Tunnel congestion node real-time estimation method and system based on digital twinning

Abstract

The invention provides a tunnel congestion node real-time estimation method and system based on digital twinning, and the construction of a digital twinning system physical scene of a tunnel is realized based on a Unity engine and SUMO software by determining the spatial characteristics, traffic information, camera information and the like of the tunnel. And then, dynamic and static data information of the vehicle is obtained through a CornerNet image detection network and a Deepsort tracking algorithm, the driving condition of the lost vehicle is deduced and simulated according to an acceleration distance control car-following model and a safe distance lane changing model, and dynamic target projection and full tracking of the digital twin system are realized based on an improved Hungary algorithm. And finally, node division of vehicles in the tunnel is realized according to the proposed intelligent traffic flow node model, node saturation is introduced, a node coupling congestion identification model is proposed, and tunnel congestion event sensing and congestion node positioning are realized. According to the invention, on one hand, traffic management resumption in the tunnel can be carried out by means of the digital twin system, and the congestion process and reasons are deeply analyzed; and on the other hand, the traffic flow congestion condition in the tunnel can be detected in real time, and the congestion node can be determined, so that the congestion traffic flow can be dredged in advance, and the accident occurrence probability in the tunnel can be reduced. And the method is of great significance to future digital and intelligent traffic flow research.

Description

Technical Field

[0001] The present invention belongs to the field of traffic flow and intelligent transportation technology. Based on image detection, simulation deduction and digital twin system, a real-time estimation method and system of tunnel congestion nodes based on digital twin are proposed for the real-time driving conditions of vehicles in the tunnel. Background Art

[0002] Tunnels are an important part of road construction. There are many mountains and hills in my country. Therefore, in the process of road construction, especially the construction of expressways and national highways, it is inevitable to dig tunnels. Tunnels have the function of shortening vehicle travel distance, improving vehicle operation efficiency, and improving the traffic capacity of the road network. However, due to the narrow and closed space in the tunnel, poor lighting conditions, low visibility and other conditions, the probability of traffic accidents in tunnel sections is higher than that in ordinary highway sections. The current tunnel traffic management methods mainly include flow control, speed control, lane control and global road network control. These methods can alleviate traffic congestion and reduce traffic accidents to a certain extent, but they require a lot of manpower and have lag.

[0003] Existing research on vehicle congestion and collision in tunnels is mainly divided into three directions: research on bottleneck congestion nodes, research on variable speed limits, and research on road diversion methods. Among them, the research on bottleneck congestion nodes is mainly devoted to analyzing the traffic flow characteristics of vehicles and the phenomenon of reduced traffic capacity at bottlenecks. However, these studies, on the one hand, do not study the traffic flow characteristics at the tunnel entrance insufficiently, and on the other hand, cannot analyze the traffic flow in real time and determine the congestion nodes.

[0004] In recent years, with the development of traffic flow theory and digital twin technology, the method of using digital twin technology to determine congestion nodes in tunnels has been realized. Digital twins mainly collect data from the real world, use data to perform three-dimensional modeling, and deeply restore the real world through tools such as automated modeling, and can perform simulations. Digital twins can be used for traffic management review, restore accident scenes, and deeply analyze congestion processes and causes; predicting congestion can guide traffic in advance, improve traffic efficiency, and reduce the probability of accidents. Therefore, the real-time estimation method and system of tunnel congestion nodes based on digital twins is a novel and practical research. Summary of the invention

[0005] In view of the shortcomings of the existing technology, the present invention proposes a real-time estimation method and system for tunnel congestion nodes based on digital twins. The physical scene of the digital twin system is constructed by collecting tunnel information, and data of vehicles in the tunnel scene is collected, detected and simulated through video streams, and sent to an adaptive collaborative simulation projection model to realize dynamic target projection and full tracking, thereby realizing the perception of tunnel congestion events and the positioning of congested nodes.

[0006] Technical solution: A method and system for real-time estimation of tunnel congestion nodes based on digital twins, comprising the following steps:

[0007] S10: Determine and record spatial characteristics of the tunnel (including tunnel structure, number of lanes, lane width, etc.) and traffic information (road capacity, etc.) and data information such as the location and number of cameras in the tunnel;

[0008] S20: Realize the physical scene construction of the digital twin system. Based on the Unity engine, reconstruct the tunnel space features obtained by S10 in the virtual scene, and connect to the SUMO software to build a collaborative simulation environment.

[0009] S30: Realize high-precision detection of vehicles in tunnel scenes. Read the video streams shot by each camera, obtain the current road image, select the CornerNet deep learning image detection network and Deepsort tracking algorithm, optimize the detection and tracking effect in dim light and special bird's-eye view of the tunnel, obtain the basic characteristics of the vehicle, traffic parameters, dynamic characteristics and other data information, and send them to the adaptive collaborative simulation projection model.

[0010] S40: Realize real-time simulation of lost vehicles. For vehicles that leave the detection area (road sections not covered by cameras), considering the impact of poor light and narrow space on driving behavior, a vehicle following model (acceleration interval control following model) and a vehicle lane changing model (safety interval lane changing model) based on tunnel conditions are proposed to optimize the simulation accuracy of vehicle driving in tunnel sections and the simulation accuracy of driving behavior, simulate and deduce vehicle driving conditions, and send them to the adaptive collaborative simulation projection model.

