Simulation-based traffic flow prediction system and method
By combining multi-source data and reinforcement learning optimization model, a simulation-based traffic flow prediction system is established, which solves the problems of dynamic adaptability and low data utilization of traffic flow prediction in the prior art, and achieves more accurate and fast traffic flow prediction.
Patent Information
- Application Number
- CN202510430566.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-08
- Publication Date
- 2025-07-08
AI Technical Summary
The existing traffic flow prediction methods are difficult to adapt to emergencies, the utilization rate of multi-source data is low, and traditional simulation models cannot respond to changes in traffic state in real time, resulting in large prediction errors.
Floating vehicle GPS, geomagnetic sensor, surveillance camera and meteorological data acquisition unit are adopted, combined with data preprocessing, dynamic simulation model and reinforcement learning module, and a simulation-based traffic flow prediction system is established. Through multi-step rolling prediction and reinforcement learning optimization model, the data is realized in space-time and space-weighted fusion and real-time response.
It improves the dynamic adaptability of traffic flow prediction, reduces prediction error, improves the utilization rate of multi-source data, and shortens the prediction response time after traffic accidents.
Smart Images
Figure CN120279708A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of road traffic, and particularly to a simulation-based traffic flow prediction system and method. Background Art
[0002] Transportation is the lifeblood of the national economy. The transportation technology faces major strategic needs. We need to achieve traffic information sharing and effective connection of various transportation modes, and improve the technical level of traffic operation management. With the continuous increase in the number of motor vehicles, the road system is under increasing pressure, and predicting road traffic flow can effectively solve or alleviate the above problems.
[0003] However, the existing technologies have the following problems:
[0004] (1) Limitation of static models: Traditional traffic flow prediction methods (such as time series models and macroscopic hydrodynamic models) rely on the statistical laws of historical data and are difficult to adapt to dynamic scenarios such as sudden traffic accidents and bad weather.
[0005] (2) Insufficient data fusion: The existing systems do not fully process the spatio-temporal correlation of floating car GPS data, geomagnetic detector data, and video recognition data, resulting in low utilization rate of multi-source heterogeneous data.
[0006] (3) Defect of fixed parameters: Traditional simulation models (such as VISSIM and SUMO) adopt fixed parameter configurations and cannot respond to traffic state changes in real time. Measured data shows that their prediction errors are relatively large during peak hours.
[0007] Therefore, the present invention proposes a simulation-based traffic flow prediction system and method. Summary of the Invention
[0008] The purpose of the present invention is to solve the deficiencies existing in the prior art, and to propose a simulation-based traffic flow prediction system and method.
[0009] In order to achieve the above purpose, the present invention adopts the following technical solutions:
[0010] A simulation-based traffic flow prediction system, comprising:
[0011] A data acquisition module, which includes a floating car GPS, a geomagnetic sensor, a monitoring camera, and a meteorological data acquisition unit;
[0012] A data processing module, which is internally provided with a data preprocessing unit and a dynamic simulation model;
[0013] A reinforcement learning optimization module, which performs reinforcement learning on the dynamic simulation model;
[0014] A prediction output module, which obtains a prediction result according to the detected data and a dynamic simulation model;
[0015] A visualization output module, which visually outputs the prediction result;
[0016] The floating car GPS is an OBD terminal installed in a taxi;
[0017] The geomagnetic sensor uses an AMR magnetoresistive array;
[0018] The monitoring camera uses a bullet-shaped network camera;
[0019] The meteorological data acquisition unit uses a roadside meteorological station.
[0020] A traffic flow prediction method based on simulation specifically includes the following steps:
[0021] S1: Data acquisition, data is acquired through a floating car GPS, a geomagnetic sensor, a monitoring camera, and a meteorological data acquisition unit;
[0022] S2: Data preprocessing, the data preprocessing unit respectively performs time synchronization and spatial mapping on the acquired multi-source data;
[0023] S3: Dynamic simulation, a dynamic simulation model is established, and the preprocessed acquired data is used as input, and then a prediction result is output;
[0024] S4: Visualization output, the prediction result is visually output.
