CN110503833A - A kind of Entrance ramp inter-linked controlling method based on depth residual error network model - Google Patents
A kind of Entrance ramp inter-linked controlling method based on depth residual error network model Download PDFInfo
- Publication number
- CN110503833A CN110503833A CN201910809931.9A CN201910809931A CN110503833A CN 110503833 A CN110503833 A CN 110503833A CN 201910809931 A CN201910809931 A CN 201910809931A CN 110503833 A CN110503833 A CN 110503833A
- Authority
- CN
- China
- Prior art keywords
- data
- model
- traffic flow
- prediction
- rate
- 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
- G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications
- G08G1/0145—Measuring and analyzing of parameters relative to traffic conditions for specific applications for active traffic flow control
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/07—Controlling traffic signals
- G08G1/075—Ramp control
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Traffic Control Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of Entrance ramp inter-linked controlling methods based on depth residual error network model, firstly, collecting traffic flow character historical data, pre-process laggard line number figure conversion；Secondly, input image data, establishes and trains the prediction model of traffic flow character value；Third collects arithmetic for real-time traffic flow characteristic, and number figure conversion input model after pretreatment exports trend figure in short-term, and convert using figure number anticipation trend figure switching to text data；4th, the text data after conversion carries out short-term prediction to road traffic features value using trained prediction model, carries out linkage control in advance to the vehicle flowrate for importing main line；Finally, carrying out simulation evaluation and the analysis of ALINEA algorithm using VB+VISSIM program, and issue traffic information.Control method of the present invention carries out number figure conversion after data prediction respectively, and number figure converts extractable two dimensional image more details feature, reduces model training and predicted time, improves precision of prediction and real time information processing speed.
Description
Technical field
The invention belongs to traffic datas to predict field, be related to a kind of prediction of short-term traffic flow characteristic and traffic flow modes
Division methods carry out number based on depth residual error network model it was predicted that and carrying out Entrance ramp linkage control using prediction data
Method.
Background technique
Road traffic flow characteristic mainly includes vehicle flowrate, flow speeds, vehicle density and vehicle occupancy rate.Road is handed over
Through-flow characteristic prediction can predict subsequent period data and divide traffic flow modes, be to carry out traffic administration and control
Prerequisite.Traffic administration and control based on forecasting traffic flow are not only convenient for traveler and formulate better plan of travel, also
Better administrative decision is made conducive to traffic management department.
In existing road traffic flow characteristic prediction technique, what use was most wide at present is shallow Model and timing mould
Type.Shallow Model cannot preferably excavate the information in traffic flow data, and temporal model only considered traffic flow in time
Feature and have ignored influence spatially.And depth residual error network can not only be based on data variation trend image extraction time
On feature, moreover it is possible to extract feature spatially, therefore the invention proposes a kind of traffic flow characters based on depth residual error network
Data predication method extracts the space-time characteristic in traffic flow character data variation tendency chart by convolution and carries out non-linear time
Return, the final prediction for realizing road traffic flow characteristic, and traffic administration and control are carried out based on prediction data.
With the rapid development of deep learning and artificial intelligence technology, the accuracy of road traffic flow prediction is increasingly mentioned
It is high.Road traffic flow prediction can make more reasonable traffic control strategy with additional transport administrative department, respond and hand over for car owner
Siphunculus control strategy provides reference, alleviates traffic congestion, reduces the waste of traffic resource.Road based on depth residual error network model
Road forecasting traffic flow is that intelligent transportation system provides the foundation data, and has pushed the development and application of intelligent transportation system.
Goal of the invention
In order to improve the deficiency of existing short-term traffic flow characteristic prediction and phase identification method precision, simultaneously
The short slab that conventional inlet ramp metering rate method does not have feed forward mechanism and forecasting mechanism is solved, the present invention provides a kind of based on deep
Spend the prediction of short-term traffic flow characteristic and the Entrance ramp inter-linked controlling method of residual error network model.
It realizes a kind of Entrance ramp inter-linked controlling method based on depth residual error network model, mainly comprises the steps that
Step 1: collect traffic flow character historical data, data prediction laggard line number figure conversion, convert data to
Time is the two dimensional image of sequence；
Step 2: the prediction model based on depth residual error network traffic flow characteristic value is established and trained to input image data,
Hyper parameter and network level are set, carry out arameter optimization by propagated forward and backpropagation, completes model training, and depth
Learn historical traffic stream characteristic value variation tendency；
Step 3: collecting arithmetic for real-time traffic flow characteristic, number figure conversion will be carried out by pretreated data, data are turned
It is changed to the two dimensional image using the time as sequence；Image data is inputted into trained depth residual error Network Prediction Model after conversion
Short-term prediction is carried out, exports trend figure in short-term, and convert using figure number and anticipation trend figure is switched into text data；
Step 4: the text data after conversion, using trained depth residual error Network Prediction Model to road traffic features
Value carries out short-term prediction, and the wagon flow state of prediction data and identification is input in On-ramp Control ALINEA algorithm, calculates
Next control period vehicle occupancy rate and maximum queuing vehicle number, carry out linkage control to the vehicle flowrate for importing main line in advance；
Step 5: carrying out simulation evaluation and the analysis of ALINEA algorithm using VB+VISSIM program, and issue traffic information.
Step 1 specifically comprises the following steps:
1.1 history data collection
Video is shot by unmanned plane and wagon detector detects two methods and collects history vehicle flowrate data, and unmanned plane is clapped
Section vehicle flowrate data and section wagon flow state change situation can be collected by taking the photograph video；Wagon detector passes through detection and processing
Collect the traffic flow character value in each lane in section；
Definition traffic flow character value is vehicle flowrate Q, averagely flow speeds v, vehicle density kmWith vehicle occupancy rate Rt, traffic
Flowing characteristic value can be by directly collecting or being calculated indirectly, and vehicle occupancy rate uses time occupancy for index, calculation formula
It is as follows:
T is the time span in each control period；tiFor i-th vehicle by the time shared by section, unit is the second；N is measurement
Pass through the vehicle number of section in time.
1.2 data prediction
Unmanned plane and the directly collected traffic flow character value of wagon detector be inevitably constantly present loss of data,
It the problems such as mistake, invalid, time drift, if problem data is directly fed back to depth residual error network model application, will cause
There is error in model prediction.Therefore, collected data must be pre-processed before the conversion of number figure, i.e. data cleansing (data
Clean), including disorder data recognition, abnormal data reparation and data de-noising and normalization；
1.2.1 disorder data recognition
The traffic flow abnormal data of acquisition mainly includes three kinds of situations:
(1) data such as the magnitude of traffic flow, speed and occupation rate have exceeded reasonable threshold range；
(2) relationship between the data such as the magnitude of traffic flow, speed and occupation rate does not meet traffic flow theory；
(3) there is missing in the data such as the magnitude of traffic flow, speed and occupation rate；
For above type of traffic flow data, data reparation can be directly deleted or carried out.
1.2.2 abnormal data reparation
Traffic flow off-note data mainly include two kinds of situations of error in data and shortage of data, for the data identified
Exceptional value is directly deleted；For missing data, counted using K minimum distance neighbour method (K-means clustering)
According to polishing, the K sample that distance has missing data sample nearest is first determined according to Euclidean distance or correlation analysis, then will
This K value is weighted and averaged the missing data to estimate the sample.
1.2.3 data de-noising
Traffic flow data usually contains relatively large number of Gaussian noise, certain in the case where collection period is shorter
The analysis and modeling of traffic data is influenced in degree, it is therefore necessary to carry out simple denoising for sampled data.；Number
Single exponential smoothing algorithm is used according to denoising, as far as possible on the basis of not increasing algorithm complexity, retains hand over to the greatest extent
The short term variations trend of through-flow data；Single Exponential Smoothing calculation formula is as follows:
In formula,It is respectively the smoothed data and real data at m moment with X (m), k is Smoothness Index, takes 0.1.
