodiments of the invention a kind of method, apparatus and computer readable storage medium of prediction of speed above
It is described in detail.Each embodiment is described by the way of progressive in specification, the highlights of each of the examples are
Difference from other examples, just to refer each other for identical similar portion between each embodiment.Embodiment is disclosed
Device for, since it is corresponded to the methods disclosed in the examples, so description is fairly simple, related place is referring to method
Part illustrates.It should be pointed out that for those skilled in the art, before not departing from the principle of the invention
It puts, can be with several improvements and modifications are made to the present invention, these improvement and modification also fall into the guarantor of the claims in the present invention
It protects in range.
Professional further appreciates that, unit described in conjunction with the examples disclosed in the embodiments of the present disclosure
And algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, in order to clearly demonstrate hardware and
The interchangeability of software generally describes each exemplary composition and step according to function in the above description.These
Function is implemented in hardware or software actually, depends on the specific application and design constraint of technical solution.Profession
Technical staff can use different methods to achieve the described function each specific application, but this realization is not answered
Think beyond the scope of this invention.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can directly be held with hardware, processor
The combination of capable software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only deposit
Reservoir (ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technology
In any other form of storage medium well known in field.
Claims (10)
1. a kind of method of prediction of speed, which is characterized in that including：
According to the track data in preset time period, each corresponding initial velocity in section in target road network is calculated；All institutes
State the initial velocity vector that the corresponding initial velocity in section constitutes the target road network；
Utilize the advance trained convolutional neural networks based on road network, initial velocity vector sum corresponding to the target road network
Adjacent section matrix carries out process of convolution, obtains corresponding eigenmatrix；Wherein, include each described in the eigenmatrix
The corresponding feature vector in section；
The feature vector of target road section is converted to time series, and using advance trained shot and long term memory network, to institute
It states time series to be handled, obtains the corresponding target velocity vector of the target road section；Wherein, the target road section is described
Any one section in all sections that target road network includes, the corresponding target velocity vector structure of all target road sections
At the target velocity matrix of the target road network.
2. according to the method described in claim 1, it is characterized in that, the convolutional neural networks based on road network and the length
The training process of phase memory network includes：
According to training set track data, each corresponding historical speed in section in the target road network is calculated；All roads
The corresponding historical speed of section constitutes the historical speed vector of the target road network；
Using the convolutional neural networks based on road network, section matrix is abutted to the corresponding historical speed vector sum of the target road network
Process of convolution is carried out, corresponding history feature matrix is obtained；Wherein, in the history feature matrix include each section
Corresponding history feature vector；
The history feature vector of target road section is converted to historical time sequence, and utilizes shot and long term memory network, to institute
It states historical time sequence to be handled, obtains the corresponding predetermined speed vector of the target road section；All target road sections pair
Predetermined speed vector answered constitutes predetermined speed matrix of the target road network；
According to predetermined speed matrix and the corresponding actual speed matrix of the target road network, adjustment is described based on road network
The value of each model parameter of convolutional neural networks and the shot and long term memory network is preset until each model parameter meets
It is required that.
3. according to the method described in claim 1, it is characterized in that, the track data according in preset time period, calculates
Going out each corresponding initial velocity in section in target road network includes：
Using following formula, target road section r corresponding initial velocities in cycle time t are calculated
In formula, T indicates to gather by all tracks of the target road section r in cycle time t in the track data；traj
Indicate a track in the track set T；F (traj, t, r) is used to indicate the track traj by the target road section r
Velocity amplitude in cycle time t；R.length indicates the total length of target road section r；Indicate track
At the beginning of traj enters the target road section r in cycle time t；Indicate track traj in the period
The time departure of the target road section r is left in time t.
4. according to the method described in claim 1-3 any one, which is characterized in that further include：
Using the full articulamentum of convolutional neural networks, it is corresponding with reference to square that the target road section is extracted from historical trajectory data
Battle array；
Fusion treatment is carried out to the target velocity matrix and the R-matrix, using obtained fusion results as the target
The newest target velocity matrix of road network.
5. according to the method described in claim 4, it is characterized in that, further including：
Obtain actual speed matrix corresponding with the target velocity matrix；
According to the target velocity matrix and the actual speed matrix obtained after fusion treatment, each mould in network model is adjusted
The value of shape parameter.
6. a kind of device of prediction of speed, which is characterized in that including computing unit, processing unit and predicting unit；
The computing unit, for according to the track data in preset time period, it is corresponding to calculate each section in target road network
Initial velocity；The corresponding initial velocity in all sections constitutes the initial velocity vector of the target road network；
The processing unit, for utilizing the advance trained convolutional neural networks based on road network, to the target road network pair
The initial velocity vector sum adjoining section matrix answered carries out process of convolution, obtains corresponding eigenmatrix；Wherein, the feature square
Include each corresponding feature vector in the section in battle array；
The predicting unit for the feature vector of target road section to be converted to time series, and utilizes advance trained length
Short-term memory network handles the time series, obtains the corresponding target velocity vector of the target road section；Wherein,
The target road section is any one section in all sections that the target road network includes, all target road sections pair
The target velocity vector answered constitutes the target velocity matrix of the target road network.