[0025] Preferably: In the S2 step, the model for time synchronization is: t raw is the original timestamp, t align is the synchronized timestamp, and T is the system reference time unit;
[0026] In the S2 step, T = 30 seconds or T = 60 seconds.
[0027] Preferably: In the S2 step, the model for spatial mapping is: where ρ seg is the average road surface density, v i is the speed of the i-th vehicle, and δ i is the residence time ratio of vehicle i in section L.
[0028] Preferably: In the S3 step, the dynamic simulation model includes a macroscopic layer based on a cellular transmission model, a microscopic layer based on an intelligent driver model, and a multi-step rolling prediction equation.
[0029] Preferably: The logical model of the macroscopic layer is: where ρ(x, t) represents the vehicle density at position x and time t, v represents the speed, which is output by the microscopic model, ε(t) is a random perturbation term, following N(0, σ 2 ), represents the rate of change of density over time, represents the gradient of the vehicle flow speed with respect to position, S in is the vehicle inflow rate at the road section entrance, S out is the vehicle outflow rate at the road section exit.
[0030] Preferably: The logical model of the microscopic layer is: where a n is the acceleration of the nth vehicle, τ is the driver's reaction time, α and β are sensitivity parameters, Δx n (t) represents the distance involved, v d is the driver's desired speed.
[0031] Preferably: In the microscopic layer, τ = (0.8s, 2.4s).
[0032] Preferably: The multi-step rolling prediction equation is: Y t+1：t+H = F(X t-T+1：t , θ * ) + ⊥·ΔD, where Y t+1：t+H is the prediction output for the next H time steps, F() is the hybrid simulation model function, whose input is the historical state sequence X t-T+1：t , θ * are the optimization parameters, ⊥ is the emergency impact matrix, and ΔD is the real-time event feature vector.
[0033] Preferably: In the S3 step, it further includes predictive model-based reinforcement learning, and the model of reinforcement learning is: where:
[0034] Q(s t , a t ) is the expected cumulative reward for executing action a t in state s t ;
[0035] η is the learning rate, η = (0.01, 0.1);
[0036] γ t is the immediate reward, q1 is the predicted flow, q2 is the observed flow, ψ is the stability penalty coefficient, θ = [τ, β, p lc ;
[0037] s t is the state vector, μ(ρ) is the average density of the road network, and σ(v) is the standard deviation of speed. is the traffic flow change rate, and Rmean is the average residual of multi-source data;
[0038] a t is the action vector, and a t = [Δβ, Δτ, Δp lc , where Δβ is the adjustment amount of the speed-density index, Δτ is the adjustment amount of the driver's reaction time, and Δp lc is the adjustment amount of the lane-changing probability.
[0039] The beneficial effects of the present invention are as follows:
[0040] 1. The dynamic adaptability of the present invention is improved. Through the online optimization of the β parameter, the speed-density relationship curve dynamically fits the actual situation, and the prediction error is reduced compared with the fixed-parameter model.
[0041] 2. In the present invention, the data fusion efficiency is broken through, and the spatio-temporal weighting mechanism improves the utilization rate of multi-source data.
[0042] 3. In the present invention, the real-time optimization ability is enhanced, and the reinforcement learning shortens the prediction response time of the simulation model after a traffic accident, which is superior to the traditional manual parameter adjustment mode. Description of the Drawings
[0043] Figure 1 is an architecture diagram of a simulation-based traffic flow prediction system proposed by the present invention;
[0044] Figure 2 is a flowchart of a simulation-based traffic flow prediction method proposed by the present invention. Detailed Embodiments
[0045] The technical solutions of the present invention will be further described in detail below in conjunction with the specific embodiments.
[0046] In the description of the present invention, it should be noted that unless otherwise clearly defined and limited, the terms "installed", "connected", "connected", and "set" should be understood in a broad sense. For example, it can be fixedly connected and set, or detachably connected and set, or integrally connected and set. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.
