CN113469425B - Deep traffic jam prediction method - Google Patents
Deep traffic jam prediction method Download PDFInfo
- Publication number
- CN113469425B CN113469425B CN202110697865.8A CN202110697865A CN113469425B CN 113469425 B CN113469425 B CN 113469425B CN 202110697865 A CN202110697865 A CN 202110697865A CN 113469425 B CN113469425 B CN 113469425B
- Authority
- CN
- China
- Prior art keywords
- road
- data
- traffic
- data set
- congestion
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 230000000295 complement effect Effects 0.000 claims abstract description 28
- 238000004891 communication Methods 0.000 claims abstract description 21
- 238000013135 deep learning Methods 0.000 claims abstract description 17
- 239000013598 vector Substances 0.000 claims abstract description 4
- 230000005540 biological transmission Effects 0.000 claims description 30
- 239000011159 matrix material Substances 0.000 claims description 20
- 230000008569 process Effects 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 5
- 230000002123 temporal effect Effects 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000011056 performance test Methods 0.000 claims description 3
- 238000005096 rolling process Methods 0.000 claims description 2
- 230000001105 regulatory effect Effects 0.000 claims 1
- 238000012549 training Methods 0.000 description 3
- 230000007423 decrease Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
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
- 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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/067—Enterprise or organisation modelling
-
- 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
- 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
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Tourism & Hospitality (AREA)
- Entrepreneurship & Innovation (AREA)
- Theoretical Computer Science (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Educational Administration (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Traffic Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for predicting deep traffic congestion, which comprises the following steps: step 1, a network server stores road traffic information, establishes a sample set of road section flow vectors, and obtains a historical data set; step 2, converting the traffic road network structure information into a topological graph structure; step 3, obtaining the degree information of the nodes and selecting the selected road section data from the historical data set according to the degree information of the nodes to obtain a missing data set; step 4, complementing the missing road section data to obtain a complement data set; and 5, inputting the complement data set into a deep learning flow prediction model to predict road congestion. The prediction method has high prediction accuracy and reduces communication overhead; the amount of data transmitted is reduced and the impact of missing data on the predictive performance is reduced.
Description
Technical Field
The invention relates to the technical field of traffic congestion prediction, in particular to a deep traffic congestion prediction method.
Background
Modern urban road traffic has serious congestion problems, and congestion prediction has become an effective means for relieving traffic pressure and avoiding congestion. And the server analyzes and processes the vehicle congestion data of different time and different road sections through big data and a deep learning algorithm, extracts characteristics and digs rules, so as to predict the congestion condition of the future road.
At present, although the traffic jam prediction technology is relatively mature, a deep learning model generally needs a large amount of data to support so as to train a model with relatively stable and relatively good prediction performance, but at the same time, the cost and the resource occupation caused in the process of sending a large amount of data from a terminal to a network server cause great pressure on communication transmission, and how to reduce the communication cost as much as possible and ensure that the accuracy of traffic jam prediction is relatively lacking in the current research.
Based on the technical defects in the prior art, the invention provides a deep traffic jam prediction method.
Disclosure of Invention
In order to solve the technical problems of the prior art, the invention provides a deep traffic jam prediction method, which comprises the steps of converting an actual traffic road into a topological network, selecting road sections based on node degree, complementing data of missing road sections, establishing and training a deep prediction model, determining the optimal number of transmission road sections according to a test result, and carrying out real-time road jam prediction.
The invention adopts the following technical scheme:
a method for deep traffic congestion prediction, comprising:
step 1, a network server stores road traffic information, establishes a sample set of road section flow vectors, and obtains a historical data set;
step 2, converting the traffic road network structure information into a topological graph structure;
step 3, obtaining the degree information of the nodes and selecting the selected road section data from the historical data set according to the degree information of the nodes to obtain a missing data set;
step 4, complementing the missing road section data to obtain a complement data set;
and 5, inputting the complement data set into a deep learning flow prediction model to predict road congestion.
Further, in step 1, the road traffic information includes traffic road network structure information and historical traffic data information.
