An operation and maintenance method for intelligent traffic monitoring system

Abstract

The present invention relates to the technical field of traffic information services, and discloses an operation and maintenance method of an intelligent traffic monitoring system, comprising the following steps: collecting traffic monitoring data and freight system data; constructing first structured data according to the collected data; constructing a model, wherein the model comprises a first hidden layer, a second hidden layer, a pre-training layer and a decoding layer, wherein the first hidden layer is used to learn and identify the time series information in the second structural dimension to obtain the update status of all objects to the second hidden layer, and the second hidden layer is used to learn and identify the correlation information related to traffic monitoring between objects in the first structural dimension to obtain the feature representation of all objects to the decoding layer, and the decoding layer maps the feature representations of all objects after splicing them to a strategy space, and outputs a truck control strategy; the present invention can prevent accidents caused by dangerous driving behaviors through traffic compulsory management measures, and can reduce the loss of freight and the loss of drivers while implementing the compulsory management measures.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic information services, and more specifically, to an operation and maintenance method of an intelligent traffic monitoring system. Background Art

[0002] Intelligent traffic monitoring system is a high-tech system that integrates the achievements of multiple disciplines such as information technology, data communication technology, sensor technology, electronic control technology and computer processing technology. It aims to effectively manage and control urban traffic through intelligent means to achieve the purpose of improving traffic efficiency, ensuring traffic safety, reducing environmental pollution and energy consumption; intelligent traffic monitoring system includes traffic signal control, electronic toll collection system, traffic video monitoring, public transportation management, accident warning and emergency response, intelligent parking management, traffic data analysis and other business sectors. Traffic data analysis is the main content of third-party services, and the purpose is to conduct in-depth analysis of the large amount of collected traffic data and provide a full range of services for traffic planning or management. The third-party services in the existing technology are simply analyzing traffic data, but the freight system involves a large number of transportation orders and lacks analysis of relevant information, which inevitably leads to the lack of humanization of traffic management related to the freight system. Summary of the invention

[0003] The present invention provides an operation and maintenance method of an intelligent traffic monitoring system, which adds general social information of a freight system to traffic data for comprehensive analysis and solves technical problems in related technologies.

[0004] The present invention provides an operation and maintenance method of an intelligent traffic monitoring system, comprising the following steps:

[0005] Step 1, collect traffic monitoring data and freight system data;

[0006] Step 2, constructing the first structured data based on the collected data. The first structured data includes two structural dimensions. The basic structural elements included in the first structural dimension are objects. The objects of the first structured data include: drivers, vehicles, goods, roads, and historical dangerous driving events; the basic structural elements included in the second structural dimension are time series units;

[0007] Step 3, build a model, which includes a first hidden layer, a second hidden layer, a pre-training layer and a decoding layer. The first hidden layer is used to learn and identify the time series information in the second structural dimension to obtain the update status of all objects to the second hidden layer. The second hidden layer is used to learn and identify the correlation information related to traffic monitoring between objects in the first structural dimension to obtain the feature representation of all objects to the decoding layer. The decoding layer concatenates the feature representations of all objects and maps them to the policy space to output the truck control policy.

[0008] Step 4, execute the truck control strategy obtained by inputting the traffic monitoring data collected at the current moment into the model; the truck control strategy is a matrix representation, the first row and the gth column of the matrix representation represent the location where the gth truck is stopped, the second row and the gth column represent the ID of the driver who is replaced after the stop, and the third row and the gth column represent the distribution center where the gth truck needs to be transferred after the stop;

[0009] The model is trained by reinforcement learning, and the state of reinforcement learning is defined as traffic monitoring data and freight system data; the action of reinforcement learning is traffic management measures, and the reward function is:

[0010] ;

[0011] in, Represents the reward value obtained by executing the ath action in the bth state;

[0012] represents an accident prevention reward, which is inversely proportional to the number of accidents or emergencies that occur;

[0013] It represents the timeliness reward, which is inversely proportional to the time of the goods’ arrival delay;

[0014] It represents the reward for the value of the goods, which is inversely proportional to the loss of the value of the goods during transportation;

[0015] , , are the first, second, and third hyperparameters.

[0016] Furthermore, the data collected in Step 1 includes: real-time truck onboard data, historical dangerous driving event data, road condition information, and cargo information.