[0047] Example 1:
[0048] A simulation-based traffic flow prediction system, which includes:
[0049] A data acquisition module, which includes a floating car GPS, a geomagnetic sensor, a monitoring camera, and a meteorological data acquisition unit;
[0050] A data processing module, which is built-in with a data preprocessing unit and a dynamic simulation model;
[0051] A reinforcement learning optimization module, which performs reinforcement learning on the dynamic simulation model;
[0052] A prediction output module, which obtains a prediction result based on the detected data and the dynamic simulation model;
[0053] A visualization output module, which visually outputs the prediction result.
[0054] The floating vehicle GPS is an OBD terminal installed in a taxi;
[0055] The geomagnetic sensor uses an AMR magnetoresistive array;
[0056] The monitoring camera uses a bullet-shaped network camera;
[0057] The meteorological data acquisition unit uses a roadside meteorological station.
[0058] Embodiment 2:
[0059] A simulation-based traffic flow prediction method, which is a method of a simulation-based traffic flow prediction system, specifically including the following steps:
[0060] S1: Data acquisition, data is acquired through a floating vehicle GPS, a geomagnetic sensor, a monitoring camera, and a meteorological data acquisition unit;
[0061] S2: Data preprocessing, the data preprocessing unit respectively performs time synchronization and spatial mapping on the acquired multi-source data;
[0062] S3: Dynamic simulation, a dynamic simulation model is established, and the preprocessed acquired data is used as input, and then a prediction result is output;
[0063] S4: Visualization output, the prediction result is visually output.
[0064] In the S2 step, the model of time synchronization is: t raw is the original timestamp, t align is the synchronized timestamp, and T is the system reference time unit.
[0065] In the S2 step, T = 30 seconds.
[0066] Embodiment 3:
[0067] A simulation-based traffic flow prediction method, which is a method of a simulation-based traffic flow prediction system, specifically including the following steps:
[0068] S1: Data collection, which is carried out through floating vehicle GPS, geomagnetic sensors, monitoring cameras, and meteorological data collection units;
[0069] S2: Data preprocessing, where the data preprocessing unit performs time synchronization and spatial mapping on the collected multi-source data respectively;
[0070] S3: Dynamic simulation, where a dynamic simulation model is established, and the preprocessed collected data is used as input, and then the prediction results are output;
[0071] S4: Visualization output, where the prediction results are visually output.
[0072] In the step S2, the model for time synchronization is: t raw is the original timestamp, t align is the synchronized timestamp, and T is the system reference time unit.
[0073] In the step S2, T = 60 seconds.
[0074] Example 4:
[0075] A traffic flow prediction method based on simulation, which is a method for a traffic flow prediction system based on simulation, specifically including the following steps:
[0076] S1: Data collection, which is carried out through floating vehicle GPS, geomagnetic sensors, monitoring cameras, and meteorological data collection units;
[0077] S2: Data preprocessing, where the data preprocessing unit performs time synchronization and spatial mapping on the collected multi-source data respectively;
[0078] S3: Dynamic simulation, where a dynamic simulation model is established, and the preprocessed collected data is used as input, and then the prediction results are output;
[0079] S4: Visualization output, where the prediction results are visually output.
[0080] In the step S2, the model for time synchronization is: t raw is the original timestamp, t align is the synchronized timestamp, and T is the system reference time unit.
[0081] In the step S2, T = 30 seconds or T = 60 seconds.
[0082] In the step S2, the model for spatial mapping is: where ρ seg is the average road surface density, v i is the speed of the i-th vehicle, δ iThe residence time ratio of vehicle i within section L.
[0083] Embodiment 5:
[0084] A simulation-based traffic flow prediction method, which is a method of a simulation-based traffic flow prediction system, specifically including the following steps:
[0085] S1: Data collection, data is collected through floating car GPS, geomagnetic sensors, surveillance cameras, and meteorological data collection units;
[0086] S2: Data preprocessing, the data preprocessing unit performs time synchronization and spatial mapping on the collected multi-source data respectively;
[0087] S3: Dynamic simulation, a dynamic simulation model is established, and the preprocessed collected data is used as input, and then the prediction result is output;
[0088] S4: Visualization output, the prediction result is visually output.
