CN107170235B - Traffic flow prediction time sequence method based on depth feature extraction network - Google Patents

Traffic flow prediction time sequence method based on depth feature extraction network Download PDF

Info

Publication number
CN107170235B
CN107170235B CN201710445486.3A CN201710445486A CN107170235B CN 107170235 B CN107170235 B CN 107170235B CN 201710445486 A CN201710445486 A CN 201710445486A CN 107170235 B CN107170235 B CN 107170235B
Authority
CN
China
Prior art keywords
traffic flow
time
series
road
feature extraction
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.)
Expired - Fee Related
Application number
CN201710445486.3A
Other languages
Chinese (zh)
Other versions
CN107170235A (en
Inventor
陈媛芳
蓝桂茂
陈法林
舒磊
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.)
Guangdong University of Petrochemical Technology
Original Assignee
Guangdong University of Petrochemical Technology
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 Guangdong University of Petrochemical Technology filed Critical Guangdong University of Petrochemical Technology
Priority to CN201710445486.3A priority Critical patent/CN107170235B/en
Publication of CN107170235A publication Critical patent/CN107170235A/en
Application granted granted Critical
Publication of CN107170235B publication Critical patent/CN107170235B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications

Landscapes

  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Traffic Control Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a traffic flow prediction time sequence method based on a depth feature extraction network, which is characterized in that a dynamic semantic graph of traffic flow correlation is obtained by obtaining traffic flow information of each road section and correlation between road sections, and a time sequence model of the traffic flow is predicted by the traffic flow of the depth feature extraction network. The invention is suitable for the optimal route design of the trip, and can play a certain role in relieving the traffic jam problem.