Further, in step 2, the converting the traffic road network structure information into the topology map structure includes: a directed graph network G (V, E) is defined from traffic roads, with the roads being nodes v= { V in the directed graph network 1 ,v 2 ,…,v N N represents the number of road segments, and the road segment intersection is taken as the edge e= { v of the node in the directed graph network i v j }，v i v j Representing road v i Can directly reach road v j 。
Further, in step 3, according to the directed graph network G (V, E), an adjacency matrix a E R of the directed graph network G is calculated N×N ，A∈R N×N Representing the communication relation between roads, wherein the matrix comprises 1 and 0 elements, wherein 1 represents that the road can be directly reached, and 0 represents that the road cannot be directly reached, namely the following formula (1):
and calculating to obtain an output matrix of the directed graph network G by the A, namely the following formula (2):
D＝∑ j Α ij ……(2)，
obtaining the degree of the node according to the above formula (1) and the above formula (2), selecting the data of the selected road sections, and setting the number of the finally selected road sections as N t Setting the selected road segment set V s And a non-selected road segment set V n To divide all road segments, including:
selecting, namely placing the road section with the highest degree of the selected node into a selected set V s In consideration of information redundancy between adjacent road segments, the neighbor road segments which can be directly reached by the largest road segment are temporarily put into a non-selected set V n Selecting the road section with the largest degree from the rest road sections, and repeatedly executing the process until only the isolated road sections without neighbors remain;
put back, non-selected set V n The road sections in the road sections are replaced in the whole road section set;
repeating the selecting and replacing steps;
selecting the road segment set V at any moment s The number of road segments in (a) is N t And when the process is finished.
Further, in step 4, a matrix P is set to indicate whether the road segment data is missing, the corresponding value of the selected road segment is 1, the corresponding value of the unselected road segment is 0, and the server side calculates the missing data set X from the complete historical data set X 1 Namely the following formula (3):
X 1 ＝X*P……(3)，
in the above formula (3), the matrix point multiplication, that is, the multiplication of corresponding position elements, divides the historical traffic data set X according to seven days of a week, adds and averages the congestion indexes of a certain road section every week and every week, and obtains the historical average value of the congestion indexes of the week X, wherein the congestion condition of the road section is affected by the congestion condition of the neighboring road section, and corrects the historical change rule of the road section according to the information of the neighboring road section, and the calculation method is as follows:
the definition missing segment completion value is composed of two parts, namely the following formula (4):
in the above formula (4), v n Representing the missing road segment, i.e. the road segment that needs to be filled,representing road segment v n Is the historical mean value data of { v g The road segment v n All selected road sections in the neighborhood, +.>The average value of the true values of the road sections selected by the neighbors is represented, and the proportionality coefficient alpha+beta=1 of the road section history information and the neighbor true information;
calculating the mean value of the true values of the neighbor selected road sections, wherein the mean value is represented by the following formula (5):
calculating the { v } of the selected road sections in the neighbor road sections g The ratio of }, i.e., the following formula (6):
in the above formula (6) { v c The road segment v n Including selecting a road segment { v }, all neighbors of g And missing road segment, coefficientFor adjusting the value of beta;
the beta value is dynamically adjusted according to the number of surrounding selected neighbor road sections to determine the influence of neighbor information on the complement value, and the N is used for different selected road sections t According to the data set X after being completed 2 Respectively calculating the minimum average absolute error of the complement data and the accurate dataThe following formula (7):
in step 5, the complement data set is input into a deep learning flow prediction model, the model is trained, performance test is carried out, and the optimal number of the transmission road sections is obtained according to the test result.
Further, in step 5, the road congestion prediction includes:
the network server obtains the optimal transmission road section number, decides whether to transmit to the terminal equipment and feeds back the optimal transmission road section number information to the vehicle or the road side unit;
the vehicle or the road side unit selectively transmits current road section data according to the indication of the network server, the network server obtains a missing data set, and the data which is not transmitted is complemented to obtain a real-time complement data set;
and the network server inputs the complement data set into a deep learning flow prediction model to predict road congestion.
