CN111951553B - Prediction method based on traffic big data platform and mesoscopic simulation model - Google Patents

Prediction method based on traffic big data platform and mesoscopic simulation model Download PDF

Info

Publication number
CN111951553B
CN111951553B CN202010825580.3A CN202010825580A CN111951553B CN 111951553 B CN111951553 B CN 111951553B CN 202010825580 A CN202010825580 A CN 202010825580A CN 111951553 B CN111951553 B CN 111951553B
Authority
CN
China
Prior art keywords
traffic
model
data
traffic flow
flow
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
CN202010825580.3A
Other languages
Chinese (zh)
Other versions
CN111951553A (en
Inventor
吴超腾
高霄
沈丹凤
蔡章辉
张璐
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.)
Shanghai Seari Intelligent System Co Ltd
Original Assignee
Shanghai Seari Intelligent System Co Ltd
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 Shanghai Seari Intelligent System Co Ltd filed Critical Shanghai Seari Intelligent System Co Ltd
Priority to CN202010825580.3A priority Critical patent/CN111951553B/en
Publication of CN111951553A publication Critical patent/CN111951553A/en
Application granted granted Critical
Publication of CN111951553B publication Critical patent/CN111951553B/en
Active 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/18—Network design, e.g. design based on topological or interconnect aspects of utility systems, piping, heating ventilation air conditioning [HVAC] or cabling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • 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
    • 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
    • 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

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Chemical & Material Sciences (AREA)
  • Geometry (AREA)
  • Analytical Chemistry (AREA)
  • Business, Economics & Management (AREA)
  • Economics (AREA)
  • Computer Hardware Design (AREA)
  • General Engineering & Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computational Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Mathematical Optimization (AREA)
  • Development Economics (AREA)
  • Mathematical Analysis (AREA)
  • Game Theory and Decision Science (AREA)
  • Artificial Intelligence (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention is suitable for the field of traffic information service and application, and provides a mesoscopic traffic simulation prediction method based on a big data platform, which comprises the following steps: the system comprises a big data platform construction module, a simulation basic platform construction module, an OD detection module, an OD correction module, a flow velocity model calibration module, a state estimation module and a deduction module based on Kalman filtering. The simulation system can fully utilize and mine the value of multi-source data, improve OD and traffic flow data accuracy based on data fusion and error estimation, provide a basis for restoration and prediction of traffic states based on supply and demand, and realize traffic flow simulation deduction under short-term and control conditions.

