CN107103754A - A kind of road traffic condition Forecasting Methodology and system - Google Patents
A kind of road traffic condition Forecasting Methodology and system Download PDFInfo
- Publication number
- CN107103754A CN107103754A CN201710326733.8A CN201710326733A CN107103754A CN 107103754 A CN107103754 A CN 107103754A CN 201710326733 A CN201710326733 A CN 201710326733A CN 107103754 A CN107103754 A CN 107103754A
- Authority
- CN
- China
- Prior art keywords
- road
- data
- traffic
- cyclic convolution
- neutral net
- 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
- G08G1/0108—Measuring and analyzing of parameters relative to traffic conditions based on the source of data
- G08G1/0112—Measuring and analyzing of parameters relative to traffic conditions based on the source of data from the vehicle, e.g. floating car data [FCD]
-
- 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
- G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
-
- 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
- G08G1/0133—Traffic data processing for classifying traffic situation
Landscapes
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Traffic Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of road traffic condition Forecasting Methodology and system, the method comprising the steps of:After the traffic data that the vehicle-mounted GPS equipment of the vehicle travelled for each road to be analyzed, acquisition on the road is recorded, the road traffic condition of the road is predicted using the cyclic convolution neutral net trained;Wherein, cyclic convolution neutral net includes cyclic convolution layer, average pond layer, exits layer, full articulamentum and output layer successively, and traffic data includes multiple GPS measuring points, and each GPS measuring points include the point current speed, coordinate and timestamp.This method is predicted using cyclic convolution neutral net, and prediction accuracy is high, and stability is high, can be widely applied in intelligent city.
Description
Technical field
The present invention relates to intelligent city field, more particularly to a kind of road traffic condition Forecasting Methodology and system.
Background technology
Explanation of nouns:
Recognition with Recurrent Neural Network:English full name Recurrent Neural Networks, are abbreviated as RNNs.At present in crowd
Immense success and extensive use are achieved in many natural language processing fields.The specific form of expression is that network can be to letter before
Breath is remembered and applied in the calculating currently exported, i.e., the node between hidden layer is no longer connectionless but has connection,
And not only the output including input layer also includes the output of last moment hidden layer for the input of hidden layer.
Convolutional neural networks:English full name Convolutional neural networks, are abbreviated as CNNs.Convolution god
One through network is exactly the characteristics of important, by convolution algorithm, can strengthen original signal feature, and reduces noise, and
And using the principle of image local correlation, sub-sample is carried out to image, data processing amount can be reduced while retaining useful information.
In intelligent transportation system, road traffic condition prediction is one of ultimate challenge that current intelligent city faces.It is accurate
True traffic status prediction, is the basis of intelligent transportation system.Mapping service provider is flowed into usually using universal network traffic
Row road condition predicting, this method assesses application implementation based on traditional segmentation method and road traffic.This mode, although certain
The prediction of traffic behavior can be realized in degree, but its precision of prediction is general, it is hard to meet the demand for development of intelligent city.
The content of the invention
In order to solve above-mentioned technical problem, it is an object of the invention to provide a kind of road traffic condition Forecasting Methodology, sheet
The purpose of invention is to provide a kind of road traffic condition forecasting system.
The technical solution adopted for the present invention to solve the technical problems is:
A kind of road traffic condition Forecasting Methodology, including step:
For each road to be analyzed, the friendship that the vehicle-mounted GPS equipment of the vehicle travelled on the road is recorded is obtained
After logical data, the road traffic condition of the road is predicted using the cyclic convolution neutral net trained;
Wherein, the cyclic convolution neutral net includes cyclic convolution layer, average pond layer, exits layer, full connection successively
Layer and output layer, the traffic data include multiple GPS measuring points, and each GPS measuring points include the current speed of the point, coordinate
And timestamp.
Further, it is further comprising the steps of:
Data acquisition step:Obtain the traffic data that multiple vehicle-mounted GPS equipments are recorded;
Data prediction step:The traffic data of acquisition is carried out after data cleansing, it is based on flow similitude that data are clear
Traffic data after washing carries out Clustering;
Characteristic vector constitution step:For each road, the history based on public track distance metric and the road is handed over
Gating condition, builds the characteristic vector of the road;
Deep learning framework constitution step:Based on deep learning technology, cyclic convolution neutral net is built, to characteristic vector
Carry out further feature extraction;
Train Optimization Steps:Based on momentum stochastic gradient descent algorithm, optimization is trained to cyclic convolution neutral net.
Further, in the data prediction step, the step of traffic data of described pair of acquisition carries out data cleansing, its
Specially:
The traffic data of acquisition is carried out successively dealing of abnormal data, invalid data delete processing, data filling processing and
Data smoothing processing.
Further, in the data prediction step, it is described based on flow similitude by the traffic data after data cleansing
The step of carrying out Clustering, including:
Each traffic data is subjected to segmentation division according to the default cycle;
Based on euclidean metric, segmentation division result is clustered;
To each cluster, the representative track of the cluster is built as the pattern of the cluster, wherein, track is represented to belong to this
The length and the average value of angle of the whole orbit segment of cluster.
Further, the characteristic vector constitution step, it is specially:For each road to be analyzed, based on common rail
Mark distance metric obtains the representative track of the road and obtained represents first K nearest representative track work of trajectory distance with this
For K neighboring modes of the road, and then according to the historical traffic condition of the K neighboring modes He the road, build the road
Characteristic vector.