Further in step 5, the deep learning traffic prediction model comprises a graph rolling network GCN for capturing data spatial correlations and a gating loop unit GRU for capturing data temporal correlations.
Further, in step 5, the complement data set X is used 2 Inputting a deep learning flow prediction model, testing, defining the communication learning efficiency eta as the ratio of square root of useful power of predicted congestion index to transmission data overhead, and calculating the communication learning efficiency eta according to the following formula (8) when the useful power is larger and the transmission data overhead is smaller:
in the above-mentioned (8),predictive value representing congestion index, +.>Representing absolute errors of true and predicted values, the transmission data overhead is the number of transmission segments N t The coefficient u represents a variable overhead, the coefficient v represents a fixed overhead, and u and v are variable according to a communication system model, that is, the number of transmission segments with the maximum communication learning efficiency is the optimal number of transmission segments, as shown in the following formula (9):
N o ＝argmax{η(N t )}……(9)。
1. according to the deep traffic congestion prediction method, the missing road section data is filled through the filling algorithm, so that the transmitted data quantity is greatly reduced, and the influence of the missing data on the prediction performance is reduced;
2. according to the deep traffic jam prediction method, through the composite structure of the GCN and the GRU, the spatial characteristics and the temporal characteristics of the data can be respectively extracted, the time-space dynamic change rule of the data is explored, and therefore the prediction accuracy is high;
3. according to the deep traffic jam prediction method, the road section selection and completion algorithm is used, so that the influence on the prediction performance can be reduced as much as possible while the transmission data quantity is reduced, and the communication overhead can be effectively reduced.
Drawings
FIG. 1 is a flow chart of a method for deep traffic congestion prediction in an embodiment of the present invention;
fig. 2 is a schematic diagram of a congestion index X around a road segment in an embodiment of the present invention;
fig. 3 is a diagram of a TGCN network structure in an embodiment of the present invention;
fig. 4 is a graph showing the average absolute error and the communication learning efficiency as a function of the number of transmission segments in the embodiment of the present invention.
Detailed Description
In order that the above-recited objects, features and advantages of the invention will be more clearly understood, a more particular description of the invention will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings. It should be noted that, without conflict, the embodiments of the present invention and features in the embodiments may be combined with each other.
Example 1
As shown in fig. 1, the depth traffic congestion prediction method includes:
step 1, a network server stores road traffic information, establishes a sample set of road section flow vectors, and obtains a historical data set, and historical traffic congestion index data X epsilon R of each road N×P Wherein R represents a real number, P represents a time sequence length of the data, and N represents the number of road segments;
step 2, converting the traffic road network structure information into a topological graph structure;
step 3, obtaining the degree information of the nodes, sequentially selecting the road section data with the largest degree in the historical data set, putting the road section data into the selected road section set, selecting the road section with the largest degree in the rest road sections, repeating the process until the finally set road section number is selected, and further obtaining the missing data set;
step 4, complementing the missing road section data to obtain a complement data set;
and 5, inputting the complement data set into a deep learning flow prediction model, and carrying out road congestion prediction on the map network.
In step 1, the road traffic information includes traffic road network structure information and historical traffic data information.
In the above embodiment, the historical traffic data information includes historical traffic time series data based on road segments within the road network, including flow values, acquisition times, road attributes, and the like.