Description

Traffic flow prediction time sequence method based on depth feature extraction network
Technical Field
The invention relates to a traffic flow prediction time sequence method based on a depth feature extraction network, and belongs to the technical field of traffic flow prediction.
Background
Urban traffic is the life pulse of urban social economic activities, has important significance for promoting the development of urban economy and facilitating the travel of people, and along with the progress of scientific technology and the development of industry, the traffic volume in cities is increased rapidly, and the original traffic mode can not meet the requirements; meanwhile, as various vehicles are provided for urban traffic by industrial development, the development of urban traffic industry is accelerated.
However, when the city is developed, a series of problems such as traffic jam, frequent traffic accidents and the like are caused. The frequent occurrence of traffic accidents is often in the congested road section, so that the prediction of the traffic flow in the next time period is very important.
At present, the research in the technical field of large-scale traffic flow prediction at home and abroad is not deep enough, particularly, the feasibility research of a road construction project is carried out, only one or a plurality of directly influenced lines are usually considered, the research on the action of the constructed road in the whole road network is lacked, and the predicted traffic volume is often far away from the actual traffic volume.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a traffic flow prediction time sequence method based on a deep feature extraction network, which predicts the traffic flow by deep learning and can predict the traffic flow in real time.
In order to solve the technical problem, the invention provides a traffic flow prediction time sequence method based on a deep feature extraction network, which comprises the following steps:
1) acquiring a traffic flow correlation dynamic semantic graph transmitted along with time;
2) on the basis of obtaining a semantic graph of traffic flow correlation degree in the step 1), a traffic flow time sequence model of a deep feature extraction network is applied to predict the traffic flow of the next time period, and the specific steps are as follows:
2-1) assuming a traffic flow X at a series of times ttComprises the following steps:
Figure BDA0001320473300000011
wherein the content of the first and second substances,the traffic flow of the ith road section at the series time t, N represents the number of the road sections, N represents the number of the time, and the traffic flow of the kth road section
Figure BDA0001320473300000013
Influenced by the traffic flow of the previous road segment connected with the link, if the previous road segment connected with the link has m segments, and the traffic flow of each segment connected with the link contributes the traffic flow to the kth road segment, then:
Figure BDA0001320473300000021
wherein the content of the first and second substances,
Figure BDA0001320473300000022
the predicted traffic flow at the series of time instants t for the kth link,
Figure BDA0001320473300000023
is that all previous links connected to the kth link contribute to
Figure BDA0001320473300000024
The sum of the traffic flows of (a),
Figure BDA0001320473300000025
is an auxiliary parameter of the jth road section, the superscript t-1 is the previous series time of the series time t,
Figure BDA0001320473300000026
is the traffic flow, epsilon, of the jth road segment at the series time t-1tIs the noise at a series of times, epsilontIs normally distributed, i.e.
Figure BDA0001320473300000027
Figure BDA0001320473300000028
Is the variance;
2-2) estimating auxiliary parameters
Figure BDA0001320473300000029
And
Figure BDA00013204733000000210
a parameter value of (d);
2-3) optimization of auxiliary parameters
Figure BDA00013204733000000211
And predicts traffic flow.
The aforementioned traffic flow correlation dynamic semantic graph is defined in the same time period, if the traffic flow of one road segment affects the traffic flow of another road segment to a certain extent, there is correlation between the two road segments, the road segments with correlation are connected together, and the real-time traffic flow dynamic semantic graph can be obtained by dynamically updating the actually observed traffic flow every time period.
The time sequence model refers to observing and measuring a certain variable or a group of variables at a series of moments t1,t2,...,tnArranged in chronological order, a mathematical expression for explaining the interrelationship between the variables and the series of moments in time.
The deep feature extraction network mentioned above means that the computation involved in generating an output from an input can be represented by a flow graph, which is a graph capable of representing the computation, in which each node represents a basic computation and a computed value, and the computed result is applied to the values of the children of this node.
In the aforementioned step 2-2), the auxiliary parameters are estimated