[0017] Furthermore, the method for constructing the first structured data includes:

[0018] extracting a first structured data object;

[0019] Extract specific features associated with the object, and divide the extracted specific features into multiple parts according to the collection period, and map one part to a time sequence unit. The order of the time sequence units is consistent with the time order of the collection period of the specific features;

[0020] Establishing a mapping between the extracted objects and their associated specific features;

[0021] For any two objects, if there is a correlation related to traffic monitoring between the associated specific features or there is a correlation related to traffic monitoring between the objects, an object relationship is established between the two objects;

[0022] The first structured data objects include:

[0023] Driver object: represents different truck drivers;

[0024] Vehicle object: represents a truck;

[0025] Cargo object: represents the goods being transported;

[0026] Road object: represents a road or road section;

[0027] Historical dangerous driving event object: represents historical dangerous driving events;

[0028] Specific characteristics associated with the driver: driver personal information, driver physiological data, continuous driving time;

[0029] Specific characteristics associated with the vehicle: load, speed, acceleration, vehicle type;

[0030] Specific characteristics associated with the goods: type of goods, shelf life, unit value, unit transportation price;

[0031] Specific features associated with roads: road monitoring images, road traffic, width, number of lanes, speed limit;

[0032] Specific features associated with historical dangerous driving events: location, time, traffic conditions, vehicle, cargo in the vehicle, and description text of the result of the congestion;

[0033] Relevance to traffic monitoring includes:

[0034] The driver and the vehicle being driven;

[0035] The vehicle and the cargo it carries;

[0036] The vehicle and the road on which it is located;

[0037] the vehicle’s association with historical dangerous driving incidents;

[0038] Driver associations with historical risky driving events.

[0039] Furthermore, the first hidden layer contains multiple parallel branches, one parallel branch inputs the temporal unit of an object, each parallel branch performs step-by-step calculation, and the expression of the t-th step calculation of the r-th parallel branch is as follows:

[0040] ;

[0041] ;

[0042] ;

[0043] ;

[0044] and They represent the activation vectors of the reset gate and update gate of the t-th step of the r-th parallel branch of the first hidden layer respectively;

[0045] represents the candidate state generated in the tth step of the rth parallel branch of the first hidden layer;

[0046] and They represent the update states of the t-th step and the t-1-th step of the r-th parallel branch of the first hidden layer respectively;

[0047] J, K, M, F, P, and Q represent the first, second, third, fourth, fifth, and sixth conversion matrices, respectively;

[0048] e, c, and v represent the first, second, and third deviation parameters, respectively;

[0049] r∈G, G represents the set of all objects, t∈{1, 2, 3, …, n}, t=1 ;

[0050] Represents the Sigmoid activation function;

[0051] Represents the tanh activation function.

[0052] Furthermore, the second hidden layer, which includes multiple serial layers, has the following expression for the l+1th layer:

[0053] ;

[0054] ;

[0055] where i∈G, , represents a set of objects that have a relationship with the i-th object related to traffic monitoring;

[0056] Representation Object In the second hidden layer The feature representation of the layer.

[0057] Represents the object To Object Related information related to traffic monitoring;

[0058] Represents an aggregate function;

[0059] represents the update function;

[0060] represents the information transfer function;

[0061] l∈{0, 1, 2, 3, ..., L-1}, L represents the total number of the second hidden layer, when l=0, , , and They represent the updated states generated by the tth step of the i-th parallel branch and the j-th parallel branch of the first hidden layer respectively.

[0062] Further, ,in Represents the weight parameter of the lth layer of the second hidden layer.

[0063] Furthermore, the decoding layer concatenates the feature representations of all objects at the Lth layer, and the concatenation of the feature representations at the Lth layer is expressed as ,in It represents the concatenation of the feature representations of all objects at layer L. represents the feature representation of the i-th object in the L-th layer of the second hidden layer, The fully connected layer of the input decoding layer is mapped to the policy space and the truck control policy is output.

[0064] Furthermore, the output of the second hidden layer is connected to the pre-training layer, and the pre-training layer outputs a result indicating whether the driver has engaged in dangerous behavior; the model is pre-trained through the loss value of the output of the pre-training layer; and then the model is retrained through the loss of the output of the decoding layer.

[0065] Further, if no accident or emergency occurs after the action is performed, In the event of an accident or emergency, ;

[0066] If the goods arrive at the destination on time, ; If the goods arrive late, , is the penalty coefficient of delay time, Indicates the delay time for all shipments, Indicates the scheduled transportation time for all shipments;

[0067] If the goods are intact during transportation, If the goods are damaged in any way, , is the penalty coefficient of the damage degree, Indicates the loss of the value of all goods. Indicates the value of all goods.

[0068] The present invention provides a computer storage medium, which is used to store instructions that can be read by a computer, and the instructions stored in the medium can execute the aforementioned operation and maintenance method of an intelligent traffic monitoring system when being read.

