CN107967532B - Urban traffic flow prediction method fusing regional vitality - Google Patents

Urban traffic flow prediction method fusing regional vitality Download PDF

Info

Publication number
CN107967532B
CN107967532B CN201711052176.1A CN201711052176A CN107967532B CN 107967532 B CN107967532 B CN 107967532B CN 201711052176 A CN201711052176 A CN 201711052176A CN 107967532 B CN107967532 B CN 107967532B
Authority
CN
China
Prior art keywords
region
vitality
traffic flow
regional
model
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.)
Active
Application number
CN201711052176.1A
Other languages
Chinese (zh)
Other versions
CN107967532A (en
Inventor
范晓亮
郑传潘
陈龙彪
王程
温程璐
李军
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xiamen University
Original Assignee
Xiamen University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xiamen University filed Critical Xiamen University
Priority to CN201711052176.1A priority Critical patent/CN107967532B/en
Publication of CN107967532A publication Critical patent/CN107967532A/en
Application granted granted Critical
Publication of CN107967532B publication Critical patent/CN107967532B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
    • G06V20/625—License plates

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Tourism & Hospitality (AREA)
  • Health & Medical Sciences (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Marketing (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Development Economics (AREA)
  • Biomedical Technology (AREA)
  • General Business, Economics & Management (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Analytical Chemistry (AREA)
  • Game Theory and Decision Science (AREA)
  • Educational Administration (AREA)
  • Primary Health Care (AREA)
  • Multimedia (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention discloses an urban traffic flow prediction method fusing regional vitality, which comprises the following steps: s1, carrying out region division on the urban road network, and calculating the traffic flow of each region; s2, designing a regional activity model: learning dynamic changes of vitality of each region in a city by using the distribution of urban interest points, holidays and weather information and a 3D convolutional neural network (3D CNN); s3, designing a flow prediction model: fusing the vitality of the region and the traffic flow, and performing flow prediction by using a convolution length-time memory network (ConvLSTM); and S4, simultaneously training the region vitality model and the flow prediction model according to the historical data, and predicting the traffic flow of each region in real time by using the trained models. According to the method, the regional vitality is fused, the influence of the driving force behind the crowd activities and external factors is considered, and high prediction accuracy can be obtained.