Figure BDA00013204733000000212
And
Figure BDA00013204733000000213
the method of the parameter values of (1) is as follows:
suppose that
Figure BDA00013204733000000214
Representing the traffic flow for the 1 st link at the series time t-1,
Figure BDA00013204733000000215
representing the traffic flow for the 2 nd road segment at the series time t-1,
Figure BDA00013204733000000216
representing the traffic flow for the 3 rd road segment at the series time t,
Figure BDA00013204733000000217
indicating the traffic flow of the 4 th road segment at the series time t, the 1 st and 2 nd road segments having a correlation with the 3 rd road segment4 road sections have correlation degrees;
let the auxiliary parameters of the 1 st and 2 nd path flows to the 3 rd path be
Figure BDA00013204733000000218
Andthe auxiliary parameters of the 1 st and 2 nd path sections flowing to the 4 th path section are
Figure BDA00013204733000000220
And
Figure BDA00013204733000000221
then according to equation (2) there is:
Figure BDA00013204733000000222
Figure BDA0001320473300000031
and due to the traffic flow of the 1 st and 2 nd road segments
Figure BDA0001320473300000032
And
Figure BDA0001320473300000033
both assigned to the 3 rd and 4 th road segments, there are:
Figure BDA0001320473300000034
then
Figure BDA0001320473300000035
Figure BDA0001320473300000036
Then
Figure BDA0001320473300000037
Multiplying both sides of formula (3) by
Figure BDA0001320473300000038
Obtaining:
Figure BDA0001320473300000039
multiplying both sides of equation (4) by
Figure BDA00013204733000000310
Obtaining:
Figure BDA00013204733000000311
then, if the formula (7) is equal to (8), the
Figure BDA00013204733000000312
And
Figure BDA00013204733000000313
viewed as a variable, there are:
Figure BDA00013204733000000314
Figure BDA00013204733000000315
the combined type (5) and (9) are as follows:
Figure BDA00013204733000000316
Figure BDA00013204733000000317
the combined type (6) and (10) are as follows:
Figure BDA00013204733000000319
the parameters can be obtained by the method
Figure BDA00013204733000000320
Figure BDA00013204733000000321
Is the sequence of model residuals εtThe variance of (f) is, therefore:
Figure BDA00013204733000000322
estimate out
Figure BDA00013204733000000323
Can be estimated according to the above formula
Figure BDA00013204733000000324
The foregoing procedure for optimizing the auxiliary parameters is: substituting the traffic flow of the road section connected with the k-th road section at the previous series of time t-1 into the modelIn (1), is calculated to obtain
Figure BDA0001320473300000042
That is, the predicted value of the kth road section is selected, data of the same series of moments in multiple days are selected for multiple times of prediction, and each predicted value and corresponding historical observation dataMakingOperation of taking
Figure BDA0001320473300000045
Is the highest value ofCorresponding to hours
Figure BDA0001320473300000046
As an auxiliary parameter after optimization.
In order to prevent the predicted value and the historical observation data from being completely equal, a constant C is introduced to correct the overfitting on the basis of the formula (2), and the following steps are provided:
Figure BDA0001320473300000047
the invention has the beneficial effects that:
the method is applied to the optimal route design process of the trip, and can play a certain role in relieving the traffic jam problem; the traffic flow prediction time sequence model based on the depth characteristic extraction network can more accurately predict the current traffic flow of each road section through analyzing the large traffic flow data.
Drawings
FIG. 1 is a traffic flow correlation dynamic semantic graph;
FIG. 2 is an example of a timing model;
fig. 3 is a schematic diagram of a depth feature extraction network.
Detailed Description
The invention is further described below. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
The invention discloses a traffic flow prediction time sequence method based on a deep feature extraction network, which comprises the following steps:
the method comprises the following steps: acquiring a traffic flow correlation dynamic semantic graph transmitted along with time:
the number X of vehicles passing through each road segment in a certain time is set as the traffic flow of the time segment, so that each road segment has a specific traffic flow in each time segment, such as X, Y, Z, and the like. If the road sections do not interfere with each other, the traffic flow of each road section is constant within a certain time period, but actually, each road section is influenced by the traffic flows of other road sections communicated with the road section, so that the traffic flow of the road section changes, and the two road sections have a correlation degree if the traffic flow of one road section influences the traffic flow of the other road section to a certain extent within the same time period. And connecting the road sections with the correlation degrees with each other, and dynamically updating the actually observed traffic flow every time period T, so that a real-time traffic flow dynamic semantic graph can be obtained.
