Safety control system and method for heavy traffic flow under adverse weather conditions

Abstract

The heavy-load traffic flow safety control system and method under adverse weather conditions comprises an information sensing module, a situation assessment module, an information publishing module and a safety gateway module; the information sensing module is used for receiving camera data; the information sensing module is used for collecting information of heavy-load traffic flow in a node disturbance area; the situation assessment module is used for receiving heavy-load traffic flow area information in the node disturbance area obtained by the information sensing module, and obtaining a road network representation by using a convolutional neural network algorithm; the information publishing module discriminates the disturbance influence degree according to the situation assessment module; and the safety gateway module is used for realizing data transmission between the information sensing module and the situation assessment module, and between the situation assessment module and the information publishing module. The perception enhancement algorithm based on artificial intelligence deep learning is used, the model generalization performance under uneven data distribution is optimized, and the perception ability of heavy-load traffic flow disturbance under adverse weather conditions is improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of intelligent traffic vehicle-road cooperation, specifically a heavy-load traffic flow safety control system and method under adverse weather conditions. BACKGROUND

[0002] With the rapid development of China's social economy, the number of cars in use continues to grow, and the contradiction between the increasing transportation demand and the limited carrying capacity of the transportation system is prominent. The terrain is complex, and the microclimate frequently occurs in some areas of the expressway, which is prone to fog and icing of the road surface, especially on bridges, culverts, shady and high-altitude sections. Under adverse weather conditions, the road capacity decreases, the driver's perception ability is limited, and the transient control response of the vehicle deviates, leading to an increase in the risk of instability of heavy-load traffic flow, which easily induces traffic flow disturbance; then the local node heavy-load traffic flow disturbance causes the traffic wave to spread continuously in the road network, affecting the entire system; if the heavy-load traffic flow disturbance cannot be actively intervened in time, it is difficult to recover to normal operation in the short term, which is manifested as a high frequency of heavy-load traffic flow disturbance induced, a wide range of propagation, and a long recovery time, leading to a decrease in the carrying capacity of the road network and impacting the overall system resilience, which brings serious challenges to the safe and efficient operation and management of the expressway.

[0003] It is necessary to focus on the complexity and particularity of heavy-load traffic on the expressway, clarify the induction mechanism of adverse weather on heavy-load traffic flow disturbance, quantify the influence degree of local heavy-load traffic flow disturbance on the road network system, and coordinate the active control means of the management end, the road control end and the driver end to prevent heavy-load traffic conflicts caused by adverse weather, limit the influence range of traffic flow disturbance, guide the rapid recovery of road network capacity, reduce the decline of the overall road network carrying capacity under local traffic conflicts, and overcome the realistic bottleneck problem of the impact of heavy-load traffic flow disturbance on the road network resilience under adverse weather conditions.

[0004] However, considering the high false detection rate of traffic flow abnormal events under adverse weather conditions, the existing detection methods are affected by environmental factors such as light conditions and visibility, and cannot meet the detection needs of actual events under adverse weather conditions. Considering that existing research focuses on exploring the periodicity, timing and long-term regularity of road network traffic spatio-temporal changes, less consideration is given to the influence of regional occasional local node disturbance on the overall trend change of road network traffic flow and the implementation of control on a single node or upstream and downstream nodes of a specific road section, and the advantages of information interconnection and overall planning of intelligent networked transportation systems are not fully utilized. SUMMARY

[0005] The application provides a heavy-load traffic flow safety control system and method under adverse weather conditions, which can quickly lock the traffic flow disturbance source through intelligent sensing, effectively estimate the disturbance influence degree by establishing a road network situation influence evaluation model, and propose a multi-layer control method to link multiple nodes of the road network for precise control and active inhibition and coordinated management of heavy-load vehicles in the node disturbance area, so as to actively intervene in the whole stage of accident prevention, impact evacuation and recovery guidance, effectively inhibit the spread of traffic flow disturbance, and ensure the stability, flexibility and safety of the traffic system, and improve the overall road network operation efficiency, so as to solve the defects in the prior art.

[0006] The application is implemented through the following technical solutions:

[0007] The heavy-load traffic flow safety control system under adverse weather conditions comprises an information sensing module, a situation evaluation module, an information publishing module and a safety gateway module.