Further, the deep learning framework constitution step, including:
For each road, the left and right contextual information for obtaining its neighboring modes is calculated as training set, and by itself and this
The corresponding traffic of road is associated;
Build after cyclic convolution neutral net, the left and right contextual information of acquisition is input to the cyclic convolution of neutral net
Layer;
Output to cyclic convolution layer is carried out after linear transformation and the calculating of tanh activation primitives, and result of calculation is averaged
Value is calculated;
The average value obtained will be calculated as the input for exiting layer, the output of output layer is finally obtained.
Further, the training Optimization Steps, it is specially:
Based on momentum stochastic gradient descent algorithm, gradient decline meter is carried out to the training parameter of cyclic convolution neutral net
Calculate, maximize the log-likelihood of training parameter so that traffic of the output of cyclic convolution neutral net closest to road reality
Situation.
The present invention solves another technical scheme for being used of its technical problem:
A kind of road traffic condition forecasting system, including processor and storage device, the storage device is stored with a plurality of
Instruction, the instruction is loaded by processor and performs following steps:
For each road to be analyzed, the friendship that the vehicle-mounted GPS equipment of the vehicle travelled on the road is recorded is obtained
After logical data, the road traffic condition of the road is predicted using the cyclic convolution neutral net trained;
Wherein, the cyclic convolution neutral net includes cyclic convolution layer, average pond layer, exits layer, full connection successively
Layer and output layer, the traffic data include multiple GPS measuring points, and each GPS measuring points include the current speed of the point, coordinate
And timestamp.
Further, processor loading instruction also performs following steps:
Data acquisition step:Obtain the traffic data that multiple vehicle-mounted GPS equipments are recorded;
Data prediction step:The traffic data of acquisition is carried out after data cleansing, it is based on flow similitude that data are clear
Traffic data after washing carries out Clustering;
Characteristic vector constitution step:For each road, the history based on public track distance metric and the road is handed over
Gating condition, builds the characteristic vector of the road;
Deep learning framework constitution step:Based on deep learning technology, cyclic convolution neutral net is built, to characteristic vector
Carry out further feature extraction;
Train Optimization Steps:Based on momentum stochastic gradient descent algorithm, optimization is trained to cyclic convolution neutral net.
The beneficial effects of the invention are as follows:The road traffic condition Forecasting Methodology of the present invention, including step:For to be analyzed
Each road, is obtained after the traffic data that the vehicle-mounted GPS equipment of the vehicle travelled on the road is recorded, using what is trained
Cyclic convolution neutral net is predicted to the road traffic condition of the road;Wherein, cyclic convolution neutral net includes successively
Cyclic convolution layer, average pond layer, layer, full articulamentum and output layer are exited, traffic data includes multiple GPS measuring points, each
GPS measuring points include the point current speed, coordinate and timestamp.This method is carried out pre- using cyclic convolution neutral net
Survey, prediction accuracy is high, and stability is high.
The another of the present invention has an effect to be:The present invention a kind of road traffic condition forecasting system, including processor and
Storage device, the storage device is stored with a plurality of instruction, and the instruction is loaded by processor and performs following steps:For treating
Each road of analysis, is obtained after the traffic data that the vehicle-mounted GPS equipment of vehicle travelled on the road is recorded, using instruction
The cyclic convolution neutral net perfected is predicted to the road traffic condition of the road;Wherein, the cyclic convolution nerve net
Network includes cyclic convolution layer, average pond layer, exits layer, full articulamentum and output layer successively, and the traffic data includes multiple
GPS measuring points, each GPS measuring points include the point current speed, coordinate and timestamp.The system is using cyclic convolution god
It is predicted through network, prediction accuracy is high, and stability is high.
Brief description of the drawings
The invention will be further described with reference to the accompanying drawings and examples.
Fig. 1 is a road sample schematic diagram of the specific embodiment of the road traffic condition Forecasting Methodology of the present invention;
Fig. 2 is the structural representation of the cyclic convolution neutral net of the road traffic condition Forecasting Methodology use of the present invention;
Fig. 3 is the result schematic diagram that performance verification is predicted to the cyclic convolution neutral net of this method.
Embodiment
Below in conjunction with the accompanying drawings and specific embodiment is described further with regard to technical scheme.It should be appreciated that this
The specific embodiment of place description is not intended to limit the present invention only to explain the present invention.
The invention provides a kind of road traffic condition Forecasting Methodology, including step:
For each road to be analyzed, the friendship that the vehicle-mounted GPS equipment of the vehicle travelled on the road is recorded is obtained
After logical data, the road traffic condition of the road is predicted using the cyclic convolution neutral net trained;
Wherein, the cyclic convolution neutral net includes cyclic convolution layer, average pond layer, exits layer, full connection successively
Layer and output layer, the traffic data include multiple GPS measuring points, and each GPS measuring points include the current speed of the point, coordinate
And timestamp.
It is further used as preferred embodiment, it is further comprising the steps of:
Data acquisition step:Obtain the traffic data that multiple vehicle-mounted GPS equipments are recorded;
Data prediction step:The traffic data of acquisition is carried out after data cleansing, it is based on flow similitude that data are clear
Traffic data after washing carries out Clustering;
Characteristic vector constitution step:For each road, the history based on public track distance metric and the road is handed over
Gating condition, builds the characteristic vector of the road;
Deep learning framework constitution step:Based on deep learning technology, cyclic convolution neutral net is built, to characteristic vector
Carry out further feature extraction;
Train Optimization Steps:Based on momentum stochastic gradient descent algorithm, optimization is trained to cyclic convolution neutral net.