In step 2, the converting the traffic road network structure information into the topological graph structure includes: the traffic road defines a directed graph network G (V, E), which is defined herein as a directed graph network having roads as nodes v= { V in the directed graph network, since the actual traffic network comprises bidirectional lanes 1 ,v 2 ,…,v N N represents the number of road segments, and the road segment intersection is taken as the edge e= { v of the node in the directed graph network i v j }，v i v j Representing road v i Can directly reach road v j ；
In step 3, according to the directed graph network G (V, E), the adjacency matrix A E R of the directed graph network G is calculated N×N ，A∈R N×N Representing the connection relationship between roads, the matrix contains 1 and 0 elements, wherein 1 represents that the road can be directly reached, and 0 represents that the road can not be directly reached, namely:
calculating to obtain an output matrix of the directed graph network G by the A:
D＝∑ j Α ij ……(2)，
because the greater the degree of the node is, the more information the node contains, the more important it is for predicting the congestion index, select the road section according to the degree of the node, put forward the road section selection algorithm based on the degree of the node, obtain the degree of the node according to the formula 1-2, select and select the road section data, set up to need to select finally in the road section data of selectingThe number of the road segments is N t Setting the selected road segment set V s And a non-selected road segment set V n To divide all road segments, including:
selecting, namely placing the road section with the highest degree of the selected node into a selected set V s In consideration of information redundancy between adjacent road segments, the neighbor road segments which can be directly reached by the largest road segment are temporarily put into a non-selected set V n Selecting the road section with the largest degree from the rest road sections, and repeatedly executing the process until only the isolated road sections without neighbors remain;
put back, non-selected set V n The road sections in the road sections are replaced in the whole road section set;
repeating the selecting and replacing steps;
selecting the road segment set V at any moment s The number of road segments in (a) is N t And when the process is finished.
In step 4, a matrix P is set to indicate whether the road segment data is missing, the corresponding value of the selected road segment is 1, the corresponding value of the unselected road segment is 0, and the server side calculates the missing data set X from the complete historical data set X 1 The method comprises the following steps:
X 1 ＝X*P……(3)，
wherein, represents matrix point multiplication, namely multiplication of corresponding position elements, dividing a historical traffic data set X according to seven days of a week, adding and averaging congestion indexes of a certain road section every week X (X= { one, two, three, four, five, six, day }) day according to the week to obtain a historical average value of the congestion indexes of the week X, as shown in figure 2, the abscissa represents the moment of 24 hours in the day, the ordinate represents the congestion index, and the historical average value of all N road sections of the week X is calculated in the same way, and finally splicing according to the format of original data to obtain historical average value data M E R N×P ；
Because the congestion condition of the road section can be influenced by the congestion condition of the neighbor road section, the historical change rule of the road section is corrected according to the information of the neighbor road section, and the calculation method is as follows:
defining the missing segment complement value consists of two parts, namely:
wherein v is n Representing the missing road segment, i.e. the road segment that needs to be filled,representing road segment v n Historical mean data of (i.e. v in M n Corresponding row), { v g The road segment v n All selected road sections in the neighborhood, +.>The average value of the true values of the road sections selected by the neighbors is represented, and the proportionality coefficient alpha+beta=1 of the road section history information and the neighbor true information;
calculating the mean value of the true values of the neighbor selected road sections, namely:
because part of the road segment data is selectively transmitted, a neighbor road segment of a road segment needing to be completed may be missing, and the selected road segment { v } in the neighbor road segment needs to be considered g The ratio of }, namely:
wherein { v c The road segment v n Including selecting a road segment { v }, all neighbors of g And missing road segment, coefficientFor adjusting the value of beta;
the beta value is dynamically adjusted according to the number of surrounding selected neighbor road sections to determine the influence of neighbor information on the complement value, and the N is used for different selected road sections t According to the data set X after being completed 2 Dividing intoSeparately calculating the mean absolute error of the complement data and the exact data to be minimizedNamely:
in step 5, inputting the complement data set into a deep learning flow prediction model, training the model and performing performance test, and obtaining the optimal number of transmission road sections according to the test result;
in step 5, the road congestion prediction includes:
the network server obtains the optimal transmission road section number, decides whether to transmit to the terminal equipment and feeds back the optimal transmission road section number information to the vehicle or the road side unit;
the vehicle or the road side unit selectively transmits current road section data according to the indication of the network server, the network server obtains a missing data set, and the data which is not transmitted is complemented to obtain a real-time complement data set;
the network server inputs the complement data set into a deep learning flow prediction model to predict road congestion;
in step 5, complement data set X is used 2 The deep learning flow prediction model is input and tested, the communication learning efficiency eta is defined as the ratio of square root of useful power of predicted congestion index to the transmitted data overhead, and when the useful power is larger (i.e. the prediction error is smaller), the transmitted data overhead is smaller, the communication learning efficiency is higher, namely:
wherein,predictive value representing congestion index, +.>Representing absolute errors of true and predicted values, the transmission data overhead is the number of transmission segments N t The coefficient u represents a variable overhead and the coefficient v represents a fixed overhead, and u and v are variable according to the communication system model, so that the number of transmission segments with the maximum communication learning efficiency, that is, the optimal number of transmission segments) is obtained:
N o ＝argmax{η(N t )}……(9)；
as shown in fig. 4, the solid line shows the average absolute error, the broken line shows the communication learning efficiency under u=0.1 and v=80, and it is clear from the graph that the average absolute error decreases with the increase of the number of transmission segments, and since the image is a monotonically decreasing curve, the optimal point for making the average absolute error and the number of transmission segments as small as possible cannot be determined, and the maximum value exists when the observation communication learning efficiency increases and decreases with the increase of the number of transmission segments, the optimal point for ensuring the prediction accuracy and making the amount of transmission data as small as possible can be obtained.