7. device according to claim 6, which is characterized in that be directed to the convolutional neural networks based on road network and institute
The training process of shot and long term memory network is stated, described device further includes adjustment unit；
The computing unit is additionally operable to according to training set track data, calculates in the target road network that each section is corresponding to be gone through
History speed；The corresponding historical speed in all sections constitutes the historical speed vector of the target road network；
The processing unit is additionally operable to utilize the convolutional neural networks based on road network, historical speed corresponding to the target road network
Vector sum abuts section matrix and carries out process of convolution, obtains corresponding history feature matrix；Wherein, in the history feature matrix
It include the corresponding history feature vector in each section；
The predicting unit is additionally operable to the history feature vector of target road section being converted to historical time sequence, and utilizes length
Short-term memory network handles the historical time sequence, obtains the corresponding predetermined speed vector of the target road section；Institute
There is the corresponding predetermined speed vector of the target road section to constitute predetermined speed matrix of the target road network；
The adjustment unit is used for according to predetermined speed matrix and the corresponding actual speed matrix of the target road network,
The value of each model parameter of the adjustment convolutional neural networks and the shot and long term memory network based on road network, until each institute
It states model parameter and meets preset requirement.
8. the device described according to claim 6 or 7, which is characterized in that further include extraction unit and integrated unit；
The extraction unit extracts the target for the full articulamentum using convolutional neural networks from historical trajectory data
The corresponding R-matrix in section；
The integrated unit melts for carrying out fusion treatment to the target velocity matrix and the R-matrix by what is obtained
Result is closed as the newest target velocity matrix of the target road network.
9. a kind of device of prediction of speed, which is characterized in that including：
Memory, for storing computer program；
Processor, for executing the computer program to realize the side of the prediction of speed as described in claim 1 to 5 any one
The step of method.
10. a kind of computer readable storage medium, which is characterized in that be stored with computer on the computer readable storage medium
Program, realizing the method for prediction of speed as described in any one of claim 1 to 5 when the computer program is executed by processor
Step.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810690697.8A CN108648457B (en) | 2018-06-28 | 2018-06-28 | Method, device and computer readable storage medium for speed prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810690697.8A CN108648457B (en) | 2018-06-28 | 2018-06-28 | Method, device and computer readable storage medium for speed prediction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108648457A true CN108648457A (en) | 2018-10-12 |
| CN108648457B CN108648457B (en) | 2021-07-13 |
Family
ID=63749776
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810690697.8A Active CN108648457B (en) | 2018-06-28 | 2018-06-28 | Method, device and computer readable storage medium for speed prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108648457B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109410575A (en) * | 2018-10-29 | 2019-03-01 | 北京航空航天大学 | A kind of road network trend prediction method based on capsule network and the long Memory Neural Networks in short-term of nested type |
| CN109544911A (en) * | 2018-10-30 | 2019-03-29 | 中山大学 | A kind of city road net traffic state prediction technique based on LSTM-CNN |
| CN109740811A (en) * | 2018-12-28 | 2019-05-10 | 斑马网络技术有限公司 | Passage speed prediction technique, device and storage medium |
| CN111152796A (en) * | 2020-04-07 | 2020-05-15 | 北京三快在线科技有限公司 | Vehicle motion state prediction method and device |
| CN111739283A (en) * | 2019-10-30 | 2020-10-02 | 腾讯科技（深圳）有限公司 | Road condition calculation method, device, equipment and medium based on clustering |
| CN111833600A (en) * | 2020-06-10 | 2020-10-27 | 北京嘀嘀无限科技发展有限公司 | Method and device for predicting transit time and data processing equipment |
| CN111862595A (en) * | 2020-06-08 | 2020-10-30 | 同济大学 | Speed prediction method, system, medium and device based on road network topological relation |
| CN111950810A (en) * | 2020-08-27 | 2020-11-17 | 南京大学 | Multivariable time sequence prediction method and device based on self-evolution pre-training |