Specifically, during being predicted, it is also desirable to carry out data prediction step and spy to the traffic data of acquisition
Vectorial constitution step is levied, then is input in cyclic convolution neutral net and is predicted, finally, uses Softmax functions to carry out accurate
True rate prediction.
It is further used as preferred embodiment, in the data prediction step, the traffic data of described pair of acquisition enters
The step of row data cleansing, it is specially:
The traffic data of acquisition is carried out successively dealing of abnormal data, invalid data delete processing, data filling processing and
Data smoothing processing.
It is further used as preferred embodiment, in the data prediction step, the flow similitude that is based on is by number
The step of Clustering being carried out according to the traffic data after cleaning, including:
Each traffic data is subjected to segmentation division according to the default cycle;
Based on euclidean metric, segmentation division result is clustered;
To each cluster, the representative track of the cluster is built as the pattern of the cluster, wherein, track is represented to belong to this
The length and the average value of angle of the whole orbit segment of cluster.
It is further used as preferred embodiment, the characteristic vector constitution step, it is specially:For to be analyzed every
Individual road, based on public track distance metric obtain the representative track of the road and obtain with this represent trajectory distance it is nearest before
K representative track as the road K neighboring modes, and then according to the K neighboring modes and the historical traffic of the road
Condition, builds the characteristic vector of the road.The historical traffic condition of the road is specifically included:In t traffic on the day of last week,
Last week on the same day t+15 road r traffic, yesterday t traffic, yesterday the traffic shape in t+15 road r
Condition etc..
It is further used as preferred embodiment, the deep learning framework constitution step, including:
For each road, the left and right contextual information for obtaining its neighboring modes is calculated as training set, and by itself and this
The corresponding traffic of road is associated;
Build after cyclic convolution neutral net, the left and right contextual information of acquisition is input to the cyclic convolution of neutral net
Layer;
Output to cyclic convolution layer is carried out after linear transformation and the calculating of tanh activation primitives, and result of calculation is averaged
Value is calculated;
The average value obtained will be calculated as the input for exiting layer, the output of output layer is finally obtained.
It is further used as preferred embodiment, the training Optimization Steps, it is specially:
Based on momentum stochastic gradient descent algorithm, gradient decline meter is carried out to the training parameter of cyclic convolution neutral net
Calculate, maximize the log-likelihood of training parameter so that traffic of the output of cyclic convolution neutral net closest to road reality
Situation.
Present invention also offers a kind of road traffic condition forecasting system, including processor and storage device, the storage
Equipment is stored with a plurality of instruction, and the instruction is loaded by processor and performs following steps:
For each road to be analyzed, the friendship that the vehicle-mounted GPS equipment of the vehicle travelled on the road is recorded is obtained
After logical data, the road traffic condition of the road is predicted using the cyclic convolution neutral net trained;
Wherein, the cyclic convolution neutral net includes cyclic convolution layer, average pond layer, exits layer, full connection successively
Layer and output layer, the traffic data include multiple GPS measuring points, and each GPS measuring points include the current speed of the point, coordinate
And timestamp.
It is further used as preferred embodiment, processor loading instruction also performs following steps:
Data acquisition step:Obtain the traffic data that multiple vehicle-mounted GPS equipments are recorded;
Data prediction step:The traffic data of acquisition is carried out after data cleansing, it is based on flow similitude that data are clear
Traffic data after washing carries out Clustering;
Characteristic vector constitution step:For each road, the history based on public track distance metric and the road is handed over
Gating condition, builds the characteristic vector of the road;
Deep learning framework constitution step:Based on deep learning technology, cyclic convolution neutral net is built, to characteristic vector
Carry out further feature extraction;
Train Optimization Steps:Based on momentum stochastic gradient descent algorithm, optimization is trained to cyclic convolution neutral net.
On the one hand this method, similar track data is combined, pre- for data as the pattern of each road
Processing, and employ the concept based on context to these patterns to construct characteristic vector as studying the defeated of framework in depth
Enter.On the other hand, a kind of deep learning framework is devised, Recognition with Recurrent Neural Network and convolutional neural networks are combined from road traffic
Condition predicting feature learning hiding information, further increases estimated performance, and then be easy to imminent traffic congestion
Administered, dredged in advance, and avoid imminent traffic congestion road.It is pre- that this method can accurately carry out traffic behavior
Survey, and the degree of accuracy is high.
In more detail, the invention mainly comprises following key step:
The first step, data acquisition step:The traffic data that multiple vehicle-mounted GPS equipments are recorded is obtained, traffic data includes
Multiple GPS measuring points, each GPS measuring points include the point current speed, coordinate and timestamp.Data are mainly derived from out
GPS device on hiring a car.
Second step, data prediction is carried out after data cleansing to the traffic data of acquisition, based on flow similitude by data
Traffic data after cleaning carries out Clustering.
Data cleansing is carried out by following four step first:
1. dealing of abnormal data:If the average speed of the speed of GPS measuring points GPS measuring points more all than contemporaneity is fast
More than 50%, then it represents that the data are abnormal data, using the average speed in contemporaneity replace it.
2. invalid data delete processing:If a certain traffic data did not record any data at intraday three hours,
Then delete the data of all day.
3. data filling is handled:If the loss of data in some timestamp, take before the timestamp with data afterwards
Average value carries out data filling processing as Filling power to the timestamp.
4. data smoothing processing:The average value that GPS on every three timestamps records spot speed is calculated, data smoothing is carried out.