1.2.4 data normalization
When constructing neural network model to traffic flow data, it usually needs data are normalized, to avoid
The phenomenon that neuron is saturated, using Logistic/Softmax transform method, all pending datas are transformed into [0,
1] in section.
1.3 number figure conversions
Two dimensional image conversion will be carried out by sequence of the time by pretreated traffic flow data, and obtain different road conditions
Under the conditions of real-time time-traffic flow character data variation tendency chart, the data being disposed carry out preservation after with graphic form
Training and test of the continuous processing as depth residual error network model.
Step 2 specifically comprises the following steps:
The input of 2.1 data and initialization
Based on TensorFlow carry out image sample data input, carry out feature set selection, different types of sample choose,
Sample vector figure and characteristic pattern conversion, TFRecords sample data set generate, data set inputs 5 parts；TFRecords is
A kind of binary file format, committed memory is small, and convenient reproduction is mobile and stores, and does not need individual label file；
The training sample for reading TFRecords format uses efficient coding (One-Hot) mode according to sample label
Data are encoded；
In order to avoid there is over-fitting, and enhance the robustness of model, original image data can by image saturation,
The data enhancement methods such as contrast conversion, by changing the information such as location of pixels, obtain more while guaranteeing feature invariant
For sufficient sample data set.
The setting of 2.2 hyper parameters
It needs to carry out hyper parameter setting before the training of depth residual error network model, the main parameter that is arranged includes batch training size
(Batch-size), learning rate (Learning rate), weight attenuation rate (weight-decay-rate), optimizer selection
(optimizer) etc.；
Batch training sizes values determine the direction of decline, are inversely proportional with the size of data set；When data set is sufficiently large,
Reduction appropriate can reduce calculation amount；If data volume is smaller, and there are when noise data, it should biggish batch instruction be arranged
Practice sizes values to reduce the interference of noise data；
Learning rate determines the amplitude of right value update, and learning rate setting is conducive to the decline of model gradient in suitable range
To optimal value；One biggish initial learning rate setting is set first, with the increase of model the number of iterations, gradually adjust to
Minimum learning rate, to obtain faster training speed and model accuracy；
The case where will appear over-fitting in depth residual error network model training process, network weight is bigger, often corresponding
Over-fitting degree is higher, therefore uses weight attenuation rate that L is arranged2Regularization term parameter, main function are adjustment model complexities
Influence to loss function prevents model over-fitting；
Momentum optimizer is selected, which is mainly based upon the movement index weighted average of gradient, the network optimization
When loss function convergence rate faster, amplitude of fluctuation is smaller.
The setting of 2.3 network model levels
Depth residual error network model (ResNet) uses the small convolution kernel normal form of 3x3, is replaced with multiple small convolution kernels one big
Convolution kernel reduces model parameter, increases the quantity of nonlinear activation function, and model calculation amount is smaller, and identification error is more
It is low.Convolutional layer identical with output characteristic pattern size for input, number of filter is constant, when characteristic pattern size halves
When, number of filter doubles, and characteristic pattern pond step-length is 2, to keep the time complexity of each interlayer.
It only needs identification image change trend to predict in the present invention, is not required to that the excessive network number of plies is arranged, in order to avoid cause
The unnecessary wasting of resources.Network first tier is convolutional layer, is responsible for extracting low-level feature, and the second layer is maximum pond layer,
Estimation mean shift is reduced, picture texture information is retained；3-6 layers are convolutional layer, are responsible for extracting high-level feature；Layer 7 is
Average pond layer, inhibits estimated value variance to increase, and calculates the feature that convolutional layer extracts and is input to full articulamentum；Full articulamentum
It is responsible for the output of class probability.
Depth residual error network is made of one group of residual block, and each residual block includes the convolutional layer of several stackings, by modified line
Property unit (Relu) and batch normalization layer (BN) as convolutional layer it is attached, avoid gradient disappear or overflow happen.
Nonlinear characteristic is introduced into prototype network by Relu activation primitive, the input feature vector of model node is converted to defeated
Feature out, and it is transferred to next operating unit.Using Relu function, part output data is made to be zeroed, prototype network can be with
It is self-introduced into sparsity, piecewise linearity can effectively overcome the problems, such as gradient disappearance.
The optimal mapping is rewritten as H (x)=F (x)+x by depth residual error network model, and it is also equivalent to approach residual error function F (x)
In approaching optimal mapping H (x).Revised residual error mapping is easier to optimize than original optimal demapping.
By increasing by one " Shortcut Connections " Lai Shixian network residual error in feedforward network, trained
Low layer error can be propagated by shortcut upper layer in journey, reduce phenomena such as gradient caused by the number of plies disappears, in calculation amount
Increase it is less in the case where, improve model training precision.
Shortcut is skipped one or more layers with different step-lengths and is converged with main diameter, and structure output is represented by
ml+1=Re lu (ml+F(ml,wl))
In formula, mlAnd ml+1It is outputting and inputting for the 1st residual block respectively, Re lu () is the linear unit function of amendment, F
Indicate residual error mapping function, wlIt is the parameter of residual error unit.
If outputting and inputting dimension difference, need to increase linear projectionCarry out matching dimensionality size, after increasing linear projection
The formula is further converted to
After being normalized using batch, with Parameters variation in model depth acceleration or training process, input data still can
It is distributed in a standard section, gradient is avoided to disappear, acceleration model convergence reduces dependence of the model to initial network weight,
Batch normalization operation formula is expressed as
In formula, xkIt is the activity of the neuron, E [xk] indicate the x that a collection of training dataset obtainskAverage value,For every a collection of training data xkStandard deviation.
Meanwhile batch normalization operation introduce two can learning parameter (γ, β), be used to transformed activation reconstruct,
Restore the feature distribution that primitive network learns.This operation will not destroy the feature that the data learn in previous layer operation,
The ability to express of network will not be impacted.
2.4 propagated forward
Propagated forward can extract the semantic feature that the high-level feature of input picture is more abstracted, it is contemplated that training
Collect actual traffic flow data between otherness and further feature ability to express, to convolutional layer feature in the present invention into
Row extracts.In propagated forward training process, desired learning objective function, function setup should be first set are as follows:
Wherein x is input feature vector value,For model prediction probability of outcome, w and b are the parameter that model training obtains；
By the sparse extraction of the feature of multiple convolutional layers, carried out using sparse convolution feature of the mean value pondization operation to extraction
It calculates, the every batch of sample image of input is often converted into sparse features, into full articulamentum, calculates, obtain by logits
To the batch sample data for each type of [batch training size × classification number] class probability matrix；
It is operated through softmax, ensure that all outputs are positive value, all line number values of matrix are stretched to [0,1] section,
And any row probability is added and is equal to 1.Softmax operates the matrix stretched, and the maximum value of every row is that output probability is maximum
Value, the as prediction result of this training.