In the above embodiment, the deep learning traffic prediction model employs a combined time-graph convolution network, including a graph convolution network GCN for capturing data spatial correlations and a gating loop unit GRU for capturing data temporal correlations;
from complement data set X 2 ∈R N×P And adjacency matrix A.epsilon.R N×N The GCN captures the spatial features between nodes through the first-order neighbors of the nodes, consisting ofI N Is an identity matrix>Is a degree matrix, the spatial dependence is captured using a 2-layer GCN in the above embodiment, and the end result can be expressed as:
wherein,is for->Normalization is carried out to prevent unstable numerical value W in operation 0 ∈R P×H Is the weight matrix from the input layer to the hidden layer, P is the length of the congestion index history time sequence, H is the number of hidden units, W 1 ∈R H×T Is the weight matrix from the hidden layer to the output layer, reLU (& gt) represents the modified linear unit, sigma (& gt) is the sigmoid nonlinear activation function, f (X) 2 ,A)∈R N×T The congestion index is predicted and output, and it can be seen that the GCN can extract the information of the first-order neighbors of each node and capture the dependence of the central road section and the adjacent road sections from the dimension of the space.
The result obtained by the GCN is input into the GRU to form the TGCN network used in the embodiment, the structure of the TGCN network is shown in figure 3, and h is shown in the specification t-1 Is the output at time t-1, h t Is the output at time t, r t And z t Reset gate and update gate representing GRU respectively, the graph convolution represents a 2-layer GCN network, usingThe complement data set representing the current time t is combined as an input to the graph convolution, and the output of the graph convolution is expressed as:
the space dependence among the roads in the traffic network is captured by the GCN module, the dependence of the road traffic jam index on the change of time is captured by the GRU module, and the dynamic acquisition of the macroscopic traffic rule is comprehensively realized to predict the road jam index in the future.
After the model training and testing process is finished, the network server side can obtain the optimal number of transmitting road segments, and when the real-time congestion of the road is predicted, the server indicates whether each vehicle or road side unit transmits the road segment data or not according to the optimal number of transmitting road segments, and the vehicle or road side unit selectively transmits the road segment data. And the server complements the received missing data, and finally inputs the complemented data into the model to predict the congestion index.
The present invention is not limited to the above-described embodiments, and the above-described embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made therein without departing from the spirit and scope of the invention, which is defined in the appended claims. The scope of the invention is defined by the appended claims.