Secondly, the remainder of track data is grouped according to flow similitude to obtain the pattern of every road.
One part is defined as a road by the application.Because a road has different traffics daily.Therefore, using cluster
Method is that each part in road generates a pattern in daily each period.The step of clustering method, is as described below:
1. each traffic data is carried out into segmentation division according to the default cycle, such as the default cycle is 5 minutes, according to
Traffic data is divided into one group of line segment by the interval of 5 minutes.
2. measure orbit segment close to each other according to certain distance is grouped into a cluster with this.In this step, make
The distance between two tracks are measured with euclidean metric.DefinitionIt is traffic data P1…nAnd Q1…nEvery
Individual sampling time t point is apart from sum, then,Euclid's degree between wherein 2 points
Amount is defined as:For this sorting procedure, using DBSCAN algorithms, because
It is to be based on Density Clustering for it, and allows cluster track to form any shape and size.
3. for each cluster, to each cluster, the representative track of the cluster is built as the pattern of the cluster, wherein,
The length and the average value of angle, the i.e. pattern for the whole orbit segment for representing track to belong to the cluster are represented within a certain period of time
Cluster feature including traffic, i.e. road conditions.
3rd step, constructs characteristic vector.In the road network of modern city, the traffic of a road is always by attached
The influence of other nearly roads.When nearby road traffic is busy, this road traffic is then relatively steady.Therefore, in predicted link
During traffic, not only need to consider the historical traffic condition of the road, also need to consider the traffic of its neighbouring road.Cause
This, will select one group of road to provide the contextual information on periphery with the road for traffic forecast to be carried out.
If using road as central point, can be used and be based on public trajectory distance degree derived from minimum boundary rectangle (MBRs)
Measure to capture the representative track of the road and representative track around it in a range of overall similarity.Make B first1
And B2It is pattern P respectively1And P2MBR.Apart from Dmin(B1,B2) represent B1And B2In minimum range between any pair point.
Then, according to Dmin(B1,B2) all patterns are ranked up, and select preceding K pattern road nearest with the road distance
K neighboring modes.
Then, related characteristic vector is built to the historical traffic condition of this road using these neighboring modes.Under
Table 1 is shown if setting K=5 and wanting the characteristic vector sample of prediction road traffic condition of 15 minutes after time t
This, the mark represents position of the feature in terms of contextual information." C " represents that the function includes information road in itself, " C's "
Different subscripts are used to distinguish different road informations.It note that due to obtaining real-time traffic very cost source, so in the application
Method in, r traffic conditions are not considered, r represents stretch, and r is dynamic, can be 100 meters, can be 300 meters and also may be used
To be 1 kilometer, r is one section 3 kilometers of road in the test data of the present embodiment." L " or " R " is represented on the left of road contextual information
Or the feature on right side, the information is depending on the geographical position residing for adjacent paths, and the subscript of " L " and " R " is for distinguishing different
Contextual information.Fig. 1 illustrates the situation of a road sample, and display where each row represents a road.If being with " AB "
Center point-rendering circle simultaneously sets K=5, then:" BH ", " BG " and " AE " can be considered as with " L " indicate feature, and " AC " and
" CD " can be considered as the feature indicated with " R ", because they are all in circle.But, although the part of " EF " is also in circle
Circle in, but apart from " AB " distance not in DminFirst 5.
Table 1:Characteristic vector sample
Mark | Feature |
C1 | In t traffic on the day of last week |
C2 | Last week on the same day t+15 r traffic |
C3 | Yesterday the traffic in t |
C4 | Yesterday the traffic in t+15 r |
C5 | Expression one day is the mark at public holiday/weekend |
C6 | One value, the working day on the day of expression |
L1 | The traffic of neighbor mode 1 at t+15 |
L2 | The traffic of neighboring modes 2 at t+15 |
L3 | The traffic of neighboring modes 3 at t+15 |
R1 | Traffic of the neighboring modes 4 at t+15 |
R2 | The traffic of neighbor mode 5 at t+15 |
It is worth noting that, the more information of this support carry out construction feature vector, such as the history of more neighbouring roads is handed over
Logical situation, but in view of calculating cost, the present embodiment preferably uses six flag signs " C " and K neighboring modes to build spy
Levy this vectorial work.
4th step, deep learning framework constitution step.In abovementioned steps, characteristic vector is based only on common locus data structure
Build.Accordingly, it would be desirable to the further feature of characteristic vector further be extracted using deep learning technology, so as to improve estimated performance.
WillWithIt is defined as road R or so contextual information.In the present embodiment,WithBe exceptWithExternal conversion corresponding neighbour's pattern traffic, be used as the value of the transportation condition directly used.R
It is also a carrier, wherein including feature road in itself.Calculated using following formulaWithWherein f is non-linear
Activation primitive, W(l)And W(r)It is the matrix that hidden layer (context) is converted to next hidden layer.