[0011] S50: Realize dynamic target projection and full tracking of digital twin system. Propose an adaptive collaborative simulation projection model. Based on real-time detection results and simulation deduction results, simulated vehicles are projected into the physical scene on sections not covered by cameras; on sections covered by cameras, based on the improved Hungarian algorithm, the detected vehicles and simulated vehicles in the sections are adaptively matched and deduplicated.

[0012] S60: Implement vehicle node division in the tunnel. Based on the vehicle information in the tunnel and the position information between vehicles, an intelligent vehicle flow node model is proposed.

[0013] S70: Realize tunnel congestion event perception and congestion node location. Couple the congestion index to identify slow-moving convoys, and then introduce node saturation based on the intelligent traffic node model, and propose a node coupling congestion identification model to determine the location of congestion nodes.

[0014] Furthermore, the road capacity refers to the maximum number of vehicles that can pass through the road in a unit of time, which is generally affected by road conditions, traffic conditions, weather conditions, control conditions, etc.

[0015] The SUMO simulation software is an open source, microscopic and continuous traffic simulation software that can simulate the driver's driving behavior and the vehicle's driving trajectory based on a self-developed model.

[0016] The CornerNet image detection algorithm is a single-stage target detection algorithm, which greatly simplifies the output of the network and does not require the design of anchor bounding boxes, and can efficiently and accurately identify vehicle information.

[0017] The Deepsort tracking algorithm is a multi-target tracking algorithm, which generally includes three parts: image feature extraction, target feature matching and target position prediction, and can obtain continuous and complete vehicle IDs.

[0018] The basic characteristics of the vehicle include static data information such as vehicle color, vehicle outline size and license plate number, the traffic flow parameters include static data information such as the number of vehicles and traffic flow density, and the dynamic characteristics include dynamic data information such as vehicle speed and vehicle position.

[0019] Furthermore, the data information of S10 is mainly collected manually except for the road capacity, and the road capacity is calculated by the capacity formula.

[0020] Furthermore, the physical scene construction of the S20 specifically includes: editing the tunnel information into the SUMO software, using the scene creation tool to build a virtual digital twin tunnel, and marking the location and coverage of the camera on the tunnel.

[0021] Further, the S30 includes the following steps:

[0022] S31: Read the video streams of each camera, obtain the current road video, and improve the video brightness through the image processing tool Photoshop.

[0023] S32: Using the CornerNet algorithm as the detection part of Deepsort, the trained Deepsort model is used to detect the processed image to obtain the detection result and its confidence. It is determined whether the obtained confidence reaches the established threshold. If so, the obtained complete and continuous vehicle data information is sent to the adaptive collaborative simulation projection model.

[0024] Furthermore, the S40 includes the following steps: the vehicle driving conditions on the road section not covered by the camera are mainly simulated by the following two models:

[0025] S41: Determine whether there is a following behavior between vehicles. The following state can be determined by a determination method based on the actual vehicle speed, the expected vehicle speed and the distance between the two vehicles. The constraint expression is:

[0026] v n-1 (t)≤v n,e (t) (1)

[0027] X n-1 (t)-X n (t)≥150 (2)

[0028] Where: n——number of vehicles;

[0029] v n-1 (t)——actual speed of the n-1th vehicle, in m / s;

[0030] v n,e (t)——the expected speed of the nth vehicle;

[0031] X n (t)——the driving displacement of the nth vehicle, in meters;

[0032] If the above constraints are met between vehicles, there is a following behavior between the vehicles.

[0033] If there is a following behavior between vehicles, on the one hand, considering that the following vehicle needs to adjust its own speed according to its own driving speed, the distance difference and speed difference with the leading vehicle, as well as the difference between the driver's psychological expectations and the actual situation during the following process; on the other hand, considering the poor light and narrow space in the tunnel, the minimum expected distance of vehicles following needs additional adjustment compared with normal roads. This paper proposes the following following model (named: acceleration distance control model): Formula (3) shows that the actual acceleration of the vehicle is affected by three factors: expected acceleration, minimum expected distance and actual distance; Formula (4) considers the impact of lane change behavior on the minimum expected distance, and uses coefficients to control the distance calculation under different behaviors.

[0034]

[0035]

[0036] Δv n (t) = v n (t)-v n-1 (t) (5)

[0037] In the formula, a n (t)——actual acceleration of the nth vehicle; a n,e (t)——expected maximum acceleration of the nth vehicle;

[0038] s1 ——Minimum expected spacing; s 2 ——actual spacing; s 3 Minimum distance required for following vehicles;

[0039] h t ——Expected headway;

[0040] m, p——adjustment coefficients; when the vehicle is changing lanes, m=0, n=1; when the vehicle is in other following states, m=1, n=0;

[0041] V——The maximum speed allowed in the tunnel;

[0042] Δv n (t)——the speed difference between the nth vehicle and the preceding vehicle;

[0043] b——comfortable deceleration;

[0044] If there is no following behavior between vehicles, the driving speed of the simulated vehicle is:

[0045] v n (t) = v n (t)+ε (6)

[0046] Where ε is the error term or random disturbance term;