[0089] In the step S2, the model of time synchronization is: t raw is the original timestamp, t align is the synchronized timestamp, and T is the system reference time unit.
[0090] In the step S2, T = 30 seconds or T = 60 seconds.
[0091] In the step S2, the model of spatial mapping is: where ρ seg is the average road surface density, v i is the speed of the i-th vehicle, δ i is the residence time ratio of vehicle i within section L.
[0092] In the step S3, the dynamic simulation model includes a macroscopic layer based on the cell transmission model, a microscopic layer based on the intelligent driver model, and a multi-step rolling prediction equation.
[0093] The logical model of the macroscopic layer is: where ρ(x, t) represents the vehicle density at position x and time t, v represents the speed and is output by the microscopic model, ε(t) is a random perturbation term and follows N(0, σ 2 ), represents the change rate of density over time, represents the gradient of the vehicle flow speed with respect to position, S in is the vehicle inflow rate at the section entrance, S out is the vehicle outflow rate at the section exit.
[0094] The logical model of the microscopic layer is: where a n is the acceleration of the nth vehicle, τ is the driver's reaction time, α and β are sensitivity parameters, and Δx n (t) represents the distance involved, and v d is the driver's desired speed.
[0095] In the micro layer, τ = 0.8 s.
[0096] Example 6:
[0097] A simulation-based traffic flow prediction method, which is a method for a simulation-based traffic flow prediction system, specifically including the following steps:
[0098] S1: Data collection, data is collected through floating car GPS, geomagnetic sensors, surveillance cameras, and meteorological data collection units;
[0099] S2: Data preprocessing, the data preprocessing unit performs time synchronization and spatial mapping on the collected multi-source data respectively;
[0100] S3: Dynamic simulation, a dynamic simulation model is established, and the preprocessed collected data is used as input, and then the prediction result is output;
[0101] S4: Visualization output, the prediction result is visually output.
[0102] In the S2 step, the time synchronization model is: t raw is the original timestamp, t align is the synchronized timestamp, and T is the system reference time unit.
[0103] In the S2 step, T = 30 seconds or T = 60 seconds.
[0104] In the S2 step, the spatial mapping model is: where ρ seg is the average road surface density, v i is the speed of the ith vehicle, and δ i is the residence time ratio of vehicle i in section L.
[0105] In the S3 step, the dynamic simulation model includes a macro layer based on the cell transmission model, a micro layer based on the intelligent driver model, and a multi-step rolling prediction equation.
[0106] The logical model of the macro layer is: where ρ(x, t) represents the vehicle density at position x and time t, v represents the speed and is output by the micro model, ε(t) is a random disturbance term and follows N(0, σ 2 ), Represents the rate of change of density over time, Represents the gradient of the traffic flow speed with respect to position, S in Is the vehicle inflow rate at the road section entrance, S out Is the vehicle outflow rate at the road section exit.
[0107] The logical model of the microscopic layer is: Where a n Is the acceleration of the nth vehicle, τ is the driver's reaction time, α and β are sensitivity parameters, Δx n (t) represents the involved distance, v d Is the driver's desired speed.
[0108] In the microscopic layer, τ = 2.4s.
[0109] Example 7:
[0110] A simulation-based traffic flow prediction method, which is a method of a simulation-based traffic flow prediction system, specifically includes the following steps:
[0111] S1: Data collection, data is collected through floating car GPS, geomagnetic sensors, surveillance cameras, and meteorological data collection units;
[0112] S2: Data preprocessing, the data preprocessing unit performs time synchronization and spatial mapping on the collected multi-source data respectively;
[0113] S3: Dynamic simulation, a dynamic simulation model is established, and the preprocessed collected data is used as input, and then the prediction result is output;
[0114] S4: Visualization output, the prediction result is visually output.