Description

Prediction method based on traffic big data platform and mesoscopic simulation model
Technical Field
The invention relates to a traffic prediction method based on a traffic big data platform and a mesoscopic simulation model, and belongs to the technical field of intelligent traffic information service and application.
Background
In order to solve the problems of traffic jam, environmental pollution and the like caused by the demand generated in the economic development process and the increase of the reserved quantity of cars, various traffic management departments build a large number of traffic guidance and management systems. On one hand, the system construction improves the efficiency of management and control measures such as traffic operation on-line monitoring, traffic guidance, ramp management and the like; and on the other hand, a great amount of traffic flow and traffic accident information is accumulated. Due to the cause of traffic congestion, the balance of traffic supply and demand caused by the need of traffic planning and traffic design for relieving and solving, and the seamless cooperation of each traffic management system, the problem of traffic congestion which can be solved or relieved by only depending on a single traffic management system is very limited. In order to provide quantitative decision basis for traffic organization management, traffic event early warning and evaluation, traffic infrastructure planning construction and traffic policy feasibility analysis research in traffic planning, design and refined cooperation management, traffic flow and service management data resources of each traffic management platform need to be integrated, a real traffic system and a management scheme are integrated and modeled based on a traffic simulation model, the effect of various measures is predicted and evaluated, and then basis is provided for decision.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: decisions in traffic planning, traffic design and collaborative management are based on experience and have no quantitative scientific basis.
In order to solve the technical problems, the technical scheme of the invention is to provide a modeling and predicting method based on a traffic big data platform and a mesoscopic simulation model, which is characterized by comprising the following steps:
the method comprises the following steps of acquiring a training set for training a mesoscopic simulation model by using a traffic big data platform, and predicting the traffic state of the trained mesoscopic simulation model based on data acquired by the traffic big data platform in real time, wherein:
the construction of the observation simulation model and the prediction by using the observation simulation model comprise the following steps:
step 1, model selection
The mesoscopic model divides the whole road into N sections of road sections by adopting traffic flow discrete state equations shown in the following formulas (1) and (2), and the length of the ith section of road section is delta i The head and tail ends of each road section are respectively provided with a traffic detector to provide measured traffic flow q in a detection period T i (k) Average velocity v i (k) Data as input to the mesoscopic model, q i (k) Traffic flow, v, for the i-th to i + 1-th route section at time kT i (k) The spatial average speed of the traffic flow at the kT moment of the ith road section comprises the following steps:
q i (k)＝αρ i (k)v i (k)+(1-α)ρ i+1 (k)v i+1 (k) (1)
Figure BDA0002636087430000021
Figure BDA0002636087430000022
Figure BDA0002636087430000023
Figure BDA0002636087430000024
in formulas (1), (2), (3), (4): ρ is a unit of a gradient i (k) The density of the traffic flow at the kT moment of the ith road section; r is i (k) The flow rate of the on-ramp vehicle at the kT time of the ith road section; s i (k) The flowrate of the off-ramp vehicle at the kT time of the ith road segment; v. of f Representing the free-running speed of the traffic flow; rho cr Representing the critical density, namely the density of the traffic flow when the traffic flow reaches the maximum; b. tau, gamma, delta, lambda and alpha are adjustment coefficients of the equation; ρ represents a density; v (ρ) represents a velocity;
step 2, state estimation model
And (3) estimating the error by adopting a Kalman filter, wherein for a nonlinear system:
x(k+1)＝f[x(k)]+Г[x(k)]w(k) (5)
y(k+1)＝h[x(k+1)]+v(k+1) (6)
in the formulas (5) and (6), w (k) and v (k) are zero mean noise vectors
Figure BDA0002636087430000025
Figure BDA0002636087430000026
Q (k) is a system noise variance matrix, and R (k) is a measurement noise matrix; r [ x (k)]Is a noise drive matrix; x (k) is a state variable; y (k) is an observed value; f [ x (k)]Is a macroscopic traffic flow model value; h [ x (k)]Is an observation model value;
will be originally systematic in
Figure BDA0002636087430000027
The process is linearized and then carried out,
Figure BDA0002636087430000028
representing the noisy traffic flow model estimate at time k, the resulting extended Kalman filter consists of the following set of recursive equations:
Figure BDA0002636087430000029
Figure BDA00026360874300000210
Figure BDA0002636087430000031
Figure BDA0002636087430000032
Figure BDA0002636087430000033
wherein:
Figure BDA0002636087430000034
Figure BDA0002636087430000035
in the formulae (7) to (13),
Figure BDA0002636087430000036
the estimated value of the traffic flow model with noise at the moment k;
Figure BDA0002636087430000037
the traffic flow model estimation value with noise at the next moment k +1 is obtained;
Figure BDA0002636087430000038
is the traffic flow model at time k;
Figure BDA0002636087430000039
an observation model estimation value with noise at the moment k;
Figure BDA00026360874300000310
to observe the model at time k; k (K + 1) is the Kalman gain at the next moment K + 1; y (k) is an observed value at time k; p (k +1 k) is a system state transition covariance matrix when the moment k is transferred to the next moment k + 1; r (k + 1) is an observation noise matrix at the next moment k + 1; p (k) is a state transition covariance matrix at the moment k; q (k + 1) is a system noise variance matrix at the next moment k + 1; x is a radical of a fluorine atom 1 (k),…,x 2N (k) Values of variables for each state; f. of 1 ,…,f 6 Is a Jacobian matrix value of the traffic flow model; h is 1 、h 2 A Jacobian matrix value of the observation model;
Figure BDA00026360874300000311
for each state variable estimate when transitioning from time k to the next time k + 1;
given the initial filtered value
Figure BDA00026360874300000312
After the covariance matrix P (0) is initially filtered, gradually calculating according to the calculation sequence of the determined recursion algorithm, and estimating the system state;
and 3, taking the traffic density of the N sections of road sections at the moment k and the traffic speed of the N sections of road sections at the moment k as state variables x n (k) N =1,2, \ 8230;, 2N, training is constructed by using measured values of the flow rate and the average speed of the head end of the first road section in the N-section road sections as input quantities, and the density and the average speed of the N-th road section in the N-section road sections as output quantitiesCalibrating parameters of the center view model by an exercise set;
step 4, calibrating parameters of an observation dynamic flow model in the traffic flow:
parameter rho of the mesoscopic model cr 、v f And b, learning by using an online learning method according to a flow density formula:
Figure BDA0002636087430000041
according to the actually measured flow and speed data, the rho is learned through a genetic algorithm cr 、v f B, the value of;
and 5: the view dynamic model of the traffic flow obtains the evolution of the traffic flow on the road in time and space;
and 6: calculating an observation model:
firstly, an observation matrix is calculated, and then the observation matrix is multiplied by a state variable to obtain an observation estimated value
Figure BDA0002636087430000045
And 7: calculating a state transition matrix F:
Figure BDA0002636087430000042
wherein
Figure BDA0002636087430000043
Figure BDA0002636087430000044
And 8: the values of the state transition covariance matrix Q, the observed noise variance matrix R are determined, and the initial values of the state covariance matrix P are determined, which are generally assigned empirically.