As shown in fig. 1: the figure depicts the traffic flow of A, B, C, D road segments in 4 time periods from bottom to top and each time period, with the first time period at the bottom and road segment A at the bottom1、B1、C1、D1Corresponding traffic flows are respectively a1、b1、c1、d1The upper layer is the second time period, and the road section is A2、B2、C2、D2Corresponding traffic flows are respectively a2、b2、c2、d2Similarly, the next layer is the third time period, and the road section is A3、B3、C3、D3Corresponding traffic flows are respectively a3、b3、c3、d3The next layer is a fourth time period, and the road section is A4、B4、C4、D4Corresponding traffic flows are respectively a4、b4、c4、d4. Suppose that a road segment A1And B2Communicate with each other, B2And C3Communicate with each other, C3And B4Communicating; at the same time, A1And C2Are also communicated with C2And D3Are in communication with each other, D3And D4Communication, known as A herein1Traffic flow a1Influence B2And C2Of traffic flow, i.e. B2And C2Traffic flow and A1Connecting the links with a certain degree of correlation, dynamically updating every time period T, and when T is 1(T is 1 refers to the first time period), A1Traffic flow a at crossing1Will choose to lead to B2Or C2(ii) a When T is 2, B2Traffic flow b obtained above2Will lead to C3，C2Traffic flow c obtained above2Will lead to D3(ii) a In the same way, when T is 3, C3Traffic flow c obtained above3Will lead to B4，D3Traffic flow d obtained above3Will lead to D4At this time, a flow transfer diagram which changes with time, namely a traffic flow correlation dynamic semantic diagram which is transferred with time, is obtained.
Step two: on the basis of obtaining the semantic graph of the traffic flow correlation degree in the first step, the traffic flow of the traffic graph in the next time period is predicted by applying a traffic flow time sequence model of a deep feature extraction network:
2-1) the definition of the timing model is: in production and scientific research, a certain variable or a group of variables are observed and measured at a series of moments t1,t2,...,tn(tiAs an argument) arranged in chronological order, a mathematical expression that may be used to interpret the interrelationship between the variable and a series of time instants.
Suppose a traffic flow X at a series of times ttComprises the following steps:
Figure BDA0001320473300000051
wherein the content of the first and second substances,
Figure BDA0001320473300000052
the traffic flow of the ith road section at the series time t, N represents the number of the road sections, N represents the number of the time, and the traffic flow of the kth road section
Figure BDA0001320473300000053
Influenced by the traffic flow of the previous road segment connected with the link, and if the previous road segment connected with the link has m sections, each section of the traffic flow connected with the link contributes a part of the traffic flow to the kth road segment, the following steps are carried out:
Figure BDA0001320473300000061
wherein the content of the first and second substances,
Figure BDA0001320473300000062
the predicted traffic flow at the series of time instants t for the kth link,is that all previous links connected to the kth link contribute to
Figure BDA0001320473300000064
The sum of the traffic flows of (a),is an auxiliary parameter of the jth road section, the superscript t-1 is the previous series time of the series time t,is the traffic flow, epsilon, of the jth road segment at the series time t-1tIs the noise at a series of times, epsilontIs normally distributed, i.e.
Figure BDA0001320473300000067
Figure BDA0001320473300000068
Is the variance.
2-2) parameter estimation: based on historical observation data
Figure BDA0001320473300000069
And
Figure BDA00013204733000000610
estimate out
Figure BDA00013204733000000611
And
Figure BDA00013204733000000612
the parameter value of (2). The method comprises the following steps:
as shown in fig. 2, in the figure,
Figure BDA00013204733000000613
representing the traffic flow for the 1 st link at the series time t-1,
Figure BDA00013204733000000614
representing the traffic flow for the 2 nd road segment at the series time t-1,
Figure BDA00013204733000000615
representing the traffic flow for the 3 rd road segment at the series time t,
Figure BDA00013204733000000616
indicating the traffic flow for the 4 th link at the series time t. In the figure, arrows indicate traffic flow directions, and it can be seen that the 3 rd link and the 4 th link are both affected by the traffic flow of the 1 st link and the 2 nd link, that is, the 1 st link and the 2 nd link have a correlation with the 3 rd link, and the 1 st link and the 2 nd link have a correlation with the 4 th link.
Let the auxiliary parameters of the 1 st and 2 nd path flows to the 3 rd path be
Figure BDA00013204733000000617
And
Figure BDA00013204733000000618
the auxiliary parameters of the 1 st and 2 nd path sections flowing to the 4 th path section are
Figure BDA00013204733000000619
And
Figure BDA00013204733000000620
then according to equation (2) there is:
Figure BDA00013204733000000622
the traffic flow in the formulas (3) and (4)
Figure BDA00013204733000000623
And