[0008] The information sensing module is used for receiving local node roadside camera video data, and calculating the traffic state parameters and abnormal event types of the node in real time through a sensing-enhanced adverse weather condition abnormal event detection algorithm, so as to provide unit node data for the subsequent modules.

[0009] The information sensing module is used for collecting the number of vehicles, vehicle type distribution, saturation and corresponding time information of the heavy-load traffic flow in the node disturbance area.

[0010] The situation evaluation module is used for receiving the heavy-load traffic flow area information in the node disturbance area obtained by the information sensing module, obtaining the road network representation by using a convolutional neural network algorithm, and evaluating the disturbance influence degree and risk level of the node in the specific road section covered by the sensor view according to the road network representation.

[0011] The information publishing module determines the disturbance influence degree according to the situation evaluation module, calculates the information publishing time and type based on a genetic algorithm, and displays the published information on the local variable information board.

[0012] The safety gateway module is used for the information transmission communication mode mainly based on mobile communication technology, and one or both of the auxiliary WiFi / BT and DSRC wireless communication modes to realize the transmission of data between the information sensing module and the situation evaluation module, and between the situation evaluation module and the information publishing module.

[0013] The heavy-load traffic flow safety control method under adverse weather conditions comprises the following steps:

[0014] S01: The information sensing module obtains the heavy-load traffic flow state data of the node disturbance area by using a heavy-load traffic flow sensing enhancement method based on artificial intelligence deep learning, including the number of heavy-load traffic flow vehicles, vehicle type distribution and corresponding time information.

[0015] S02: The information perception module analyzes and processes the observation data, including data analysis, feature extraction, and information fusion;

[0016] S03: According to the collected and processed observation data of the information perception module, a heavy traffic flow data set is constructed, including the number of heavy vehicles, the degree of road congestion, and the type of abnormal events;

[0017] S04: The security gateway module securely and stably transmits the heavy traffic flow data set constructed in step S03 to the situation assessment module;

[0018] S05: The situation assessment module receives the heavy traffic flow data set and uses a convolutional neural network algorithm to evaluate the disturbance situation of the heavy traffic flow, and obtains the road network representation g;

[0019] S06: The information release module receives the road network representation g data from the situation assessment module, selects the information release time and type based on the genetic algorithm, stores the selected release information, and transmits it to the intelligence board, and displays the release information on the local variable information board;

[0020] S07: The variable information board guides the change of the traffic state in the node disturbance area, and feeds back the heavy traffic state data according to the information perception area, forming a closed-loop guidance control.

[0021] The heavy traffic flow safety control method under adverse weather conditions as described above, the information perception module based on artificial intelligence deep learning heavy traffic flow perception enhancement method in S01 includes the following steps:

[0022] Step one: According to the actual traffic video data and abnormal traffic flow characteristic parameter performance, the perception loss function is L p , expressed as:

[0023]

[0024] In the formula, represents the feature map of the i-th layer of the high-speed heavy traffic flow image feature extraction network, and the size C i ×H i ×W i . J represents a clear high-speed heavy traffic flow image, and F(x) represents a detected image;

[0025] Step two: Under the condition of meeting step one: further construct a multi-scale structural similarity loss represented by L m :

[0026]

[0027] In the formula, F(x), J respectively represent the detection image and the clear image, u J and σ J are the mean and variance of the detection image, u F(x) and σ F(x) are the mean and variance of the clear image, σ F(x)J indicates the covariance, C1 and C2 are adjustable variables.

[0028] The heavy traffic flow safety control method under adverse weather conditions as described above, the step two minimizes L m , remove the image noise caused by adverse weather, adjust the contrast and brightness of the processed image, get the best quality image g(x).

[0029] The heavy traffic flow safety control method under adverse weather conditions as described above, the gradient descent method minimizes L m in the formula is as follows: In the formula, θ represents the optimal parameter, α represents the learning rate, indicates the gradient.