Fig. 2 illustrates the structural representation of the cyclic convolution neutral net of the present invention, for loop structure, first by fixed
Context is captured to Recognition with Recurrent Neural Network.By taking table 1 as an example,It is the contextual information of adjacent pattern, positioned at closest
The left side of road.WithIt is closely upper with the adjacent modes of the 3rd minimum distance with second on the left of road
Context information.Similarly,WithBe have on the right side of road first closely with the second adjacent modes closely
Contextual information.Based on above formula, context vector captures all left and right contextual informations.Secondly, definition represents x such as with R
Shown in following formula, x is all left context informationAll right context informationWith the company of R insertion characteristic information
Connect.Using this contextual information, compared with the traditional neural model using only stationary window, this Recognition with Recurrent Neural Network can be more preferable
Ground learns R hiding information:
In the scanning forward of road contextual information, loop structure can obtain all during road is scanned from left to rightAnd road scan from right to left in it is allDuring this, time complexity is O (n).Obtaining x
Afterwards, linear transformation is carried out to it and tanh activation primitives are calculated, and result of calculation y is sent to next layer, wherein y is bag
The output of matrix containing various information, also referred to as cyclic convolution layer, y formula is as follows:
Y=tanh (WX+b)
Wherein, the convolutional neural networks in the framework that the application is proposed are defined as representing road, therefore, from convolutional Neural
From the perspective of network, loop structure noted earlier is convolutional layer.Then an average pond layer is applied using following formula, with
This consolidates the feature of last layer learning and expression:
y*=average (y)
Wherein, average functions are first prime functions, using the average value of y each element as next layer of input value, with
Promotional features are represented, and reduce the over-fitting to training data by model.The application herein without using maximum pond, because
For in the case of only one of which convolutional layer, average pond is more suitable for capturing information.Pondization by the use of loop structure output as
Input, time complexity is O (n).Block mold is the series connection of loop structure and average pond layer, therefore entirety time complexity
Remain as O (n).
Add one to exit layer and full articulamentum to reduce overfitting, to create the group with feature and activation primitive
Close, be predicted so as to after by network.The last part of the application framework is output layer.With traditional neural network class
Seemingly, it is defined as y**=Wy*+b。
During checking, the output of Softmax function pair output layers can be used to calculate prediction probability, i.e. checking prediction is accurate
Rate.
5th step, trains Optimization Steps:Based on momentum stochastic gradient descent algorithm, cyclic convolution neutral net is instructed
Practice optimization:
If all parameter definitions that will be trained are θ, then θ logarithm is maximized using the training objective of network seemingly
So it is worth:
WhereinTraining set is represented, D represents training setIn training sample, classDIt is the correct of road traffic condition
Classification, training objective is optimized using momentum stochastic gradient descent algorithm.In each step, one example of random selection (D,
classD) and perform gradient decline processing:
Its ∝ is learning rate, it is contemplated that the affordable actual computing resource of server, is full articulamentum there is provided 100
Internal storage location is used as circulation layer and 256 hidden units.And carried out 10,000 training iteration.By adjusting several times, if
∝ initial values are put for 0.2, and carry out by γ=0.1 successively decreasing for every 1000 iteration.In addition, momentum stochastic gradient descent
Factor of momentum in algorithm is set to 0.9, and the rate of exiting is set to 0.2, and pond length is set to after 2, repetitive exercise, training
Collect the penalty values with checking collection substantially close to 0, it can be seen that, this method, prediction accuracy is high.
In addition, building after cyclic convolution neutral net, the training verification step of detail is included how input data
Choose a part to be trained as training set, the mode that another part as checking collection verify etc., with other nerve nets
Network training step is similar, and the present invention is repeated no more.
Road traffic condition prediction can provide suggestion to driver, be predicted the outcome so being necessary to ensure that close to reality
Road traffic condition.This method is verified using a data set, the data set includes 30, Beijing of in May, 2013 road
44580 track data samples of 30 roads in section.Training process, using 100 periods come training network, and is used
Softmax functions carry out final accuracy rate prediction.Finally, it is estimated using 10 times of cross validations.Finally, calculate and obtain
This method can reach nearly 90% precision of prediction, and 5% is improved than traditional convolutional neural networks.This is due to that convolutional layer can be with
The information of more distinguishing characteristic situations is selected by average pond and capture.
In addition, the ability of the seizure contextual information in order to verify this method, sets K=1,5,10,15,20 distinguish
Convolutional neural networks and this method are tested, test result is as shown in Figure 3, it is seen that, in all K values, this method is better than traditional
Convolutional neural networks, in addition, as K=10, two models all reach optimum performance, and the performance changed with K of this method
It is so obvious that change is not as convolutional neural networks, therefore, and this method will not extremely rely on K, job stability more preferably because following
Ring structure can retain longer contextual information and reduce noise.
Above is the preferable implementation to the present invention is illustrated, but the invention is not limited to the implementation
Example, those skilled in the art can also make a variety of equivalent variations or replace on the premise of without prejudice to spirit of the invention
Change, these equivalent modifications or replacement are all contained in the application claim limited range.
Claims (9)
1. a kind of road traffic condition Forecasting Methodology, it is characterised in that including step:
For each road to be analyzed, the traffic number that the vehicle-mounted GPS equipment of the vehicle travelled on the road is recorded is obtained
According to rear, the road traffic condition of the road is predicted using the cyclic convolution neutral net trained;
Wherein, the cyclic convolution neutral net include successively cyclic convolution layer, average pond layer, exit layer, full articulamentum and
Output layer, the traffic data includes multiple GPS measuring points, each GPS measuring points comprising the current speed of the point, coordinate and
Timestamp.
2. a kind of road traffic condition Forecasting Methodology according to claim 1, it is characterised in that further comprising the steps of:
Data acquisition step:Obtain the traffic data that multiple vehicle-mounted GPS equipments are recorded;
Data prediction step:The traffic data of acquisition is carried out after data cleansing, based on flow similitude by after data cleansing
Traffic data carry out Clustering;
Characteristic vector constitution step:For each road, the historical traffic bar based on public track distance metric and the road
Part, builds the characteristic vector of the road;
Deep learning framework constitution step:Based on deep learning technology, cyclic convolution neutral net is built, characteristic vector is carried out
Further feature is extracted;
Train Optimization Steps:Based on momentum stochastic gradient descent algorithm, optimization is trained to cyclic convolution neutral net.