Claims (8)
1. A method for predicting deep traffic congestion, comprising:
step 1, a network server stores road traffic information, establishes a sample set of road section flow vectors, and obtains a historical data set;
step 2, converting the traffic road network structure information into a topological graph structure;
step 3, obtaining the degree information of the nodes and selecting the selected road section data from the historical data set according to the degree information of the nodes to obtain a missing data set;
step 4, completing the missing road segment data to obtain a complete data set, setting a matrix P to indicate whether the road segment data is missing, selecting a road segment corresponding value of 1, selecting a non-selected road segment corresponding value of 0, and calculating by the server side from the complete historical data set X to obtain the missing data set X 1 Namely the following formula (1):
X 1 ＝X*P……(1)，
in the above formula (1), the matrix point multiplication, that is, the multiplication of the corresponding position elements, divides the historical data set X according to seven days of a week, adds the congestion indexes of any one day of each week of a certain road section according to the weeks, and then averages the congestion indexes to obtain the historical average value of the congestion indexes of any one day, and the congestion condition of the road section is affected by the congestion condition of the neighboring road section, and corrects the historical change rule of the road section according to the information of the neighboring road section, and the calculation method is as follows:
the definition missing segment completion value is composed of two parts, namely the following formula (2):
above (2), v n Representing the missing road segment, i.e. the road segment that needs to be filled,representing road segment v n Is the historical mean value data of { v g -representing the selected road segment among all neighbors of the road segment vn +.>The average value of the true values of the road sections selected by the neighbors is represented, and the proportionality coefficient alpha+beta=1 of the road section history information and the neighbor true information;
calculating the mean value of the true values of the neighbor selected road sections, wherein the mean value is represented by the following formula (3):
calculating the { v } of the selected road sections in the neighbor road sections g The ratio of }, i.e., the following formula (4):
in the above formula (4) { v c The road segment v n Including selecting a road segment { v }, all neighbors of g And missing road segment, coefficientFor adjusting the value of beta;
selecting neighbors according to the surroundingThe beta value is dynamically regulated by the number of the living road segments to determine the influence of neighbor information on the complement value, and the number N of the living road segments is selected according to different numbers t According to the data set X after being completed 2 Respectively calculating the minimum average absolute error of the complement data and the accurate dataThe following formula (5):
and 5, inputting the complement data set into a deep learning flow prediction model to predict road congestion.
2. The deep traffic congestion prediction method according to claim 1, wherein in step 1, the road traffic information includes traffic road network structure information and historical traffic data information.
3. The deep traffic congestion prediction method according to claim 1, wherein in step 2, converting traffic road network structure information into a topology map structure includes: a directed graph network G (V, E) is defined from traffic roads, with the roads being nodes v= { V in the directed graph network 1 ,v 2 ,…,v N N represents the number of road segments, and the road segment intersection is taken as the edge e= { v of the node in the directed graph network i v j }，v i v j Representing road v i Can directly reach road v j 。
4. The method for predicting deep traffic congestion according to claim 3, wherein in step 3, the adjacency matrix a E R of the directed graph network G is calculated according to the directed graph network G (V, E) N×N ，A∈R N×N Representing the communication relationship between roads, wherein the matrix contains 1 and 0 elements, wherein 1 represents that the road can be directly reached, and 0 represents that the road cannot be directly reached, namely the following formula (6):
and calculating to obtain an output matrix of the directed graph network G by the A, namely the following formula (7):
D＝∑ j Α ij ……(7)，
obtaining the degree of the node according to the above formula (6) and the above formula (7), selecting the data of the selected road sections, and setting the number of the finally selected road sections as N t Setting the selected road segment set V s And a non-selected road segment set V n To divide all road segments, including:
selecting, namely placing the road section with the highest degree of the selected node into a selected set V s In consideration of information redundancy between adjacent road segments, the neighbor road segments which can be directly reached by the largest road segment are temporarily put into a non-selected road segment set V n Selecting the road section with the largest degree from the rest road sections, and repeatedly executing the process until only the isolated road sections without neighbors remain;
put back, gather the non-selected road segments V n The road sections in the road sections are replaced in the whole road section set;
repeating the selecting and replacing steps;
selecting the road segment set V at any moment s The number of road segments in (a) is N t And when the process is finished.
5. The deep traffic congestion prediction method according to claim 1, wherein in step 5, the complement data set is input into a deep learning traffic prediction model, the model is trained and performance test is performed, and the optimal number of transmission road segments is obtained according to the test result.