[0047] S42: Considering the lighting conditions and limited vision in the tunnel, vehicles need to keep a greater distance from other vehicles when changing lanes and overtaking compared to normal roads. Based on this, this paper introduces the following lane-changing model (named: safe distance lane-changing model):

[0048] First of all, the lane-changing behavior of vehicles is mostly due to the fact that the speed of the front vehicle is lower than the expected speed of the rear vehicle. Therefore, the first premise of lane-changing is:

[0049] v n-1 (t)≤v n,e (t) (7)

[0050] In addition, the vehicle lane change needs to ensure that there is at least a safe distance between the vehicle and the vehicle in the adjacent lane to prevent collision. Therefore, the second prerequisite for the vehicle lane change is:

[0051]

[0052]

[0053]

[0054] In the formula, s n,a(t)——the distance between the nth vehicle and the preceding vehicle in the adjacent lane at time t; s n,b (t)——the distance between the nth vehicle and the vehicle behind it in the adjacent lane at time t; s n (t)——the distance between the nth vehicle and the preceding vehicle in the lane at time t;

[0055] D n,a (t)——the safe distance between the nth vehicle and the preceding vehicle in the adjacent lane at time t; D n,b (t)——the safe distance between the nth vehicle and the vehicle behind it in the adjacent lane at time t;

[0056] l n ——The length of the nth vehicle;

[0057] t r ——Driver reaction time, generally 0.5s to 1.5s;

[0058] v n-1,a (t)——the speed of the preceding vehicle in the adjacent lane at time t; v n+1,b (t)——the speed of the vehicle behind in the adjacent lane at time t;

[0059] a n-1,a (t)——the acceleration of the preceding vehicle in the adjacent lane at time t; a n+1,b (t)——the acceleration of the vehicle behind in the adjacent lane at time t;

[0060] If the vehicle meets the above two prerequisites during driving, it will change lanes.

[0061] Furthermore, the adaptive collaborative simulation projection model based on tunnel driving mainly realizes the dynamic target projection of the digital twin system, providing a basis for determining tunnel congestion nodes. The model performs different dynamic target projections on different sections of the tunnel based on the above-mentioned real-time detection results and simulation deduction results: for sections not covered by cameras, the simulated vehicles deduced are projected into the physical scene; for sections covered by cameras, based on the improved Hungarian algorithm, the simulated vehicles in the sections covered by cameras are matched with the detected vehicles for deduplication.

[0062] Furthermore, the improved Hungarian algorithm solves the problem of matching and deduplication of simulated vehicles and detected vehicles. The main improvement is that in order to better improve the matching accuracy of simulated vehicles and detected vehicles, this paper introduces the expression E representing the position relationship between the simulated vehicle and the projected vehicle: mn , this formula fully considers the influence of the position information of the simulated vehicle and the projected vehicle on the position relationship between the two under different coefficients. mn The smaller it is, the closer the distance between the simulated vehicle and the projected vehicle is, and the better the match is. So let the objective function be E mn With variables The minimum value of the product. The improved algorithm model is as follows:

[0063] Introducing E mn To represent the positional relationship between the mth simulated vehicle and the nth projected vehicle.

[0064]

[0065] Define the decision variables as:

[0066]

[0067] The matching problem is transformed into:

[0068]

[0069]

[0070] In the formula, and Respectively represent the position information of the mth simulated vehicle and the nth projected vehicle. j1 represents the set of simulated vehicles, D j2 Represents the set of projected vehicles.

[0071] Furthermore, the intelligent traffic flow node model of the S60 is as follows:

[0072] According to the vehicle data information in the adaptive collaborative simulation projection model, this paper regards different vehicles as different movable nodes, constructs a road network, and proposes an intelligent traffic flow node model Node=(V,E,d ij ), where V is a node set, representing the vehicles in the road network; E is an edge set, representing the positional relationship between two vehicles in the road network; d ij is the weight, which represents the position information between the i-th vehicle and the j-th vehicle. Since the vehicle information of the adaptive collaborative simulation projection model is updated in real time, the node set, edge set and weight are also updated in real time.

[0073] Further, the S70 includes the following steps:

[0074] S71: Determine the traffic congestion situation.

[0075] Considering that the measurement of traffic congestion is related to people's psychological tolerance, a custom quantitative indicator "congestion degree" is proposed to measure the congestion situation. In this article, congestion degree refers to the loss ratio of traffic flow during driving compared with normal smooth driving. Its formula can be expressed as follows:

[0076]

[0077]

[0078] Where: C——congestion degree;

[0079] S 1 ——Current traffic volume in the tunnel; S 2 Tunnel capacity;

[0080] i——density ratio, which represents the current traffic density k and traffic congestion density k j The ratio of

[0081] n——parameter, which can be regarded as a fixed constant;

[0082] If the convoy congestion degree C is greater than 80%, the road section is deemed to be congested and the process proceeds to S52.