[0115] In the S2 step, the time synchronization model is: t raw Is the original timestamp, t align Is the synchronized timestamp, and T is the system reference time unit.
[0116] In the S2 step, T = 30 seconds or T = 60 seconds.
[0117] In the S2 step, the spatial mapping model is: Where ρ seg Is the average road surface density, v i Is the speed of the ith vehicle, δ i Is the residence time ratio of vehicle i within the road section L.
[0118] In the S3 step, the dynamic simulation model includes a macroscopic layer based on the cell transmission model, a microscopic layer based on the intelligent driver model, and a multi-step rolling prediction equation.
[0119] The logical model of the macroscopic layer is as follows: where ρ(x, t) represents the vehicle density at position x and time t, v represents the speed, which is output by the microscopic model, and ε(t) is a random perturbation term, following N(0, σ 2 ), represents the change rate of density over time, represents the gradient of the vehicle flow speed with respect to position, S in is the vehicle inflow rate at the road section entrance, and S out is the vehicle outflow rate at the road section exit.
[0120] The logical model of the microscopic layer is as follows: where a n is the acceleration of the nth vehicle, τ is the driver's reaction time, α and β are sensitivity parameters, Δx n (t) represents the involved distance, and v d is the driver's desired speed.
[0121] In the microscopic layer, τ = 1.6 s.
[0122] Example 8:
[0123] A simulation-based traffic flow prediction method, which is a method of a simulation-based traffic flow prediction system, specifically includes the following steps:
[0124] S1: Data collection, collecting data through floating vehicle GPS, geomagnetic sensors, surveillance cameras, and meteorological data collection units;
[0125] S2: Data preprocessing, and the data preprocessing unit performs time synchronization and spatial mapping on the collected multi-source data respectively;
[0126] S3: Dynamic simulation, establishing a dynamic simulation model, and using the preprocessed collected data as input, and then outputting a prediction result;
[0127] S4: Visualization output, visually outputting the prediction result.
[0128] In the S2 step, the model of time synchronization is: t raw is the original timestamp, t align is the synchronized timestamp, and T is the system reference time unit.
[0129] In the S2 step, T = 30 seconds or T = 60 seconds.
[0130] In the S2 step, the model of spatial mapping is: where ρ seg is the average road surface density, vi is the speed of the i-th vehicle, and δ i is the residence time ratio of vehicle i in section L.
[0131] In the step S3, the dynamic simulation model includes a macroscopic layer based on the cell transmission model, a microscopic layer based on the intelligent driver model, and a multi-step rolling prediction equation.
[0132] The logical model of the macroscopic layer is: where ρ(x, t) represents the vehicle density at position x and time t, v represents the speed and is output by the microscopic model, ε(t) is a random perturbation term and follows N(0, σ 2 ), represents the change rate of density with time, represents the gradient of the vehicle flow speed with respect to position, and S in is the vehicle inflow rate at the section entrance, and S out is the vehicle outflow rate at the section exit.
[0133] The logical model of the microscopic layer is: where a n is the acceleration of the n-th vehicle, τ is the driver's reaction time, α and β are sensitivity parameters, and Δx n (t) represents the distance involved, and v d is the driver's desired speed.
[0134] In the microscopic layer, τ = (0.8s, 2.4s).
[0135] The multi-step rolling prediction equation is: Y t+1：t+H = F(X t-T+1：t , θ * ) + ⊥·ΔD, where Y t+1：t+H is the prediction output for the next H time steps, F() is the hybrid simulation model function, whose input is the historical state sequence X t-T+1：t , θ * are the optimization parameters, ⊥ is the emergency impact matrix, and ΔD is the real-time event feature vector.