And step 9: performing extended Kalman Filter calculation: performing iterative computation of extended Kalman filtering according to equations (5) to (13) in the extended Kalman filtering EKF state estimation principle to obtain a final solution;
step 10: the OD and the traffic capacity in a future period are estimated based on the influence analysis and calibration of traffic accidents and control measures, and the traffic state in the future period can be predicted by repeating the processes.
Preferably, the construction of the traffic big data platform comprises the following steps:
step 1, automatic modeling of a simulation area road network and a traffic improvement scheme traffic simulation scene based on a standardized map layer: establishing an integrated data standard according to a person-vehicle-road-traffic environment to generate a unified road network GIS layer; modeling traffic scheme scene data such as traffic organization, induction, management and control and the like, and performing spatial association with a road object;
step 2, designing a standardization layer based on the converged traffic demands, traffic flow data and vehicle data, and extracting, converting and loading scattered, isolated and irregular collected data of each platform to form uniform, standard and layered dynamic original data;
step 3, aiming at the data loss, inconsistency and error problem characteristics in the dynamic original data, combining data conditions, designing a data quality index and a discrimination method, realizing comprehensive identification on the data quality problem, and repairing the problem data based on the spatial correlation and node conservative principle to form complete, consistent and accurate dynamic data;
step 4, establishing a microscopic index system, a mesoscopic index system and a mesoscopic index system for simulation input and output, uniformly defining each index, and designing a standardized processing method for each index;
step 5, designing a simulation basic platform to build for an actual simulation modeling prediction scene, verifying the platform precision, and finally realizing index output and application flow;
and 6, designing a service scene, and butting simulation output and user requirements.
Compared with the prior art, the traffic simulation prediction method provided by the invention utilizes the relation between traffic supply and demand and road section observed quantity on the basis of fully utilizing and mining the value of multi-source data, generates road network traffic demand, traffic flow indexes and OD (origin-destination) data on the basis of partial detection data, realizes short, medium and long-term traffic flow prediction under the condition of no intervention and various control conditions, and evaluates the effects of various schemes.
Other beneficial effects brought by the traffic simulation method also comprise the following contents:
(1) Fully dig according to current facility potentiality, effectively reduce government's traffic foundation investment, turn to the higher traffic control level of "prevention": the current situation and the traffic plan are subjected to dynamic traffic simulation, and the optimization scheme of the mesoscopic simulation system is established according to the experience of developed foreign cities under the condition of reaching the same target, so that the time can be saved by at least 40 percent, and the investment can be saved by more than 50 percent.
(2) Realize intelligent transportation system integration and linkage, helping hand intelligent transportation system construction and development: the modeling is carried out for urban traffic operation, modeling tracking is carried out on each running vehicle, the digitization of the real world is realized, and a tamping foundation is built for an intelligent traffic system.
(3) Transition of traffic from passive bold management to active fine management: the evaluation of the traffic state is changed from the observation level to the middle microscopic level, the basis of the traffic evaluation can be changed from the original local traffic survey to the simulation according to the regional road network model, the cost can be effectively saved, and more precise and real-time traffic management can be realized.
Drawings
FIG. 1 is a traffic simulation model scheme based on a big data platform;
FIG. 2 is an algorithm framework;
FIG. 3 is a traffic mesoscopic simulation prediction implementation route;
FIG. 4 is a GIS layer of the expressway network;
FIG. 5 is a GIS layer of the Zhonghuajia region;
FIG. 6 is a diagram of detection device location information;
FIG. 7 is a comparison of license plate matching flow and coil collection flow;
FIG. 8 shows the distribution of license plate OD of an upstream main line;
FIG. 9 shows the distribution of the license plate OD on the upstream ramp;
FIG. 10 is a flow density velocity calibration;
FIG. 11 is a flow simulation deduction;
fig. 12 is a velocity simulation deduction.
Detailed Description
The invention will be further illustrated with reference to the following specific examples. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the present invention. Further, it should be understood that various changes or modifications of the present invention can be made by those skilled in the art after reading the teaching of the present invention, and these equivalents also fall within the scope of the claims appended to the present application.
The invention discloses a modeling and predicting method based on a traffic big data platform and a mesoscopic simulation model.
The technical scheme disclosed by the invention comprises the following two aspects:
in a first aspect, the construction of a big data platform comprises the following steps:
step 1, automatically modeling a simulation area road network and a traffic improvement scheme traffic simulation scene based on a standardized map layer. Establishing an integrated data standard according to a human-vehicle-road-traffic environment to generate a unified road network GIS layer; and modeling traffic scheme scene data such as traffic organization, induction, management and control and the like, and performing spatial association with the road object.
And 2, designing a standardization layer based on the converged traffic demands, traffic flow data and vehicle data, and extracting, converting and loading scattered, isolated and irregular collected data of each platform to form uniform, standard and layered dynamic original data.
And 3, aiming at the data loss, inconsistency and error problem characteristics in the dynamic original data, designing a data quality index and a judgment method by combining data conditions, realizing comprehensive identification on the data quality problem, and repairing the problem data based on the spatial correlation and node conservative principle to form complete, consistent and accurate dynamic data.
And 4, establishing a micro, mesoscopic and mesoscopic index system for simulation input and output, uniformly defining each index, and designing a standardized processing method for each index.
And 5, designing a simulation basic platform to build for an actual simulation modeling prediction scene, verifying the platform precision, and finally realizing index output and an application process.
And 6, designing a service scene, and butting simulation output and user requirements.
In a second aspect, a traffic flow simulation deduction method:
step 1, model selection
The mesoscopic model adopts the traffic flow discrete state equations of scholars such as Payne and Papageorgiou, and is shown in the following formula (1) and formula (2). Consider a piece, which is suitably divided into N segments, each segment length Δ i In the order of hundreds of meters. The head and tail ends of the whole road are provided with traffic detectors which provide measured data as the input of a model, including traffic flow q i (k) Average velocity v i (k) .1. The The detection period T is 10-60 seconds.
q i (k)＝αρ i (k)v i (k)+(1-α)ρ i+1 (k)v i+1 (k) (1)
Figure BDA0002636087430000071
Figure BDA0002636087430000072
Figure BDA0002636087430000073
Figure BDA0002636087430000074