Description

Urban traffic flow prediction method fusing regional vitality
Technical Field
The invention relates to the field of cross technology application of deep learning and traffic flow prediction, in particular to an urban traffic flow prediction method fusing regional vitality.
Background
With the acceleration of the urbanization process, a large number of people are rushed into cities, and the problem of hidden traffic safety hazards is increasingly highlighted while the economy is prosperous. If a holiday comes, a large number of people gather and serious traffic jam occurs near a shopping mall, so that great potential safety hazard is brought. Therefore, the prediction of the traffic flow has an extremely important meaning for city safety, and is a common concern all over the world. If the traffic flow of each area in the city can be accurately predicted, the traffic management department can dredge the traffic in time, and the public can selectively bypass the congested area, so that the potential safety hazard is reduced.
In view of the importance of the traffic flow prediction problem, much research work has been done on this. Earlier prediction models were: autoregressive model (AR), historical average model (HA), autoregressive moving average model (ARIMA), and the like. These models mainly study the spatiotemporal regularity of traffic flow changes on the basis of mathematical statistics and focus on solving the traffic flow prediction problem for a road segment or several road segments. With the development of machine learning, especially deep learning techniques, a large number of models based on artificial intelligence have appeared in recent years, such as a BP neural network model, a stacked self-encoding model (SAE), a convolutional neural network model (CNN), a recurrent neural network model (RNN), and the like. The models can obtain better prediction accuracy by means of strong learning capability of machine learning.
The defects of the prior method or the invention are as follows: 1) the existing method mostly only focuses on traffic flow prediction of one road section or a plurality of road sections, and lacks of measurement on crowd gathering degree of each region of a city; 2) most of the existing methods only concern traffic flow data, and the flow data lack semantic information and cannot reflect the driving force of crowd movement; 3) the existing method does not well deal with the influence of the change of external factors on the activity rule of people, for example, the influence of heavy rain on commuting behavior is small, but the influence can have a great influence on tourism.
Disclosure of Invention
In view of the above, the present invention provides a method for predicting urban traffic flow by merging region vitality, which applies deep learning to urban region traffic flow prediction, utilizes the spatio-temporal feature learning capability of a 3D convolutional neural network to mine the dynamic change feature of region vitality from data of interest points, time, weather, etc., merges region vitality and traffic flow, and utilizes the powerful spatio-temporal sequence prediction capability of a convolutional memory network to obtain a prediction result. The method integrates the vitality of the areas, well combines the internal regularity of crowd movement and the influence characteristics of external factors, realizes the flow prediction of each area of the city, and provides a new thought for the traffic flow prediction problem.
Firstly, grid area division is carried out on an urban road network according to the longitude and latitude, and the traffic flow of each area is calculated by utilizing the data of a license plate recognition device; then, designing a regional activity model and a flow prediction model by using a deep neural network; and finally, training a model by using historical data and predicting the urban traffic flow in real time by using the trained model.
The method comprises the following specific steps:
s1, dividing the urban road network into M × N grid areas according to the longitude and latitude, and calculating the traffic flow in each area according to the data recorded by the license plate recognition equipment;
s2, designing a regional activity model: learning dynamic changes of vitality of each region in a city by using the distribution of urban interest points, holidays and weather information and a 3D convolutional neural network;
s3, designing a flow prediction model: fusing the activity of the region and the traffic flow, and performing flow prediction by using a convolution duration memory network;
and S4, simultaneously training the region vitality model and the flow prediction model according to the historical data, and then predicting the traffic flow of each region in real time by using the trained models.
Further, step S1 specifically includes:
s11, dividing the urban road network into M × N grid areas according to the longitude and latitude, so that the area (M, N) represents the grid area of the M-th row and the N-th column;
s12, extracting recording data < d, v, tau > of the license plate recognition equipment, wherein d is the number of the license plate recognition equipment, v is the number of the shot license plate, and tau represents recording time;
s13, mapping the license plate recognition devices into grid areas according to the longitude and latitude, and calculating the number of vehicles recorded by the license plate recognition devices in each area in each time period as a traffic flow:
Figure GDA0002452805400000021
Figure GDA0002452805400000022
represents the traffic flow of the region (m, n) during the time period t;
s14, normalizing the traffic flow in each area:
Figure GDA0002452805400000023
wherein x0For original traffic flow, xminAnd xmaxRespectively representing the minimum value and the maximum value of the traffic flow, wherein x is a normalized value;