[0083] S72: Determine congestion nodes and issue warnings. Based on the intelligent traffic node model, we introduce node saturation and propose a node coupling congestion identification model to determine the location of congestion nodes:

[0084] x n (t) = |(1-β)f(x n (t))+βf(x n-1 (tt* ) )|+b i H (17)

[0085] f(x n (t)) = x n (t)(1-x n (t)) (18)

[0086]

[0087] Where: x n (t)——Whether the nth node (nth vehicle) is congested at time t, with a value of [0,1]. When it is greater than 0.8, the node is considered congested;

[0088] β——coupling strength between vehicles, value is (0,1];

[0089] f(x n (t))——a mapping function of the nth node;

[0090] t * ——Delay time: the driving behavior of the n-1th vehicle has a hysteresis effect on the driving behavior of the nth vehicle;

[0091] ——initial saturation of the nth node;

[0092] H——disturbance value, indicating the disturbance value of the tunnel to the node, simulating the degree of congestion;

[0093] When x n When the (t) value is greater than 0.8, it indicates that congestion occurs at the nth vehicle, and the system issues an early warning for the location of the congested node.

[0094] Beneficial effects: The present invention proposes a real-time estimation method and system for tunnel congestion nodes based on digital twins, implements the construction of the physical scene of the digital twin system based on the Unity engine and SUMO software, obtains vehicle dynamic and static data information through the CornerNet image detection network and the Deepsort tracking algorithm, deduces and simulates vehicle conditions based on the acceleration interval control following model and the safe interval lane change model, and implements dynamic target projection and full tracking of the digital twin system based on the improved Hungarian algorithm. Finally, according to the proposed intelligent traffic node model and node coupling congestion identification model, the perception of tunnel congestion events and the positioning of congestion nodes are realized. On the one hand, the present invention can use the digital twin system to review the traffic management in the tunnel and deeply analyze the congestion process and causes; on the other hand, it can detect the traffic congestion in the tunnel in real time and determine the congestion nodes, so as to realize the early diversion of congested traffic and reduce the probability of accidents in the tunnel. It is of great significance to the future research on digital and intelligent traffic flow. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 It is a schematic diagram of the process of the present invention;

[0096] Figure 2 This is a bird’s-eye view of the digital twin system scene;

[0097] Figure 3 This is a side view of the digital twin system scenario;

[0098] Figure 4 This is the vehicle ID map obtained by tracking based on CornerNet and Deepsort algorithms. DETAILED DESCRIPTION

[0099] The present invention will now be further described with reference to the embodiments and the accompanying drawings.

[0100] Example:

[0101] The real-time estimation method and system of tunnel congestion nodes based on digital twins in this embodiment, the process is shown in Figure 1 Specifically, it can be divided into static physical scene construction, vehicle adaptive collaborative simulation, and tunnel congestion event perception and congestion node positioning.

[0102] Step 1: Implement the physical scene construction of the digital twin system, which is divided into the following two steps:

[0103] Information collection: This embodiment takes a tunnel in China from 10:00 to 11:00 on November 2, 2020 as the specific observation tunnel. The basic information of the tunnel is as follows:

[0104] Table 1 Tunnel specifications

[0105] Total length of tunnel 5853m Lane Design Six lanes in both directions Tunnel Width Two lanes are 3.5 meters wide and one lane is 3.75 meters wide Design speed 80 km / h road capacity 1800pcu / h

[0106] The number and location of cameras in the tunnel are shown in the following table. The effective distance for camera ID recognition is approximately 60m, and the speed measurement distance is approximately 300m.

[0107] Table 2 Cameras built into tunnels

[0108] Camera Name Camera Position Camera 1 Tunnel entrance Camera 2 943m from the tunnel entrance Camera 3 1925m from the tunnel entrance Camera 4 3357m from the tunnel entrance Camera 5 4883m from the tunnel entrance Camera 6 5853 meters from the tunnel entrance (i.e. exit)

[0109] Physical scene construction: Based on the Unity engine, the information about the tunnel and the camera is edited and written into the SUMO software. The scene creation tool is used to build a virtual digital twin tunnel, and the location and coverage of the camera are marked on the tunnel.

[0110] Step 2: Implement dynamic target projection and full tracking of the digital twin system, which includes the following three steps:

[0111] Get traffic data information:

[0112] The video streams of each camera are read in, the current road video is obtained, and the video brightness is improved through the image processing tool Photoshop. Then, the CornerNet algorithm is used as the detection part of Deepsort, and the trained Deepsort model is used to detect the processed video to obtain the detection result and its confidence. It is judged whether the obtained confidence reaches the established threshold. If it does, the obtained complete and continuous vehicle data information (including basic vehicle characteristics, traffic flow parameters and dynamic characteristics) is sent to the adaptive collaborative simulation projection model.

[0113] Realize real-time simulation of lost vehicles:

[0114] For vehicles on roads not covered by cameras, their driving conditions are mainly simulated by the following two models:

[0115] First, to determine whether there is a following behavior between vehicles, the following state can be determined by a discrimination method based on the actual vehicle speed, the expected vehicle speed and the distance between the two vehicles. The constraint expression is:

[0116] v n-1 (t)≤v n,e (t) (1)

[0117] X n-1 (t)-Xn (t)≥150 (2)

[0118] Where: n——number of vehicles;

[0119] v n-1 (t)——actual speed of the n-1th vehicle, in m / s;

[0120] v n,e (t)——the expected speed of the nth vehicle;

[0121] X n (t)——the driving displacement of the nth vehicle, in meters;

[0122] If the above constraints are met between vehicles, there is a following behavior between the vehicles.