6. The deep traffic congestion prediction method according to claim 5, wherein in step 5, the road congestion prediction includes:
the network server obtains the optimal transmission road section number, decides whether to transmit to the terminal equipment and feeds back the optimal transmission road section number information to the vehicle or the road side unit;
the vehicle or the road side unit selectively transmits current road section data according to the indication of the network server, the network server obtains a missing data set, and the data which is not transmitted is complemented to obtain a real-time complement data set;
and the network server inputs the complement data set into a deep learning flow prediction model to predict road congestion.
7. The deep traffic congestion prediction method according to claim 1, wherein in step 5, the deep learning traffic prediction model includes a graph rolling network GCN for capturing data spatial correlations and a gating loop unit GRU for capturing data temporal correlations.
8. The deep traffic congestion prediction method according to claim 6, wherein in step 5, the complement data set X is used 2 Inputting a deep learning flow prediction model, testing, defining the communication learning efficiency eta as the ratio of square root of useful power of predicted congestion index to transmission data overhead, and calculating the communication learning efficiency eta according to the following formula (8):
in the above-mentioned (8),predictive value representing congestion index, +.>Representing absolute errors of true and predicted values, the transmitted data overhead is for different selected road segment numbers N t The coefficient u represents a variable overhead, the coefficient v represents a fixed overhead, and u and v are variable according to a communication system model, that is, the number of transmission segments with the maximum communication learning efficiency is the optimal number of transmission segments, as shown in the following formula (9):
N o ＝arg max{η(N t )}……(9)。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110697865.8A CN113469425B (en) | 2021-06-23 | 2021-06-23 | Deep traffic jam prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110697865.8A CN113469425B (en) | 2021-06-23 | 2021-06-23 | Deep traffic jam prediction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113469425A CN113469425A (en) | 2021-10-01 |
| CN113469425B true CN113469425B (en) | 2024-02-13 |
Family
ID=77869230
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110697865.8A Active CN113469425B (en) | 2021-06-23 | 2021-06-23 | Deep traffic jam prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113469425B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113936462A (en) * | 2021-10-18 | 2022-01-14 | 广州交信投科技股份有限公司 | Bus road condition prediction method and system based on ASTGCN algorithm |
| CN115063054B (en) * | 2022-08-17 | 2022-11-29 | 深圳市索菱实业股份有限公司 | Internet of vehicles big data analysis method and server |
| CN115620524B (en) * | 2022-12-15 | 2023-03-28 | 中南大学 | Traffic jam prediction method, system, equipment and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007103180A2 (en) * | 2006-03-03 | 2007-09-13 | Inrix, Inc. | Assessing road traffic conditions using data from mobile data sources |
| CN104091081A (en) * | 2014-07-15 | 2014-10-08 | 中国科学院自动化研究所 | Traffic data make-up method |
| CN109584553A (en) * | 2018-11-29 | 2019-04-05 | 中电海康集团有限公司 | A kind of section degree of association missing complementing method based on space time information |
| CN110555018A (en) * | 2019-07-29 | 2019-12-10 | 浙江工业大学 | traffic flow completion and prediction method |
| CN111694830A (en) * | 2020-06-12 | 2020-09-22 | 复旦大学 | Missing data completion method based on deep ensemble learning |
| CN111724601A (en) * | 2020-06-30 | 2020-09-29 | 西安电子科技大学 | Motor vehicle traffic flow prediction method suitable for data loss |
| CN111899510A (en) * | 2020-07-28 | 2020-11-06 | 南京工程学院 | Intelligent traffic system flow short-term prediction method and system based on divergent convolution and GAT |
| CN112085947A (en) * | 2020-07-31 | 2020-12-15 | 浙江工业大学 | Traffic jam prediction method based on deep learning and fuzzy clustering |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11715369B2 (en) * | 2016-08-15 | 2023-08-01 | University Of Southern California | Latent space model for road networks to predict time-varying traffic |
-
2021