In formulas (1), (2), (3), (4): rho i (k) The density of the traffic flow at the kT moment of the ith road section; v. of i (k) The space average speed of the traffic flow at the kT moment of the ith road section (note: the space average speed of the traffic flow speed in the invention); q. q.s i (k) The traffic flow from the ith road section to the (i + 1) th road section at the moment of kT; delta i The length is a spatial sampling length, namely the length of the ith road section; r is i (k) The flow rate of the on-ramp vehicle at the kT time of the ith road section; s i (k) The flowrate of the traffic on the exit ramp at the kT time of the ith road section; v. of f Representing the free-running speed of the traffic flow; rho cr Representing the critical density, namely the density of the traffic flow when the traffic flow reaches the maximum; b. tau, gamma, delta, lambda and alpha are adjustment coefficients of the equation; ρ represents a density; v (ρ) represents a vehicle speed. Equation (4) is a steady-state traffic flow model describing the speed-density relationship of steady-state traffic flow.
Step 2, state estimation model
Because the acquired data has errors, the errors need to be estimated. The Kalman filter is a minimum variance filter in a linear system, for which there are:
x(k+1)＝f[x(k)]+Г[x(k)]w(k) (5)
y(k+1)＝h[x(k+1)]+v(k+1) (6)
in the formulas (5) and (6), w (k) and v (k) are zero-mean noise vectors, and
Figure BDA0002636087430000081
Figure BDA0002636087430000082
q (k) is a system noise variance matrix, and R (k) is a measurement noise matrix; r [ x (k)]Is a noise drive matrix; x (k) is a state variable; y (k) is an observed value; f [ x (k)]Is a macroscopic traffic flow model value; h [ x (k)]To observe the model value.
Will be originally systemized in
Figure BDA0002636087430000083
The process is linearized in the process of the chemical synthesis,
Figure BDA0002636087430000084
representing the noisy traffic flow model estimate at time k, resulting in an Extended Kalman Filter (EKF) consisting of the following set of recursion equations:
Figure BDA0002636087430000085
Figure BDA0002636087430000086
Figure BDA0002636087430000087
Figure BDA0002636087430000088
Figure BDA0002636087430000089
wherein:
Figure BDA00026360874300000810
Figure BDA00026360874300000811
in the formulae (7) to (13),
Figure BDA00026360874300000812
the estimated value of the traffic flow model with noise at the moment k;
Figure BDA00026360874300000813
the traffic flow model estimation value with noise at the next moment k +1 is obtained;
Figure BDA00026360874300000814
is the traffic flow model at time k;
Figure BDA00026360874300000815
an observation model estimation value with noise at the moment k;
Figure BDA00026360874300000816
to observe the model at time k; k (K + 1) is the Kalman gain at the next moment K + 1; y (k) is an observed value at time k; p (k +1 k) is a system state transition covariance matrix when transitioning from the time k to the next time k + 1; r (k + 1) is an observation noise matrix at the next moment k + 1; p (k) is a state transition covariance matrix at the moment k; q (k + 1) is a system noise variance matrix at the next moment k + 1; x is the number of 1 (k),…,x 2N (k) Values of variables for each state; f. of 1 ,…,f 6 The Jacobian matrix value of the traffic flow model is obtained; h is a total of 1 、h 2 Is the Jacobian matrix value of the observation model;
Figure BDA0002636087430000091
for each state variable estimate at the transition from time k to the next time k + 1.
Given the initial filtered value
Figure BDA0002636087430000092
And after the initial filtering covariance matrix P (0), the calculation order of the recursion algorithm is:
Figure BDA0002636087430000093
P(0)->p (1; 0) and
Figure BDA0002636087430000094
and
Figure BDA0002636087430000095
and P (2 < 1 >)>\8230, and the method can be used for estimating the system state step by step and realizing a recurrence equation on line.
Step 3, algorithm framework
Taking an express way with three equal-length road sections, an entrance ramp and an exit ramp as an example, the traffic flow state estimation algorithm framework is explained in detail with reference to fig. 2.
The road section has 3 road sections and 4 cross sections. Only the head end of the first link (i.e., section 1) and the end of the third link (i.e., section 4) are equipped with detection devices that provide flow, average speed detection data as input and output quantities (observed values) of the traffic flow state estimation problem. Taking the measured values of the flow q0 and the average velocity v0 at the section 1 (the head end of the first section) as input quantities; taking the density ρ of the section 4 (end of the third section) 3 Average velocity v 3 The measured value of (d) is used as the output quantity. The purpose of the traffic flow state estimation is to accurately estimate the traffic variable value of the road section without the detector in the middle, such as the speed, the density and the flow at the section 1 and the section 2, through filtering operation according to the measured value of the detector.
According to the mathematical model of the extended Kalman filter and the mesoscopic traffic flow dynamic model, the number of state variables x (k) is 6, and the state variables are as follows:
Figure BDA0002636087430000096
there are 4 input variables, which are:
Figure BDA0002636087430000097
wherein ρ 1 (k)、ρ 2 (k)、ρ 3 (k) The traffic flow densities of the first road section, the second road section and the third road section at the moment k are respectively; v. of 1 (k)、v 2 (k)、v 3 (k) The traffic flow speed of the first road section, the second road section and the third road section at the moment k respectively is unknown, and what the model needs to solve is rho at each moment 1 、ρ 2 、ρ 3 、v 1 、v 2 、v 3 The value is obtained. v. of 0 (k) And q is 0 (k) The velocity and the flow rate at the cross section 1 are known as observed values. r (k) and s (k) are the ratios of the traffic flows on the on-ramp and the off-ramp, which are derived from past historical data and experience, and these four variables are used as the inputs of the model.
And 4, step 4: calibrating parameters of an observation dynamic flow model in traffic flow:
the traffic flow model used in this study is rich in superparameters, including ρ cr 、v f B, τ, γ, δ, λ, α, etc., where the 3 key parameters that have significant impact on the model are: ρ is a unit of a gradient cr 、v f And b, because the 3 parameters determine the traffic capacity, the values of the parameters of the dynamic model in the traffic flow are greatly different for two different states of congestion and unblocked, and therefore an online learning method is needed for parameter learning.
According to the flow density formula:
Figure BDA0002636087430000101
rho can be learned through a genetic algorithm according to the actually measured flow and speed data of the previous 30 minutes cr 、v f B, value of b.
And 5: calculating the evolution process of the traffic flow: according to the dynamic model of the traffic flow in the view, the evolution of the traffic flow in the space and time on the road can be obtained, and the evolution is as follows:
Figure BDA0002636087430000102
Figure BDA0002636087430000103
Figure BDA0002636087430000104
Figure BDA0002636087430000105
Figure BDA0002636087430000111
Figure BDA0002636087430000112
in equations (14) to (19), Δ is a calculation time interval.
Step 6: calculating an observation model:
firstly, an observation matrix is calculated:
Figure BDA0002636087430000113
then multiplying the observation matrix by the state variable to obtain an observation estimated value
Figure BDA0002636087430000117
And 7: calculating a state transition matrix F:
Figure BDA0002636087430000114
wherein
Figure BDA0002636087430000115
Figure BDA0002636087430000116
And step 8: the values of the state transition covariance matrix Q, the observed noise variance matrix R, and the initial values of the state covariance matrix P are determined, and these values are generally assigned empirically.
And step 9: performing extended Kalman Filter calculation: and (4) performing iterative computation of the extended Kalman filtering according to equations (5) to (13) in the extended Kalman filtering EKF state estimation principle to obtain a final solution.
Step 10: the OD and the traffic capacity in a future period are estimated based on the influence analysis and calibration of traffic accidents and control measures, and the traffic state in the future period can be predicted by repeating the processes.