[0123] If there is a following behavior between vehicles, on the one hand, the following vehicle needs to adjust its own speed according to its own driving speed, the distance difference and speed difference with the leading vehicle, as well as the difference between the driver's psychological expectation and the actual situation during the following process; on the other hand, considering the poor light and narrow space in the tunnel, the minimum expected distance of vehicles following needs additional adjustment compared with normal roads. This paper proposes an acceleration distance control model: Formula (3) shows that the actual acceleration of the vehicle is affected by three factors: expected acceleration, minimum expected distance and actual distance; Formula (4) considers the impact of lane change behavior on the minimum expected distance, and uses coefficients to control the distance calculation under different behaviors.

[0124]

[0125]

[0126] Δv n (t) = v n (t)-v n-1 (t) (5)

[0127] In the formula, a n (t)——actual acceleration of the nth vehicle; a n,e (t)——expected maximum acceleration of the nth vehicle;

[0128] s 1 ——Minimum expected spacing; s 2 ——actual spacing; s 3 Minimum distance required for following vehicles;

[0129] h t ——Expected headway;

[0130] m, p——adjustment coefficients; when the vehicle is changing lanes, m=0, p=1; when the vehicle is in other following states, m=1, p=0;

[0131] V——The maximum speed allowed in the tunnel, here it is 80km / h;

[0132] Δv n (t)——the speed difference between the nth vehicle and the preceding vehicle;

[0133] b——Comfortable deceleration, the value range is (0,1], and the specific value depends on the road conditions;

[0134] If there is no following behavior between vehicles, the driving speed of the simulated vehicle is:

[0135] v n (t) = v n (t)+ε (6)

[0136] Where ε is the error term or random disturbance term;

[0137] Secondly, considering the lighting conditions and limited vision in the tunnel, vehicles need to keep a greater distance from other vehicles when changing lanes and overtaking compared to normal roads. Based on this, this paper proposes a safe distance lane changing model:

[0138] First of all, the lane-changing behavior of vehicles is mostly due to the fact that the speed of the front vehicle is lower than the expected speed of the rear vehicle. Therefore, the first premise of lane-changing is:

[0139] v n-1 (t)≤v n,e (t) (7)

[0140] In addition, the vehicle lane change needs to ensure that there is at least a safe distance between the vehicle and the vehicle in the adjacent lane to prevent collision. Therefore, the second prerequisite for the vehicle lane change is:

[0141]

[0142]

[0143]

[0144] In the formula, s n,a (t)——the actual distance between the nth vehicle and the preceding vehicle in the adjacent lane at time t; s n,b (t)——the actual distance between the nth vehicle and the vehicle behind it in the adjacent lane at time t; s n (t)——the actual distance between the nth vehicle and the vehicle ahead in the lane at time t;

[0145] D n,a(t)——the safe distance between the nth vehicle and the preceding vehicle in the adjacent lane at time t; D n,b (t)——the safe distance between the nth vehicle and the vehicle behind it in the adjacent lane at time t;

[0146] l n ——The length of the nth vehicle;

[0147] t r ——Driver reaction time, generally 0.5s to 1.5s, here 0.8s;

[0148] v n-1,a (t)——the speed of the preceding vehicle in the adjacent lane at time t; v n+1,b (t)——the speed of the vehicle behind in the adjacent lane at time t;

[0149] a n-1,a (t)——the acceleration of the preceding vehicle in the adjacent lane at time t; a n+1,b (t)——the acceleration of the vehicle behind in the adjacent lane at time t;

[0150] If the vehicle meets the above two prerequisites during driving, lane changing will be performed.

[0151] Realize dynamic target projection and full tracking of digital twin system:

[0152] After simulation, an adaptive collaborative simulation projection model is proposed. This adaptive collaborative simulation projection model based on tunnel driving mainly realizes the dynamic target projection of the digital twin system and provides a basis for determining tunnel congestion nodes. The model performs different dynamic target projections on different sections of the tunnel based on the above-mentioned real-time detection results and simulation results: for sections not covered by cameras, the simulated vehicles obtained by simulation are projected into the physical scene; for sections covered by cameras, based on the improved Hungarian algorithm, the simulated vehicles in the sections covered by cameras are matched with the detected vehicles for deduplication.

[0153] The improved Hungarian algorithm solves the problem of matching and deduplication of simulated vehicles and detected vehicles. The main improvement is that in order to better improve the matching accuracy of simulated vehicles and detected vehicles, this paper introduces the expression E that represents the position relationship between the simulated vehicle and the projected vehicle. mn , this formula fully considers the influence of the position information of the simulated vehicle and the projected vehicle on the position relationship between the two under different coefficients. mn The smaller it is, the closer the distance between the simulated vehicle and the projected vehicle is, and the better the match is. So let the objective function be E mn With variables The minimum value of the product. The improved algorithm model is as follows:

[0154] Introducing E mnTo represent the positional relationship between the mth simulated vehicle and the nth projected vehicle.

[0155]

[0156] Define the decision variables as:

[0157]

[0158] The matching problem is transformed into:

[0159]

[0160]

[0161] In the formula, and Respectively represent the position information of the mth simulated vehicle and the nth projected vehicle. j1 represents the set of simulated vehicles, D j2 Represents the set of projected vehicles.