[0136] In the step S3, it also includes predictive model-based reinforcement learning, and the model of reinforcement learning is: where:
[0137] Q(s t , a t ) is the expected cumulative reward for executing action a t in state s t ;
[0138] η is the learning rate, η = 0.01;
[0139] γt For immediate rewards, q1 is the predicted traffic flow, q2 is the observed traffic flow, ψ is the stability penalty coefficient, and θ = [τ, β, p lc ;
[0140] s t is the state vector, μ(ρ) is the average density of the road network, σ(v) is the standard deviation of speed, is the traffic flow change rate, and Rmean is the average residual of multi-source data;
[0141] a t is the action vector, a t = [Δβ, Δτ, Δp lc , Δβ is the adjustment amount of the speed-density index, Δτ is the adjustment amount of the driver's reaction time, and Δp lc is the adjustment amount of the lane-changing probability.
[0142] Example 9:
[0143] A simulation-based traffic flow prediction method, which is a method of a simulation-based traffic flow prediction system, specifically including the following steps:
[0144] S1: Data collection, data is collected through floating vehicle GPS, geomagnetic sensors, monitoring cameras, and meteorological data collection units;
[0145] S2: Data preprocessing, the data preprocessing unit performs time synchronization and spatial mapping on the collected multi-source data respectively;
[0146] S3: Dynamic simulation, a dynamic simulation model is established, and the preprocessed collected data is used as input, and then the prediction result is output;
[0147] S4: Visualization output, the prediction result is visually output.
[0148] In the step S2, the model of time synchronization is: t raw is the original timestamp, t align is the synchronized timestamp, and T is the system reference time unit.
[0149] In the step S2, T = 30 seconds or T = 60 seconds.
[0150] In the step S2, the model of spatial mapping is: where ρ seg is the average road surface density, v i is the speed of the i-th vehicle, and δ i is the residence time ratio of vehicle i in section L.
[0151] In the step S3, the dynamic simulation model includes a macroscopic layer based on the cell transmission model, a microscopic layer based on the intelligent driver model, and a multi-step rolling prediction equation.
[0152] The logical model of the macroscopic layer is: where ρ(x, t) represents the vehicle density at position x and time t, v represents the speed, which is output by the microscopic model, ε(t) is a random perturbation term, obeying N(0, σ 2 ), represents the change rate of density over time, represents the gradient of the vehicle flow speed with respect to position, S in is the vehicle inflow rate at the road section entrance, S out is the vehicle outflow rate at the road section exit.
[0153] The logical model of the microscopic layer is: where a n is the acceleration of the nth vehicle, τ is the driver's reaction time, α and β are sensitivity parameters, Δx n (t) represents the distance involved, v d is the driver's desired speed.
[0154] In the microscopic layer, τ = (0.8s, 2.4s).
[0155] The multi-step rolling prediction equation is: Y t+1：t+H = F(X t-T+1：t , θ * ) + ⊥·ΔD, where Y t+1：t+H is the prediction output for the next H time steps, F() is the hybrid simulation model function, whose input is the historical state sequence X t-T+1：t , θ * is the optimization parameter, ⊥ is the emergency impact matrix, and ΔD is the real-time event feature vector.
[0156] In the step S3, it also includes predictive model-based reinforcement learning, and the model of reinforcement learning is: where:
[0157] Q(s t , a t ) is the expected cumulative reward for executing action a t in state s t ;
[0158] η is the learning rate, η = 0.05;
[0159] γ t is the immediate reward, q1 is the predicted flow, q2 is the observed flow, ψ is the stability penalty coefficient, θ = [τ, β, p lc ;
[0160] s t is the state vector, μ(ρ) is the average density of the road network, and σ(v) is the standard deviation of speed, is the flow rate change rate, and Rmean is the average residual of multi-source data;
[0161] a t is the action vector, a t = [Δβ, Δτ, Δp lc , Δβ is the adjustment amount of the speed-density index, Δτ is the adjustment amount of the driver's reaction time, and Δp lc is the adjustment amount of the lane-changing probability.
[0162] Example 10:
[0163] A simulation-based traffic flow prediction method, which is a method of a simulation-based traffic flow prediction system, specifically including the following steps:
[0164] S1: Data collection, data is collected through floating car GPS, geomagnetic sensors, monitoring cameras, and meteorological data collection units;
[0165] S2: Data preprocessing, the data preprocessing unit performs time synchronization and spatial mapping on the collected multi-source data respectively;
[0166] S3: Dynamic simulation, a dynamic simulation model is established, and the preprocessed collected data is used as input, and then the prediction result is output;
[0167] S4: Visualization output, the prediction result is visually output.