The invention is further illustrated below by a specific example:
the method comprises the following steps of a first stage, big data platform construction and simulation input data processing:
step 1, building a simulation initial platform, including two aspects of a road network and a service scene. And extracting the road network range from the GIS layer as a simulation model road network modeling construction according to the road network range of the simulation actual region. The GIS layer comprises element contents such as road sections, nodes, intersections, steering relations and the like, and is uniformly converted according to element definitions based on the navigation node layer. The simulation scene is provided or researched and obtained by a user according to the user requirements, and generally comprises traffic events or traffic control strategies such as traffic guidance, ramp control, road construction and restriction. Taking the control simulation of the Hujia regional ramp in the expressway as an example, as shown in fig. 4 and 5, the road network modeling can be completed by extracting the corresponding spatial part from the unified GIS layer of the expressway according to the spatial range of the service scene.
And 2, verifying and perfecting the simulation platform. Based on the large data platform convergence and processing of traffic demand (OD data), traffic flow data and a road network model which are subjected to quality control, verification and correction are carried out on an initial OD (origin-destination) based on the relation between the traffic demand and the road section flow, the relation between flow-density-speed and the relation between traffic flow evolution and the road section speed change, the relation between the flow density and the speed is calibrated, and estimation, error estimation and correction are carried out on the road section speed based on a traffic distribution-path selection-deduction model.
For the Zhonghuajia area as an example, license plate recognition devices are arranged at the entrance and exit of a main line of the area and on upper and lower ramps, and annular coils and radar traffic flow detection devices are arranged at intervals of 200 meters on the main line, and the device distribution is shown in fig. 6. The license plate data can obtain a single vehicle license plate, the section OD can be obtained through matching, and the section flow and speed statistical information can be obtained through the annular coil and the radar. The data collected by the equipment are converged to a big data platform according to the flow to be uniformly stored, standardized, quality controlled and index data, and an interface is provided for the simulation model.
Because the license plate data is influenced by weather and illumination environment, the license plate data is lost to a certain extent, the coil and the radar are not influenced by the weather and the illumination environment, the flow acquisition precision is high, and the conditions of the license plate and the coil acquisition flow and errors are shown in figure 7. Therefore, the OD acquired by the license plate is the initial OD, and the high-precision road section flow acquired by the annular coil and the radar is corrected. The OD proportion can be calculated through OD flow of each path of an upstream main line and a ramp by matching license plates on the cross sections of the upstream license plate and the downstream license plate, and the OD flow of the region can be calculated through the flow fusion of the coils, which is shown in figures 8 and 9.
And in the second stage, deduction model calibration and simulation prediction application:
step 1, test protocol
Taking the Zhonghua Shanghai area as an example, the road section has 5 sections, coil detection devices are uniformly distributed on the sections, and the time period is selected from 3 months in 2020, 16 days in the morning and 6 am: 00-7: 00, measured data comprises the speed, density and flow of 5 sections. The accuracy of the model can be verified by comparing the model estimation value with the actual measurement value, and analyzing the model estimation value, which is obtained by operating the traffic flow state estimation model with the measurement speed and flow rate of the sections 1, 3, 4, and 5 in fig. 1 as the input of the model and the measurement speed and density of the section 2 as the observed value of the model, and with the time interval of 20 seconds in 1 hour.
Step 2, calibrating model parameters
For this case section, the distance per section is 400 meters, i.e., Δ =400m; the time interval is 20 seconds, i.e. T =20s; the adjustment coefficients of the viewing dynamic model in the traffic flow are tau =0.027, gamma =0.9, delta =1, lambda =5 and alpha =0.97. The other parameter fitting method adopts a genetic algorithm, and the calibration steps are as follows:
step a1, determining lane traffic flow parameter initial population, wherein the size of the population is 20, and the scale is shown in the following table:
Figure BDA0002636087430000131
and a2, determining a fitness function, and fitting the speed and the actual speed standard deviation by using a traffic flow model.
Figure BDA0002636087430000132
Wherein λ (i, d) is fitness, i is lane number, d is date, t is time stamp, n is sample time sequence number in one day, V (i, t) is actual acquisition speed, V (i, t) is time sequence number in one day, and ′ and (i, t) fitting the speed by using a traffic flow model.
And a3, determining a population updating rule, reserving 5 population samples with the highest fitness value, eliminating 5 population samples with the lowest fitness value, and randomly generating 5 new population samples. For the 10 samples with the fitness value in the middle, the parameter average is taken two by two to yield 10 samples.
A4, model iteration is carried out according to a population updating rule;
step a5, determining an iteration termination condition, wherein the difference value of the model parameter with the lowest fitness value in the two previous and next times is smaller than a specified value, the difference value of the free flow speed is smaller than 1, the difference value of the critical density is smaller than 1, the difference value of the index parameter is smaller than 0.05, and the result output is updated to a database.
Step a5, outputting a fitting result
For this segment, the vehicle speed v learned in the congestion time segment f =80, critical density ρ cr ＝50、b＝1.8。
Step 3, model precision evaluation
Fig. 2 is a comparison between the density extended kalman filter model value and the measured value of the traffic flow of section 2 in fig. 1. The abscissa is time, from 6 in 2019, 17, 6:00 start, 6 month by 2019, 17 day 7:00 ends with a time interval of 20 seconds. The ordinate is the density of the section of the traffic stream, the blue solid line is the coil measured value, the red dotted line is the extended kalman filter estimation reduction value, and it can be seen from the figure that the trends of the two values are relatively close and consistent, and meanwhile, according to the calculation, the average absolute error (MAE) between the model value and the measured value is 6.2, and the Root Mean Square Error (RMSE) is 7.8.
Fig. 3 is a comparison of the model estimated value and the observed value of the traffic flow velocity at the section 3 in fig. 1. The abscissa is time, from 6 months 6, 17 days 6:00 start, 6 month by 2019, 17 day 7:00 ends with a time interval of 20 seconds. The ordinate is the density of the section of the traffic flow, the blue solid line is the coil measured value, the red dotted line is the extended kalman filter estimated reduction value, and it can be seen from the figure that the trends of the two are relatively close and consistent, and meanwhile, according to the calculation, the Mean Absolute Error (MAE) between the model value and the measured value is 3.9, and the Root Mean Square Error (RMSE) is 7.6.
Step 4, scheme simulation prediction
And according to the service scene, predicting and estimating the traffic state evolution in a future short term or under the traffic control condition based on supply and demand prediction and estimation.