[0162] Step 3: Realize tunnel congestion event perception and congestion node location, which is divided into the following two steps:

[0163] Establishing intelligent traffic flow node model:

[0164] In order to better determine the movement status and position of the traffic flow, the intelligent traffic flow node model is as follows: According to the vehicle data information in the adaptive collaborative simulation projection model, the present invention regards different vehicles as different movable nodes, constructs a road network, and proposes an intelligent traffic flow node model Node = (V, E, d ij ), where V is a node set, representing the vehicles in the road network; E is an edge set, representing the positional relationship between two vehicles in the road network; d ij is the weight, which represents the position information between the i-th vehicle and the j-th vehicle. Since the vehicle information of the adaptive collaborative simulation projection model is updated in real time, the node set, edge set and weight are also updated in real time.

[0165] Determine traffic congestion, identify congestion nodes and issue warnings:

[0166] The first step is to determine the traffic congestion. Considering that the measurement of traffic congestion is related to people's psychological tolerance, a self-defined quantitative indicator "congestion degree" is proposed to measure the congestion. In the present invention, congestion degree refers to the loss ratio of traffic flow during driving compared with normal smooth driving. Its formula can be expressed as follows:

[0167]

[0168]

[0169] Where: C——congestion degree;

[0170] S 1 ——Current traffic volume in the tunnel; S 2 Tunnel capacity, here taken as 5400pcu / h;

[0171] i——density ratio, which represents the current traffic density k and traffic congestion density k j The ratio of

[0172] n——parameter, which can be regarded as a fixed constant;

[0173] If the fleet congestion degree C is greater than 80%, the road section is considered congested and proceed to the next step.

[0174] The second step is to identify the congestion nodes and issue warnings. Based on the intelligent traffic node model, we introduce node saturation and propose a node coupling congestion identification model to determine the location of congestion nodes:

[0175] x n (t) = |(1-β)f(x n (t))+βf(x n-1 (tt * ))|+b i H (17)

[0176] f(x n (t)) = x n (t)(1-x n (t)) (18)

[0177]

[0178] Where: x n (t)——Whether the nth node (nth vehicle) is congested at time t, with a value of [0,1]. When it is greater than 0.8, the node is considered congested;

[0179] β——coupling strength between vehicles, value is (0,1];

[0180] f(x n (t))——a mapping function of the nth node;

[0181] t * ——Delay time: the driving behavior of the n-1th vehicle has a hysteresis effect on the driving behavior of the nth vehicle;

[0182] ——initial saturation of the nth node;

[0183] H——disturbance value, indicating the disturbance value of the tunnel to the node, simulating the degree of congestion;

[0184] When x n When the (t) value is greater than 0.8, it indicates that congestion occurs at the nth vehicle, and the system issues an early warning for the location of the congested node.

[0185] In this example, since the congestion degree C of the fleet is proportional to x n (t) have all changed. Due to space limitations, this article only lists the x obtained by the tunnel vehicle deduced in the digital twin system from 10 to 11 o'clock. n The four times and locations where the (t) value is greater than 0.8 (a continuous congestion in a certain fleet is considered as one congestion) are shown in the following table:

[0186] time Vehicle ID Location 10:07 153 2319m from the entrance 10:27 517 4035m from the entrance 10:29 495 1121m from the entrance 10:48 1254 1374m from the entrance

Claims

1. A real-time estimation method and system for tunnel congestion nodes based on digital twins, characterized by: The following steps are involved: S10: Determine and record spatial characteristics of the tunnel (including tunnel structure, number of lanes, lane width, etc.) and traffic information (road capacity, etc.) and data information such as the location and number of cameras in the tunnel; S20: Realize the physical scene construction of the digital twin system. Based on the Unity engine, reconstruct the tunnel space features obtained by S10 in the virtual scene, and connect to the SUMO software to build a collaborative simulation environment. S30: Realize high-precision detection of vehicles in tunnel scenes. Read the video streams shot by each camera, obtain the current road image, select the CornerNet deep learning image detection network and Deepsort tracking algorithm, optimize the detection and tracking effect in dim light and special bird's-eye view of the tunnel, obtain the basic characteristics of the vehicle, traffic parameters, dynamic characteristics and other data information, and send them to the adaptive collaborative simulation projection model. S40: Realize real-time simulation of lost vehicles. For vehicles that leave the detection area (road sections not covered by cameras), considering the impact of poor light and narrow space on driving behavior, a vehicle following model (acceleration interval control following model) and a vehicle lane changing model (safety interval lane changing model) based on tunnel conditions are proposed to optimize the simulation accuracy of vehicle driving in tunnel sections and the simulation accuracy of driving behavior, simulate and deduce vehicle driving conditions, and send them to the adaptive collaborative simulation projection model. S50: Realize dynamic target projection and full tracking of digital twin system. Propose an adaptive collaborative simulation projection model. Based on the real-time detection results and simulation results, the simulated vehicle is projected into the physical scene on the road section not covered by the camera; On the road sections covered by the cameras, the detected vehicles and simulated vehicles on the road sections are adaptively matched and deduplicated based on the improved Hungarian algorithm. S60: Implement vehicle node division in the tunnel. Based on the vehicle information in the tunnel and the position information between vehicles, an intelligent vehicle flow node model is proposed. S70: Realize tunnel congestion event perception and congestion node location. Couple the congestion index to identify slow-moving convoys, and then introduce node saturation based on the intelligent traffic node model, and propose a node coupling congestion identification model to determine the location of congestion nodes.