s15, representing the flow rate of the normalized time period t obtained in the step S14 as a 3-dimensional matrix of M rows, N columns and 1 channel, and recording the matrix as Xt∈RM×N×1Flow X in historical λ time periodshistory＝{Xt1,2, …, λ } as input, the next time period flow rate Xtrue＝Xλ+1As output, construct the sample { Xhistory,xtrue}；
And S16, constructing samples of all the flow data according to the step S15, and dividing the samples into a training set, a verification set and a test set according to a certain proportion.
Further, step S2 specifically includes:
s21, counting the number of various interest points POI in each area, wherein the number of the ith interest point in the area (m, n) is recorded as POI (m, n, i), wherein i ∈ [1, k ], k represents the category number of the interest points;
s22, calculating the inherent influence of various interest points in each region according to the distribution condition of the interest points, and recording the inherent influence as IIF ∈ RM×N×k；
S23, dividing time into working days, weekends and holidays, and converting the working days, weekends and holidays into certain-dimension vector representations by using a trained Word2Vec model;
s24, converting the weather condition into vector representation of a certain dimension by using a Word2Vec model, and normalizing other numerical weather data;
s25, inputting the results obtained in S23 and S24 into a two-layer full-connection networkTo obtain CIF ∈ R(λ+1)×kThe sensitivity degree of each type of interest point to external changes in the historical lambda time periods and the predicted 1 time period is represented;
s26, combining the result of S22 and the result of S25 to obtain the comprehensive influence of ICIF (t, m, n, i) ═ IIF (m, n, i) × CIF (t, i), t ∈ [1, lambda +1 ]],i∈[1,k]Where ICIF is a 4-dimensional matrix, denoted ICIF ∈ R(λ+1)×M×N×k；
S27, learning dynamic change of region vitality by using the 3D convolutional neural network, inputting the result of S26 into the 3D convolutional neural network to obtain the region vitality Vit ∈ R(λ+1)×M×N×1Representing the activity of each region in the historical lambda time periods and the predicted 1 time period;
s28, splitting the result obtained in S27 into region vitality Vit of historical time periodhistory∈Rλ×M×N×1And predicting the regional vitality Vit of the time periodpred∈RM×N×1。
Further, the method for calculating the intrinsic influence of each type of interest point in each region in step S22 is:
s221, calculating the proportion of each interest point in each region in the region:
Figure GDA0002452805400000031
s222, calculating the proportion of each interest point in each region in the whole city:
Figure GDA0002452805400000032
s223, calculating shannon entropy of each interest point:
Figure GDA0002452805400000041
the maximum value is Smax(POIi)＝log(M×N)；
S224, calculating the imbalance of the distribution of various interest points:
Figure GDA0002452805400000042
s225, calculating the inherent influence of various interest points in each region:
IIF(m,n,i)＝Den(m,n,i)×Int(m,n,i)×Equi。
further, in step S25, the first layer of the fully-connected network extracts the characteristics of time and weather by using 64 nodes, and the second layer simulates the sensitivity of each type of interest point to external changes by using k nodes.
Further, in step S27, the number of layers of the 3D convolutional neural network is 4, zero padding is used, the size of the convolutional kernel is set to 3 × 3 × 3, the first three layers use 64 convolutional kernels, and the output layer uses 1 convolutional kernel to fuse the influence of each type of interest point to obtain the region activity.
Further, step S3 specifically includes:
s31, subtracting the activity of the historical region by the historical traffic flow:
Figure GDA0002452805400000043
s32, predicting a space-time sequence by using a convolution duration memory network: inputting the result of S31 into a convolution duration memory network, and recording the output result as
Figure GDA0002452805400000044
S33, adding the result of the S32 and the regional vitality of the prediction time period to obtain a final prediction result:
Figure GDA0002452805400000045
further, in step S32, the number of layers of the convolutional long-term memory network is 4, the size of the convolution kernel is set to 5 × 5 using zero padding, the first 3 layers use 64 convolution kernels, and the output layer uses 1 convolution kernel.
Further, step S4 specifically includes:
s41 training by use ofSimultaneously training the region activity model and the flow prediction model in the steps S2 and S3, wherein the loss function is a mean square error function MSE:
Figure GDA0002452805400000046
wherein XiThe flow rate of the fluid is represented as a real flow rate,
Figure GDA0002452805400000047
representing the predicted flow rate;
s42, selecting a model with the minimum MSE as an optimal model according to the verification set and the test set;
and S43, normalizing the real-time traffic data, inputting the normalized real-time traffic data into the optimal model obtained in S42, obtaining an output result, performing inverse normalization to obtain a final prediction result, and realizing real-time prediction of traffic flow of each region of the city.
After adopting the technical scheme, compared with the background technology, the invention has the following advantages:
1. most of the existing methods only pay attention to the traffic flow prediction of one road section or a plurality of road sections, and the measurement of crowd gathering degree of each region of a city is lacked. The method provided by the invention divides the urban road network into regions, designs a model to realize simultaneous prediction of traffic flow of each region of the city, and can better reflect the range and degree of crowd aggregation in the city.
2. Most of the existing methods only pay attention to traffic flow data lacking semantic information, and cannot mine driving force behind crowd movement. The method provided by the invention combines the interest point data with rich semantic information, can better reflect the purpose and regularity of the crowd activities in the city, and reveals the driving force behind the crowd movement.