- 2021-06-23 CN CN202110697865.8A patent/CN113469425B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007103180A2 (en) * | 2006-03-03 | 2007-09-13 | Inrix, Inc. | Assessing road traffic conditions using data from mobile data sources |
| CN104091081A (en) * | 2014-07-15 | 2014-10-08 | 中国科学院自动化研究所 | Traffic data make-up method |
| CN109584553A (en) * | 2018-11-29 | 2019-04-05 | 中电海康集团有限公司 | A kind of section degree of association missing complementing method based on space time information |
| CN110555018A (en) * | 2019-07-29 | 2019-12-10 | 浙江工业大学 | traffic flow completion and prediction method |
| CN111694830A (en) * | 2020-06-12 | 2020-09-22 | 复旦大学 | Missing data completion method based on deep ensemble learning |
| CN111724601A (en) * | 2020-06-30 | 2020-09-29 | 西安电子科技大学 | Motor vehicle traffic flow prediction method suitable for data loss |
| CN111899510A (en) * | 2020-07-28 | 2020-11-06 | 南京工程学院 | Intelligent traffic system flow short-term prediction method and system based on divergent convolution and GAT |
| CN112085947A (en) * | 2020-07-31 | 2020-12-15 | 浙江工业大学 | Traffic jam prediction method based on deep learning and fuzzy clustering |
Non-Patent Citations (4)
| Title |
|---|
| Intelligent Prediction of Mobile Vehicle Trajectory Based on Space-Time Information;Dong Guan;IEEE 89th Vehicular Technology Conference;全文 * |
| Short-Term Traffic Prediction Based on DeepCluster in Large-Scale Road Networks;Lingyi Han;IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY;第VOL. 68卷(第NO. 12期);全文 * |
| 城市实时交通流预测研究及应用;刘安琪;中国优秀硕士学位论文全文数据库;全文 * |
| 基于GRU-RNN模型的交叉口短时交通流预测研究;于德新;邱实;周户星;王卓睿;;公路工程(04);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113469425A (en) | 2021-10-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113469425B (en) | Deep traffic jam prediction method | |
| US11270579B2 (en) | Transportation network speed foreeasting method using deep capsule networks with nested LSTM models | |
| CN109285346B (en) | Urban road network traffic state prediction method based on key road sections | |
| CN110517482B (en) | Short-term traffic flow prediction method based on 3D convolutional neural network | |
| CN113313947A (en) | Road condition evaluation method of short-term traffic prediction graph convolution network | |
| CN113762595B (en) | Traffic time prediction model training method, traffic time prediction method and equipment | |
| CN116468186B (en) | Flight delay time prediction method, electronic equipment and storage medium | |
| CN112037539B (en) | Method and system for recommending signal control scheme for saturated urban traffic network | |
| CN113919231A (en) | PM2.5 concentration space-time change prediction method and system based on space-time diagram neural network | |
| CN114925836B (en) | Urban traffic flow reasoning method based on dynamic multi-view graph neural network | |
| CN101706888A (en) | Method for predicting travel time | |
| CN105471647A (en) | Power communication network fault positioning method | |
| CN114240796B (en) | Remote sensing image cloud and fog removing method, equipment and storage medium based on GAN | |
| CN115376317A (en) | Traffic flow prediction method based on dynamic graph convolution and time sequence convolution network | |
| CN115099328A (en) | Traffic flow prediction method, system, device and storage medium based on countermeasure network | |
| CN113627676A (en) | Traffic prediction method and system based on multi-attention causal relationship | |
| CN117610734A (en) | Deep learning-based user behavior prediction method, system and electronic equipment | |
| CN112465055A (en) | Network fault diagnosis method based on convolutional neural network | |
| CN115936238B (en) | Method, system, equipment and medium for predicting medium-long-term output of global wind power | |
| CN112907971B (en) | Urban road network short-term traffic flow prediction method based on genetic algorithm optimization space-time residual error model | |
| CN115905997A (en) | Wind turbine generator meteorological disaster early warning method and system based on prediction deviation optimization | |
| CN114611753A (en) | Flight airplane trajectory data prediction system and method | |
| CN113515890A (en) | Renewable energy day-ahead scene generation method based on federal learning | |
| CN114785701A (en) | Method and device for predicting optical transport network traffic, electronic device and storage medium | |
| CN113516163A (en) | Vehicle classification model compression method and device based on network pruning and storage medium |
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 |