Claims (2)

1. A modeling and prediction method based on a traffic big data platform and a mesoscopic simulation model is characterized by comprising the following steps:
acquiring a training set for training a mesoscopic simulation model by using a traffic big data platform, and predicting the traffic state of the trained mesoscopic simulation model based on data acquired by the traffic big data platform in real time, wherein:
the construction of the mesoscopic simulation model and the prediction by utilizing the mesoscopic simulation model comprise the following steps:
step 1, model selection
The mesoscopic model divides the whole road into N sections of road sections by adopting traffic flow discrete state equations shown in the following formulas (1) and (2), and the length of the ith section of road section is delta i The head and tail ends of each road section are respectively provided with a traffic detector to provide measured traffic flow q in a detection period T i (k) Average velocity v i (k) Data as input to the mesoscopic model, q i (k) Traffic flow at time k, v, for the i-th to i + 1-th road section i (k) The average speed of the traffic flow space at the moment k of the ith road section comprises the following components:
q i (k)＝αρ i (k)v i (k)+(1-α)ρ i+1 (k)v i+1 (k) (1)
Figure FDA0003783365530000011
Figure FDA0003783365530000012
Figure FDA0003783365530000013
in formulas (1), (2), (3), (4): ρ is a unit of a gradient i (k) The density of the traffic flow at the moment k of the ith road section; r is a radical of hydrogen i (k) The flow rate of the entrance ramp at the moment k of the ith road section; s i (k) The flow rate of the exit ramp at the moment k of the ith road section; v. of f Representing the free-running speed of the traffic flow; rho cr Representing the critical density, namely the density of the traffic flow when the traffic flow reaches the maximum; b. tau, gamma, delta, lambda and alpha are adjustment coefficients of the equation; ρ represents a density; v (ρ) represents a vehicle speed;
step 2, state estimation model
And (3) estimating the error by adopting a Kalman filter, wherein for a nonlinear system:
x(k+1)＝f[x(k)]+Γ[x(k)]w(k) (5)
y(k+1)＝h[x(k+1)]+v(k+1) (6)
in the formulas (5) and (6), w (k) and v (k) are zero-mean noise vectors, and
Figure FDA0003783365530000014
Figure FDA0003783365530000015
q (k) is a system noise variance matrix, and R (k) is a measurement noise matrix; Γ [ x (k))]Is a noise drive matrix; x (k) is a state variable; y (k) is an observed value; f [ x (k)]Is a macroscopic traffic flow model value; h [ x (k)]Is an observation model value;
will be originally systematic in
Figure FDA0003783365530000021
The process is linearized and then carried out,
Figure FDA0003783365530000022
representing the traffic flow model estimate with noise at time k, the resulting extended Kalman filter consists of the following set of recursive equations:
Figure FDA0003783365530000023
Figure FDA0003783365530000024
Figure FDA0003783365530000025
Figure FDA0003783365530000026
Figure FDA0003783365530000027
wherein:
Figure FDA0003783365530000028
Figure FDA0003783365530000029
in the formulae (7) to (13),
Figure FDA00037833655300000210
the estimated value of the traffic flow model with noise at the moment k;
Figure FDA00037833655300000211
the traffic flow model estimation value with noise at the next moment k +1 is obtained;
Figure FDA00037833655300000212
is the traffic flow model at time k;
Figure FDA00037833655300000213
an observation model estimation value with noise at the moment k;
Figure FDA00037833655300000214
to observe the model at time k; k (K + 1) is the Kalman gain at the next moment K + 1; y (k) is an observed value at time k; p (k +1 k) is a system state transition covariance matrix when transitioning from the time k to the next time k + 1; r (k + 1) is an observation noise matrix at the next moment k + 1; p (k) is a state transition covariance matrix at the moment k; q (k + 1) is a system noise variance matrix at the next moment k + 1; x is a radical of a fluorine atom 1 (k)，...，x 2N (k) Values of variables for each state; f. of 1 ，...，f 2N The Jacobian matrix value of the traffic flow model is obtained; h is 1 、h 2 Is the Jacobian matrix value of the observation model;
Figure FDA0003783365530000031
for each state variable estimate at the transition from time k to the next time k + 1;
at the given initial filtered value
Figure FDA0003783365530000032
After the covariance matrix P (0) is initially filtered, gradually calculating according to the calculation sequence of the determined recursion algorithm, and estimating the system state;
and 3, taking the traffic density of the N sections of road sections at the moment k and the traffic speed of the N sections of road sections at the moment k as state variables x n (k) N =1, 2N, using measured values of flow and average speed of a head end of a first road section in the N sections of road sections as input quantities, and using density and average speed of an nth road section in the N sections of road sections as output quantities, thereby constructing a training set to perform parameter calibration on the mesoscopic model;
step 4, calibrating parameters of an observation dynamic flow model in the traffic flow:
parameter rho of the mesoscopic model cr 、v f And b, learning by using an online learning method according to a flow density formula:
Figure FDA0003783365530000033
according to the actually measured flow and speed data, the rho is learned through a genetic algorithm cr 、v f B, the value of;
and 5: the view dynamic model of the traffic flow obtains the evolution of the traffic flow on the road in time and space;
step 6: calculating an observation model:
firstly, an observation matrix is calculated, and then the observation matrix is multiplied by a state variable to obtain an observation estimation value
Figure FDA0003783365530000034
And 7: calculating a state transition matrix F:
Figure FDA0003783365530000035
wherein
Figure FDA0003783365530000036
Figure FDA0003783365530000037
And step 8: determining values of a state transition covariance matrix Q and an observation noise variance matrix R, and determining an initial value of a state covariance matrix P, wherein the values are assigned according to experience;
and step 9: performing extended Kalman Filter calculation: performing iterative computation of extended Kalman filtering according to equations (5) to (13) in the extended Kalman filtering EKF state estimation principle to obtain a final solution;
step 10: and (4) estimating the OD and the traffic capacity in a future period of time based on the analysis and calibration of the influence of the traffic accidents and the control measures, and repeating the processes to realize the prediction of the traffic state in the future period of time.
2. The modeling and prediction method based on the traffic big data platform and the mesoscopic simulation model as claimed in claim 1, wherein the construction of the traffic big data platform comprises the following steps:
step 1, automatic modeling of a simulation area road network and a traffic improvement scheme traffic simulation scene based on a standardized map layer: establishing an integrated data standard according to a human-vehicle-road-traffic environment to generate a unified road network GIS layer; modeling traffic organization, induction and control traffic scheme scene data, and performing spatial association with a road object;
step 2, designing a standardization layer based on the converged traffic demands, traffic flow data and vehicle data, and extracting, converting and loading scattered, isolated and irregular collected data of each platform to form uniform, standard and layered dynamic original data;
step 3, aiming at the data loss, inconsistency and error problem characteristics in the dynamic original data, combining data conditions, designing a data quality index and a discrimination method, realizing comprehensive identification on the data quality problem, and repairing the problem data based on the spatial correlation and node conservative principle to form complete, consistent and accurate dynamic data;
step 4, establishing a mesoscopic simulation model and a mesoscopic index system for simulation input and output, uniformly defining each index, and designing a standardized processing method for each index;
step 5, designing a simulation basic platform to build for an actual simulation modeling prediction scene, verifying the platform precision, and finally realizing index output and application flow;
and 6, designing a service scene, and butting simulation output and user requirements.
CN202010825580.3A 2020-08-17 2020-08-17 Prediction method based on traffic big data platform and mesoscopic simulation model Active CN111951553B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010825580.3A CN111951553B (en) 2020-08-17 2020-08-17 Prediction method based on traffic big data platform and mesoscopic simulation model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010825580.3A CN111951553B (en) 2020-08-17 2020-08-17 Prediction method based on traffic big data platform and mesoscopic simulation model