3. The existing method does not well deal with the influence of changes of external factors on the activity rules of people. The method provided by the invention considers different sensitivity degrees of different interest points to external changes, and meanwhile, the correlation between the external influences can be better learned by using Word2Vec to code time and weather.
In conclusion, the method provided by the invention can be used for predicting the urban regional traffic flow by fusing regional vitality through the deep learning technology, and simultaneously, the influence of the driving force behind the crowd activities and external factors is considered, so that higher prediction precision can be obtained.
Drawings
FIG. 1 is a flow chart of a method for urban traffic flow prediction fusing regional vitality;
FIG. 2 is a schematic diagram of urban road network grid area division;
FIG. 3 is a schematic view of traffic flow calculation;
FIG. 4 is a schematic diagram of a region activity model;
FIG. 5 is a schematic view of a flow prediction model;
fig. 6 is a schematic diagram of model training and real-time traffic prediction.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Examples
Fig. 1 is a flow chart of an urban traffic flow prediction method for merging regional vitality, which includes traffic flow calculation, model design, model training, and real-time flow prediction. In the traffic flow calculation, grid area division is firstly carried out on an urban road network according to the longitude and latitude, and the traffic flow of each area is calculated according to the data of the license plate recognition equipment. In model design, dynamic change of regional vitality is learned through a 3D convolutional neural network (3D CNN) based on information such as distribution of urban interest points, weather, holidays and the like, regional vitality and traffic flow are fused, and flow prediction is carried out by utilizing a convolutional long-term memory network (ConvLSTM). In the model training and the real-time flow prediction, firstly, the region activity model and the flow prediction model are trained simultaneously by using historical data, and then the real-time flow prediction is carried out by using the trained model.
The method comprises the following specific steps:
s1, dividing the urban road network into M × N grid areas according to the longitude and latitude, and calculating the traffic flow in each area according to the data recorded by the license plate recognition equipment:
s11, dividing the urban road network into M × N grid areas according to the longitude and latitude, so that the area (M, N) represents the grid area of the M-th row and the N-th column;
s12, extracting the recording data of the license plate recognition equipment: d, v, tau >, wherein d is the number of the license plate recognition equipment, v is the number of the shot license plate, and tau represents the recording time;
s13, mapping the license plate recognition devices into grid areas according to the longitude and latitude, and calculating the number of vehicles recorded by the license plate recognition devices in each area in each time period as a traffic flow:
Figure GDA0002452805400000061
Figure GDA0002452805400000062
represents the traffic flow of the region (m, n) during the time period t;
s14, normalizing the traffic flow in each area:
Figure GDA0002452805400000063
wherein x0For original traffic flow, xminAnd xmaxRespectively representing the minimum value and the maximum value of the traffic flow, wherein x is a normalized value;
s15, obtaining the flow of the normalized time period t according to the step S14, and representing the flow as a 3-dimensional matrix of M rows, N columns and 1 channels, and marking the matrix as Xt∈RM×N×1Flow X in historical λ time periodshistory＝{Xt1,2, …, λ } as input, the next time period flow rate Xtrue＝Xλ+1As output, a sample { X } may be constructedhistory,Xtrue}；
And S16, constructing samples of all the flow data according to the step S15, and dividing the samples into a training set, a verification set and a test set according to the proportion of 3:1: 1.
S2, designing a regional activity model: by utilizing the information of the distribution of urban interest points, festivals, holidays, weather and the like, the dynamic change of the vitality of each area in the city is learned by applying volume 3 DCNN:
s21, counting the number of various interest Points (POIs) (such as restaurants, scenic spots and the like) in each area, wherein the number of the ith interest point in the area (m, n) is recorded as the POI (m, n, i), and i ∈ [1, k ], k represents the category number of the interest points;
s22, calculating the inherent influence of various interest points in each region according to the distribution condition of the interest points, and recording the inherent influence as IIF ∈ RM×N×k：
S221, calculating the proportion of each interest point in each region in the region:
Figure GDA0002452805400000071
s222, calculating the proportion of each interest point in each region in the whole city:
Figure GDA0002452805400000072
s223, calculating shannon entropy of each interest point:
Figure GDA0002452805400000073
the maximum value is Smax(POIi)＝log(M×N)；
S224, calculating the imbalance of the distribution of various interest points:
Figure GDA0002452805400000074
Equithe larger the distribution of the i-th interest points in the city is, the more unbalanced the distribution is, the larger the influence is. For example, the distribution of airports in cities is extremely unbalanced, and the whole city can be influenced;
s225, calculating the inherent influence of various interest points in each region:
IIF(m,n,i)＝Den(m,n,i)×Int(m,n,i)×Equi。
s23, dividing time into working days, weekends and holidays, and converting the working days, weekends and holidays into 200-dimensional vector representation by using a trained Word2Vec model;