2.5 backpropagations and arameter optimization
In depth Remanent Model training process, convolutional layer extracts each batch sample data and successively calculates sparse features simultaneously
Corresponding parametric values are recorded, the sparse features that the bottom extracts are input to logits layers, calculate sample classification value；
Trained loss function is calculated as the cross entropy of sample actual types Yu model prediction result, every batch of sample every time
Training loss function calculate such as following formula
In formula, tkiBelong to the probability of classification i, y for sample kkiBelong to the model prediction probability of classification i for sample k；
By comparing really with model prediction and identification classification results, model loss function is calculated, models fitting misses
Poor backpropagation, each parameter constantly adjust during depth Remanent Model iteration, effectively increase the robustness of model,
Reduce the probability of happening of over-fitting；
Training set data is inputted, carries out arameter optimization, and select corresponding optimizer.Common mini-batch SGD instruction
Practice algorithm and easily fall into local optimum, and is influenced by learning rate.Therefore movement index of the selection based on gradient is average weighted
Momentum optimizer, is smoothed network parameter, can solve mini-batch SGD optimization algorithm and update amplitude pendulum
Excessive problem is moved, while accelerating the convergence rate of network；
If current iterative step is t, as follows based on Momentum optimization algorithm calculation formula:
vdw=β vdw+(1-β)dW
vdb=β vdb+(1-β)db
W=W- α vdw
B=b- α vdb
In above formula, vdwAnd vdbIt is that the gradient momentum β that loss function is accumulated in preceding t-1 wheel iterative process is respectively
One index value of gradient accumulation, is set as 0.9；DW and db obtained gradient when being loss function backpropagation respectively, W,
B is the more new formula of network weight vector sum bias vector, and α is the learning rate of network；
After arameter optimization, the final step of model training, input verifying collection data, test model performance, hand are carried out
The hyper parameters numerical value such as dynamic fine tuning learning rate.
After step 3 specifically comprises the following steps: the training of completion depth residual error Network Prediction Model, depth residual error can be based on
Network model carries out short-term prediction to road traffic features value；Video is shot by unmanned plane and two methods of wagon detector are received
Collect arithmetic for real-time traffic flow characteristic, and phase transition decision condition be set for the track of vehicle move in traffic flow: more than to
Determine in threshold time interval, is got lower than along the car speed of rail running or fast higher than the given threshold value of the phase transitions point
Degree；
Data entry time window is set, the real-time traffic characteristic value that input wagon detector is collected, depth is residual after input
Poor network model is by data prediction and is converted to time series variation figure, and exports subsequent period traffic characteristic prediction number
According to trend chart, predict that the two dimensional image of output can be converted to text data by several figures；
In conjunction with road traffic flow space-time characteristic, traffic flow is divided into free flow (F), synchronous stream (S) and Wide moving jam
(J) three-phase；Traffic flow phase change can be considered as one and freely flow to the phase transition process step by step that synchronous stream arrives wide movement obstruction again
(F →S→J)；Referring to three-phase traffic flow theory, in conjunction with China's road status, each phase transitions point is set by threshold value of speed.
The model of design of the invention can also be according to the speed threshold of setting other than carrying out short-term prediction to traffic characteristic value
Value divides each phase interval automatically.Be conducive to execute the road control in step 4 to the division of prediction data phase interval.
Short-term prediction is carried out to road traffic features value described in step 4, short-term prediction includes that flow speeds, wagon flow are close
The prediction data of degree, vehicle occupancy rate.
On-ramp Control is to be widely used and a kind of effective alleviation road congested traffic control mode, is based on
The On-ramp Control strategy of ALINEA is simple, efficient and easy to implement.With ALENIA algorithm, control red light can be passed through
Duration adjusts clearance vehicle number per minute to control the regulation rate of ring road, to achieve the purpose that control Entrance ramp flow.
When traditional ALENIA algorithm carries out being lined up control, according to the vehicle occupancy rate of main line downstream road section and a upper period
Enter circle conciliation rate to calculate current period enters circle conciliation rate, does not have feed forward mechanism and forecasting mechanism.And the present invention is by traffic
Feature prediction data substitutes into algorithm, take short-term prediction time window as the control period of ALENIA algorithm, can be realized more accurate
Road control in advance, compensate for the deficiency of traditional algorithm.Adjusting efficiency is not only increased, and effectively reduces congestion hair
Raw probability, while the variation after capable of controlling road feeds back to the control of model realization re-optimization, improves degree of regulation.
Using three upstream section speed, downstream lane occupancy ratio and main line downstream flow setup measures On-ramp Controls
State.We provide that upstream section speed mutually divides unanimously with upper section traffic flow, i.e., by three-phase traffic flow theory and entrance circle
Channel control method links, one of the index by the division of traffic flow phase as On-ramp Control.
On-ramp Control method described in step 4, specifically comprises the following steps:
Before ALINEA algorithm, downstream vehicle occupation rate O in the k-1 control period is definedout(k-1) by wagon detector
Acquisition, k control downstream vehicle occupation rate O in the periodout(k) and Entrance ramp vehicle arriving rate d (k) is by depth residual error network mould
Type prediction；
R (k) is by O in the k-1 periodout(k-1) data are calculated, then by r (k) and Oout(k) the k+1 period can be obtained
Metering rate predicted value；
ALINEA algorithm long green light time is fixed, and the time interval by adjusting adjacent green light starting in per minute leads remittance
The vehicle of line carries out flow control；
Within a control period, regulation rate calculation formula is
In formula, r (k+1) is the metering rate that kth+1 controls period calculating；R (k) is the ring road in the kth control period
Regulation rate r (k) is calculated by vehicle occupancy rate measured data in the k-1 control period and is obtained, and regulation rate is green in a control period
Lamp duration, unit s；KRParameter adjusts external disturbance fixed in feedback control；It is the expectation occupation rate in main line downstream；
Oout(k) be kth control the period in main line downstream vehicle occupancy rate predicted value.
Simulation evaluation described in step 5 and analysis are specifically developed program using VB+VISSIM 4.3COM, are based on
ALINEA algorithm realizes ramp metering rate；Load road net model is emulated in a program, and the simulation run time takes 3600s, corresponding
Practical peak one hour, in simulation process, each control period will return to simulation status data, including the green light end time,
Split and occupation rate；
If letter controls the period as t, unit s, program obtains occupying for data detector return at the t-1 moment of each cycle
Rate, detector returns to an occupation rate at interval of t-2, and calculates next cycle according to adjustment factor and best occupation rate
Split；
When split is more than or equal to 0.8, ring road letter, which is controlled, continues green light；Less than 0.2, ring road letter, which is controlled, continues red light；Between
Split between 0.8 and 0.2 carries out fixed cycle control, and the long green light time of ramp metering rate is calculated according to the split after optimization.
The present invention is directed to predict traffic characteristic value, identifies terrain vehicle stream mode and use On-ramp Control method to alleviate
Traffic congestion, therefore need to evaluate On-ramp Control effect.It is right since embedded wagon detector process is complex
Pavement destruction is larger, before control is adjusted to road in the present invention, should be evaluated by simulation software its effect.
Depth residual error network robustness is good, and gradient disappearance problem is effectively relieved.The present invention is based on depth residual error network models
Entrance ramp inter-linked controlling method, by collecting traffic flow character historical data and real time data, after data prediction respectively
Number figure conversion is carried out, number figure converts extractable two dimensional image more details feature, reduces model training and predicted time, improves
Precision of prediction and real time information processing speed.It is special that the space-time in traffic flow character data variation tendency chart is also extracted by convolution
Nonlinear regression is levied and carries out, the final prediction for realizing road traffic flow characteristic, and traffic pipe is carried out based on prediction data
Reason and control, provide the foundation data, and pushed the development and application of intelligent transportation system for intelligent transportation system.
Detailed description of the invention
Fig. 1 is control method overview flow chart of the present invention；
Fig. 2 is depth residual error network model of the present invention training flow chart；
Fig. 3 is traffic flow character value variation tendency and prediction error analysis figure in embodiment.
Specific embodiment
The content of present invention is further described below with reference to embodiment and attached drawing, but is not limitation of the invention.
Referring to Fig.1-3, the Entrance ramp inter-linked controlling method based on depth residual error network model, includes the following steps:
1. data collection and processing
1.1 data collections and screening
Traffic flow data sampling method mainly includes that UAV Video shooting and wagon detector are collected, by wagon detector
The traffic flow character data of acquisition are set as training set for model training and parameter optimization, and unmanned plane shoots video data conduct
Verifying collection is used for the manual tuning of model hyper parameter.