[0168] In the step S2, the model of time synchronization is: t raw is the original timestamp, t align is the synchronized timestamp, and T is the system reference time unit.
[0169] In the step S2, T = 30 seconds or T = 60 seconds.
[0170] In the step S2, the model of spatial mapping is: where ρ seg is the average density of the road surface, v i is the speed of the i-th vehicle, and δ i is the residence time ratio of vehicle i in section L.
[0171] In the step S3, the dynamic simulation model includes a macroscopic layer based on the cell transmission model, a microscopic layer based on the intelligent driver model, and a multi-step rolling prediction equation.
[0172] The logical model of the macroscopic layer is as follows: where ρ(x, t) represents the vehicle density at position x and time t, v represents the speed, which is output by the microscopic model, and ε(t) is a random perturbation term, following N(0, σ 2 ). represents the rate of change of density over time, represents the gradient of the vehicle flow speed with respect to position, and S in is the vehicle inflow rate at the road section entrance, and S out is the vehicle outflow rate at the road section exit.
[0173] The logical model of the microscopic layer is as follows: where a n is the acceleration of the nth vehicle, τ is the driver's reaction time, α and β are sensitivity parameters, and Δx n (t) represents the involved distance, and v d is the driver's desired speed.
[0174] In the microscopic layer, τ = (0.8s, 2.4s).
[0175] The multi-step rolling prediction equation is: Y t+1：t+H = F(X t-T+1：t , θ * ) + ⊥·ΔD, where Y t+1：t+H is the prediction output for the next H time steps, F() is the hybrid simulation model function, whose input is the historical state sequence X t-T+1：t , θ * is the optimization parameter, ⊥ is the emergency impact matrix, and ΔD is the real-time event feature vector.
[0176] In the S3 step, it also includes predictive model-based reinforcement learning, and the model of reinforcement learning is: where:
[0177] Q(s t , a t ) is the expected cumulative reward for executing action a t in state s t ;
[0178] η is the learning rate, η = 0.1;
[0179] γ t is the immediate reward, q1 is the predicted flow, q2 is the observed flow, ψ is the stability penalty coefficient, and θ = [τ, β, p lc ;
[0180] s t is the state vector, μ(ρ) is the average density of the road network, and σ(v) is the standard deviation of the speed, is the flow rate change rate, and Rmean is the average residual of multi-source data;
[0181] a t is the action vector, a t = [Δβ, Δτ, Δp lc , Δβ is the speed-density index adjustment amount, Δτ is the driver reaction time adjustment amount, and Δp lc is the lane-changing probability adjustment amount
[0182] The dynamic adaptability of the present invention is improved. Through the online optimization of the β parameter (for example, the β value is adjusted from 1.2 to 2.3 during congestion), the speed-density relationship curve dynamically fits the actual situation, and the prediction error is reduced compared with the fixed parameter model;
[0183] The data fusion efficiency of the present invention is broken through, and the spatio-temporal weighting mechanism improves the utilization rate of multi-source data.
[0184] The real-time optimization ability of the present invention is enhanced. Reinforcement learning shortens the prediction response time of the simulation model after a traffic accident, which is better than the traditional manual parameter adjustment mode.
[0185] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.
Claims
1. A simulation-based traffic flow prediction system, characterized in that, Including: A data acquisition module, which includes a floating vehicle GPS, a geomagnetic sensor, a monitoring camera, and a meteorological data acquisition unit; A data processing module, which is built-in with a data preprocessing unit and a dynamic simulation model; A reinforcement learning optimization module, which performs reinforcement learning on the dynamic simulation model; A prediction output module, which obtains a prediction result based on the detected data and the dynamic simulation model; A visualization output module, which visually outputs the prediction result; The floating vehicle GPS is an OBD terminal installed in a taxi; The geomagnetic sensor uses an AMR magnetoresistive array; The monitoring camera uses a bullet-type network camera; The meteorological data acquisition unit uses a roadside meteorological station.