s24, converting weather conditions (sunny, cloudy and the like) into 200-dimensional vector representation by using a Word2Vec model, and normalizing other numerical weather data (visibility, rainfall and the like);
s25, inputting the results obtained in S23 and S24 into a two-layer full-connection network, extracting the characteristics of time and weather by using 64 nodes in the first layer of the network, simulating the sensitivity of each type of interest point to external change by using k nodes in the second layer, and outputting the result which is recorded as CIF ∈ R(λ+1)×kWhere λ is the number of history periods. The CIF represents the sensitivity of each type of interest point to external changes in historical lambda time periods and predicted 1 time period;
s26, combining the result of S22 and the result of S25 to obtain the comprehensive influence of ICIF (t, m, n, i) ═ IIF (m, n, i) × CIF (t, i), t ∈ [1, lambda +1 ]],i∈[1,k]Where ICIF is a 4-dimensional matrix, denoted ICIF ∈ R(λ+1)×M×N×k；
S27, utilizing dynamic change of 3D CNN learning region vitality, inputting the result of S26 into a 3D CNN network of 4 layers, using zero-padding operation (zero-padding) for the 3D CNN, setting the size of a convolution kernel to be 3 × 3 × 3, using 64 convolution kernels for the first three layers, using 1 convolution kernel for an output layer to fuse influence of various interest points to obtain region vitality, which is recorded as Vit ∈ R(λ+1)×M×N×1Representing the activity of each region of the historical lambda time periods and the predicted 1 time period;
s28, splitting the result obtained in S27 into region vitality Vit of historical time periodhistory∈Rλ×M×N×1And predicting the regional vitality Vit of the time periodpred∈RM×N×1。
S3, designing a flow prediction model: fusing the vitality of the region and the traffic flow and performing flow prediction by using a convolution length-time memory network (ConvLSTM);
s31, subtracting the activity of the historical region by the historical flow:
Figure GDA0002452805400000081
s32, using ConvLSTM to predict space-time sequence, inputting the result of S31 into 4 layers of ConvLSTM network, using zero-filling operation, setting the size of convolution kernel to be 5 × 5, using 64 convolution kernels in the first 3 layers, using 1 convolution kernel in the output layer, and recording the output result as
Figure GDA0002452805400000082
S33, adding the result of the S32 and the regional vitality of the prediction time period to obtain a final prediction result:
Figure GDA0002452805400000083
s4, simultaneously training the region vitality model and the flow prediction model according to historical data, and then predicting the traffic flow of each region in real time by using the trained models, specifically:
s41, simultaneously training the region activity model and the flow prediction model in the steps S2 and S3 by utilizing a training set, wherein a loss function is a mean square error function MSE:
Figure GDA0002452805400000084
wherein XiThe flow rate of the fluid is represented as a real flow rate,
Figure GDA0002452805400000085
representing the predicted flow rate;
s42, selecting a model with the minimum MSE as an optimal model according to the verification set and the test set;
and S43, normalizing the real-time traffic data, inputting the normalized real-time traffic data into the optimal model obtained in S42, obtaining an output result, performing inverse normalization to obtain a final prediction result, and realizing real-time prediction of traffic flow of each region of the city.
Some steps in the method are described in detail below with reference to specific examples.
Traffic flow calculation
Fig. 2 is a schematic diagram showing division of a building gate island road network 18 × 18 in building gate city of Fujian province, China, and fig. 3 is a schematic diagram showing calculation of traffic flow, wherein the building gate island road network is divided into 18 × 18 grid regions according to longitude and latitude, and then data recorded by a license plate recognition device is extracted to calculate the number of vehicles shot in each time period in each region as the traffic flow.
Secondly, designing a regional vitality model
Fig. 4 is a schematic diagram of a region activity model. Firstly, the inherent influence of various interest points in each region is calculated according to the distribution condition of the interest points in the city; secondly, simulating the sensitivity of various interest points to external changes according to external factors (weather and time); the comprehensive influence of various interest points in each region is obtained by combining the inherent influence and the sensitivity; and finally, learning the space-time characteristics of the influence of the interest points by using a 3D CNN network and fusing the influence of various interest points to obtain the regional activity.
Thirdly, designing a flow prediction model
Fig. 5 is a schematic diagram of a flow prediction model. The method comprises the steps of subtracting historical regional vitality from historical flow, inputting the historical regional vitality into a ConvLSTM network, and adding the regional vitality of a prediction time period to an output result to obtain a final prediction result.
Model training and real-time flow prediction
Fig. 6 is a schematic diagram of model training and real-time traffic prediction. In model training, after historical flow data are normalized, a region activity model and a flow prediction model are trained simultaneously, a loss function is a mean square error function (MSE), and an optimal model is selected according to a verification set and a test set. In the real-time flow prediction, after the real-time flow data is normalized, an output result is obtained through an optimal model, and then the final predicted flow of each area is obtained through reverse normalization.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (7)