The high-incidence section of one congestion of Nanjing is selected, which has ring road remittance, and the same day is working day, and state of weather is good.Root
According to wagon detector real-time data collection, the data of our a length of 350s when intercepting wherein, which reflects traffic flow congestion
Forming process, have apparent variation characteristic.
1.2 data prediction
The data of interception are pre-processed, deletes and repairs beyond reasonable threshold range, do not meet traffic flow theory
With the traffic flow character data of missing, noise reduction is carried out to data using single exponential smoothing algorithm, retains traffic flow character data
Short term variations trend and data are normalized using Logistic/Softmax transform method.It is pretreated
Each traffic flow character data are as shown in the table:
1 arithmetic for real-time traffic flow characteristic table of table
1.3 number figure conversions
After data prediction, data text messages are converted to using the time as " when m- vehicle number " of sequence variation
Trend image, " when m- flow speeds " variation tendency image, " when m- vehicle density " variation tendency image and " when m- vehicle
Occupation rate " variation tendency image sets four groups of image variation tendency time windows as 50s, i.e., according in the current 50s period
Traffic flow character data predict the traffic flow character data variation trend in next 50s.
Number figure conversion finishes, and starts to train depth residual error network traffic flow characteristic prediction model.
2. the training of depth residual error network traffic flow characteristic prediction model
The setting of 2.1 hyper parameters
It needs to carry out hyper parameter setting before the training of depth residual error network model, the main parameter that is arranged includes batch training size
(Batch-size), learning rate (Learning rate), weight attenuation rate (weight-decay-rate), optimizer selection
(optimizer) etc..Specific hyper parameter setting is as shown in the table.
The setting of 2 depth residual error network hyper parameter of table
When model training, reduces the convergence rate that Batch is brought and promote effect far below property caused by introducing much noise
Can fall, and GPU can play more preferably performance to the Batch of 2 power, therefore the Batch value used is 256.
The biggish initial learning rate setting of setting one, with the increase of model the number of iterations, gradually adjusts to most primary school
Habit rate, to obtain faster training speed and model accuracy.Using Momentum optimizer, shifting of the optimizer based on gradient
Dynamic exponent-weighted average, faster, amplitude of fluctuation is smaller for loss function convergence rate when the network optimization.0.0001 weight is set
Attenuation rate adjusts influence of the model complexity to loss function, prevents model over-fitting.
The input of 2.2 data and initialization
2.2.1 training set and test set divide
The means and type that the division of training set, verifying collection and test set is acquired based on data.
The data volume that wagon detector is collected is big, can be used as training set for depth Remanent Model arameter optimization, parameter can
It is updated by gradient decline, function to achieve the objective minimizes.
The data volume of unmanned plane acquisition is less, and has referential, therefore the data set is surpassed as verifying collection for model
Parameter manually adjusts, implementation model re-optimization.
Data acquisition system of the test set as test model predictablity rate is acquired in real time by wagon detector, and processing is simultaneously
It uploads, the predictable subsequent period data out of depth residual error network model.
2.2.2 data conversion and enhancing
Image sample data input is carried out based on TensorFlow, feature set is chosen, different types of sample is chosen, sample
Polar plot and characteristic pattern conversion, TFRecords sample data set generate and data set inputs.
The training sample for reading TFRecords format uses efficient coding (One-Hot) mode according to sample label
Data are encoded, and original image data is carried out the data such as image saturation, contrast conversion to enhance.
The configuration of 2.3 depth residual error network levels
Based on training mission and complexity of the invention, the depth residual error network model that depth is 18 layers is set, is not being accounted for
Precision of prediction is improved while with a large amount of training resources, following table is each level and its feature.
The configuration of 3 depth residual error network level of table
Model uses the small convolution kernel normal form of 3x3, replaces a big convolution kernel with multiple small convolution kernels, reduces model ginseng
Number, increases the quantity of nonlinear activation function.Convolutional layer identical with output characteristic pattern size for input, filtering
Device number is constant, and when characteristic pattern size halves, number of filter is doubled, and characteristic pattern pond step-length is 2, to keep each interlayer
Time complexity.
Depth residual error network is made of one group of residual block, and each residual block includes the convolutional layer of several stackings, by modified line
Property unit (Relu) and batch normalization layer (BN) as convolutional layer it is attached, avoid gradient disappear or overflow happen.
The optimal mapping is rewritten as H (x)=F (x)+x by depth residual error network model, and it is also equivalent to approach residual error function F (x)
In approaching optimal mapping H (x).Revised residual error mapping is easier to optimize than original optimal demapping.By in feedforward network
Middle " Shortcut Connections " Lai Shixian network residual error for increasing by one.Shortcut with different step-lengths skip one or
Multiple layers converge with main diameter, and structure output is represented by
ml+1=Re lu (ml+F(ml,wl))
In formula, mlAnd ml+1It is outputting and inputting for the 1st residual block respectively, Re lu () is the linear unit function of amendment, F
Indicate residual error mapping function, wlIt is the parameter of residual error unit.
If outputting and inputting dimension difference, need to increase linear projectionCarry out matching dimensionality size, after increasing linear projection
The formula is further converted to
2.4 propagated forward
Propagated forward can extract the semantic feature that the high-level feature of input picture is more abstracted, it is contemplated that training
Collect actual traffic flow data between otherness and further feature ability to express, to convolutional layer feature in the present invention into
Row extracts.In propagated forward training process, desired learning objective function, function setup should be first set are as follows:
Wherein x is input feature vector data,For model prediction probability of outcome, w and b are the parameter that model training obtains.
By the sparse extraction of the feature of multiple convolutional layers, carried out using sparse convolution feature of the mean value pondization operation to extraction
It calculates, the every batch of sample image of input is often converted into sparse features, into full articulamentum, calculates, obtain by logits
To the batch sample data for each type of [batch training size × classification number] class probability matrix, then pass through
Softmax operation ensure that all outputs are positive value, all line number values of matrix be stretched to [0,1] section, and any row
Probability, which is added, is equal to 1.The matrix stretched is operated through softmax, the maximum value of every row is the maximum value of output probability, as
The prediction result of this training.
2.5 backpropagations and arameter optimization
In depth Remanent Model training process, convolutional layer extracts each batch sample data and successively calculates sparse features simultaneously
Corresponding parametric values are recorded, the sparse features extracted from the bottom are input to logits layers, calculate sample classification value.
Trained loss function is calculated as the cross entropy of sample actual types Yu model prediction result, every batch of sample every time
Training loss function calculate such as following formula:
In formula, tkiBelong to the probability of classification i, y for sample kkiBelong to the model prediction probability of classification i for sample k.
By comparing really with model prediction and identification classification results, model loss function is calculated, models fitting misses
Poor backpropagation, each parameter constantly adjust during depth Remanent Model iteration, effectively increase the robustness of model,
Reduce the probability of happening of over-fitting.
Training set data is inputted, carries out arameter optimization, and select corresponding optimizer.Common mini-batch SGD instruction
Practice algorithm and easily fall into local optimum, and is influenced by learning rate.Therefore movement index of the selection based on gradient is average weighted
Momentum optimizer, is smoothed network parameter, can solve mini-batch SGD optimization algorithm and update amplitude pendulum
Excessive problem is moved, while accelerating the convergence rate of network.