[0030] The heavy traffic flow safety control method under adverse weather conditions as described above, the heavy traffic flow disturbance trend influence evaluation method using convolutional neural network algorithm deep learning in S05 includes the following steps:

[0031] Step 1: disturbance area vehicle selection: the vehicle causing congestion source in the disturbance node area is set as the disturbance vehicle, and the node disturbance area traffic state parameters (including node disturbance area heavy traffic flow density, speed, saturation) and abnormal event type are obtained from g(x);

[0032] Step 2: build heavy traffic flow model, the construction algorithm is as follows:

[0033]

[0034] In the formula, u, ρ, f are the speed, density and flow of heavy traffic wave, ρ e is the effective density of the disturbance area, ρ crit is the critical density of the disturbance area, ρ jam is the jam density of the disturbance area, u crit is the critical speed of heavy traffic flow, u max is the maximum speed of heavy traffic flow w is the wave speed of heavy traffic wave, which can be represented as:

[0035] Step 3: establishment of regional road network representation model: according to the bidirectional distribution characteristics of road network traffic flow, the road network is represented as g, and the regional characteristics and traffic space-time characteristics of the node disturbance area are preliminarily represented;

[0036] Step 4: The abnormal event is characterized by the disturbance type, and the road network representation g is introduced into the decoupling representation framework to construct the mechanism of analyzing the node disturbance area overload traffic flow on the global road network.

[0037] The heavy traffic flow safety control method under adverse weather conditions as described above, the S05 deep learning uses convolutional neural network algorithm to solve, the learning process includes the following steps:

[0038] Step 1): The spatiotemporal representation model constructs an adaptive static adjacency matrix A S , the expression is: S = ReLU(tanh(a1(EE T ))), where E∈R N×D represents the random initialization of the N nodes of the heavy traffic flow disturbance node embedding, which projects the heavy traffic flow disturbance node ID into a vector with heavy traffic flow node information. D is the dimension of the disturbance node information embedding, a1 is a control parameter for controlling the saturation rate of the activation function;

[0039] Step 2): In the case of meeting step 1), use graph convolution to process heavy flow data and hidden state at each step, use current heavy flow data X :,t and hidden state as the input I t , the expression is: Where I t ∈R B×N×C , B represents the size of the batch heavy traffic flow data, [·,·] represents the concatenation operation;

[0040] Step 3): Another adjacency matrix A is generated in the same way as step 2) with another control parameter a2, the expression is: Graph convolution is performed on the input I t to obtain the heavy traffic flow dynamic feature map DF t , the expression is: Where W d ∈R C×C is the feature weight matrix of the highway network;

[0041] Step 4): Under the condition of meeting step 3), the dynamic feature map DF t obtains the road network representation g=(V,E,A,W), where V(vertices) represents the set of bidirectional nodes of the highway network, E is the edge set, which represents the connection relationship between the nodes of the highway network, A S is the node adjacency matrix obtained in 1), W represents the feature weight matrix of the highway network in 3), and the heavy vehicle proportion is introduced to represent the importance of different nodes.

[0042] The heavy traffic flow safety control method under adverse weather conditions as described above, the S05 state evaluation module transmits the required road network representation g to the information release module, and the release information includes the disturbance type, the disturbance location, and the duration. The selection of the release information affects the control effect of the node disturbance area, and the lack of constraints has very limited guiding significance for real-time data exploration. Therefore, a target function with constraints is further proposed, and a penalty function is introduced to optimize the target function while supplementing the constraints of data exploration.

[0043] The heavy traffic flow safety control method under adverse weather conditions as described above, the S06 solution method based on genetic algorithm includes the following steps:

[0044] Step (1): Construct the node disturbance area penalty function Penalty m : According to the queuing length of the upstream, disturbance node, and downstream of the path (i, j) in the road network representation g and Q l , the increase of the congestion degree of the disturbance node area will cause the penalty function Penalty m to increase;

[0045] Step (2): Optimize the node disturbance area target function: according to the obtained heavy vehicle throughput v of the node disturbance area and the heavy vehicle path r under the node disturbance area, maximize the target function Objective m .

[0046] Step (3): Genetic algorithm target function Objective m check: according to the check of the heavy vehicle path and the queuing length of the path (i, j) in the node disturbance area under each time length, continuously improve the accuracy of target generation and information release.

[0047] The heavy traffic flow safety control method under adverse weather conditions as described above, the expression of the penalty function Penalty m is:

[0048]

[0049] In the formula, M1, M k are weight factors, and Q l are the upstream, downstream, and node disturbance area queuing lengths of the path (i, j), respectively.

[0050] The expression of the target function Objective m is:

[0051]

[0052] In the formula, v is the area heavy vehicle throughput, and r represents different vehicle paths.