1. The urban traffic flow prediction method fusing regional vitality is characterized by comprising the following steps: the method comprises the following steps:
s1, dividing the urban road network into M × N grid areas according to the longitude and latitude, and calculating the traffic flow in each area according to the data recorded by the license plate recognition equipment;
s2, designing a regional activity model: learning dynamic changes of vitality of each region in a city by using the distribution of urban interest points, holidays and weather information and a 3D convolutional neural network;
s3, designing a flow prediction model: fusing the activity of the region and the traffic flow, and performing flow prediction by using a convolution duration memory network;
s4, simultaneously training the region vitality model and the flow prediction model according to historical data, and then predicting the traffic flow of each region in real time by using the trained models;
step S1 specifically includes:
s11, dividing the urban road network into M × N grid areas according to the longitude and latitude, so that the area (M, N) represents the grid area of the M-th row and the N-th column;
s12, extracting recording data < d, v, tau > of the license plate recognition equipment, wherein d is the number of the license plate recognition equipment, v is the number of the shot license plate, and tau represents recording time;
s13, mapping the license plate recognition devices into grid areas according to the longitude and latitude, and calculating the number of vehicles recorded by the license plate recognition devices in each area in each time period as a traffic flow:
Figure FDA0002485024700000011
Figure FDA0002485024700000012
Figure FDA0002485024700000013
represents the traffic flow of the region (m, n) during the time period t;
s14, for each zoneNormalizing the traffic flow in the domain:
Figure FDA0002485024700000014
wherein x0For original traffic flow, xminAnd xmaxRespectively representing the minimum value and the maximum value of the traffic flow, wherein x is a normalized value;
s15, representing the flow rate of the normalized time period t obtained in the step S14 as a 3-dimensional matrix of M rows, N columns and 1 channel, and recording the matrix as Xt∈RM×N×1Flow X in historical λ time periodshistory＝{XtI t1, 2.. lambda. } as input, the flow X for the next time periodtrue＝Xλ+1As output, construct the sample { Xhistory，Xtrue}；
S16, constructing samples of all flow data according to the step S15, and dividing the samples into a training set, a verification set and a test set according to a certain proportion;
step S2 specifically includes:
s21, counting the number of various interest points POI in each area, wherein the number of the ith interest point in the area (m, n) is recorded as POI (m, n, i), wherein i ∈ [1, k ], k represents the category number of the interest points;
s22, calculating the inherent influence of various interest points in each region according to the distribution condition of the interest points, and recording the inherent influence as IIF ∈ RM ×N×k；
S23, dividing time into working days, weekends and holidays, and converting the working days, weekends and holidays into certain-dimension vector representations by using a trained Word2Vec model;
s24, converting the weather condition into vector representation of a certain dimension by using a Word2Vec model, and normalizing other numerical weather data;
s25, inputting the results obtained in S23 and S24 into a two-layer full-connection network to obtain CIF ∈ R (λ+1)×kThe sensitivity degree of each type of interest point to external changes in the historical lambda time periods and the predicted 1 time period is represented;
s26, combining the result of S22 and the result of S25 to obtain the comprehensive image of various interest pointsICIF (t, m, n, i) ═ IIF (m, n, i) × CIF (t, i), t ∈ [1, lambda +1]，i∈[1，k]Where ICIF is a 4-dimensional matrix, denoted ICIF ∈ R(λ+1)×M×N×k；
S27, learning dynamic change of region vitality by using the 3D convolutional neural network, inputting the result of S26 into the 3D convolutional neural network to obtain the region vitality Vit ∈ R(λ+1)×M×N×1Representing the activity of each region in the historical lambda time periods and the predicted 1 time period;
s28, splitting the result obtained in S27 into region vitality Vit of historical time periodhistory∈Rλ×M×N×1And predicting the regional vitality Vit of the time periodpred∈RM×N×1。
2. The method for predicting urban traffic flow by fusing regional vitality according to claim 1, wherein the method for calculating the intrinsic influence of each type of interest point in each region in step S22 is as follows:
s221, calculating the proportion of each interest point in each region in the region:
Figure FDA0002485024700000021
s222, calculating the proportion of each interest point in each region in the whole city:
Figure FDA0002485024700000022
s223, calculating shannon entropy of each interest point:
Figure FDA0002485024700000023
the maximum value is Smax(POIi)＝log(M×N)；
S224, calculating the imbalance of the distribution of various interest points:
Figure FDA0002485024700000031
s225, calculating the inherent influence of various interest points in each region:
IIF(m，n，i)＝Den(m，n，i)×Int(m，n，i)×Equi。
3. the urban traffic flow prediction method for merging regional vitality according to claim 1, characterized in that: in step S25, the first layer of the fully-connected network uses 64 nodes to extract the characteristics of time and weather, and the second layer uses k nodes to simulate the sensitivity of each type of interest point to external changes.
4. The urban traffic flow prediction method fusing regional vitality according to claim 1, wherein in step S27, the number of layers of the 3D convolutional neural network is 4, the 3D convolutional neural network uses zero padding, the size of the convolutional kernel is set to 3 × 3 × 3, the first three layers use 64 convolutional kernels, and the output layer uses 1 convolutional kernel to fuse the influence of various interest points to obtain regional vitality.
5. The urban traffic flow prediction method for merging regional vitality according to claim 1, wherein step S3 specifically includes:
s31, subtracting the activity of the historical region by the historical traffic flow:
Figure FDA0002485024700000032
s32, predicting a space-time sequence by using a convolution duration memory network: inputting the result of S31 into a convolution duration memory network, and recording the output result as
Figure FDA0002485024700000033
S33, adding the result of the S32 and the regional vitality of the prediction time period to obtain a final prediction result:
Figure FDA0002485024700000034
6. the method for predicting urban traffic flow based on fusion region vitality according to claim 5, wherein in step S32, zero padding is used when the number of layers of the convolution long-and-short memory network is 4, the convolution kernel size is set to 5 × 5, 64 convolution kernels are used for the first 3 layers, and 1 convolution kernel is used for the output layer.
7. The urban traffic flow prediction method for merging regional vitality according to claim 1, wherein step S4 specifically includes:
s41, simultaneously training the region activity model and the flow prediction model in the steps S2 and S3 by utilizing a training set, wherein a loss function is a mean square error function MSE:
Figure FDA0002485024700000035
wherein XiThe flow rate of the fluid is represented as a real flow rate,
Figure FDA0002485024700000036
representing the predicted flow rate;
s42, selecting a model with the minimum MSE as an optimal model according to the verification set and the test set;
and S43, normalizing the real-time traffic data, inputting the normalized real-time traffic data into the optimal model obtained in S42, obtaining an output result, performing inverse normalization to obtain a final prediction result, and realizing real-time prediction of traffic flow of each region of the city.
CN201711052176.1A 2017-10-30 2017-10-30 Urban traffic flow prediction method fusing regional vitality Active CN107967532B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201711052176.1A CN107967532B (en) 2017-10-30 2017-10-30 Urban traffic flow prediction method fusing regional vitality