If current iterative step is t, as follows based on Momentum optimization algorithm calculation formula:
vdw=β vdw+(1-β)dW
vdb=β vdb+(1-β)db
W=W- α vdw
B=b- α vdb
In above formula, vdwAnd vdbIt is that the gradient momentum β that loss function is accumulated in preceding t-1 wheel iterative process is respectively
One index value of gradient accumulation, is set as 0.9；DW and db obtained gradient when being loss function backpropagation respectively, W,
B is the more new formula of network weight vector sum bias vector, and α is the learning rate of network；
After arameter optimization, the final step of model training, input verifying collection data, test model performance, hand are carried out
The hyper parameters numerical value such as the learning rate being arranged in dynamic fine tuning table 2.
3. subsequent period traffic characteristic value is predicted
After each arameter optimization, road traffic flow characteristic is become using the depth residual error network model that training is completed
Change trend carries out short-term prediction.Input data time point, then output data time point was from 51s to 401s from 1s to 351s.
The two-dimentional variation tendency image that prediction obtains is again converted to text data through number figure conversion.Flow speeds and vehicle
Current density prediction data is mutually divided for three-phase traffic flow stream, and vehicle arriving rate and vehicle occupancy rate data are used for Entrance ramp
Control.Prediction data is as shown in the table:
4 predicting traffic flow characteristic table of table
According to three-phase traffic flow theory, decision condition, that is, threshold velocity is arranged for phase transition in combining road actual road conditions.
Identify corresponding with calibration traffic flow phase interval according to the threshold speed of setting, the division of phase interval be also in step 4 into
Line entry ramp metering rate provides decision condition.Specific threshold velocity and interval division are given in Table 5:
5 threshold velocity interval division table of table
When i.e. flow speeds are less than or equal to 50km/h, free flow is changed into synchronous stream, and speed is less than synchronous stream when 22km/h
It is changed into Wide moving jam；When flow speeds are greater than 25km/h, Wide moving jam is changed into synchronous stream, and speed is greater than 70km/h
When, synchronous circulation becomes free flow.
4 ramp metering rates and information are issued
4.1 ring road inter-linked controlling methods
Using the threshold speed being arranged in previous step as the judgement item for determining that open ring road, ramp metering and ring road are closed
One of part.Traffic flow free flow, synchronous stream in wide movement obstruction mutually conversion and control the opening and closing of ring road, road
When road is in synchronous stream and Wide moving jam state, corresponding On-ramp Control method is taken.
Using three upstream section speed, downstream lane occupancy ratio and main line downstream flow setup measures On-ramp Controls
State.We manage using the threshold speed for dividing traffic flow phase as one of index of On-ramp Control in conjunction with three-phase traffic flow
By with two kinds of theories of On-ramp Control method, realize road linkage control, specific On-ramp Control state see the table below.
6 On-ramp Control condition table of table
When carrying out ramp metering, we use ALINEA algorithm, are improved in the present invention the algorithm, utilize prediction
Next control period vehicle occupancy calculate metering rate so that ALINEA algorithm is had feed forward mechanism and forecasting mechanism.
Regulating cycle and short-term prediction window are taken identical value i.e. 50s by us.This assignment mode can be improved depth residual error mould
The compatible degree of type, three-phase traffic flow and ALINEA algorithm three, becoming three kinds of algorithms progress Entrance ramp linkage controls can
Energy.
If current period is k-1 period, Oout(k-1) etc. data are acquired in real time by wagon detector, OoutIt (k) is next week
Phase k vehicle occupancy rate predicted value, r (k) is by O in the k-1 periodout(k-1) data are calculated, then by r (k) and Oout(k) may be used
Obtain the metering rate predicted value in k+1 period.ALINEA algorithm long green light time is fixed, by adjusting adjacent green light in per minute
The time interval of starting carries out flow control to the vehicle for importing main line.
ALINEA algorithm is as follows:
Before calculating metering rate, occupation rate, regulating cycle and K need to it is expected to downstreamRThree parameters are demarcated.According to
Correlative study experience, it is proposed that expectation occupation rate is set as 0.3, KR70 are set as to obtain Optimal Control effect.Vehicle occupies
Rate Oout(k-1)=0.37, Oout(k)=0.44.
It is 20s that period metering rate r (k) value, which is calculated, by k-1 period vehicle collector measured data, brings improvement into
ALINEA algorithm calculates to obtain r (the k+1)=≈ of 20-70 × 0.14 10s.Green light total duration is 10s, signal lamp i.e. in regulating cycle
It must be switched by traffic lights and vehicle is controlled, each scintillation time of green light is fixed as 2s, remaining 40s in regulating cycle
Red light is distributed to, red light flashes 5 times altogether, and each duration is 8s.
The publication of 4.2 information
There are mainly three types of modes for information publication, the first is entered by ring road message signal lamp using algorithm calculated result
Mouth ramp metering rate, control effect are best；Other two kinds are issued as driver's trip decision-making by congestion information and provide reference.
Ring road signal lamp.Using ring road message signal lamp, vehicle entrance ramp is controlled.The green light period allows vehicle from ring road
Drive into major trunk roads；Red signal interval, vehicle must wait for parking in ring road, not allow to drive into major trunk roads.
Section upstream LED display board.The networking LED display board of upstream setting may be notified that section driver's congestion ahead, guide
Driver selects other sections or replacement mode of transportation trip to avoid congestion.
Mobile device.Road conditions publication is carried out by mobile devices such as broadcast, navigation softwares, section congestion is carried out to driver and is mentioned
Show, provides reference for driver's trip decision-making.
4.3 error analysis
The dispersion degree and actual error size of prediction data are able to reflect using tri- indexs of RMSE, MAE and MAPE, it can
Measure the prediction effect of prediction model.
The formula of three evaluation indexes is as follows:
In formula, xiIndicate the i-th moment actual traffic flow data, x 'iIndicate the pre- of the traffic flow of the i-th moment model output
Measured data, N indicate the length of Traffic Flow Time Series to be assessed.
It is as shown in the table that three index sizes are calculated in the present embodiment:
Table 7 predicts error analysis table
By the calculating to each index, it is not difficult to find out that, the dispersion degree of this embodiment prediction data is smaller, data prediction
Accuracy be up to 97.06%.Traffic flow character value variation tendency and prediction error analysis are as shown in figure 3, explanation utilizes depth
Residual error network model carries out data and predicts that complexity is low, and accuracy is high, and prediction numerical value change is relatively stable.
4.4 emulation and effect assessment
Since embedded wagon detector process is complex, road pavement destruction is larger, and road is adjusted in the present invention
Before control, its performance should be assessed by simulation software.Most using downstream Main Line Speed maximum, minimum value and flow
Greatly, minimum value, downstream Main Line Speed transverse direction, longitudinal fluctuation, downstream main line flow six parameters pair of lateral, longitudinal fluctuation
Road control effect is evaluated.Evaluation index uses sequence difference absolute mean and sequence criteria deviation, respectively characterize data sequence
Arrange the amplitude of variation and the speed (or frequency) of data sequence variation, i.e. transversal wave movement and longitudinal fluctuation, evaluation index meter
It is as follows to calculate formula:
In formula, xiFor i-th of value in data, Δ xi=xi-xi-1；For data mean value.Evaluation of simulation result such as table 8, table 9
It is shown.
8 downstream Main Line Speed Simulation Evaluation result of table
9 downstream main line Traffic simulation assessment result of table
As can be seen from Table 8, under ramp metering rate state, main line velocity of downstream, which has, to be obviously improved.In addition, with general
ALINEA algorithm is compared, and after the improved ring road inter-linked controlling method of the present invention, downstream Main Line Speed minimum value has significantly
Promotion.