2. According to a method and system for estimating tunnel congestion nodes based on digital twins according to claim 1, it is characterized by: The road capacity refers to the maximum number of vehicles that can pass through the road in a unit of time, which is generally affected by road conditions, traffic conditions, weather conditions, control conditions, etc. The SUMO simulation software is an open source, microscopic and continuous traffic simulation software that can simulate the driver's driving behavior and the vehicle's driving trajectory based on a self-developed model. The CornerNet image detection algorithm is a single-stage target detection algorithm, which greatly simplifies the output of the network and does not require the design of anchor bounding boxes, and can efficiently and accurately identify vehicle information. The Deepsort tracking algorithm is a multi-target tracking algorithm, which generally includes three parts: image feature extraction, target feature matching and target position prediction, and can obtain continuous and complete vehicle IDs. The basic characteristics of the vehicle include static data information such as vehicle color, vehicle outline size and license plate number, the traffic flow parameters include static data information such as the number of vehicles and traffic flow density, and the dynamic characteristics include dynamic data information such as vehicle speed and vehicle position.

3. According to a method and system for estimating tunnel congestion nodes based on digital twins according to claim 1, it is characterized by: The data information of S10 is mainly collected manually except for the road capacity, and the road capacity is calculated by the capacity formula.

4. According to a method and system for estimating tunnel congestion nodes based on digital twins according to claim 1, it is characterized by: The physical scene construction of the S20 specifically includes: editing the tunnel information into the SUMO software, using the scene creation tool to build a virtual digital twin tunnel, and marking the location and coverage of the camera on the tunnel.

5. According to a method and system for estimating tunnel congestion nodes based on digital twins according to claim 1, it is characterized by: The S30 includes the following steps: S31: Read the video streams of each camera, obtain the current road video, and improve the video brightness through the image processing tool Photoshop. S32: Using the CornerNet algorithm as the detection part of Deepsort, the trained Deepsort model is used to detect the processed image to obtain the detection result and its confidence. It is determined whether the obtained confidence reaches the established threshold. If so, the obtained complete and continuous vehicle data information is sent to the adaptive collaborative simulation projection model.

6. A tunnel congestion node estimation method and system based on digital twin according to claims 1 and 5, characterized in that: The S40 includes the following steps: the vehicle driving conditions on the road section not covered by the camera are mainly simulated by the following two models: S41: Determine whether there is a following behavior between vehicles. The following state can be determined by a determination method based on the actual vehicle speed, the expected vehicle speed and the distance between the two vehicles. The constraint expression is: v n-1 (t)≤v n，e (t) (1) X n-1 (t)-X n (t)≥150 (2) Where: n——number of vehicles; v n-1 (t)——actual speed of the n-1th vehicle, in m / s; v n，e (t)——the expected speed of the nth vehicle; X n (t)——the driving displacement of the nth vehicle, in meters; If the above constraints are met between vehicles, there is a following behavior between the vehicles. If there is a following behavior between vehicles, on the one hand, considering that the rear vehicle needs to adjust its own speed according to its own driving speed, the distance difference and speed difference with the front vehicle, as well as the difference between the driver's psychological expectation and the actual situation during the following process; on the other hand, considering the poor light and narrow space in the tunnel, the minimum expected distance of the vehicle following needs additional adjustment compared with the normal road. The present invention proposes the following following model (named: acceleration distance control model): Formula (3) shows that the actual acceleration of the vehicle is affected by three factors: expected acceleration, minimum expected distance and actual distance; Formula (4) takes into account the influence of lane change behavior on the minimum expected distance, and uses coefficients to control the distance calculation under different behaviors. Δv n (t)=v n (t)-v n-1 (t) (5) In the formula, a n (t)——actual acceleration of the nth vehicle; a n，e (t)——expected maximum acceleration of the nth vehicle; s1——minimum expected distance; s2——actual distance; s3——minimum distance required for following vehicles; h t ——Expected headway; m, p——adjustment coefficients; when the vehicle is changing lanes, m=0, p=1; when the vehicle is in other following states, m=1, p=0; V——The maximum speed allowed in the tunnel; Δv n (t)——the speed difference between the nth vehicle and the preceding vehicle; b——Comfortable deceleration, the value range is (0, 1], and the specific value depends on the road conditions; If there is no following behavior between vehicles, the driving speed of the simulated vehicle is: v n (t)=v n (t)+ε (6) Where ε is the error term or random disturbance term; S42: Considering the lighting conditions and visual field limitations in the tunnel, vehicles need to maintain a greater distance from other vehicles when changing lanes or overtaking compared to normal roads. Based on this, the present invention proposes the following lane changing model (named: safe distance lane changing model): First of all, the lane-changing behavior of vehicles is mostly due to the fact that the speed of the front vehicle is lower than the expected speed of the rear vehicle. Therefore, the first premise of lane-changing is: v n-1 (t)≤v n，e (t) (7) In addition, the vehicle lane change needs to ensure that there is at least a safe distance between the vehicle and the vehicle in the adjacent lane to prevent collision. Therefore, the second prerequisite for the vehicle lane change is: In the formula, s n，a (t)——the actual distance between the nth vehicle and the preceding vehicle in the adjacent lane at time t; s n，b (t)——the actual distance between the nth vehicle and the vehicle behind it in the adjacent lane at time t; s n (t)——the actual distance between the nth vehicle and the vehicle ahead in the lane at time t; D n，a (t)——the safe distance between the nth vehicle and the preceding vehicle in the adjacent lane at time t; D n，b (t)——the safe distance between the nth vehicle and the vehicle behind it in the adjacent lane at time t; l n ——The length of the nth vehicle; t r ——Driver reaction time, generally 0.5s to 1.5s; v n-1，a (t)——the speed of the preceding vehicle in the adjacent lane at time t; v n+1，b (t)——the speed of the vehicle behind in the adjacent lane at time t; a n-1，a (t)——the acceleration of the preceding vehicle in the adjacent lane at time t; a n+1，b (t)——the acceleration of the vehicle behind in the adjacent lane at time t; If the vehicle meets the above two prerequisites during driving, it will change lanes.