Figure BDA00013204733000000624
the specific numerical value of the historical observation data contains the noise parameter, so that the noise parameter epsilon is not required to be added in the formulas (3) and (4)tOtherwise, the calculation is repeated.
And due to the traffic flow of the 1 st and 2 nd road segments
Figure BDA00013204733000000625
And
Figure BDA00013204733000000626
both assigned to the 3 rd and 4 th road segments, there are:
Figure BDA00013204733000000627
then
Figure BDA00013204733000000628
Figure BDA00013204733000000629
Then
Figure BDA00013204733000000630
Multiplying both sides of formula (3) by
Figure BDA00013204733000000631
Obtaining:
multiplying both sides of equation (4) by
Figure BDA0001320473300000071
Obtaining:
Figure BDA0001320473300000072
then, if the formula (7) is equal to (8), the
Figure BDA0001320473300000073
And
Figure BDA0001320473300000074
viewed as a variable, there are:
Figure BDA0001320473300000075
Figure BDA0001320473300000076
the combined type (5) and (9) are as follows:
Figure BDA0001320473300000077
Figure BDA0001320473300000078
the combined type (6) and (10) are as follows:
Figure BDA0001320473300000079
Figure BDA00013204733000000710
the parameters can be obtained by the method
Figure BDA00013204733000000711
Figure BDA00013204733000000712
Is the sequence of model residuals εtThe variance of (f) is, therefore:
estimate out
Figure BDA00013204733000000714
Can be estimated according to the above formula
Figure BDA00013204733000000715
2-3) deep feature extraction network:
the computation involved in producing an output from an input can be represented by a flow graph, which is a graph that can represent the computation, in which each node represents a basic computation and a computed value, and the results of the computation are applied to the values of the children of that node.
2-3.1) deep feature extraction network As shown in FIG. 3, this means "feeding" data to the timing model and then continually optimizing the parameters to make the prediction of the model more accurate. Applying the depth feature extraction network to a traffic flow time sequence model, and inputting one
Figure BDA00013204733000000716
Through the step I, a dynamic semantic graph of the relevance of the traffic flow is obtained, and the traffic flow of the previous road section connected with the road section is found
Figure BDA0001320473300000081
According toAnd
Figure BDA0001320473300000083
estimating parameters from historical observation data
Figure BDA0001320473300000084
And
Figure BDA0001320473300000085
the obtained parameters and the traffic flow of the current timeSubstituting formula (2)In order to obtain the output layer
Figure BDA0001320473300000088
2-3.2) optimization of model parameters
Figure BDA0001320473300000089
And predicting traffic flow:
dividing one day into P moments, wherein each day has a series of moments t, excluding the influence of some non-ideal conditions, and assuming that the traffic flow of a road section at each day and at the same moment is relatively similar, for example, when people go to work and leave work on each day, the road section which people pass on and off work is basically constant, and the traffic flow of the k road section at the last series of moments t-1 of the road section connected with the k road section is substituted into a model
Figure BDA00013204733000000810
In (1), is calculated to obtainThat is, the predicted value of the kth road section is selected, data of the same series of moments in multiple days are selected for multiple times of prediction, and each predicted value and corresponding historical data are used for conducting multiple times of prediction
Figure BDA00013204733000000812
Operation of taking
Figure BDA00013204733000000813
The smaller the value of (A) is, the corresponding
Figure BDA00013204733000000814
Is optimizedUsing the optimization
Figure BDA00013204733000000816
And calculated
Figure BDA00013204733000000817
To predict the traffic flow at the present time.
If the traffic flow of the next series time of a road section is predicted, the traffic flow of the current time of all the road sections connected with the road section is only required to be predicted
Figure BDA00013204733000000818
The traffic flow of the next series of moments of the road section to be predicted can be calculated by substituting the calculation
Figure BDA00013204733000000819
In order to prevent the predicted value and the actual historical value from being completely equal, because the completely equal value is over-fitted, and therefore, on the basis of the formula (2), a constant C is introduced to correct the over-fitting, the following steps are provided:
by inputting actual observations
Figure BDA00013204733000000821
Finding out the dynamic semantic graph of the traffic flow at the t-1 moment of the last road section connected with each road section corresponding to each road section, and outputting the predicted dynamic semantic graph through a depth feature extraction network
Figure BDA00013204733000000822
The obtained time sequence model is used for predicting the traffic flow, so that a relatively accurate traffic flow value is predicted.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.