3. a kind of road traffic condition Forecasting Methodology according to claim 2, it is characterised in that the data prediction step
In rapid, the step of traffic data of described pair of acquisition carries out data cleansing, it is specially:
Dealing of abnormal data, invalid data delete processing, data filling processing and data are carried out successively to the traffic data of acquisition
Smoothing processing.
4. a kind of road traffic condition Forecasting Methodology according to claim 2, it is characterised in that the data prediction step
It is described the step of the traffic data after data cleansing is carried out into Clustering based on flow similitude in rapid, including:
Each traffic data is subjected to segmentation division according to the default cycle;
Based on euclidean metric, segmentation division result is clustered;
To each cluster, the representative track for building the cluster is used as the pattern of the cluster.
5. a kind of road traffic condition Forecasting Methodology according to claim 2, it is characterised in that the characteristic vector construction
Step, it is specially:For each road to be analyzed, the representative track of the road is obtained simultaneously based on public track distance metric
Obtain and represent first K nearest representative track of trajectory distance as K neighboring modes of the road with this, and then according to the K
The historical traffic condition of individual neighboring modes and the road, builds the characteristic vector of the road.
6. a kind of road traffic condition Forecasting Methodology according to claim 2, it is characterised in that the deep learning framework
Constitution step, including:
For each road, the left and right contextual information for obtaining its neighboring modes is calculated as training set, and by itself and the road
Corresponding traffic is associated;
Build after cyclic convolution neutral net, the left and right contextual information of acquisition is input to the cyclic convolution layer of neutral net;
Output to cyclic convolution layer is carried out after linear transformation and the calculating of tanh activation primitives, and average value meter is carried out to result of calculation
Calculate;
The average value obtained will be calculated as the input for exiting layer, the output of output layer is finally obtained.
7. a kind of road traffic condition Forecasting Methodology according to claim 2, it is characterised in that the training optimization step
Suddenly, it is specially:
Based on momentum stochastic gradient descent algorithm, gradient descent algorithm is carried out to the training parameter of cyclic convolution neutral net, most
The log-likelihood of bigization training parameter so that traffic of the output of cyclic convolution neutral net closest to road reality.
8. a kind of road traffic condition forecasting system, it is characterised in that including processor and storage device, the storage device is deposited
A plurality of instruction is contained, the instruction is loaded by processor and performs following steps:
For each road to be analyzed, the traffic number that the vehicle-mounted GPS equipment of the vehicle travelled on the road is recorded is obtained
According to rear, the road traffic condition of the road is predicted using the cyclic convolution neutral net trained;
Wherein, the cyclic convolution neutral net include successively cyclic convolution layer, average pond layer, exit layer, full articulamentum and
Output layer, the traffic data includes multiple GPS measuring points, each GPS measuring points comprising the current speed of the point, coordinate and
Timestamp.
9. a kind of road traffic condition forecasting system according to claim 8, it is characterised in that processor loading instruction is also
Perform following steps:
Data acquisition step:Obtain the traffic data that multiple vehicle-mounted GPS equipments are recorded;
Data prediction step:The traffic data of acquisition is carried out after data cleansing, based on flow similitude by after data cleansing
Traffic data carry out Clustering;
Characteristic vector constitution step:For each road, the historical traffic bar based on public track distance metric and the road
Part, builds the characteristic vector of the road;
Deep learning framework constitution step:Based on deep learning technology, cyclic convolution neutral net is built, characteristic vector is carried out
Further feature is extracted;
Train Optimization Steps:Based on momentum stochastic gradient descent algorithm, optimization is trained to cyclic convolution neutral net.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710326733.8A CN107103754B (en) | 2017-05-10 | 2017-05-10 | Road traffic condition prediction method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710326733.8A CN107103754B (en) | 2017-05-10 | 2017-05-10 | Road traffic condition prediction method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107103754A true CN107103754A (en) | 2017-08-29 |
CN107103754B CN107103754B (en) | 2020-05-22 |
Family
ID=59669105
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710326733.8A Active CN107103754B (en) | 2017-05-10 | 2017-05-10 | Road traffic condition prediction method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107103754B (en) |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108205889A (en) * | 2017-12-29 | 2018-06-26 | 长春理工大学 | Freeway traffic flow Forecasting Methodology based on convolutional neural networks |
CN108648457A (en) * | 2018-06-28 | 2018-10-12 | 苏州大学 | A kind of method, apparatus and computer readable storage medium of prediction of speed |