Publications (2)

Publication Number Publication Date
CN111951553A CN111951553A (en) 2020-11-17
CN111951553B true CN111951553B (en) 2022-11-11

Family

ID=73342532

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010825580.3A Active CN111951553B (en) 2020-08-17 2020-08-17 Prediction method based on traffic big data platform and mesoscopic simulation model

Country Status (1)

Country Link
CN (1) CN111951553B (en)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113053106B (en) * 2021-03-05 2022-08-26 交通运输部公路科学研究所 Traffic state estimation method and device based on multi-sensor information fusion
CN113191528B (en) * 2021-04-02 2022-02-22 东南大学 Traffic state deduction method integrating macroscopic traffic simulation and short-term traffic prediction
CN114004077A (en) * 2021-10-28 2022-02-01 腾讯科技（深圳）有限公司 Traffic simulation conversion method, device, computer equipment and storage medium
CN113781785B (en) * 2021-11-10 2022-02-08 禾多科技（北京）有限公司 Random traffic flow control method for simulation test
CN114333335A (en) * 2022-03-15 2022-04-12 成都交大大数据科技有限公司 Lane-level traffic state estimation method, device and system based on track data
CN114937366B (en) * 2022-07-22 2022-11-25 深圳市城市交通规划设计研究中心股份有限公司 Traffic flow calculation method based on multi-scale traffic demand and supply conversion
CN116306318B (en) * 2023-05-12 2023-08-01 青岛哈尔滨工程大学创新发展中心 Three-dimensional ocean thermal salt field forecasting method, system and equipment based on deep learning