After ring road linkage control, longitudinal fluctuation of main line velocity of downstream and flow, when not controlling with Traditional control
Speed and longitudinal fluctuation of flow are enhanced, and illustrate the traffic condition in main line downstream to the ramp metering rate strategy
Regulation it is quite sensitive.By regulation, main line downstream is transitioned into synchronous stream mode via blocked state；It is linked using ring road
After control, the variation of main line downstream flow transversal wave movement less, shows the control strategy in the normal trip for not influencing vehicle
In the case of traffic congestion more efficiently alleviated using control strategy, improve the traffic condition in main line downstream.
Claims (6)
1. a kind of Entrance ramp inter-linked controlling method based on depth residual error network model, which comprises the steps of:
Step 1: collecting traffic flow character historical data, the laggard line number figure conversion of data prediction was converted data to the time
For the two dimensional image of sequence；
Step 2: the prediction model based on depth residual error network traffic flow characteristic value, setting are established and trained to input image data
Hyper parameter and network level carry out arameter optimization by propagated forward and backpropagation, complete model training, and deep learning is gone through
History traffic flow character value variation tendency；
Step 3: collecting arithmetic for real-time traffic flow characteristic, number figure conversion will be carried out by pretreated data, is converted data to
Using the time as the two dimensional image of sequence；After conversion by image data input trained depth residual error Network Prediction Model carry out it is short
When predict, output trend figure in short-term, and converted using figure number and anticipation trend figure is switched into text data；
Step 4: the text data after conversion, using trained depth residual error Network Prediction Model to road traffic features value into
The wagon flow state of prediction data and identification is input in On-ramp Control ALINEA algorithm by row short-term prediction, is calculated next
Period vehicle occupancy rate and maximum queuing vehicle number are controlled, linkage control is carried out in advance to the vehicle flowrate for importing main line；
Step 5: carrying out simulation evaluation and the analysis of ALINEA algorithm using VB+VISSIM program, and issue traffic information.
2. the Entrance ramp inter-linked controlling method according to claim 1 based on depth residual error network model, feature exist
In step 1 specifically comprises the following steps:
1.1 history data collection
Video is shot by unmanned plane and two methods of wagon detector collect history vehicle flowrate data, and unmanned plane shoots video can
To collect section vehicle flowrate data and section wagon flow state change situation；Wagon detector collects section by detection and processing
The traffic flow character value in each lane；
Definition traffic flow character value is vehicle flowrate Q, averagely flow speeds v, vehicle density kmWith vehicle occupancy rate Rt, traffic flow spy
Value indicative can be by directly collecting or being calculated indirectly, and vehicle occupancy rate uses time occupancy for index, and calculation formula is as follows:
T is the time span in each control period；tiIt is i-th vehicle by the time shared by section, unit is the second；N is minute
The interior vehicle number by section；
1.2 data prediction
Collected data must be pre-processed before number figure conversion, including disorder data recognition, abnormal data reparation and data
Denoising and normalization；
1.2.1 disorder data recognition
The traffic flow abnormal data of acquisition mainly includes three kinds of situations:
(1) magnitude of traffic flow, speed and occupation rate data have exceeded reasonable threshold range；
(2) relationship between the magnitude of traffic flow, speed and occupation rate data does not meet traffic flow theory；
(3) there is missing in the magnitude of traffic flow, speed and occupation rate data；
For above type of traffic flow data, data reparation can be directly deleted or carried out；
1.2.2 abnormal data reparation
Traffic flow off-note data mainly include two kinds of situations of error in data and shortage of data, for the data exception identified
Value, is directly deleted；For missing data, Data-parallel language is carried out using K minimum distance neighbour's method, first according to Euclidean distance or phase
Analysis is closed to determine the distance K sample nearest with missing data sample, then this K value is weighted and averaged to estimate the sample
Missing data；
1.2.3 data de-noising
Denoising uses single exponential smoothing algorithm, and calculation formula is as follows:
In formula,It is respectively the smoothed data and real data at m moment with X (m), k is Smoothness Index, takes 0.1；
1.2.4 data normalization
Using Logistic/Softmax transform method, all pending datas are transformed into [0,1] section；
1.3 number figure conversions
Two dimensional image conversion will be carried out by sequence of the time by pretreated traffic flow data, and obtain different road condition conditions
Lower real-time time-traffic flow character data variation tendency chart, the data being disposed are saved with graphic form through subsequent place
Manage the training and test as depth residual error network model.
3. the Entrance ramp inter-linked controlling method according to claim 1 based on depth residual error network model, feature exist
In step 2 specifically comprises the following steps:
The input of 2.1 data and initialization
Image sample data input is carried out based on TensorFlow, carries out feature set selection, different types of sample is chosen, sample
Polar plot and characteristic pattern conversion, TFRecords sample data set generate, data set inputs；
The training sample for reading TFRecords format uses an efficient coding One-Hot mode by data according to sample label
It is encoded；
The setting of 2.2 hyper parameters
It needs to carry out hyper parameter setting before the training of depth residual error network model, the main parameter that is arranged includes batch training size, study
Rate, weight attenuation rate, optimizer selection；
Learning rate setting is conducive to model gradient in suitable range and drops to optimal value, and biggish initial is arranged first
The setting of habit rate, with the increase of model the number of iterations, gradually adjusts to minimum learning rate, to obtain faster training speed and mould
Type precision；L is arranged using weight attenuation rate2Regularization term parameter adjusts influence of the model complexity to loss function, prevents
Model over-fitting；Momentum optimizer is selected, the movement index weighted average based on gradient improves loss function convergence speed
Degree；
The setting of 2.3 network model levels
Network model first layer is convolutional layer, is responsible for extracting low-level feature；
The second layer is maximum pond layer, reduces estimation mean shift, retains picture texture information；
3-6 layers are convolutional layer, are responsible for extracting high-level feature；
Layer 7 is average pond layer, calculates the feature that convolutional layer extracts and is input to full articulamentum；
Full articulamentum is responsible for the output of class probability；
2.4 propagated forward
In propagated forward training process, desired learning objective function, function setup should be first set are as follows:
Wherein x is input feature vector value,For model prediction probability of outcome, w and b are the parameter that model training obtains；Pass through multiple volumes
The sparse extraction of the feature of lamination calculates the sparse convolution feature of extraction using the operation of mean value pondization, the every batch of of input
Sample image is converted into sparse features, into full articulamentum, calculates by logits, obtains the batch sample data for every
[batch training size × classification number] class probability matrix of seed type；
Softmax operation ensure that all outputs are positive value, all line number values of matrix is stretched to [0,1] section, and any
Row probability, which is added, is equal to the matrix that 1, softmax operation stretched, and the maximum value of every row is the maximum value of output probability, as originally
The prediction result of secondary training；
2.5 backpropagations and arameter optimization
In depth Remanent Model training process, convolutional layer, which extracts each batch sample data, successively to be calculated sparse features and records
Corresponding parametric values, the sparse features extracted from the bottom, are input to logits layers, calculate sample classification value；
Trained loss function is calculated as the cross entropy of sample actual types Yu model prediction result, the instruction of every batch of sample every time
Practice loss function and calculates such as following formula
In formula, tkiBelong to the probability of classification i, y for sample kkiBelong to the model prediction probability of classification i for sample k；It is true by comparing
Real and model prediction and identification classification results, are calculated model loss function, models fitting error back propagation；
Training set data is inputted, arameter optimization is carried out, selects the average weighted Momentum optimization of the movement index based on gradient
Device is smoothed network parameter；
If current iterative step is t, as follows based on Momentum optimization algorithm calculation formula:
vdw=β vdw+(1-β)dW
vdb=β vdb+(1-β)db
W=W- α vdw
B=b- α vdb
In above formula, vdwAnd vdbThe gradient momentum β that be loss function respectively accumulate in preceding t-1 wheel iterative process is that gradient is tired
A long-pending index value, is set as 0.9；DW and db obtained gradient when being loss function backpropagation respectively, W, b are networks
The more new formula of weight vectors and bias vector, α are the learning rates of network；After arameter optimization, model training is carried out most
Latter step, input verifying collection data, test model performance, manual fine-tuning hyper parameter numerical value.