Claims (6)

1. The traffic flow prediction time sequence method based on the depth feature extraction network is characterized by comprising the following steps of:
1) acquiring a traffic flow correlation dynamic semantic graph transmitted along with time; the dynamic semantic graph of the traffic flow correlation degree is defined in the same time period, if the traffic flow of one road section influences the traffic flow of another road section to a certain extent, the two road sections have the correlation degree, the road sections with the correlation degree are connected with each other, and the actually observed traffic flow is dynamically updated every time period, so that the real-time dynamic semantic graph of the traffic flow can be obtained;
2) on the basis of obtaining a traffic flow correlation dynamic semantic graph in the step 1), a traffic flow time sequence model of a deep feature extraction network is applied to predict the traffic flow of the next time period, and the specific steps are as follows:
2-1) assuming a traffic flow X at a series of times ttComprises the following steps:
wherein the content of the first and second substances,
Figure FDA0002169054860000012
the traffic flow of the ith road section at the series time t, N represents the number of the road sections, N represents the number of the time, and the traffic flow of the kth road section
Figure FDA0002169054860000013
Influenced by the traffic flow of the previous road segment connected with the link, if the previous road segment connected with the link has m segments, and the traffic flow of each segment connected with the link contributes the traffic flow to the kth road segment, then:
Figure FDA0002169054860000014
wherein the content of the first and second substances,the predicted traffic flow at the series of time instants t for the kth link,
Figure FDA0002169054860000016
is that all previous links connected to the kth link contribute to
Figure FDA0002169054860000017
The sum of the traffic flows of (a),
Figure FDA0002169054860000018
is an auxiliary parameter of the jth road section, the superscript t-1 is the previous series time of the series time t,
Figure FDA0002169054860000019
is the traffic flow, epsilon, of the jth road segment at the series time t-1tIs the noise at a series of times, epsilontIs normally distributed, i.e.
Figure FDA00021690548600000110
Is the variance;
2-2) estimating auxiliary parameters
Figure FDA00021690548600000112
And
Figure FDA00021690548600000113
a parameter value of (d);
2-3) optimization of auxiliary parameters
Figure FDA00021690548600000114
And predicts traffic flow.
2. The depth feature extraction network-based traffic flow prediction time sequence method according to claim 1, wherein the time sequence model refers to observing and measuring a certain variable or a group of variables and performing a series of time t1,t2,...,tnArranged in chronological order, a mathematical expression for explaining the interrelationship between the variables and the series of moments in time.
3. The traffic flow prediction time-series method based on the deep feature extraction network according to claim 1, wherein the computation involved in generating an output from an input is represented by a flow graph, the flow graph is a graph capable of representing the computation, each node in the graph represents a basic computation and a value of the computation, and the result of the computation is applied to values of sub-nodes of the node.
4. The depth feature extraction network-based traffic flow prediction time sequence method according to claim 1, wherein in the step 2-2), auxiliary parameters are estimated
Figure FDA0002169054860000021
Andthe method of the parameter values of (1) is as follows:
suppose that
Figure FDA0002169054860000023
Representing the traffic flow for the 1 st link at the series time t-1,
Figure FDA0002169054860000024
representing the traffic flow for the 2 nd road segment at the series time t-1,
Figure FDA0002169054860000025
representing the traffic flow for the 3 rd road segment at the series time t,
Figure FDA0002169054860000026
representing the traffic flow of the 4 th road segment at the series time t, wherein the 1 st road segment and the 2 nd road segment have correlation with the 3 rd road segment, and the 1 st road segment and the 2 nd road segment have correlation with the 4 th road segment;
let the auxiliary parameters of the 1 st and 2 nd path flows to the 3 rd path be
Figure FDA0002169054860000027
And
Figure FDA0002169054860000028
the auxiliary parameters of the 1 st and 2 nd path sections flowing to the 4 th path section areAnd
Figure FDA00021690548600000210
then according to equation (2) there is:
Figure FDA00021690548600000211
Figure FDA00021690548600000212
and due to the traffic flow of the 1 st and 2 nd road segments
Figure FDA00021690548600000213
And
Figure FDA00021690548600000214
both assigned to the 3 rd and 4 th road segments, there are:
Figure FDA00021690548600000215
Figure FDA00021690548600000216
multiplying both sides of formula (3) by
Figure FDA00021690548600000217
Obtaining:
Figure FDA00021690548600000218
multiplying both sides of equation (4) byObtaining:
then, if the formula (7) is equal to (8), the
Figure FDA00021690548600000221
And
Figure FDA00021690548600000222
viewed as a variable, there are:
Figure FDA00021690548600000223
Figure FDA00021690548600000224
the combined type (5) and (9) are as follows:
Figure FDA00021690548600000225
Figure FDA0002169054860000031
the combined type (6) and (10) are as follows:
Figure FDA0002169054860000033
the parameters can be obtained by the method
Figure FDA0002169054860000035
Is the sequence of model residuals εtThe variance of (f) is, therefore:
Figure FDA0002169054860000036
estimate out
Figure FDA0002169054860000037
Can be estimated according to the above formula
Figure FDA0002169054860000038
5. The depth feature extraction network-based traffic flow prediction timing method according to claim 1, wherein the process of optimizing the auxiliary parameters is as follows: substituting the traffic flow of the road section connected with the k-th road section at the previous series of time t-1 into the model
Figure FDA0002169054860000039
In (1), is calculated to obtain
Figure FDA00021690548600000310
That is, the predicted value of the kth road section is selected, data of the same series of moments in multiple days are selected for multiple times of prediction, and each predicted value and corresponding historical observation data
Figure FDA00021690548600000311
Making
Figure FDA00021690548600000312
Operation of takingCorresponding to the value of (A) being minimum
Figure FDA00021690548600000314
As an auxiliary parameter after optimization.
6. The traffic flow prediction time sequence method based on the depth feature extraction network as claimed in claim 5, wherein in order to prevent the predicted value and the historical observation data from being completely equal, a constant C is introduced to correct the over-fitting on the basis of the formula (2), and then:
Figure FDA00021690548600000315
CN201710445486.3A 2017-06-13 2017-06-13 Traffic flow prediction time sequence method based on depth feature extraction network Expired - Fee Related CN107170235B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710445486.3A CN107170235B (en) 2017-06-13 2017-06-13 Traffic flow prediction time sequence method based on depth feature extraction network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710445486.3A CN107170235B (en) 2017-06-13 2017-06-13 Traffic flow prediction time sequence method based on depth feature extraction network