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201711052176.1A CN107967532B (en) 2017-10-30 2017-10-30 Urban traffic flow prediction method fusing regional vitality

Publications (2)

Publication Number Publication Date
CN107967532A CN107967532A (en) 2018-04-27
CN107967532B true CN107967532B (en) 2020-07-07

Family

ID=61999972

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201711052176.1A Active CN107967532B (en) 2017-10-30 2017-10-30 Urban traffic flow prediction method fusing regional vitality

Country Status (1)

Country Link
CN (1) CN107967532B (en)

Families Citing this family (48)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108417040B (en) * 2018-05-14 2020-09-08 武汉理工大学 Characteristic town traffic distribution prediction method
CN108985475B (en) * 2018-06-13 2021-07-23 厦门大学 Network taxi appointment and taxi calling demand prediction method based on deep neural network
CN108876045A (en) * 2018-06-25 2018-11-23 上海应用技术大学 Emergency tender optimal route recommended method based on LSTM model prediction
CN109034460A (en) * 2018-07-03 2018-12-18 深圳市和讯华谷信息技术有限公司 Prediction technique, device and system for scenic spot passenger flow congestion level
CN109299401B (en) * 2018-07-12 2022-02-08 中国海洋大学 Metropolitan area space-time flow prediction method based on mixed deep learning model LSTM-ResNet
CN108961816B (en) * 2018-07-19 2020-03-24 泰华智慧产业集团股份有限公司 Road parking berth prediction method based on optimized LSTM model
CN110929543B (en) * 2018-08-30 2023-08-25 阿里巴巴集团控股有限公司 Meteorological data processing method and device
CN109345296A (en) * 2018-09-20 2019-02-15 深圳市东部公共交通有限公司 Common people's Travel Demand Forecasting method, apparatus and terminal
CN109523082B (en) * 2018-11-20 2023-12-22 广东机场白云信息科技股份有限公司 CNN-LSTM flight normal clearance rate prediction method
CN109886444B (en) * 2018-12-03 2023-07-11 深圳市北斗智能科技有限公司 Short-time traffic passenger flow prediction method, device, equipment and storage medium
CN109697590A (en) * 2018-12-07 2019-04-30 东南大学 A kind of city distribution vehicle driving monitoring system based on big data
CN109711004B (en) * 2018-12-11 2023-03-28 重庆邮电大学 Optical fiber refractive index big data prediction method
CN109887272B (en) * 2018-12-26 2021-08-13 创新先进技术有限公司 Traffic pedestrian flow prediction method and device
CN110046787A (en) * 2019-01-15 2019-07-23 重庆邮电大学 A kind of urban area charging demand for electric vehicles spatio-temporal prediction method
CN110147904B (en) * 2019-04-23 2021-06-18 深圳先进技术研究院 Urban gathering event prediction and positioning method and device
CN110210656B (en) * 2019-05-21 2021-07-13 上海交通大学 Shared bicycle flow prediction method and system based on station behavior analysis
CN110517482B (en) * 2019-07-29 2021-06-29 杭州电子科技大学 Short-term traffic flow prediction method based on 3D convolutional neural network
CN110288170B (en) * 2019-07-29 2020-04-14 重庆电子工程职业学院 Prediction model construction method, traffic flow prediction device and electronic equipment
CN110414747B (en) * 2019-08-08 2022-02-01 东北大学秦皇岛分校 Space-time long-short-term urban pedestrian flow prediction method based on deep learning
CN110570035B (en) * 2019-09-02 2023-04-07 上海交通大学 People flow prediction system for simultaneously modeling space-time dependency and daily flow dependency
CN114365205A (en) * 2019-09-19 2022-04-15 北京嘀嘀无限科技发展有限公司 System and method for determining estimated time of arrival in online-to-offline service
CN110633871B (en) * 2019-09-25 2022-09-16 大连理工大学 Regional traffic demand prediction method based on convolution long-term and short-term memory network
CN110766589A (en) * 2019-10-28 2020-02-07 电子科技大学 Method for deducing city function based on communication data and interest point data
CN110991713B (en) * 2019-11-21 2022-04-01 杭州电子科技大学 Irregular area flow prediction method based on multi-graph convolution sum GRU
CN113095535A (en) * 2020-01-08 2021-07-09 普天信息技术有限公司 Flow prediction method and device based on deep space-time residual error network
CN111343650B (en) * 2020-02-14 2021-11-30 山东大学 Urban scale wireless service flow prediction method based on cross-domain data and loss resistance
CN111192457A (en) * 2020-03-20 2020-05-22 青岛海信网络科技股份有限公司 Method and device for controlling urban-level integral traffic signals
CN111599170B (en) * 2020-04-13 2021-12-17 浙江工业大学 Traffic running state classification method based on time sequence traffic network diagram
CN111696345A (en) * 2020-05-08 2020-09-22 东南大学 Intelligent coupled large-scale data flow width learning rapid prediction algorithm based on network community detection and GCN
TWI775123B (en) * 2020-07-29 2022-08-21 財團法人資訊工業策進會 Traffic condition prediction system and traffic condition prediction method
CN112215059B (en) * 2020-08-26 2023-10-27 厦门大学 Deep learning-based urban village identification and population estimation method, system and computer-readable storage medium
CN112017436B (en) * 2020-09-09 2021-09-28 中国科学院自动化研究所 Method and system for predicting urban traffic travel time
CN112183824B (en) * 2020-09-11 2022-07-19 武汉大学 Online and offline correlated urban passenger flow prediction method
CN112101682B (en) * 2020-09-25 2024-04-09 北京百度网讯科技有限公司 Traffic pattern prediction method, traffic pattern prediction device, server and readable medium
CN112216108B (en) * 2020-10-12 2021-06-29 中南大学 Traffic prediction method based on attribute-enhanced space-time graph convolution model
CN112396218B (en) * 2020-11-06 2024-01-09 南京航空航天大学 Crowd flow prediction method based on urban area multi-mode fusion
CN112669595B (en) * 2020-12-10 2022-07-01 浙江大学 Network taxi booking flow prediction method based on deep learning
CN113077053B (en) * 2021-03-18 2022-09-02 中国电子科技集团公司第五十四研究所 Multi-group activity mode characterization method and target geographic activity area identification method
CN113341419B (en) * 2021-05-25 2022-11-01 成都信息工程大学 Weather extrapolation method and system based on VAN-ConvLSTM
CN113313303A (en) * 2021-05-28 2021-08-27 南京师范大学 Urban area road network traffic flow prediction method and system based on hybrid deep learning model
CN113673752A (en) * 2021-08-03 2021-11-19 湖北三江航天万峰科技发展有限公司 BP neural network-based variable lane traffic flow prediction method
CN113780857A (en) * 2021-09-17 2021-12-10 清华大学 Method and device for measuring land utilization vitality fitness and electronic equipment
CN114187759B (en) * 2021-11-19 2023-01-03 东南大学 Road side unit driving assistance method and device based on data driving model
CN114299727B (en) * 2021-12-28 2022-12-09 杭州滨电信息技术有限公司 Traffic flow prediction system based on Internet of things and edge computing and cloud platform
CN114613135B (en) * 2022-03-06 2023-04-21 南京理工大学 Urban road network traffic flow prediction method and system considering influence of building sites
CN116703008B (en) * 2023-08-02 2023-10-31 山东高速股份有限公司 Traffic volume prediction method, equipment and medium for newly built highway
CN117273287B (en) * 2023-11-23 2024-03-08 山东科技大学 Tensor-matrix coupling-based subway passenger flow prediction method
CN117423238A (en) * 2023-12-19 2024-01-19 北京华录高诚科技有限公司 Integrated traffic flow prediction device and prediction method based on convolution switching network