4. the Entrance ramp inter-linked controlling method according to claim 1 based on depth residual error network model, feature exist
In step 3 specifically comprises the following steps:
Video is shot by unmanned plane and two methods of wagon detector collect arithmetic for real-time traffic flow characteristic, and in traffic flow
Phase transition decision condition is arranged in mobile track of vehicle: being more than in given threshold value time interval, along the vehicle of rail running
Speed get lower than or higher than the phase transitions point given threshold value speed；
Set data entry time window, the real-time traffic characteristic value that input wagon detector is collected, depth residual error net after input
Network model is by data prediction and is converted to time series variation figure, and exports the variation of subsequent period traffic characteristic prediction data
Tendency chart predicts that the two dimensional image of output is converted to text data by figure number；
Traffic flow is divided into free flow referring to the three-phase traffic flow theory that Kona proposes in conjunction with road traffic flow space-time characteristic
F, synchronous stream S and Wide moving jam J three-phase；
Traffic flow phase change is considered as one and freely flows to the phase transition process step by step that synchronous stream arrives wide movement obstruction again: F → S → J；
In conjunction with China's road status, each phase transitions point is set by threshold value of speed.
5. the Entrance ramp inter-linked controlling method according to claim 1 based on depth residual error network model, feature exist
In: short-term prediction is carried out to road traffic features value described in step 4, short-term prediction includes flow speeds, vehicle density, vehicle
The prediction data of occupation rate；
The Entrance ramp inter-linked controlling method, specifically comprises the following steps:
Before ALINEA algorithm, main line downstream vehicle occupation rate O in the k-1 control period is definedout(k-1) by wagon detector
Acquisition, k control downstream vehicle occupation rate O in the periodout(k) and Entrance ramp vehicle arriving rate d (k) is by depth residual error network mould
Type prediction；
R (k) is by O in the k-1 periodout(k-1) data are calculated, then by r (k) and Oout(k) ring road in k+1 period can be obtained
Regulation rate predicted value；
ALINEA algorithm long green light time is fixed, by adjusting the time interval of adjacent green light starting in per minute to remittance main line
Vehicle carries out flow control；
Within a control period, regulation rate calculation formula is
In formula, r (k+1) is the metering rate that kth+1 controls period calculating；R (k) is the ramp metering in the kth control period
Rate r (k) is calculated by vehicle occupancy rate measured data in the k-1 control period and is obtained, when regulation rate is green light in a control period
It is long, unit s；KRParameter adjusts external disturbance fixed in feedback control；It is the expectation occupation rate in main line downstream；Oout(k)
It is the vehicle occupancy rate predicted value in main line downstream in the kth control period.
6. the Entrance ramp inter-linked controlling method according to claim 1 based on depth residual error network model, feature exist
In, simulation evaluation described in step 5 and analysis, specifically
Program is developed using 4.3 COM of VB+VISSIM, ramp metering rate is realized based on ALINEA algorithm；Road network is loaded in a program
Model is emulated, and the simulation run time takes 3600s, and corresponding practical peak one hour, in simulation process, each control period will
Simulation status data, including green light end time, split and occupation rate can be returned；
If letter controls the period as t, unit s, program obtains the occupation rate of data detector return at the t-1 moment of each cycle,
Detector returns to an occupation rate at interval of t-2, and the green letter of next cycle is calculated according to adjustment factor and best occupation rate
Than；
When split is more than or equal to 0.8, ring road letter, which is controlled, continues green light；Less than 0.2, ring road letter, which is controlled, continues red light；Between 0.8 with
Split between 0.2 carries out fixed cycle control, and the signal lamp duration of ramp metering rate is calculated according to the split after optimization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910809931.9A CN110503833B (en) | 2019-08-29 | 2019-08-29 | Entrance ramp linkage control method based on depth residual error network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910809931.9A CN110503833B (en) | 2019-08-29 | 2019-08-29 | Entrance ramp linkage control method based on depth residual error network model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110503833A true CN110503833A (en) | 2019-11-26 |
| CN110503833B CN110503833B (en) | 2021-06-08 |
Family
ID=68590510
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910809931.9A Active CN110503833B (en) | 2019-08-29 | 2019-08-29 | Entrance ramp linkage control method based on depth residual error network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110503833B (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111260922A (en) * | 2020-01-20 | 2020-06-09 | 浙江工业大学 | Ramp control method based on congestion situation classification |
| CN111476191A (en) * | 2020-04-15 | 2020-07-31 | 陈建 | Artificial intelligent image processing method based on intelligent traffic and big data cloud server |
| CN111785088A (en) * | 2020-06-23 | 2020-10-16 | 大连理工大学 | Double-layer collaborative optimization method for merging network vehicle ramps |
| CN111949703A (en) * | 2020-07-09 | 2020-11-17 | 广东工业大学 | Unmanned aerial vehicle deployment and flight trajectory optimization method and system for intelligent transportation |
| CN112148730A (en) * | 2020-06-30 | 2020-12-29 | 网络通信与安全紫金山实验室 | Method for extracting product data characteristics in batches by using matrix generalized inverse |
| CN112669594A (en) * | 2020-12-11 | 2021-04-16 | 国汽(北京)智能网联汽车研究院有限公司 | Method, device, equipment and storage medium for predicting traffic road conditions |
| CN112734703A (en) * | 2020-12-28 | 2021-04-30 | 佛山市南海区广工大数控装备协同创新研究院 | PCB defect optimization method by utilizing AI cloud collaborative detection |
| CN113125992A (en) * | 2021-04-23 | 2021-07-16 | 合肥工业大学 | NPC three-level inverter fault diagnosis method and system based on DBN |
| CN113221442A (en) * | 2020-12-24 | 2021-08-06 | 山东鲁能软件技术有限公司 | Construction method and device of health assessment model of power plant equipment |
| CN113256973A (en) * | 2021-05-11 | 2021-08-13 | 青岛海信网络科技股份有限公司 | Peak start time prediction method, device, equipment and medium |
| CN113379156A (en) * | 2021-06-30 | 2021-09-10 | 南方科技大学 | Speed prediction method, device, equipment and storage medium |
| CN114141029A (en) * | 2021-11-25 | 2022-03-04 | 东南大学 | Ramp control method based on offline reinforcement learning and macroscopic model |
| CN114501010A (en) * | 2020-10-28 | 2022-05-13 | Oppo广东移动通信有限公司 | Image encoding method, image decoding method and related device |
| CN115083038A (en) * | 2022-06-29 | 2022-09-20 | 南京理工大学泰州科技学院 | Device and method for detecting vehicle driving abnormity |
| CN115544901A (en) * | 2022-11-28 | 2022-12-30 | 西南石油大学 | Intelligent thick oil working condition fault identification method for small sample |
| CN115662114A (en) * | 2022-10-08 | 2023-01-31 | 广州玩鑫信息科技有限公司 | Intelligent traffic system for relieving congestion based on big data and operation method thereof |
| CN117807378A (en) * | 2023-12-01 | 2024-04-02 | 太极计算机股份有限公司 | Intelligent wind power data restoration method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180089994A1 (en) * | 2016-09-27 | 2018-03-29 | International Business Machines Corporation | Predictive traffic management using virtual lanes |