[0053] The advantages of the present application are:

[0054] 1. The present application is aimed at the problem of reduced heavy traffic flow anomaly sensing ability under adverse weather conditions, and uses a perception enhancement algorithm based on artificial intelligence deep learning to optimize the model generalization performance under uneven data distribution and improve the perception ability of heavy traffic flow disturbance under adverse weather conditions.

[0055] 2. The present application adopts a heavy traffic flow disturbance trend influence evaluation method based on artificial intelligence deep learning, uses deep learning CNN (convolutional neural network algorithm) to solve, obtains the corresponding road network representation g from the known data, and realizes quantitative interpretation of the heavy traffic flow node disturbance propagation process under adverse weather conditions.

[0056] 3. The present application introduces a method for selecting information release time and type based on genetic algorithm, greatly improves the accuracy and robustness of information selection, and also ensures the data processing efficiency, displays the accurate released information on the local variable information board, and effectively improves the overall system resilience of the road network under heavy traffic flow disturbance. BRIEF DESCRIPTION OF DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings described below are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0058] Figure 1 is a simulation schematic diagram of the safety control system of the present application;

[0059] Figure 2 is a flow chart of the safety control method of the present application. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0061] The method provided by the application is based on the following assumptions: 1) the node disturbance area includes the abnormal event source area road section, and the node disturbance area control range is the communication range of the road test unit; 2) the delay existing in information transmission, data processing and calculation, and instruction execution is ignored, that is, the speed of the operation of each module is sufficient to support the system operation; 3) the length of the node disturbance observation area is fixed, that is, Q m = Q max -Q min ; 4) the compliance degree of the driver of the heavy vehicle to the information published by the variable information board is ignored; 5) the road test unit high-definition camera, the situation deduction, and the variable information board need to have the necessary information transmission capability, that is, through the information interaction capability of each module, and the heavy vehicle completely executes the control instruction.

[0062] As Figure 1 shown, the heavy traffic flow safety control system under adverse weather conditions comprises an information perception module, a situation assessment module, an information publishing module, and a safety gateway module;

[0063] The information perception module is used to receive local node road side camera video data, and to solve the traffic state parameters and abnormal event types of the road section where the node is located in real time through a perception enhanced adverse weather condition abnormal event detection algorithm, so as to provide unit node data for subsequent modules;

[0064] The information perception module is used to collect the number of vehicles, the vehicle type distribution, the saturation, and the corresponding time information of the heavy traffic flow in the node disturbance area;

[0065] The information perception module comprises a road side observation unit and a data processing unit;

[0066] The road side observation unit is responsible for collecting traffic state information, such as the speed, position, abnormal event type, and corresponding time information of the vehicle in the region;

[0067] The data processing unit is responsible for analyzing and processing the related data information from the road side observation unit to obtain the number of vehicles, the vehicle type distribution, the abnormal event type, and the traffic flow saturation;

[0068] The situation assessment module is used to receive the heavy traffic flow area information in the node disturbance area obtained by the information perception module, to obtain the road network representation using a convolutional neural network algorithm, and to evaluate the disturbance influence degree and risk level of the node sensor visual field covering a specific road section according to the road network representation; the situation assessment module constructs a heavy traffic flow model according to the traffic state information provided by the information perception module, realizes the overall heavy traffic flow model representation g, and evaluates the disturbance influence degree and risk level of the node sensor visual field covering a specific road section according to the road network representation g. Meanwhile, the disturbance influence degree and risk level information set are packaged and sent to the information publishing module;

[0069] The information publishing module displays the published information on the local variable bulletin board according to the determination of the disturbance influence degree by the situation assessment module, and selects the information publishing time and type based on a genetic algorithm solution.

[0070] The security gateway module is used for information transmission communication mode mainly based on mobile communication technology (5G), and one or two of auxiliary WiFi / BT, DSRC wireless communication modes to realize data transmission between the information perception module and the situation assessment module, and the situation assessment module and the information publishing module; the security gateway module introduces Kafka as a data source and data routing to provide access control mechanism based on users and ACL (access control list), effectively limiting the access of unauthorized users to traffic information, and reducing the risk of unauthorized access and network security threats.