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101656020A (en) * 2009-09-23 2010-02-24 北京交通大学 System and method for evaluating urban road traffic zone servings levels based on actual measurements
CN103413443A (en) * 2013-07-03 2013-11-27 太原理工大学 Short-term traffic flow forecasting method based on hidden Markov model
CN104866654A (en) * 2015-05-06 2015-08-26 广州市交通规划研究院 Construction method for integrated dynamic traffic simulation platform of city
CN105787858A (en) * 2016-03-29 2016-07-20 交通运输部公路科学研究所 Situation deduction method for expressway network
CN109255948A (en) * 2018-08-10 2019-01-22 昆明理工大学 A kind of divided lane wagon flow scale prediction method based on Kalman filtering
US10684598B1 (en) * 2019-01-04 2020-06-16 Johnson Controls Technology Company Building management system with efficient model generation for system identification

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4116262B2 (en) * 2001-03-16 2008-07-09 ダイキン工業株式会社 Air conditioner, air conditioning load prediction method, and program
JP3797949B2 (en) * 2002-03-28 2006-07-19 株式会社東芝 Image processing apparatus and method
TWI446129B (en) * 2012-02-15 2014-07-21 Nat Applied Res Laboratories Method of real-time correction of water stage forecast
CN102663887B (en) * 2012-04-13 2014-06-11 浙江工业大学 Implementation system and method for cloud calculation and cloud service of road traffic information based on technology of internet of things
CN104408913B (en) * 2014-11-03 2016-03-16 东南大学 A kind of traffic flow three parameter real-time predicting method considering temporal correlation
CN105809958A (en) * 2016-03-29 2016-07-27 中国科学院深圳先进技术研究院 Traffic control method and system based on intersection group
CN106251630B (en) * 2016-10-13 2018-09-07 东南大学 A kind of progressive Extended Kalman filter traffic status of express way method of estimation based on multi-source data
CN107103142B (en) * 2017-07-11 2019-12-03 交通运输部公路科学研究所 Emulation technology is deduced towards the comprehensive traffic network operation situation of highway and the railway network
CN110020475B (en) * 2019-04-03 2023-10-10 北京工业大学 Markov particle filtering method for traffic flow prediction
CN110164128B (en) * 2019-04-23 2020-10-27 银江股份有限公司 City-level intelligent traffic simulation system
CN111260131B (en) * 2020-01-16 2023-06-02 汕头大学 Short-term traffic flow prediction method and device

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101656020A (en) * 2009-09-23 2010-02-24 北京交通大学 System and method for evaluating urban road traffic zone servings levels based on actual measurements
CN103413443A (en) * 2013-07-03 2013-11-27 太原理工大学 Short-term traffic flow forecasting method based on hidden Markov model
CN104866654A (en) * 2015-05-06 2015-08-26 广州市交通规划研究院 Construction method for integrated dynamic traffic simulation platform of city
CN105787858A (en) * 2016-03-29 2016-07-20 交通运输部公路科学研究所 Situation deduction method for expressway network
CN109255948A (en) * 2018-08-10 2019-01-22 昆明理工大学 A kind of divided lane wagon flow scale prediction method based on Kalman filtering
US10684598B1 (en) * 2019-01-04 2020-06-16 Johnson Controls Technology Company Building management system with efficient model generation for system identification

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
城市道路交通实时在线仿真平台优化与提升;陈振武等;《交通与运输》;20190731;第160-165页 *

Also Published As

Publication number Publication date
CN111951553A (en) 2020-11-17

Similar Documents

Publication Publication Date Title
CN111951553B (en) Prediction method based on traffic big data platform and mesoscopic simulation model
CN109285346B (en) Urban road network traffic state prediction method based on key road sections
Kurzhanskiy et al. Active traffic management on road networks: a macroscopic approach
Schultz et al. Analysis of distribution and calibration of car-following sensitivity parameters in microscopic traffic simulation models
Ghanim et al. Estimating turning movements at signalized intersections using artificial neural networks
CN110956807B (en) Highway flow prediction method based on combination of multi-source data and sliding window
CN104778837A (en) Multi-time scale forecasting method for road traffic running situation
CN113011455B (en) Air quality prediction SVM model construction method
CN111583628B (en) Road network heavy truck traffic flow prediction method based on data quality control
Padiath et al. Prediction of traffic density for congestion analysis under Indian traffic conditions
KR102063404B1 (en) system for managing traffic based on Platform
Bae et al. Multicontextual machine-learning approach to modeling traffic impact of urban highway work zones
ChikkaKrishna et al. Short-term traffic prediction using fb-prophet and neural-prophet
CN103870890A (en) Prediction method for traffic flow distribution of expressway network
JPH0877485A (en) Predicting device for required travel time
Juri et al. Integrated traffic simulation–statistical analysis framework for online prediction of freeway travel time
CN107730882A (en) Congestion in road forecasting system and method based on artificial intelligence
Liu et al. A streamlined network calibration procedure for california sr41 corridor traffic simulation study
Weng et al. Freeway travel speed calculation model based on ETC transaction data
Islek et al. Use of LSTM for Short-Term and Long-Term Travel Time Prediction.
Hasan et al. Modeling of travel time variations on urban links in London
Bie et al. Improving traffic state prediction model for variable speed limit control by introducing stochastic supply and demand
CN106127664A (en) Method for controlling passenger flow in peak period of urban rail transit transfer station
Van Grol et al. DACCORD: On-line travel time prediction
Wang Real time toll optimization based on predicted traffic conditions

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