| CN112265546A (en) * | 2020-10-26 | 2021-01-26 | 吉林大学 | Networked automobile speed prediction method based on time-space sequence information |
| CN112863180A (en) * | 2021-01-11 | 2021-05-28 | 腾讯大地通途（北京）科技有限公司 | Traffic speed prediction method, device, electronic equipment and computer readable medium |
| CN113160570A (en) * | 2021-05-27 | 2021-07-23 | 长春理工大学 | Traffic jam prediction method and system |
| CN113971885A (en) * | 2020-07-06 | 2022-01-25 | 华为技术有限公司 | Vehicle speed prediction method, device and system |
| CN115410386A (en) * | 2022-09-05 | 2022-11-29 | 同盾科技有限公司 | Short-time speed prediction method and device, computer storage medium and electronic equipment |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105654729A (en) * | 2016-03-28 | 2016-06-08 | 南京邮电大学 | Short-term traffic flow prediction method based on convolutional neural network |
| CN106779050A (en) * | 2016-11-24 | 2017-05-31 | 厦门中控生物识别信息技术有限公司 | The optimization method and device of a kind of convolutional neural networks |
| CN106886023A (en) * | 2017-02-27 | 2017-06-23 | 中国人民解放军理工大学 | A kind of Radar Echo Extrapolation method based on dynamic convolutional neural networks |
| CN106981198A (en) * | 2017-05-24 | 2017-07-25 | 北京航空航天大学 | Deep learning network model and its method for building up for predicting travel time |
| CN107103754A (en) * | 2017-05-10 | 2017-08-29 | 华南师范大学 | A kind of road traffic condition Forecasting Methodology and system |
| CN107180530A (en) * | 2017-05-22 | 2017-09-19 | 北京航空航天大学 | A kind of road network trend prediction method based on depth space-time convolution loop network |
| CN107273782A (en) * | 2016-04-08 | 2017-10-20 | 微软技术许可有限责任公司 | Detected using the online actions of recurrent neural network |
| US20180129910A1 (en) * | 2016-11-08 | 2018-05-10 | Nec Laboratories America, Inc. | Landmark localization on objects in images using convolutional neural networks |
| CN108196535A (en) * | 2017-12-12 | 2018-06-22 | 清华大学苏州汽车研究院（吴江） | Automated driving system based on enhancing study and Multi-sensor Fusion |
-
2018
- 2018-06-28 CN CN201810690697.8A patent/CN108648457B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105654729A (en) * | 2016-03-28 | 2016-06-08 | 南京邮电大学 | Short-term traffic flow prediction method based on convolutional neural network |
| CN107273782A (en) * | 2016-04-08 | 2017-10-20 | 微软技术许可有限责任公司 | Detected using the online actions of recurrent neural network |
| US20180129910A1 (en) * | 2016-11-08 | 2018-05-10 | Nec Laboratories America, Inc. | Landmark localization on objects in images using convolutional neural networks |
| CN106779050A (en) * | 2016-11-24 | 2017-05-31 | 厦门中控生物识别信息技术有限公司 | The optimization method and device of a kind of convolutional neural networks |
| CN106886023A (en) * | 2017-02-27 | 2017-06-23 | 中国人民解放军理工大学 | A kind of Radar Echo Extrapolation method based on dynamic convolutional neural networks |
| CN107103754A (en) * | 2017-05-10 | 2017-08-29 | 华南师范大学 | A kind of road traffic condition Forecasting Methodology and system |
| CN107180530A (en) * | 2017-05-22 | 2017-09-19 | 北京航空航天大学 | A kind of road network trend prediction method based on depth space-time convolution loop network |
| CN106981198A (en) * | 2017-05-24 | 2017-07-25 | 北京航空航天大学 | Deep learning network model and its method for building up for predicting travel time |
| CN108196535A (en) * | 2017-12-12 | 2018-06-22 | 清华大学苏州汽车研究院（吴江） | Automated driving system based on enhancing study and Multi-sensor Fusion |
Non-Patent Citations (5)
| Title |
|---|
| MICHAËL DEFFERRARD: "Convolutional Neural Networks on Graphs", 《NIPS》 * |
| YISHENG LV: "Traffic Flow Prediction With Big Data:", 《IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS》 * |
| ZHIZHAO ZHANG， PENG CHEN: "A Hybrid Deep Learning approach for Urban Expressway Travel Time Prediction", 《 2017 IEEE 20TH INTERNATIONAL CONFERENCE ON INTELLIGENT TRANSPORTATION SYSTEMS (ITSC)》 * |
| ZHONGTAO DUAN,YUN YANG,: "Improved Deep Hybrid Networks for Urban", 《IEEE ACCESS》 * |
| 许佳捷: "轨迹大数据：数据、应用与技术现状", 《通信学报》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109410575A (en) * | 2018-10-29 | 2019-03-01 | 北京航空航天大学 | A kind of road network trend prediction method based on capsule network and the long Memory Neural Networks in short-term of nested type |
| CN109544911A (en) * | 2018-10-30 | 2019-03-29 | 中山大学 | A kind of city road net traffic state prediction technique based on LSTM-CNN |
| CN109740811A (en) * | 2018-12-28 | 2019-05-10 | 斑马网络技术有限公司 | Passage speed prediction technique, device and storage medium |
| CN111739283A (en) * | 2019-10-30 | 2020-10-02 | 腾讯科技（深圳）有限公司 | Road condition calculation method, device, equipment and medium based on clustering |
| CN111739283B (en) * | 2019-10-30 | 2022-05-20 | 腾讯科技（深圳）有限公司 | Road condition calculation method, device, equipment and medium based on clustering |