[0071] As shown in Figure 2 The heavy traffic flow safety control method under adverse weather conditions comprises the following steps:

[0072] S01: The information perception module acquires heavy traffic flow state data of the node disturbance area based on the heavy traffic flow perception enhancement method of artificial intelligence deep learning, including the number of heavy traffic flow vehicles, vehicle type distribution, and corresponding time information;

[0073] S02: The information perception module analyzes and processes the observation data, including data analysis, feature extraction, and information fusion;

[0074] S03: According to the collected and processed observation data of the information perception module, a heavy traffic flow data set is constructed, including the number of heavy vehicles, road congestion degree, and abnormal event type;

[0075] S04: The security gateway module safely and stably transmits the heavy traffic flow data set constructed in step S03 to the situation assessment module;

[0076] S05: The situation assessment module uses the heavy traffic flow disturbance situation influence evaluation method of convolutional neural network algorithm deep learning according to the received heavy traffic flow data set to evaluate the road network representation g.

[0077] S06: The information publishing module selects the information publishing time and type based on a genetic algorithm solution according to the g data of the situation assessment module transmitted by the information publishing module, stores and transmits the selected published information to the bulletin board, and displays the published information on the local variable bulletin board.

[0078] S07: The information guidance of the variable information board changes the traffic state of the node disturbance area, feeds back the heavy load traffic state data according to the information perception area, forms a closed-loop guidance control, and thus the optimal information publishing strategy can be obtained, so as to realize the optimal control of the heavy load traffic flow information collaborative publishing safety control method and system for the highway scene under adverse weather conditions through each module, and realize efficient, safe and energy-saving driving of the relevant part of the road section after the disturbance.

[0079] Specifically, the heavy load traffic flow perception enhancement method based on artificial intelligence deep learning of the information perception module in S01 described in the embodiment includes the following steps:

[0080] Step one: according to the actual traffic video data and abnormal traffic flow characteristic parameter performance, the loss function is L p , which is expressed as:

[0081]

[0082] In the formula, represents the feature map of the i-th layer of the high-speed heavy load traffic flow image feature extraction network, and the size is C i ×H i ×W i . J represents a clear image of high-speed heavy load traffic flow, and F(x) represents a detected image;

[0083] Step two: under the condition of step one: further construct a multi-scale structural similarity loss represented by L m :

[0084]

[0085] In the formula, F(x) and J represent the detected image and the clear image respectively, u J and σ J are the mean and variance of the detected image, u F(x) and σ F(x) are the mean and variance of the clear image, σ F(x)J represents the covariance, and C1 and C2 are adjustable variables.

[0086] Specifically, the gradient descent method is used to minimize L m in step two described in the embodiment to remove image noise caused by adverse weather, adjust the contrast and brightness of the processed image, and obtain the best quality image g(x).

[0087] More specifically, the calculation formula of the gradient descent method for minimizing L m in the embodiment is: In the formula, θ represents the optimal parameter, α represents the learning rate, and represents the gradient.

[0088] More specifically, the overload traffic flow disturbance trend influence evaluation method using convolutional neural network algorithm deep learning in S05 of the embodiment includes the following steps:

[0089] Step 1: Select the vehicle in the disturbance area: the vehicle causing the congestion source in the disturbance node area is set as the disturbance vehicle, and the node disturbance area traffic state parameters (including node disturbance area overload traffic flow density, speed, saturation) and abnormal event type are obtained from g(x);

[0090] Step 2: Construct the overload traffic flow model, and the algorithm is:

[0091]

[0092] In the formula, u, p, and f are the speed, density and flow of the overload traffic wave, p e is the effective density of the disturbance area, p crit is the critical density of the disturbance area, p jam is the jam density of the disturbance area, u crit is the critical speed of the overload traffic flow, u max is the maximum speed of the overload traffic flow, and w is the wave speed of the overload traffic wave, which can be expressed as:

[0093] Step 3: Establishment of regional road network representation model: according to the bidirectional distribution characteristics of road network traffic flow, the road network is represented as g, and the regional characteristics and traffic space-time characteristics of the node disturbance area are preliminarily represented;

[0094] Step 4: The disturbance type under the represented abnormal event is complex, the road network representation g is introduced into the decoupling representation framework, and the action mechanism of the overload traffic flow in the node disturbance area on the global road network is constructed.