| CN109035779A (en) * | 2018-08-30 | 2018-12-18 | 南京邮电大学 | Freeway traffic flow prediction technique based on DenseNet |
| CN109816983A (en) * | 2019-02-26 | 2019-05-28 | 昆明理工大学 | A kind of short-term traffic flow forecast method based on depth residual error network |
| CN109903557A (en) * | 2019-03-04 | 2019-06-18 | 南京邮电大学 | Based on the freeway traffic flow prediction technique for improving independent loops neural network |
-
2019
- 2019-08-29 CN CN201910809931.9A patent/CN110503833B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180089994A1 (en) * | 2016-09-27 | 2018-03-29 | International Business Machines Corporation | Predictive traffic management using virtual lanes |
| CN109035779A (en) * | 2018-08-30 | 2018-12-18 | 南京邮电大学 | Freeway traffic flow prediction technique based on DenseNet |
| CN109816983A (en) * | 2019-02-26 | 2019-05-28 | 昆明理工大学 | A kind of short-term traffic flow forecast method based on depth residual error network |
| CN109903557A (en) * | 2019-03-04 | 2019-06-18 | 南京邮电大学 | Based on the freeway traffic flow prediction technique for improving independent loops neural network |
Non-Patent Citations (1)
| Title |
|---|
| 方传武: "城市快速路入口匝道流量控制研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》 * |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111260922B (en) * | 2020-01-20 | 2021-01-29 | 浙江工业大学 | Ramp control method based on congestion situation classification |
| CN111260922A (en) * | 2020-01-20 | 2020-06-09 | 浙江工业大学 | Ramp control method based on congestion situation classification |
| CN111476191A (en) * | 2020-04-15 | 2020-07-31 | 陈建 | Artificial intelligent image processing method based on intelligent traffic and big data cloud server |
| CN111785088A (en) * | 2020-06-23 | 2020-10-16 | 大连理工大学 | Double-layer collaborative optimization method for merging network vehicle ramps |
| CN112148730A (en) * | 2020-06-30 | 2020-12-29 | 网络通信与安全紫金山实验室 | Method for extracting product data characteristics in batches by using matrix generalized inverse |
| CN111949703B (en) * | 2020-07-09 | 2023-06-06 | 广东工业大学 | Unmanned aerial vehicle deployment and flight trajectory optimization method and system for intelligent traffic |
| CN111949703A (en) * | 2020-07-09 | 2020-11-17 | 广东工业大学 | Unmanned aerial vehicle deployment and flight trajectory optimization method and system for intelligent transportation |
| CN114501010A (en) * | 2020-10-28 | 2022-05-13 | Oppo广东移动通信有限公司 | Image encoding method, image decoding method and related device |
| CN114501010B (en) * | 2020-10-28 | 2023-06-06 | Oppo广东移动通信有限公司 | Image encoding method, image decoding method and related devices |
| CN112669594A (en) * | 2020-12-11 | 2021-04-16 | 国汽(北京)智能网联汽车研究院有限公司 | Method, device, equipment and storage medium for predicting traffic road conditions |
| CN113221442A (en) * | 2020-12-24 | 2021-08-06 | 山东鲁能软件技术有限公司 | Construction method and device of health assessment model of power plant equipment |
| CN113221442B (en) * | 2020-12-24 | 2022-08-30 | 山东鲁能软件技术有限公司 | Method and device for constructing health assessment model of power plant equipment |
| CN112734703A (en) * | 2020-12-28 | 2021-04-30 | 佛山市南海区广工大数控装备协同创新研究院 | PCB defect optimization method by utilizing AI cloud collaborative detection |
| CN113125992A (en) * | 2021-04-23 | 2021-07-16 | 合肥工业大学 | NPC three-level inverter fault diagnosis method and system based on DBN |
| CN113125992B (en) * | 2021-04-23 | 2022-07-19 | 合肥工业大学 | NPC three-level inverter fault diagnosis method and system based on DBN |
| CN113256973A (en) * | 2021-05-11 | 2021-08-13 | 青岛海信网络科技股份有限公司 | Peak start time prediction method, device, equipment and medium |
| CN113379156A (en) * | 2021-06-30 | 2021-09-10 | 南方科技大学 | Speed prediction method, device, equipment and storage medium |
| CN114141029A (en) * | 2021-11-25 | 2022-03-04 | 东南大学 | Ramp control method based on offline reinforcement learning and macroscopic model |
| CN114141029B (en) * | 2021-11-25 | 2022-11-18 | 东南大学 | Ramp control method based on offline reinforcement learning and macroscopic model |
| CN115083038A (en) * | 2022-06-29 | 2022-09-20 | 南京理工大学泰州科技学院 | Device and method for detecting vehicle driving abnormity |
| CN115083038B (en) * | 2022-06-29 | 2024-02-27 | 南京理工大学泰州科技学院 | Device and method for detecting abnormal running of vehicle |
| CN115662114A (en) * | 2022-10-08 | 2023-01-31 | 广州玩鑫信息科技有限公司 | Intelligent traffic system for relieving congestion based on big data and operation method thereof |
| CN115662114B (en) * | 2022-10-08 | 2024-07-05 | 北京中软政通信息技术有限公司 | Intelligent traffic system for relieving congestion based on big data and operation method thereof |
| CN115544901A (en) * | 2022-11-28 | 2022-12-30 | 西南石油大学 | Intelligent thick oil working condition fault identification method for small sample |
| CN115544901B (en) * | 2022-11-28 | 2023-02-21 | 西南石油大学 | Intelligent thick oil working condition fault identification method for small sample |
| CN117807378A (en) * | 2023-12-01 | 2024-04-02 | 太极计算机股份有限公司 | Intelligent wind power data restoration method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110503833B (en) | 2021-06-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110503833A (en) | A kind of Entrance ramp inter-linked controlling method based on depth residual error network model | |
| CN108197739B (en) | Urban rail transit passenger flow prediction method | |
| CN111210633B (en) | Short-term traffic flow prediction method based on deep learning | |
| CN109754597A (en) | A kind of urban road area congestion regulating strategy recommender system and method | |
| CN109785618B (en) | Short-term traffic flow prediction method based on combinational logic | |
| CN110299011A (en) | A kind of traffic flow forecasting method of the highway arbitrary cross-section based on charge data | |
| CN109902880A (en) | A kind of city stream of people's prediction technique generating confrontation network based on Seq2Seq | |
| CN110555990A (en) | effective parking space-time resource prediction method based on LSTM neural network | |
| CN107704970A (en) | A kind of Demand-side load forecasting method based on Spark | |
| CN110310479A (en) | A kind of Forecast of Urban Traffic Flow forecasting system and method | |
| CN113362598A (en) | Traffic flow prediction method for expressway service area | |
| CN107862877A (en) | A kind of urban traffic signal fuzzy control method | |
| CN114187766B (en) | Road service level evaluation method based on saturation rate | |
| CN111242395A (en) | Method and device for constructing prediction model for OD (origin-destination) data | |
| CN112906945A (en) | Traffic flow prediction method, system and computer readable storage medium | |
| CN113112823A (en) | Urban road network traffic signal control method based on MPC | |
| CN114565187A (en) | Traffic network data prediction method based on graph space-time self-coding network | |
| CN114582131B (en) | Monitoring method and system based on ramp intelligent flow control algorithm | |
| CN113688200B (en) | Decision tree-based special population action track collection method and system | |
| Zhang et al. | Improved genetic algorithm optimized LSTM model and its application in short-term traffic flow prediction | |
| CN107274086A (en) | A kind of gridding governance information approach based on hidden Markov model | |
| CN112927507B (en) | Traffic flow prediction method based on LSTM-Attention | |
| Ye et al. | Demand forecasting of online car‐hailing by exhaustively capturing the temporal dependency with TCN and Attention approaches | |
| Zhang et al. | Short-term Traffic Flow Prediction With Residual Graph Attention Network. | |
| Li et al. | Cycle-based signal timing with traffic flow prediction for dynamic environment |
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 |