CN108986453A (en) * | 2018-06-15 | 2018-12-11 | 华南师范大学 | A kind of traffic movement prediction method based on contextual information, system and device |
CN109068174A (en) * | 2018-09-12 | 2018-12-21 | 上海交通大学 | Video frame rate upconversion method and system based on cyclic convolution neural network |
CN109064748A (en) * | 2018-09-18 | 2018-12-21 | 浙江工业大学 | Traffic average speed prediction method based on temporal clustering analysis and variable convolution neural network |
CN109767622A (en) * | 2019-02-14 | 2019-05-17 | 重庆邮电大学 | The traffic flow forecasting method with Recognition with Recurrent Neural Network is divided towards bayonet similitude |
CN109887279A (en) * | 2019-02-26 | 2019-06-14 | 北京隆普智能科技有限公司 | A kind of traffic congestion prediction technique and system |
CN110232820A (en) * | 2019-05-20 | 2019-09-13 | 北京世纪高通科技有限公司 | A kind of method for building up and device of road condition predicting model |
CN110570658A (en) * | 2019-10-23 | 2019-12-13 | 江苏智通交通科技有限公司 | Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering |
WO2020024319A1 (en) * | 2018-08-01 | 2020-02-06 | 苏州大学张家港工业技术研究院 | Convolutional neural network based multi-point regression forecasting model for traffic flow forecasting |
CN110942637A (en) * | 2019-12-17 | 2020-03-31 | 浙江工业大学 | SCATS system road traffic flow prediction method based on airspace map convolutional neural network |
CN111524348A (en) * | 2020-04-14 | 2020-08-11 | 长安大学 | Long-short term traffic flow prediction model and method |
CN111566709A (en) * | 2017-11-30 | 2020-08-21 | 大众汽车股份公司 | System and method for predicting and maximizing traffic flow |
CN111986490A (en) * | 2020-09-18 | 2020-11-24 | 北京百度网讯科技有限公司 | Road condition prediction method and device, electronic equipment and storage medium |
WO2020244066A1 (en) * | 2019-06-04 | 2020-12-10 | 平安科技(深圳)有限公司 | Text classification method, apparatus, device, and storage medium |
CN112330953A (en) * | 2020-10-10 | 2021-02-05 | 杭州翔毅科技有限公司 | Traffic control method, device and equipment based on cloud data and storage medium |
CN113379754A (en) * | 2020-12-02 | 2021-09-10 | 哈尔滨理工大学 | Road center line extraction method based on vehicle-mounted GPS data and neural network |
CN113450562A (en) * | 2021-05-08 | 2021-09-28 | 浙江工业大学 | Road network traffic state discrimination method based on clustering and graph convolution network |
CN114764977A (en) * | 2022-04-06 | 2022-07-19 | 成都亿盟恒信科技有限公司 | Road congestion calculation method and system based on driving data |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP4016412A1 (en) * | 2020-12-18 | 2022-06-22 | Imec VZW | Traffic prediction |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104572892A (en) * | 2014-12-24 | 2015-04-29 | 中国科学院自动化研究所 | Text classification method based on cyclic convolution network |
CN106205126A (en) * | 2016-08-12 | 2016-12-07 | 北京航空航天大学 | Large-scale Traffic Network based on convolutional neural networks is blocked up Forecasting Methodology and device |
CN106250103A (en) * | 2016-08-04 | 2016-12-21 | 东南大学 | A kind of convolutional neural networks cyclic convolution calculates the system of data reusing |
CN106503055A (en) * | 2016-09-27 | 2017-03-15 | 天津大学 | A kind of generation method from structured text to iamge description |
CN106611162A (en) * | 2016-12-20 | 2017-05-03 | 西安电子科技大学 | Method for real-time detection of road vehicle based on deep learning SSD frame |
-
2017
- 2017-05-10 CN CN201710326733.8A patent/CN107103754B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104572892A (en) * | 2014-12-24 | 2015-04-29 | 中国科学院自动化研究所 | Text classification method based on cyclic convolution network |
CN106250103A (en) * | 2016-08-04 | 2016-12-21 | 东南大学 | A kind of convolutional neural networks cyclic convolution calculates the system of data reusing |
CN106205126A (en) * | 2016-08-12 | 2016-12-07 | 北京航空航天大学 | Large-scale Traffic Network based on convolutional neural networks is blocked up Forecasting Methodology and device |
CN106503055A (en) * | 2016-09-27 | 2017-03-15 | 天津大学 | A kind of generation method from structured text to iamge description |
CN106611162A (en) * | 2016-12-20 | 2017-05-03 | 西安电子科技大学 | Method for real-time detection of road vehicle based on deep learning SSD frame |
Non-Patent Citations (1)
Title |
---|
周记 等: "BFGS训练的深度循环卷积神经网络目标检测与分类", 《第35届中国控制会议论文集》 * |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111566709A (en) * | 2017-11-30 | 2020-08-21 | 大众汽车股份公司 | System and method for predicting and maximizing traffic flow |
CN108205889A (en) * | 2017-12-29 | 2018-06-26 | 长春理工大学 | Freeway traffic flow Forecasting Methodology based on convolutional neural networks |
CN108205889B (en) * | 2017-12-29 | 2021-04-27 | 长春理工大学 | Method for predicting highway traffic flow based on convolutional neural network |
CN108986453A (en) * | 2018-06-15 | 2018-12-11 | 华南师范大学 | A kind of traffic movement prediction method based on contextual information, system and device |
CN108648457A (en) * | 2018-06-28 | 2018-10-12 | 苏州大学 | A kind of method, apparatus and computer readable storage medium of prediction of speed |
WO2020024319A1 (en) * | 2018-08-01 | 2020-02-06 | 苏州大学张家港工业技术研究院 | Convolutional neural network based multi-point regression forecasting model for traffic flow forecasting |
CN109068174B (en) * | 2018-09-12 | 2019-12-27 | 上海交通大学 | Video frame rate up-conversion method and system based on cyclic convolution neural network |
CN109068174A (en) * | 2018-09-12 | 2018-12-21 | 上海交通大学 | Video frame rate upconversion method and system based on cyclic convolution neural network |