2. A simulation-based traffic flow prediction method, which is the method of a simulation-based traffic flow prediction system described in claim 1, characterized in that, Specifically, it includes the following steps: S1: Data acquisition, data is acquired through the floating vehicle GPS, the geomagnetic sensor, the monitoring camera, and the meteorological data acquisition unit; S2: Data preprocessing, the data preprocessing unit performs time synchronization and spatial mapping on the acquired multi-source data respectively; S3: Dynamic simulation, a dynamic simulation model is established, and the preprocessed acquired data is used as input, and then a prediction result is output; S4: Visualization output, the prediction result is visually output.
3. The traffic flow prediction method based on simulation according to claim 2, wherein In the step S2, the model for time synchronization is as follows: t raw is the original timestamp, and t align is the synchronized timestamp, where T is the system reference time unit; In the S2 step, T = 30 seconds or T = 60 seconds.
4. A simulation-based traffic flow prediction method according to claim 3, characterized in that In the step S2, the model of spatial mapping is as follows: where ρ seg is the average density of the road surface, v i is the speed of the i-th vehicle, and δ i is the proportion of the residence time of vehicle i in section L.
5. The traffic flow prediction method based on simulation according to claim 2, wherein In the S3 step, the dynamic simulation model includes a macroscopic layer based on the cell transmission model, a microscopic layer based on the intelligent driver model, and a multi-step rolling prediction equation.
6. The traffic flow prediction method based on simulation according to claim 5, wherein The logical model at the macroscopic level is as follows: where ρ(x, t) represents the vehicle density at position x and time t, v represents the speed and is output by the microscopic model, and ε(t) is a random perturbation term that follows N(0, σ 2 ), represents the rate of change of density over time, represents the gradient of the vehicle flow speed with respect to position, and S in is the vehicle inflow rate at the section entrance, and S out is the vehicle outflow rate at the section exit.
7. A simulation-based traffic flow prediction method according to claim 6, characterized in that The logical model of the micro layer is as follows: where a n is the acceleration of the nth vehicle, τ is the driver's reaction time, α and β are sensitivity parameters, and Δx n (t) represents the distance involved, and v d is the driver's desired speed.
8. A simulation-based traffic flow prediction method according to claim 7, characterized in that In the microscopic layer, τ = (0.8s, 2.4s).
9. A simulation-based traffic flow prediction method according to claim 7, characterized in that, The multi-step rolling prediction equation is: Y t+1：t+H = F(X t-T+1：t , θ * ) + ⊥·ΔD, where Y t+1：t+H is the prediction output for the next H time steps, F() is the hybrid simulation model function, whose input is the historical state sequence X t-T+1：t , θ * are the optimization parameters, ⊥ is the emergency impact matrix, and ΔD is the real-time event feature vector.
10. A traffic flow prediction method based on simulation according to claim 2, characterized in that, In the step S3, it further includes predictive model-based reinforcement learning, and the model of reinforcement learning is: Q(s t , a t ) ← Q(s t , a t ) + η[γ t + λ·maxQ(s t+1 , a) - Q(s t , a t )], where: a Q(s t , a t ) is the expected cumulative reward for executing action a t in state s t ; η is the learning rate, η = (0.01, 0.1); γ t For immediate reward, q1 is the predicted flow rate, q2 is the observed flow rate, ψ is the stability penalty coefficient, and θ = [τ, β, p lc ; s t is the state vector, μ(ρ) is the average density of the road network, and σ(v) is the standard deviation of speed, is the flow rate change rate, and Rmean is the average residual of multi-source data; a t is the action vector, a t = [Δβ, Δτ, Δp lc , where Δβ is the speed-density index adjustment amount, Δτ is the driver reaction time adjustment amount, and Δp lc is the lane change probability adjustment amount.