[0095] Further, the deep learning in S05 of the embodiment is solved by using convolutional neural network algorithm, and the learning process includes the following steps:

[0096] Step 1): Construct a space-time representation model to generate an adaptive static adjacency matrix A S , the expression is: A S = ReLU(tanh(a1(EE T ))), wherein E∈R N×D represents the random initialization node embedding of N nodes of the overload traffic flow disturbance, which projects the overload traffic flow disturbance node ID into a vector with overload traffic flow node information. D is the disturbance node information embedding dimension, a1 is a control parameter for controlling the saturation rate of the activation function;

[0097] Step 2): Under the condition of step 1), the overloaded flow data and hidden state of each step are processed using graph convolution, and the current overloaded flow data X :,t and hidden state are taken as the input I t , the expression is: wherein I t ∈R B×N×C , B represents the size of the batch of overloaded traffic flow data, and [·,·] represents the concatenation operation;

[0098] Step 3): Another adjacent matrix A2 is generated in the same way as step 2) using another control parameter a2 The expression is: The graph convolution is performed on the input I t to obtain the overloaded traffic flow dynamic feature map DF t , the expression is: wherein W d ∈R C×C is the feature weight matrix of the highway network;

[0099] Step 4): Under the condition of step 3), the dynamic feature map DF t is obtained, and the highway network representation g=(V, E, A, W) is obtained, wherein V (vertices) represents a set of bidirectional nodes of the highway network, E is an edge set, represents a connection relationship between nodes of the highway network, A S is the node adjacent matrix obtained in 1), W represents the feature weight matrix of the highway network in 3), and the proportion of overloaded vehicles is introduced to represent the importance of different nodes.

[0100] Further, the situation assessment module S05 described in the embodiment transmits the required highway network representation g to the information publishing module, and the published information includes the disturbance type, the disturbance position, and the duration. The selection of the published information affects the management and control effect of the node disturbance area, and the lack of constraints has very limited guiding significance in real time. Therefore, a target function with constraints is further proposed, and a penalty function is introduced to realize the constraint supplement of data exploration while optimizing the target function.

[0101] Further, the solving method based on the genetic algorithm S06 described in the embodiment includes the following steps:

[0102] Step (1): Construct the node disturbance area penalty function Penalty m : According to the queue length of the upstream path (i, j) in the disturbance area, the disturbance node, and the downstream in the highway network representation g and Q l , the increase of the congestion degree of the disturbance node area will cause the penalty function Penalty m to increase;

[0103] Step (2): Optimization of node disturbance area objective function: according to the obtained heavy vehicle throughput v of the node disturbance area, and the heavy vehicle path r under the node disturbance area, the objective function Objective m is maximized.

[0104] Step (3): Genetic algorithm objective function Objective m checking: according to the heavy vehicle path under each time length and the queue length of the path (i,j) in the node disturbance area, the accuracy of the target generation and information release is continuously improved.

[0105] Further, the expression of the penalty function Penalty m of the embodiment is:

[0106]

[0107] In the formula, M1, M k is a weight factor, and Q l is the upstream, downstream and node disturbance area queue length of the path (i,j) respectively.

[0108] The expression of the objective function Objective m is:

[0109]

[0110] In the formula, v is the area heavy vehicle throughput, and r represents different vehicle paths.

[0111] Compared with the prior art, the advantages of the present application are:

[0112] 1) Unlike the prior art of simply controlling the road network node disturbance under good weather, the present application provides a method for coordinating and releasing information for safety control of heavy traffic flow on expressways under adverse conditions, and the present application provides a key technical link for controlling the node disturbance of the real expressway.

[0113] 2) The present application further optimizes the method of accurately sensing under adverse weather conditions, and enriches the sensing algorithm of the node disturbance area abnormal event. Compared with the simple selection mode in the prior art, the selection scheme of the present application can obtain the optimal abnormal event sensing scheme, further improve the running efficiency of the algorithm, and the sensing ability of the abnormal event is more accurate, efficient and stable.

[0114] 3) The application is aimed at the problem that the influence of local node disturbance of heavy-load large-flow road section on road network situation is difficult to quantify, adopts an evaluation method based on convolutional neural network, quantitatively characterizes the regional space-time characteristics of the node disturbance area according to the space-time propagation law of heavy-load traffic flow nodes under adverse weather, is different from the fuzzy characterization of the prior art, realizes quantitative interpretation and characterization of the heavy-load traffic flow node disturbance propagation process under adverse weather conditions, and further improves the evaluation accuracy of the application.

[0115] 4) The application further optimizes the information release strategy selection algorithm provided by the prior art, introduces a genetic algorithm settlement method, is different from the setting of a simple objective function and a penalty function, guarantees that the utilization efficiency of data is further improved, and the directionality and correctness of information release selection are greatly improved.

[0116] 5) The application fully considers the vehicle characteristics of heavy-load vehicles, and takes the travel cost, travel waiting time, convenience and personalized service as constraint conditions in the distribution optimization control of the traffic characteristics of heavy-load traffic flow. Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the application, and not to limit them; although the application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: they can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the application.

Claims

1. A heavy traffic flow safety control system under adverse weather conditions, characterized in that: The information perception module, the situation assessment module, the information publishing module and the security gateway module are comprised; The information perception module is used for receiving local node road side camera video data, and solving traffic state parameters and abnormal event types of a node through a perception enhanced adverse weather condition abnormal event detection algorithm in real time, so as to provide unit node data for subsequent modules; The information perception module is used for collecting the number of vehicles, vehicle type distribution, saturation and corresponding time information of heavy traffic flow in the node disturbance area; The situation assessment module is used for receiving the heavy traffic flow area information in the node disturbance area obtained by the information perception module, obtaining a road network representation by using a convolutional neural network algorithm, and evaluating the disturbance influence degree and risk level of the node in covering a specific road section according to the road network representation; The information publishing module is used for publishing information on a local variable information board according to the disturbance influence degree determined by the situation assessment module, and based on a genetic algorithm solution to select information publishing time and type. The security gateway module is used for transmitting data between the information perception module and the situation assessment module, and between the situation assessment module and the information publishing module in one or two of the following modes: a mobile communication technology mode, a WiFi / BT mode and a DSRC wireless communication mode.

2. A method for safe control of heavy traffic flow under adverse weather conditions, characterized in that: The method comprises the following steps: S01: The information perception module obtains heavy traffic flow state data of a node disturbance area based on an artificial intelligence deep learning heavy traffic flow perception enhancement method, including the number of heavy traffic flow vehicles, vehicle type distribution and corresponding time information; S02: The information perception module analyzes and processes observation data, including data analysis, feature extraction and information fusion; S03: According to the observation data collected and processed by the information perception module, a heavy traffic flow data set is constructed, including the number of heavy vehicles, road congestion degree and abnormal event type; S04: The security gateway module safely and stably transmits the heavy traffic flow data set constructed in S03 to the situation assessment module; S05: The situation assessment module adopts a convolutional neural network algorithm deep learning heavy traffic flow disturbance situation influence evaluation method to evaluate a road network representation g according to the received heavy traffic flow data set; S06: The information publishing module stores and transmits the selected publishing information to an information board according to the road network representation g data transmitted by the situation assessment module in S05, and displays the publishing information on a local variable information board based on a genetic algorithm solution to select information publishing time and type; S07: The traffic state of the node disturbance area is changed under the guidance of the information of the variable information board, and the heavy traffic state data is fed back according to the information perception area to form a closed loop guidance control.

3. The method for safe control of heavy traffic flow in adverse weather conditions according to claim 2, characterized in that: The information perception module based on the artificial intelligence deep learning heavy traffic flow perception enhancement method in S01 comprises the following steps: Step one: According to the actual traffic video data and abnormal traffic flow characteristic parameter performance, the loss function is L p , which is represented as: In the formula, represents the feature map of the i-th layer of the high-speed heavy traffic flow image feature extraction network, and the size is C i × H i × W i , J represents a clear image of a high-speed heavy traffic flow, and F(x) represents a detected image; Step two: further construct the multi-scale structural similarity loss representation L under the conditions of step one: m : where F(x), J represent the detected image and the sharp image, respectively, u J and σ J are the mean and variance of the detected image, u F(x) and σ F(x) are the mean and variance of the sharp image, σ F(x)J denotes the covariance, and C1 and C2 are adjustable variables.

4. The method of claim 3, wherein the method further comprises: The step two minimizes L by gradient descent method m , remove image noise caused by bad weather, adjust the contrast and brightness of the processed image, and get the best quality image g(x).

5. The method for safe control of heavy traffic flow in adverse weather conditions according to claim 4, characterized in that: The gradient descent method minimizes L m The calculation formula is: In the formula, θ represents the optimal parameter, α represents the learning rate, represents the gradient.

6. The method of claim 2, wherein the method further comprises: The heavy traffic flow disturbance situation influence evaluation method based on the convolutional neural network algorithm deep learning in S05 comprises the following steps: Step 1: Disturbance area vehicle selection: the vehicle that causes the disturbance node area to be a congestion source is set as the disturbance vehicle, and the node disturbance area traffic state parameters (including node disturbance area heavy traffic flow density, speed, saturation) and abnormal event types are obtained from g(x); Step 2: Construction of heavy traffic flow model, the construction algorithm is: where u, p, f are the heavy vehicle wave speed, density and flow, p e is the effective density of the perturbed region, p crit is the critical density of the perturbed region, p jam is the jam density of the perturbed region, u crit is the critical speed of the heavy vehicle traffic flow, u max is the maximum speed of the heavy vehicle traffic flow w is the heavy vehicle wave speed, which can be expressed as: Step 3: Establishment of regional road network representation model: according to the bidirectional distribution characteristics of road network traffic flow, the road network is represented as g, and the regional characteristics and traffic space-time characteristics of the node disturbance area are preliminarily represented; Step 4: The disturbance type under the represented abnormal event is complex, the road network representation g is introduced into the decoupling representation framework, and the action mechanism of the node disturbance area heavy traffic flow on the global road network is constructed.

7. The method of claim 2, wherein the method further comprises: The S05 deep learning adopts a convolutional neural network algorithm for solving, and the learning process includes the following steps: Step 1): Construction of spatio-temporal representation model to generate adaptive static adjacency matrix A S , the expression is: A S = ReLU(tanh(a1(EE T )), where E ∈ R N×D represents the random initialization node embedding of N nodes disturbed by overload traffic flow, which projects the overload traffic flow disturbance node ID into a vector with overload traffic node information, D is the dimension of disturbance node information embedding, a1 is a control parameter for controlling the saturation rate of the activation function; Step 2): In the case of step 1), use the graph convolution to process the overload flow data and the hidden state of each step, and use the current overload flow data X :,t and the hidden state as the input I t , the expression is: where I t ∈R B×N×C , B represents the batch size of the overload traffic flow data, and [·,·] represents the concatenation operation; Step 3): generating another adjacency matrix in the same way as in step 2) with another control parameter a2 The expression is: The input I t The graph convolution is performed to obtain the dynamic feature map DF of the heavy traffic flow t The expression is: Wherein, W d ∈R C×C is the highway network feature weight matrix; Step 4): a dynamic feature map DF satisfying the condition of step 3) t A road network representation g=(V, E, A, W) is obtained, wherein V (vertices) represents a two-way node set of the highway network, E is an edge set, representing the connection relationship between nodes of the highway network, A S is the node adjacency matrix obtained in 1), W represents the highway network feature weight matrix in 3), and the proportion of heavy vehicles is introduced to represent the importance of different nodes.

8. The method of claim 2, wherein the method further comprises: The S05 situation assessment module transmits the required road network representation g to the information release module, and the release information includes disturbance type, disturbance location, duration, the selection of release information affects the management and control effect of the node disturbance area, and the lack of constraints has very limited guiding significance for real-time, therefore, further propose a target function with constraints, introduce a penalty function, optimize the target function while realizing the constraint supplement of data exploration.

9. The method of claim 2, wherein the method further comprises: The S06 genetic algorithm-based solving method includes the following steps: Step (1): Constructing the node perturbation area penalty function Penalty m : According to the road network representation g, obtain the queuing length of the upstream, perturbation node, and downstream of the path (i, j) in the perturbation area and Q l , the increase of the congestion degree of the perturbation node area will cause the penalty function Penalty m to increase; Step (2): Optimizing the node perturbation area objective function: according to the obtained heavy vehicle throughput v of the node perturbation area, and the heavy vehicle path r under the node perturbation area, the objective function Objective m is maximized; Step (3): Genetic algorithm objective function Objective m Verification: According to the queue length of the overload vehicle path and the path (i, j) in the node disturbance area under each time length, the accuracy of target generation and information release is continuously improved.

10. The method of claim 9, wherein the method further comprises: The expression of the penalty function Penalty m is: In the formula, M1, M k is a weight factor, and Q l is the upstream, downstream, and node perturbation region queue length of the respective path (i,j). The objective function Objective m is expressed as: In the formula, v is the regional heavy vehicle throughput, and r represents different vehicle paths.

Images