CN109064748A (en) * | 2018-09-18 | 2018-12-21 | 浙江工业大学 | Traffic average speed prediction method based on temporal clustering analysis and variable convolution neural network |
CN109767622A (en) * | 2019-02-14 | 2019-05-17 | 重庆邮电大学 | The traffic flow forecasting method with Recognition with Recurrent Neural Network is divided towards bayonet similitude |
CN109767622B (en) * | 2019-02-14 | 2021-12-17 | 重庆邮电大学 | Traffic flow prediction method for bayonet similarity division and recurrent neural network |
CN109887279A (en) * | 2019-02-26 | 2019-06-14 | 北京隆普智能科技有限公司 | A kind of traffic congestion prediction technique and system |
CN110232820A (en) * | 2019-05-20 | 2019-09-13 | 北京世纪高通科技有限公司 | A kind of method for building up and device of road condition predicting model |
WO2020244066A1 (en) * | 2019-06-04 | 2020-12-10 | 平安科技(深圳)有限公司 | Text classification method, apparatus, device, and storage medium |
CN110570658A (en) * | 2019-10-23 | 2019-12-13 | 江苏智通交通科技有限公司 | Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering |
CN110570658B (en) * | 2019-10-23 | 2022-02-01 | 江苏智通交通科技有限公司 | Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering |
CN110942637A (en) * | 2019-12-17 | 2020-03-31 | 浙江工业大学 | SCATS system road traffic flow prediction method based on airspace map convolutional neural network |
CN111524348A (en) * | 2020-04-14 | 2020-08-11 | 长安大学 | Long-short term traffic flow prediction model and method |
CN111986490A (en) * | 2020-09-18 | 2020-11-24 | 北京百度网讯科技有限公司 | Road condition prediction method and device, electronic equipment and storage medium |
CN112330953A (en) * | 2020-10-10 | 2021-02-05 | 杭州翔毅科技有限公司 | Traffic control method, device and equipment based on cloud data and storage medium |
CN113379754A (en) * | 2020-12-02 | 2021-09-10 | 哈尔滨理工大学 | Road center line extraction method based on vehicle-mounted GPS data and neural network |
CN113450562A (en) * | 2021-05-08 | 2021-09-28 | 浙江工业大学 | Road network traffic state discrimination method based on clustering and graph convolution network |
CN114764977A (en) * | 2022-04-06 | 2022-07-19 | 成都亿盟恒信科技有限公司 | Road congestion calculation method and system based on driving data |
Also Published As
Publication number | Publication date |
---|---|
CN107103754B (en) | 2020-05-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107103754A (en) | A kind of road traffic condition Forecasting Methodology and system | |
CN108197739B (en) | Urban rail transit passenger flow prediction method | |
CN110264709A (en) | The prediction technique of the magnitude of traffic flow of road based on figure convolutional network | |
CN111080029B (en) | Urban traffic road speed prediction method and system based on multi-path segment space-time correlation | |
CN109190507A (en) | A kind of passenger flow crowding calculation method and device based on rail transit train | |
CN112085163A (en) | Air quality prediction method based on attention enhancement graph convolutional neural network AGC and gated cyclic unit GRU | |
CN111275962B (en) | Vehicle track data aggregation effect prediction method and device | |
CN108986453A (en) | A kind of traffic movement prediction method based on contextual information, system and device | |
CN106874688A (en) | Intelligent lead compound based on convolutional neural networks finds method | |
CN112365708B (en) | Scenic spot traffic volume prediction model establishing and predicting method based on multi-graph convolution network | |
CN110826244B (en) | Conjugated gradient cellular automaton method for simulating influence of rail transit on urban growth | |
CN111737826B (en) | Rail transit automatic simulation modeling method and device based on reinforcement learning | |
CN108734225A (en) | A kind of transmission line construction subject image detection method based on deep learning | |
Zhang et al. | TSR-GAN: Generative adversarial networks for traffic state reconstruction with time space diagrams | |
CN111047078A (en) | Traffic characteristic prediction method, system and storage medium | |
CN115271373A (en) | Method and system for defining elastic development boundary of urban group | |
Dai et al. | Spatio-temporal deep learning framework for traffic speed forecasting in IoT | |
CN112598165A (en) | Private car data-based urban functional area transfer flow prediction method and device | |
CN112884014A (en) | Traffic speed short-time prediction method based on road section topological structure classification | |
CN110988263B (en) | Vehicle exhaust concentration estimation method based on improved Stacking model | |
CN112419711A (en) | Closed parking lot parking demand prediction method based on improved GMDH algorithm | |
CN117271899A (en) | Interest point recommendation method based on space-time perception | |
CN113158084B (en) | Method, device, computer equipment and storage medium for processing movement track data | |
Zheng et al. | A deep learning–based approach for moving vehicle counting and short-term traffic prediction from video images | |
CN108876144B (en) | Transformer substation pre-addressing method based on deep learning algorithm |
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 | ||
TR01 | Transfer of patent right |
Effective date of registration: 20200820 Address after: 518000 Room 408, Huatong business building, No. 3, Xili Guan Long Village, Nanshan District, Shenzhen, Guangdong Patentee after: SHENZHEN LINGHANGZHE AUTOMOBILE INTELLIGENT TECHNOLOGY DEVELOPMENT Co.,Ltd. Address before: 510631 School of computer science, South China Normal University, 55 West Zhongshan Road, Guangzhou, Guangdong, Tianhe District Patentee before: SOUTH CHINA NORMAL University |
|
TR01 | Transfer of patent right |