Publications (2)

Publication Number Publication Date
CN107170235A CN107170235A (en) 2017-09-15
CN107170235B true CN107170235B (en) 2020-03-03

Family

ID=59825863

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710445486.3A Expired - Fee Related CN107170235B (en) 2017-06-13 2017-06-13 Traffic flow prediction time sequence method based on depth feature extraction network

Country Status (1)

Country Link
CN (1) CN107170235B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115641718B (en) * 2022-10-24 2023-12-08 重庆邮电大学 Short-time traffic flow prediction method based on bayonet flow similarity and semantic association

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101673463A (en) * 2009-09-17 2010-03-17 北京世纪高通科技有限公司 Traffic information predicting method based on time series and device thereof
CN104882006A (en) * 2014-07-03 2015-09-02 中国科学院沈阳自动化研究所 Message-based complex network traffic signal optimization control method
CN105096614A (en) * 2015-09-23 2015-11-25 南京遒涯信息技术有限公司 Newly established crossing traffic flow prediction method based on generating type deep belief network
CN105160866A (en) * 2015-08-07 2015-12-16 浙江高速信息工程技术有限公司 Traffic flow prediction method based on deep learning nerve network structure

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7463937B2 (en) * 2005-11-10 2008-12-09 William Joseph Korchinski Method and apparatus for improving the accuracy of linear program based models

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101673463A (en) * 2009-09-17 2010-03-17 北京世纪高通科技有限公司 Traffic information predicting method based on time series and device thereof
CN104882006A (en) * 2014-07-03 2015-09-02 中国科学院沈阳自动化研究所 Message-based complex network traffic signal optimization control method
CN105160866A (en) * 2015-08-07 2015-12-16 浙江高速信息工程技术有限公司 Traffic flow prediction method based on deep learning nerve network structure
CN105096614A (en) * 2015-09-23 2015-11-25 南京遒涯信息技术有限公司 Newly established crossing traffic flow prediction method based on generating type deep belief network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
"Deep Architecture for Traffic Flow Prediction：Deep Belief Networks With Multitask Learning";W Huang et.al;《IEEE Transactions on Intelligent Transportation Systems》;20141031;正文全文 *
"基于时空特性的城市道路短时交通流预测研究";邱世崇;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;20160415;正文全文 *

Also Published As

Publication number Publication date
CN107170235A (en) 2017-09-15

Similar Documents

Publication Publication Date Title
Cigizoglu et al. Rainfall-runoff modelling using three neural network methods
Abdi et al. Forecasting of short-term traffic-flow based on improved neurofuzzy models via emotional temporal difference learning algorithm
Zaman Zad Ghavidel et al. Application of different data-driven methods for the prediction of total dissolved solids in the Zarinehroud basin
CN104408913B (en) A kind of traffic flow three parameter real-time predicting method considering temporal correlation
CN107274030B (en) Runoff Forecast method and system based on hydrology variable year border and monthly variation characteristic
Galavi et al. Klang River–level forecasting using ARIMA and ANFIS models
CN111652425A (en) River water quality prediction method based on rough set and long and short term memory network
Samantaray et al. Evaluation of suspended sediment concentration using descent neural networks
CN108877224B (en) A kind of Short-time Traffic Flow Forecasting Methods carrying out Estimating Confidence Interval
CN104050547A (en) Non-linear optimization decision-making method of planning schemes for oilfield development
CN113326919A (en) Traffic travel mode selection prediction method based on computational graph
CN116307152A (en) Traffic prediction method for space-time interactive dynamic graph attention network
CN107170235B (en) Traffic flow prediction time sequence method based on depth feature extraction network
Li et al. Urban water consumption prediction based on CPMBNIP
CN113343601A (en) Dynamic simulation method for water level and pollutant migration of complex water system lake
TW201738859A (en) Speed prediction method
Xie et al. Surface water quality evaluation based on Bayesian network
Xie et al. A method of flood forecasting of chaotic radial basis function neural network
Joshi et al. Rainfall-runoff modeling using Artificial Neural Network (a literature review)
Lyu et al. Water level prediction model based on GCN and LSTM
Chunmei et al. The research of method of short-term traffic flow forecast based on ga-bp neural network and chaos theory
Zhang et al. A refined rank set pair analysis model based on wavelet analysis for predicting temperature series
Wu et al. An attention mechanism-based method for predicting traffic flow by GCN
Aljumaily Predicating the Durations of Irregation Channels Projects in Iraq By Using Ann Modelling
Li et al. Short-Time Traffic Flow Prediction Based on K-means++ and LSTM

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
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20200303

Termination date: 20210613

CF01 Termination of patent right due to non-payment of annual fee