7. A method and system for estimating tunnel congestion nodes based on digital twins according to claims 1, 5 and 6, characterized in that: The adaptive collaborative simulation projection model based on tunnel driving mainly realizes the dynamic target projection of the digital twin system and provides a basis for determining tunnel congestion nodes. The model performs different dynamic target projections on different sections of the tunnel based on the above-mentioned real-time detection results and simulation deduction results: for sections not covered by cameras, the simulated vehicles obtained by deduction are projected into the physical scene; For road sections covered by cameras, the simulated vehicles and detected vehicles in the road sections covered by cameras are matched and deduplicated based on the improved Hungarian algorithm.

8. A tunnel congestion node estimation method and system based on digital twin according to claims 1 and 7, characterized in that: The improved Hungarian algorithm solves the problem of matching and deduplication of simulated vehicles and detected vehicles. The main improvement is that in order to better improve the matching accuracy of simulated vehicles and detected vehicles, this paper introduces the expression E that represents the position relationship between the simulated vehicle and the projected vehicle. mn , this formula fully considers the influence of the position information of the simulated vehicle and the projected vehicle on the position relationship between the two under different coefficients. mn The smaller it is, the closer the distance between the simulated vehicle and the projected vehicle is, and the better the match is. So let the objective function be E mn With variables The minimum value of the product. The improved algorithm model is as follows: Introducing E mn To represent the positional relationship between the mth simulated vehicle and the nth projected vehicle. Define the decision variables as: The matching problem is transformed into: In the formula, and Respectively represent the position information of the mth simulated vehicle and the nth projected vehicle. j1 represents the set of simulated vehicles, D j2 Represents the set of projected vehicles.

9. According to a method and system for estimating tunnel congestion nodes based on digital twins according to claim 1, it is characterized by: In order to better determine the movement status and position of the traffic flow, the intelligent traffic flow node model of the S60 is as follows: According to the vehicle data information in the adaptive collaborative simulation projection model, this paper regards different vehicles as different movable nodes, constructs a road network, and proposes an intelligent traffic flow node model Node = (V, E, d ij ), where V is a set of nodes, representing vehicles in the road network; E is an edge set, which represents the position relationship between two vehicles in the road network; d ij is the weight, which represents the position information between the i-th vehicle and the j-th vehicle. Since the vehicle information of the adaptive collaborative simulation projection model is updated in real time, the node set, edge set and weight are also updated in real time.

10. A tunnel congestion node estimation method and system based on digital twin according to claims 1 and 9, characterized in that: The S70 includes the following steps: S71: Determine the traffic congestion situation. Considering that the measurement of traffic congestion is related to people's psychological tolerance, a custom quantitative indicator "congestion degree" is proposed to measure the congestion situation. In this article, congestion degree refers to the loss ratio of traffic flow during driving compared with normal smooth driving. Its formula can be expressed as follows: Where: C——congestion degree; S1——current traffic volume in the tunnel; S2——the tunnel capacity; i——density ratio, which represents the current traffic density k and traffic congestion density k j The ratio of n——parameter, which can be regarded as a fixed constant; If the convoy congestion degree C is greater than 80%, the road section is deemed to be congested and the process proceeds to S52. S72: Determine congestion nodes and issue warnings. Based on the intelligent traffic node model, we introduce node saturation and propose a node coupling congestion identification model to determine the location of congestion nodes: x n (t)=|(1-β)f(x n (t))+βf(x n-1 (t-t * ))|+b i H (17) f(x n (t))=x n (t)(1-x n (t)) (18) Where: x n (t)——Whether the nth node (nth vehicle) is congested at time t, with a value of [0, 1]. When it is greater than 0.8, the node is considered congested; β——coupling strength between vehicles, value is (0, 1]; f(x n (t))——a mapping function of the nth node; t * ——Delay time: the driving behavior of the n-1th vehicle has a hysteresis effect on the driving behavior of the nth vehicle; ——initial saturation of the nth node; H——disturbance value, indicating the disturbance value of the tunnel to the node, simulating the degree of congestion; When x n When the (t) value is greater than 0.8, it indicates that congestion occurs at the nth vehicle, and the system issues an early warning for the location of the congested node.

Images