| CN111152796A (en) * | 2020-04-07 | 2020-05-15 | 北京三快在线科技有限公司 | Vehicle motion state prediction method and device |
| CN111862595B (en) * | 2020-06-08 | 2021-12-31 | 同济大学 | Speed prediction method, system, medium and device based on road network topological relation |
| CN111862595A (en) * | 2020-06-08 | 2020-10-30 | 同济大学 | Speed prediction method, system, medium and device based on road network topological relation |
| CN111833600A (en) * | 2020-06-10 | 2020-10-27 | 北京嘀嘀无限科技发展有限公司 | Method and device for predicting transit time and data processing equipment |
| CN113971885A (en) * | 2020-07-06 | 2022-01-25 | 华为技术有限公司 | Vehicle speed prediction method, device and system |
| CN111950810A (en) * | 2020-08-27 | 2020-11-17 | 南京大学 | Multivariable time sequence prediction method and device based on self-evolution pre-training |
| CN111950810B (en) * | 2020-08-27 | 2023-12-15 | 南京大学 | Multi-variable time sequence prediction method and equipment based on self-evolution pre-training |
| CN112265546A (en) * | 2020-10-26 | 2021-01-26 | 吉林大学 | Networked automobile speed prediction method based on time-space sequence information |
| CN112863180A (en) * | 2021-01-11 | 2021-05-28 | 腾讯大地通途（北京）科技有限公司 | Traffic speed prediction method, device, electronic equipment and computer readable medium |
| CN112863180B (en) * | 2021-01-11 | 2022-05-06 | 腾讯大地通途（北京）科技有限公司 | Traffic speed prediction method, device, electronic equipment and computer readable medium |
| CN113160570A (en) * | 2021-05-27 | 2021-07-23 | 长春理工大学 | Traffic jam prediction method and system |
| CN115410386A (en) * | 2022-09-05 | 2022-11-29 | 同盾科技有限公司 | Short-time speed prediction method and device, computer storage medium and electronic equipment |
| CN115410386B (en) * | 2022-09-05 | 2024-02-06 | 同盾科技有限公司 | Short-time speed prediction method and device, computer storage medium and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108648457B (en) | 2021-07-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108648457A (en) | A kind of method, apparatus and computer readable storage medium of prediction of speed | |
| CN108197739B (en) | Urban rail transit passenger flow prediction method | |
| CN111462484B (en) | Congestion state determination method, device, equipment and computer readable storage medium | |
| CN111653088B (en) | Vehicle driving quantity prediction model construction method, prediction method and system | |
| CN109035761B (en) | Travel time estimation method based on auxiliary supervised learning | |
| CN111696355A (en) | Dynamic graph convolution traffic speed prediction method | |
| CN103646187B (en) | Method for obtaining vehicle travel path and OD (Origin-Destination) matrix in statistic period | |
| CN103794053B (en) | Vague predicting method for city short-distance logistics simple target delivering time | |
| Dunne et al. | Regime-based short-term multivariate traffic condition forecasting algorithm | |
| CN107492113A (en) | A kind of moving object in video sequences position prediction model training method, position predicting method and trajectory predictions method | |
| CN107018493B (en) | Geographic position prediction method based on continuous time sequence Markov model | |
| CN108427965A (en) | A kind of hot spot region method for digging based on road network cluster | |
| CN104408203A (en) | Method for predicting path destination of moving object | |
| CN112071062B (en) | Driving time estimation method based on graph convolution network and graph attention network | |
| CN108230020B (en) | Method for mining space-time frequent region based on multi-dimensional time granularity | |
| CN112863182B (en) | Cross-modal data prediction method based on transfer learning | |
| CN106504548A (en) | Traffic lights intelligent control method and system | |
| CN112766600B (en) | Urban area crowd flow prediction method and system | |
| CN108629970A (en) | Intersection signal parameter optimization method based on the search of Monte Carlo tree | |
| CN108805392B (en) | Reachability evaluation method for integrating human trip behaviors based on trajectory data | |
| CN111127879B (en) | Urban traffic flow prediction method based on generative countermeasure network | |
| CN113327417A (en) | Traffic flow prediction method based on 3D dynamic space-time residual convolution associated network | |
| CN115691167A (en) | Single-point traffic signal control method based on intersection holographic data | |
| CN109325657A (en) | A kind of measures and procedures for the examination and approval of credit operation, storage medium and server | |
| CN109712402B (en) | Mobile object running time prediction method and device based on meta-path congestion mode mining |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |