Highway traffic condition detection system based on artificial intelligence

Through the highway traffic condition detection system based on artificial intelligence, risks are evaluated in real time and traffic control measures are generated, the problem of lack of long-term traffic situation analysis in the existing technology is solved, and the monitoring and management capabilities of highway traffic safety are improved.

CN118887802AInactive Publication Date: 2024-11-01张文浩 +1
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410914979.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-09
Publication Date
2024-11-01
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing intelligent driving technology lacks the ability to analyze the highway traffic situation for a long time, resulting in the inability to implement effective traffic control measures for rear vehicles in a timely manner, increasing the risk of accidents.

Method used

A highway traffic condition detection system based on artificial intelligence was designed, through traffic data acquisition module, risk level generation module, traffic control measures generation module, graph network data construction module, vehicle speed limit measures generation module and vehicle drainage measures generation module , combine risk assessment models and prediction models to detect risk levels in real time and generate traffic control measures.

Benefits of technology

It realizes the real-time generation of risk assessment and traffic control measures in the highway monitoring area, ensures the driving safety of vehicles and reduces the risk of accidents.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118887802A_ABST
    Figure CN118887802A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of traffic condition detection, and discloses a highway traffic condition detection system based on artificial intelligence, and the system comprises a traffic data collection module which collects traffic data in a monitoring area, and generates a feature sequence; the risk level generation module inputs the feature sequence into a risk assessment model and outputs a risk level; the traffic control measure generation module is used for generating traffic control measures for each vehicle in the control area; the vehicle speed limiting measure generation module outputs a vehicle speed limiting measure through a first prediction model; the vehicle drainage measure generation module outputs a vehicle drainage measure through a second prediction model; according to the invention, the risk level in the monitoring area of the expressway is detected through the risk assessment model, the driving data of each vehicle in the control area is comprehensively analyzed through the first prediction model and the second prediction model, and effective traffic control measures are generated for each vehicle in the control area, so that the driving safety of the vehicles is guaranteed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of traffic condition detection, and more specifically, to a highway traffic condition detection system based on artificial intelligence. Background Art

[0002] Vehicles on highways generally travel at high speeds. Once an abnormal situation occurs, such as a sudden traffic accident, road damage, or obstructed vision caused by bad weather conditions, if the following vehicles cannot obtain abnormal situation information in time and take effective measures, secondary accidents such as chain rear-end collisions are very likely to occur. In such emergency situations, the rapid implementation of effective traffic control measures, such as road diversion and vehicle speed limits, is crucial to preventing subsequent accidents.

[0003] In response to the above, cameras and lidars are currently used to capture information about the vehicle's surrounding environment, and intelligent driving technology is used to promptly obtain information about abnormal situations and take effective measures, such as emergency braking and lane changing. However, existing intelligent driving technology is limited by the detection range of cameras and lidars, and can usually only perform short-term trajectory prediction and short-distance road condition recognition. It lacks the ability to comprehensively analyze traffic conditions long in advance, resulting in the inability to quickly implement effective traffic control measures for vehicles behind in a timely manner. Summary of the invention

[0004] The present invention provides an artificial intelligence-based highway traffic condition detection system to solve the technical problem that the existing intelligent driving technology in the relevant technology lacks the ability to comprehensively analyze the traffic situation long in advance, resulting in the inability to promptly implement effective traffic control measures for the rear vehicles.

[0005] The present invention provides a highway traffic condition detection system based on artificial intelligence, comprising:

[0006] Traffic data collection module, which is used to collect traffic data in the monitoring area within a preset time period T and at preset time intervals t, and generate a feature sequence;

[0007] The feature sequence includes N sequence units, and the i-th sequence unit represents the traffic data at the i-th time point, where N = T / t, 1≤i≤N; the traffic data includes: traffic flow, vehicle density and weather conditions;

[0008] A risk level generation module is used to input the feature sequence into the risk assessment model, and the output value represents the risk level in the monitoring area; the risk levels include: low risk level, medium risk level and high risk level;

[0009] A traffic control measure generation module, which is used to generate traffic control measures for each vehicle in the control area according to the risk level in the monitoring area;

[0010] Low risk levels do not generate traffic control measures;

[0011] The traffic control measures corresponding to the medium risk level are vehicle speed limit measures;

[0012] Traffic control measures corresponding to high risk levels are vehicle diversion measures;

[0013] A graph network data construction module, which is used to collect the driving data of each vehicle in the control area and construct graph network data;

[0014] The graph network data includes: nodes, initial features of nodes and edges between nodes; data mapping is established between nodes and each vehicle in the control area; the initial features of nodes are represented according to the driving data of the vehicle mapped by the node; the driving data includes: vehicle type, maximum speed limit of the lane, vehicle speed and acceleration;

[0015] A vehicle speed limit measure generation module, which is used to input the graph network data into the first prediction model and output the vehicle speed limit measure; the vehicle speed limit measure represents the maximum speed of each vehicle in the control area;

[0016] The vehicle diversion measure generation module is used to input the graph network data into the second prediction model and output the vehicle diversion measure; the vehicle diversion measure represents the lane change position of each vehicle in the control area, wherein the lane change position is represented by longitude and latitude.

[0017] Furthermore, the traffic volume represents the number of vehicles in the monitoring area; the vehicle density is equal to the traffic volume divided by the area of ​​the monitoring area; the weather conditions include: rainfall, snowfall, visibility and weather type, among which the weather types include: sunny, rainy, snowy and haze, and the weather types are represented by real number codes.

[0018] Furthermore, the monitoring area and the control area are both rectangular areas, the monitoring area and the control area are adjacent to each other, and the widths of the monitoring area and the control area are both lane widths;

[0019] The length of the monitoring area L monitor The calculation formula includes:

[0020] L monitor =speed max *time react +dis brake ;

[0021]

[0022] Where speed max Indicates the maximum speed of vehicles in the monitoring area, time reactIndicates the driver's reaction time, time react The fixed value is 2 seconds, dis brake Indicates the braking distance at the maximum speed of the vehicle in the monitoring area, acc brake The braking acceleration at the maximum speed of the vehicle in the monitoring area, where dis brake and acc brake All are custom parameters;

[0023] The length of the control area L control The calculation formula is as follows:

[0024] L control =speed max *time adjust ;

[0025] where time adjust Indicates the time it takes for the maximum speed of the vehicle in the monitoring area to decelerate to 0, and time adjust is a custom parameter.

[0026] Furthermore, the risk assessment model includes N time steps, the i-th time step inputs the i-th sequence unit of the feature sequence and outputs the hidden state;

[0027] The hidden state output at the Nth time step is input to the first classifier, and the classification space of the first classifier represents the risk level in the monitoring area;

[0028] The sample labels of the training samples used to train the risk assessment model are obtained by annotation by relevant technical experts.

[0029] Furthermore, the calculation formula of the risk assessment model includes:

[0030]

[0031] h i =tanh(W2*x i +W3*(g i ⊙h i-1 )+b2);

[0032] g i =simgoid(W4*x i +W5*h i-1 +b3);

[0033] in represents the hidden state output at the i-th time step, x i represents the i-th sequence unit of the feature sequence input at the i-th time step, h i and g idenote the candidate hidden state and gated unit at the i-th time step, respectively, and h i-1 represents the candidate hidden state at the i-1th time step, x k represents the kth sequence unit of the feature sequence input at the kth time step, W1, W2, W3, W4 and W5 represent the first weight parameter, the second weight parameter, the third weight parameter, the fourth weight parameter and the fifth weight parameter respectively, b1, b2 and b3 represent the first bias parameter, the second bias parameter and the third bias parameter respectively, concat represents a concatenation operation, MLP represents a multilayer perceptron, ⊙ represents element-by-element multiplication, tanh represents a hyperbolic tangent activation function, and sigmoid represents a sigmoid activation function.

[0034] Furthermore, a scanning area is generated with the vehicle mapped by each node as the center and a preset distance as the radius, and edges are constructed between the nodes corresponding to all vehicles in the scanning area, wherein the preset distance is a custom parameter.

[0035] Further, the first prediction model includes: a first hidden layer, a first extraction layer and a second classifier;

[0036] The first hidden layer inputs the graph network data and outputs the first graph update matrix;

[0037] A row vector of the first graph update matrix represents a first updated feature of a node;

[0038] The first extraction layer is used to extract the first update feature of each node of the first graph update matrix and input it into the second classifier, and the classification space of the second classifier represents the maximum speed of the vehicle;

[0039] The calculation formula for the first hidden layer includes:

[0040]

[0041] Where S1 represents the first graph update matrix, represents the first updated feature of the u-th node, Col u represents the set of nodes that have edges connected to the u-th node, H u and H v They represent the initial features of the u-th node and the v-th node respectively, U1 represents the first weight parameter corresponding to the first hidden layer, Represents the first distance coefficient between the uth node and the vth node, dis uv represents the Euclidean distance between the vehicle mapped by the data of the u-th node and the vehicle mapped by the data of the v-th node, pile represents a stacking operation, concat represents a concatenation operation, MLP represents a multi-layer perceptron, e represents a natural constant, and σ represents the standard deviation of the Gaussian kernel.

[0042] Furthermore, obtaining sample labels of training samples for training the first prediction model through a simulation platform includes the following steps:

[0043] Step S201, generating a speed within a speed range of the lane where the vehicle is located for each vehicle in the control area through a simulation platform;

[0044] Step S202, taking the speed of the vehicle that has not reached the center point of the monitoring area within the time threshold as a sample label of a training sample, wherein the time threshold is a custom parameter;

[0045] Step S203, repeating steps S201 and S202 until P training samples are obtained, where P is a custom parameter.

[0046] Further, the second prediction model includes: a second hidden layer, a second extraction layer, a third classifier and a conversion layer;

[0047] The second hidden layer inputs the graph network data and outputs the second graph update matrix;

[0048] A row vector of the second graph update matrix represents a second updated feature of a node;

[0049] The second extraction layer is used to extract the second update feature of each node of the second graph update matrix and input it into the third classifier, and the classification space of the third classifier represents the Euclidean distance between the lane change position of the vehicle and the traffic data collection point;

[0050] The conversion layer is used to convert the Euclidean distance between the lane-changing position of the vehicle and the traffic data collection point into longitude and latitude representation according to the longitude and latitude of the traffic data collection point;

[0051] The calculation formula for the second hidden layer includes:

[0052]

[0053] Where S2 represents the second graph update matrix, represents the second updated feature of the u-th node, Col u represents the set of nodes that have edges connected to the u-th node, H u and H v Represent the initial features of the u-th node and the v-th node respectively, U2 represents the second weight parameter corresponding to the second hidden layer, Represents the second distance coefficient between the uth node and the vth node, dis uv represents the Euclidean distance between the vehicle mapped by the u-th node and the vehicle mapped by the v-th node, dis min and dis maxIt represents the minimum and maximum values ​​of the Euclidean distance between the vehicles mapped by the node data, SAttion represents the self-attention mechanism function, pile represents the stacking operation, and concat represents the concatenation operation.

[0054] Furthermore, obtaining sample labels of training samples for training the second prediction model through a simulation platform includes the following steps:

[0055] Step S301, randomly generating a lane-changing position for each vehicle in a control area through a simulation platform;

[0056] Step S302, if it is determined that the Euclidean distance between the lane-changing positions of each vehicle is less than a distance threshold, the lane-changing position of each vehicle is used as a sample label of a training sample, wherein the distance threshold is a custom parameter;

[0057] Step S303, repeating steps S301 and S302 until Q training samples are obtained, where Q is a custom parameter.

[0058] The beneficial effects of the present invention are as follows: the present invention detects the risk level in the monitoring area of ​​the highway through a risk assessment model, and comprehensively analyzes the driving data of each vehicle in the control area through a first prediction model and a second prediction model, and generates effective traffic control measures for each vehicle in the control area, thereby ensuring the driving safety of the vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is a schematic diagram of a highway traffic condition detection system based on artificial intelligence of the present invention;

[0060] Figure 2 It is a flow chart of obtaining sample labels of training samples for training a first prediction model through a simulation platform of the present invention;

[0061] Figure 3 It is a flow chart of obtaining sample labels of training samples for training a second prediction model through a simulation platform of the present invention.

[0062] In the figure: traffic data collection module 101, risk level generation module 102, traffic control measures generation module 103, graph network data construction module 104, vehicle speed limit measures generation module 105, vehicle diversion measures generation module 106. DETAILED DESCRIPTION

[0063] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0064] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0065] like Figure 1 to Figure 3 As shown, a highway traffic condition detection system based on artificial intelligence includes:

[0066] Traffic data collection module 101, which is used to collect traffic data in the monitoring area within a preset time period T and at preset time intervals t, and generate a feature sequence;

[0067] The feature sequence includes N sequence units, and the i-th sequence unit represents the traffic data at the i-th time point, where N = T / t, 1≤i≤N;

[0068] Traffic data includes: traffic volume, vehicle density and weather conditions;

[0069] A risk level generating module 102, which is used to input the feature sequence into the risk assessment model, and the output value represents the risk level in the monitoring area;

[0070] Risk levels include: low risk level, medium risk level and high risk level;

[0071] A traffic control measure generation module 103, which is used to generate traffic control measures for each vehicle in the control area according to the risk level in the monitoring area;

[0072] Low risk levels do not generate traffic control measures;

[0073] The traffic control measures corresponding to the medium risk level are vehicle speed limit measures;

[0074] Traffic control measures corresponding to high risk levels are vehicle diversion measures;

[0075] A graph network data construction module 104, which is used to collect the driving data of each vehicle in the control area and construct graph network data;

[0076] Graph network data includes: nodes, initial features of nodes, and edges between nodes;

[0077] The node establishes data mapping with each vehicle in the control area;

[0078] The initial features of the node are represented according to the driving data of the vehicle to which the node is data mapped;

[0079] Driving data includes: vehicle type, maximum speed limit of the lane, vehicle speed and acceleration;

[0080] A vehicle speed limit measure generation module 105, which is used to input the graph network data into the first prediction model and output the vehicle speed limit measure;

[0081] Vehicle speed limit measures indicate the maximum speed of each vehicle in the control area;

[0082] A vehicle diversion measure generation module 106, which is used to input the graph network data into the second prediction model and output the vehicle diversion measure;

[0083] The vehicle diversion measure represents the lane-changing position of each vehicle in the control area, where the lane-changing position is represented by longitude and latitude.

[0084] In one embodiment of the present invention, the traffic volume represents the number of vehicles in the monitoring area; the vehicle density is equal to the traffic volume divided by the area of ​​the monitoring area; the weather conditions include: rainfall, snowfall, visibility and weather type, wherein the weather types include: sunny, rainy, snowy and haze, and the weather type is represented by a real number code, for example, a sunny weather type is represented by the real number 1.1, and a rainy weather type is represented by the real number 1.2.

[0085] In one embodiment of the present invention, the vehicle density can also be obtained by dividing the vehicle flow by the total lane length and the total lane width in the monitoring area.

[0086] In one implementation of the present invention, the monitoring area and the control area are both rectangular areas, the monitoring area and the control area are adjacent, and the widths of the monitoring area and the control area are both lane widths;

[0087] The length of the monitoring area Lmonitor The calculation formula includes:

[0088] L monitor =speed max *time react +dis brake ;

[0089]

[0090] Where speed max Indicates the maximum speed of vehicles in the monitoring area, time react Indicates the driver's reaction time, time react The fixed value is 2 seconds, dis brake Indicates the braking distance at the maximum speed of the vehicle in the monitoring area, acc brake The braking acceleration at the maximum speed of the vehicle in the monitoring area, where dis brake and acc brake All are custom parameters;

[0091] The length of the control area L control The calculation formula is as follows:

[0092] L control =speed max *time adjust ;

[0093] where time adjust Indicates the time it takes for the maximum speed of the vehicle in the monitoring area to decelerate to 0, and time adjust is a custom parameter.

[0094] It should be noted that the length of the monitoring area is calculated based on the maximum speed of the vehicle and the driver's reaction time, so that there is enough time to identify and deal with abnormal situations before the vehicle driver reacts; the length of the control area is calculated by taking into account the deceleration time of the vehicle, so that the driver can take speed limit and lane change measures in time; in addition, the braking distance, braking acceleration and time for the maximum speed to decelerate to 0 of the vehicle at the maximum speed in the monitoring area can be obtained through the vehicle database provided by the vehicle manufacturer, which will not be elaborated here.

[0095] In one implementation of the present invention, the risk assessment model includes N time steps, the i-th time step inputs the i-th sequence unit of the feature sequence and outputs a hidden state;

[0096] The hidden state output at the Nth time step is input to the first classifier, and the classification space of the first classifier represents the risk level in the monitoring area.

[0097] In one implementation of the present invention, sample labels of training samples used to train a risk assessment model are obtained by annotation by relevant technical experts.

[0098] In one embodiment of the present invention, the calculation formula of the risk assessment model includes:

[0099]

[0100] h i =tanh(W2*x i +W3*(g i ⊙h i-1 )+b2);

[0101] g i =simgoid(W4*x i +W5*h i-1 +b3);

[0102] in represents the hidden state output at the i-th time step, x i represents the i-th sequence unit of the feature sequence input at the i-th time step, h i and g i denote the candidate hidden state and gated unit at the i-th time step, respectively, and h i-1 represents the candidate hidden state at the i-1th time step, x k represents the kth sequence unit of the feature sequence input at the kth time step, W1, W2, W3, W4 and W5 represent the first weight parameter, the second weight parameter, the third weight parameter, the fourth weight parameter and the fifth weight parameter respectively, b1, b2 and b3 represent the first bias parameter, the second bias parameter and the third bias parameter respectively, concat represents a concatenation operation, MLP represents a multilayer perceptron, ⊙ represents element-by-element multiplication, tanh represents a hyperbolic tangent activation function, and sigmoid represents a sigmoid activation function.

[0103] In one embodiment of the present invention, the time step of the risk assessment model can also be constructed based on RNN (recurrent neural network) units, LSTM (long short-term memory neural network) units or GRU (gated network) units.

[0104] In one embodiment of the present invention, vehicle types include: buses and trucks. Buses include: Type 1 buses (≤7 seats), Type 2 buses (8 to 19 seats), Type 3 buses (20 to 39 seats) and Type 4 buses (≥40 seats), and are represented by real number codes, for example, Type 1 buses are represented by the real number 2.1, and Type 2 buses are represented by the real number 2.2; trucks include: Type 1 trucks (≤2 tons), Type 2 trucks (2 tons to 5 tons inclusive), Type 3 trucks (5 tons to 10 tons inclusive), Type 4 trucks (10 tons to 15 tons inclusive) and Type 5 trucks (>15 tons), and are represented by real number codes, for example, Type 1 trucks are represented by the real number 3.1, and Type 2 trucks are represented by 3.2.

[0105] In one embodiment of the present invention, a scanning area is generated with the vehicle mapped by each node as the center and a preset distance as the radius, and edges are constructed between the nodes corresponding to all vehicles in the scanning area, wherein the preset distance is a custom parameter. Preferably, the preset distance is set to 200 meters.

[0106] In one implementation of the present invention, the first prediction model includes: a first hidden layer, a first extraction layer, and a second classifier;

[0107] The first hidden layer inputs the graph network data and outputs the first graph update matrix;

[0108] A row vector of the first graph update matrix represents a first updated feature of a node;

[0109] The first extraction layer is used to extract the first update feature of each node of the first graph update matrix and input it into the second classifier, and the classification space of the second classifier represents the maximum speed of the vehicle.

[0110] In one implementation of the present invention, the calculation formula of the first hidden layer includes:

[0111]

[0112] Where S1 represents the first graph update matrix, represents the first updated feature of the u-th node, Col u represents the set of nodes that have edges connected to the u-th node, H u and H v They represent the initial features of the u-th node and the v-th node respectively, U1 represents the first weight parameter corresponding to the first hidden layer, Represents the first distance coefficient between the uth node and the vth node, dis uv represents the Euclidean distance between the vehicle mapped by the data of the u-th node and the vehicle mapped by the data of the v-th node, pile represents a stacking operation, concat represents a concatenation operation, MLP represents a multi-layer perceptron, e represents a natural constant, and σ represents the standard deviation of the Gaussian kernel.

[0113] In one embodiment of the present invention, Figure 2 As shown, obtaining sample labels of training samples for training the first prediction model through a simulation platform includes the following steps:

[0114] Step S201, generating a speed within a speed range of the lane where the vehicle is located for each vehicle in the control area through a simulation platform;

[0115] For example, the speed range of the leftmost lane is 110-120 km / h, the speed range of the middle lane is 90-100 km / h, and the speed range of the rightmost lane is 60-90 km / h;

[0116] Step S202, taking the speed of the vehicle that has not reached the center point of the monitoring area within the time threshold as a sample label of a training sample, wherein the time threshold is a custom parameter, and preferably, the time threshold is set to 10 seconds;

[0117] Step S203, repeating step S201 and step S202 until P training samples are obtained, where P is a custom parameter, and preferably, P is set to 2000.

[0118] In one implementation of the present invention, the second prediction model includes: a second hidden layer, a second extraction layer, a third classifier, and a conversion layer;

[0119] The second hidden layer inputs the graph network data and outputs the second graph update matrix;

[0120] A row vector of the second graph update matrix represents a second updated feature of a node;

[0121] The second extraction layer is used to extract the second update feature of each node of the second graph update matrix and input it into the third classifier, and the classification space of the third classifier represents the Euclidean distance between the lane change position of the vehicle and the traffic data collection point;

[0122] The conversion layer is used to convert the Euclidean distance between the lane-changing position of the vehicle and the traffic data collection point into longitude and latitude representation according to the longitude and latitude of the traffic data collection point.

[0123] In one embodiment of the present invention, the Euclidean distance between the lane-changing position of the vehicle and the traffic data collection point is converted into longitude and latitude representation according to the longitude and latitude of the traffic data collection point by GIS (Geographic Information System) software.

[0124] In one implementation of the present invention, the calculation formula of the second hidden layer includes:

[0125]

[0126] Where S2 represents the second graph update matrix, represents the second updated feature of the u-th node, Col u represents the set of nodes that have edges connected to the u-th node, H u and H v Represent the initial features of the u-th node and the v-th node respectively, U2 represents the second weight parameter corresponding to the second hidden layer, Represents the second distance coefficient between the uth node and the vth node, dis uv represents the Euclidean distance between the vehicle mapped by the u-th node and the vehicle mapped by the v-th node, dis min and dis max It represents the minimum and maximum values ​​of the Euclidean distance between the vehicles mapped by the node data, SAttion represents the self-attention mechanism function, pile represents the stacking operation, and concat represents the concatenation operation.

[0127] In one embodiment of the present invention, Figure 3 As shown, obtaining sample labels of training samples for training the second prediction model through a simulation platform includes the following steps:

[0128] Step S301, randomly generating a lane-changing position for each vehicle in a control area through a simulation platform;

[0129] Step S302, if it is determined that the Euclidean distance between the lane-changing positions of each vehicle is less than a distance threshold, the lane-changing position of each vehicle is used as a sample label of a training sample, wherein the distance threshold is a custom parameter, and preferably, the distance threshold is set to 50 meters;

[0130] Step S303, repeating step S301 and step S302 until Q training samples are obtained, where Q is a custom parameter, and preferably, Q is set to 2000.

[0131] It should be noted that during the training process of the first prediction model and the second prediction model, the difference between the value output by the first prediction model and the second prediction model in each iteration and the value of the sample label of the training sample is specified as a loss function, such as a mean square error function, and then the parameters in the first prediction model and the second prediction model are updated by a gradient descent algorithm.

[0132] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A highway traffic condition detection system based on artificial intelligence, characterized in that: include: Traffic data collection module, which is used to collect traffic data in the monitoring area within a preset time period T and at preset time intervals t, and generate a feature sequence; The feature sequence includes N sequence units, and the i-th sequence unit represents the traffic data at the i-th time point, where N = T / t, 1≤i≤N; the traffic data includes: traffic flow, vehicle density and weather conditions; A risk level generation module is used to input the feature sequence into the risk assessment model, and the output value represents the risk level in the monitoring area; the risk levels include: low risk level, medium risk level and high risk level; A traffic control measure generation module, which is used to generate traffic control measures for each vehicle in the control area according to the risk level in the monitoring area; Low risk levels do not generate traffic control measures; The traffic control measures corresponding to the medium risk level are vehicle speed limit measures; Traffic control measures corresponding to high risk levels are vehicle diversion measures; A graph network data construction module, which is used to collect the driving data of each vehicle in the control area and construct graph network data; The graph network data includes: nodes, initial features of nodes and edges between nodes; data mapping is established between nodes and each vehicle in the control area; the initial features of nodes are represented according to the driving data of the vehicle mapped by the node; the driving data includes: vehicle type, maximum speed limit of the lane, vehicle speed and acceleration; A vehicle speed limit measure generation module, which is used to input the graph network data into the first prediction model and output a vehicle speed limit measure; the vehicle speed limit measure represents the maximum speed of each vehicle in the control area; The vehicle diversion measure generation module is used to input the graph network data into the second prediction model and output the vehicle diversion measure; the vehicle diversion measure represents the lane change position of each vehicle in the control area, wherein the lane change position is represented by longitude and latitude.

2. The highway traffic condition detection system based on artificial intelligence according to claim 1 is characterized in that: Traffic volume indicates the number of vehicles in the monitoring area; Vehicle density is equal to the traffic volume divided by the area of ​​the monitoring area; weather conditions include: rainfall, snowfall, visibility and weather type, among which weather types include: sunny, rainy, snowy and haze, and weather types are represented by real number codes.

3. The highway traffic condition detection system based on artificial intelligence according to claim 1 is characterized in that: The monitoring area and the control area are both rectangular areas, adjacent to each other, and the widths of the monitoring area and the control area are both lane widths; The length of the monitoring area L monitor The calculation formula include: L monitor =speed max *time react +dis brake ; Where speed max Indicates the maximum speed of vehicles in the monitoring area, time react Indicates the driver's reaction time, time react The fixed value is 2 seconds, dis brake Indicates the braking distance at the maximum speed of the vehicle in the monitoring area, acc brake The braking acceleration at the maximum speed of the vehicle in the monitoring area, where dis brake and acc brake All are custom parameters; The length of the control area L control The calculation formula is as follows: L control =speed max *time adjust ; where time adjust Indicates the time it takes for the maximum speed of the vehicle in the monitoring area to decelerate to 0, and time adjust is a custom parameter.

4. The highway traffic condition detection system based on artificial intelligence according to claim 1 is characterized in that: The risk assessment model includes N time steps, where the i-th sequence unit of the feature sequence is input at the i-th time step and the hidden state is output; The hidden state output at the Nth time step is input to the first classifier, and the classification space of the first classifier represents the risk level in the monitoring area; The sample labels of the training samples used to train the risk assessment model are obtained by annotation by relevant technical experts.

5. The highway traffic condition detection system based on artificial intelligence according to claim 4 is characterized in that: The calculation formula of the risk assessment model includes: h i =tanh(W2*x i +W3*(g i ⊙h i-1 )+b2); g i =simgoid(W4*x i +W5*h i-1 +b3); in represents the hidden state output at the i-th time step, x i represents the i-th sequence unit of the feature sequence input at the i-th time step, h i and g i denote the candidate hidden state and gated unit at the i-th time step, respectively, and h i-1 represents the candidate hidden state at the i-1th time step, x k represents the kth sequence unit of the feature sequence input at the kth time step, W1, W2, W3, W4 and W5 represent the first weight parameter, the second weight parameter, the third weight parameter, the fourth weight parameter and the fifth weight parameter respectively, b1, b2 and b3 represent the first bias parameter, the second bias parameter and the third bias parameter respectively, concat represents a concatenation operation, MLP represents a multilayer perceptron, ⊙ represents element-by-element multiplication, tanh represents a hyperbolic tangent activation function, and sigmoid represents a sigmoid activation function.

6. The highway traffic condition detection system based on artificial intelligence according to claim 1 is characterized in that: A scanning area is generated with the vehicle mapped by each node as the center and the preset distance as the radius, and edges are built between the nodes corresponding to all vehicles in the scanning area, where the preset distance is a custom parameter.

7. The highway traffic condition detection system based on artificial intelligence according to claim 1 is characterized in that: The first prediction model includes: a first hidden layer, a first extraction layer, and a second classifier; The first hidden layer inputs the graph network data and outputs the first graph update matrix; A row vector of the first graph update matrix represents a first updated feature of a node; The first extraction layer is used to extract the first update feature of each node of the first graph update matrix and input it into the second classifier, and the classification space of the second classifier represents the maximum speed of the vehicle; The calculation formula for the first hidden layer includes: Where S1 represents the first graph update matrix, represents the first updated feature of the u-th node, Col u represents the set of nodes that have edges connected to the u-th node, H u and H v They represent the initial features of the u-th node and the v-th node respectively, U1 represents the first weight parameter corresponding to the first hidden layer, Represents the first distance coefficient between the uth node and the vth node, dis uv represents the Euclidean distance between the vehicle mapped by the data of the u-th node and the vehicle mapped by the data of the v-th node, pile represents a stacking operation, concat represents a concatenation operation, MLP represents a multi-layer perceptron, e represents a natural constant, and σ represents the standard deviation of the Gaussian kernel.

8. The highway traffic condition detection system based on artificial intelligence according to claim 1 is characterized in that: Obtaining sample labels of training samples for training a first prediction model through a simulation platform includes the following steps: Step S201, generating a speed within a speed range of the lane where the vehicle is located for each vehicle in the control area through a simulation platform; Step S202, taking the speed of the vehicle that has not reached the center point of the monitoring area within the time threshold as a sample label of a training sample, wherein the time threshold is a custom parameter; Step S203, repeating steps S201 and S202 until P training samples are obtained, where P is a custom parameter.

9. The highway traffic condition detection system based on artificial intelligence according to claim 1 is characterized in that: The second prediction model includes: a second hidden layer, a second extraction layer, a third classifier and a conversion layer; The second hidden layer inputs the graph network data and outputs the second graph update matrix; A row vector of the second graph update matrix represents a second updated feature of a node; The second extraction layer is used to extract the second update feature of each node of the second graph update matrix and input it into the third classifier, and the classification space of the third classifier represents the Euclidean distance between the lane change position of the vehicle and the traffic data collection point; The conversion layer is used to convert the Euclidean distance between the lane-changing position of the vehicle and the traffic data collection point into longitude and latitude representation according to the longitude and latitude of the traffic data collection point; The calculation formula for the second hidden layer includes: Where S2 represents the second graph update matrix, represents the second updated feature of the u-th node, Col u represents the set of nodes that have edges connected to the u-th node, H u and H v Represent the initial features of the u-th node and the v-th node respectively, U2 represents the second weight parameter corresponding to the second hidden layer, Represents the second distance coefficient between the uth node and the vth node, dis uv represents the Euclidean distance between the vehicle mapped by the u-th node and the vehicle mapped by the v-th node, min and dis max It represents the minimum and maximum values ​​of the Euclidean distance between the vehicles mapped by the node data, SAttion represents the self-attention mechanism function, pile represents the stacking operation, and concat represents the concatenation operation.

10. The highway traffic condition detection system based on artificial intelligence according to claim 1 is characterized in that: Obtaining sample labels of training samples for training a second prediction model through a simulation platform includes the following steps: Step S301, randomly generating a lane-changing position for each vehicle in a control area through a simulation platform; Step S302, if it is determined that the Euclidean distance between the lane-changing positions of each vehicle is less than a distance threshold, the lane-changing position of each vehicle is used as a sample label of a training sample, wherein the distance threshold is a custom parameter; Step S303, repeating steps S301 and S302 until Q training samples are obtained, where Q is a custom parameter.

Citation Information

Cited By

  • Traffic guidance system for highway engineering construction

    CN119694140A

  • Internet of Things security protection system and method based on artificial intelligence

    CN119995999A