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106935034A (en) * 2017-05-08 2017-07-07 西安电子科技大学 Towards the regional traffic flow forecasting system and method for car networking
CN107103758A (en) * 2017-06-08 2017-08-29 厦门大学 A kind of city area-traffic method for predicting based on deep learning

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106935034A (en) * 2017-05-08 2017-07-07 西安电子科技大学 Towards the regional traffic flow forecasting system and method for car networking
CN107103758A (en) * 2017-06-08 2017-08-29 厦门大学 A kind of city area-traffic method for predicting based on deep learning

Also Published As

Publication number Publication date
CN107967532A (en) 2018-04-27

Similar Documents

Publication Publication Date Title
CN107967532B (en) Urban traffic flow prediction method fusing regional vitality
CN109285346B (en) Urban road network traffic state prediction method based on key road sections
CN108629978A (en) A kind of traffic trajectory predictions method based on higher-dimension road network and Recognition with Recurrent Neural Network
CN112561146A (en) Large-scale real-time traffic flow prediction method based on fuzzy logic and depth LSTM
CN106971547A (en) A kind of Short-time Traffic Flow Forecasting Methods for considering temporal correlation
CN112257934A (en) Urban people flow prediction method based on space-time dynamic neural network
US20160125307A1 (en) Air quality inference using multiple data sources
CN108108844A (en) A kind of urban human method for predicting and system
CN109840660A (en) A kind of vehicular characteristics data processing method and vehicle risk prediction model training method
CN109410577A (en) Adaptive traffic control sub-area division method based on Spatial Data Mining
Ballestores Jr et al. An integrated parcel-based land use change model using cellular automata and decision tree
CN109359166B (en) Space growth dynamic simulation and driving force factor contribution degree synchronous calculation method
CN113505521B (en) Urban waterlogging rapid forecasting method based on neural network-numerical simulation
Doglioni et al. Inferring groundwater system dynamics from hydrological time-series data
CN116227834A (en) Intelligent scenic spot digital platform based on three-dimensional point cloud model
CN114897444A (en) Method and system for identifying service facility requirements in urban subarea
CN110070720B (en) Calculation method for improving fitting degree of traffic capacity model of intersection road occupation construction area
CN112633602B (en) Traffic congestion index prediction method and device based on GIS map information
Ahmed Modelling spatio-temporal urban land cover growth dynamics using remote sensing and GIS techniques: A case study of Khulna City
CN114973657A (en) Urban traffic noise pollution analysis and evaluation method based on trajectory data
Alhasoun et al. Streetify: Using street view imagery and deep learning for urban streets development
CN116913088A (en) Intelligent flow prediction method for expressway
Mohan et al. A brief review of recent developments in the integration of deep learning with GIS
Huber et al. Modelling bicycle route choice in German cities using open data, MNL and the bikeSim web-app
CN117037461A (en) Road network traffic jam prediction method based on multi-weight graph three-dimensional convolution

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