[0069] The beneficial effects of the present invention are as follows: the present invention comprehensively analyzes the information and traffic data of the freight system, which can not only prevent accidents caused by dangerous driving behaviors to a certain extent through compulsory traffic management measures, but also can reduce the losses of freight and drivers while implementing compulsory management measures, thereby improving the humanization of management. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a flow chart of an operation and maintenance method of an intelligent traffic monitoring system of the present invention;

[0071] Figure 2 This is Example 1 of the desensitized data sample of the present invention;

[0072] Figure 3 This is Example 2 of the desensitized data sample of the present invention;

[0073] Figure 4 It is the change of the average Q value during the model training process of the present invention. DETAILED DESCRIPTION

[0074] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations 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 present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0075] At least one embodiment of the present invention discloses an operation and maintenance method of an intelligent traffic monitoring system, such as Figure 1 As shown, the following steps are included:

[0076] Step 1: Collect traffic monitoring data and freight system data:

[0077] Real-time truck onboard data (such as speed, acceleration, navigation path, driver physiological data, driver personal information, etc.);

[0078] Historical dangerous driving incident data (location, time, cargo conditions, traffic conditions, etc.);

[0079] Traffic information (road congestion, alternative roads, etc.);

[0080] Cargo information (type, timeliness, emergency measures, etc.);

[0081] Of course, generally, data preprocessing, including feature engineering, is performed while collecting data. The features described later are the results after data preprocessing, which is another form of data expression. However, the corresponding data is still identified by the meaning represented by the data in the embodiments.

[0082] Step 2, constructing the first structured data based on the collected data. The first structured data includes two structural dimensions. The basic structural elements included in the first structural dimension are objects. The objects of the first structured data include: drivers, vehicles, goods, roads, and historical dangerous driving events; the basic structural elements included in the second structural dimension are time series units;

[0083] The method of constructing the first structured data includes:

[0084] extracting a first structured data object;

[0085] Driver object: represents different truck drivers;

[0086] Vehicle object: represents a truck;

[0087] Cargo object: represents the goods being transported;

[0088] Road object: represents a road or road section;

[0089] Historical dangerous driving event object: represents historical dangerous driving events;

[0090] Extract specific features associated with the object, and divide the extracted specific features into multiple parts according to the collection period, and map one part to a time sequence unit. The order of the time sequence units is consistent with the time order of the collection period of the specific features;

[0091] Establishing a mapping between the extracted objects and their associated specific features;

[0092] Specific characteristics associated with the driver: driver personal information, driver physiological data, continuous driving time, etc.;

[0093] Specific characteristics associated with the vehicle: load, speed, acceleration, vehicle type, etc.;

[0094] Specific characteristics associated with the goods: type of goods, shelf life, unit value, unit transportation price, etc.;

[0095] Specific features associated with roads: road monitoring images, road traffic, width, number of lanes, speed limit, etc.

[0096] Specific features associated with historical dangerous driving events: location, time, traffic conditions, vehicle, cargo in the vehicle, description text of the result causing congestion, etc.

[0097] For any two objects, if there is a correlation related to traffic monitoring between the associated specific features or there is a correlation related to traffic monitoring between the objects, an object relationship is established between the two objects;

[0098] In some embodiments of the present invention, the associations related to traffic monitoring include:

[0099] The driver and the vehicle being driven;

[0100] The vehicle and the cargo it carries;

[0101] The vehicle and the road on which it is located;

[0102] the vehicle’s association with historical dangerous driving incidents;

[0103] Driver associations with historical risky driving events.

[0104] Step 3, build a model, which includes a first hidden layer, a second hidden layer, a pre-training layer and a decoding layer. The first hidden layer is used to learn and identify the time series information in the second structural dimension to obtain the update status of all objects to the second hidden layer. The second hidden layer is used to learn and identify the correlation information related to traffic monitoring between objects in the first structural dimension to obtain the feature representation of all objects to the decoding layer. The decoding layer concatenates the feature representations of all objects and maps them to the policy space to output the truck control policy.

[0105] The policy space contains all possible truck control policies.

[0106] The output of the second hidden layer is connected to the pre-training layer, and the pre-training layer outputs the result indicating whether the driver's object has dangerous behavior. The model is pre-trained through the loss value of the output of the pre-training layer; and then the model is retrained (reinforcement learning) through the loss of the output of the decoding layer.

[0107] In one embodiment of the present invention, traffic management measures include the location where the truck is stopped, the replacement driver after the stop (used to replace the current driver to continue the freight mission), and the distribution center where the truck needs to be transferred after the stop (used to distribute the goods to other trucks for continued transportation).

[0108] Step 4: Execute the truck control strategy obtained by inputting the traffic monitoring data collected at the current moment into the model.

[0109] In one embodiment of the present invention, a first hidden layer for learning to recognize the time series information in the second structural dimension is provided, which includes multiple parallel branches, one parallel branch inputs a time series unit of an object, each parallel branch performs step-by-step calculation, and the expression of the t-th step calculation of the r-th parallel branch is as follows:

[0110]

[0111]

[0112]

[0113]

[0114] and They represent the activation vectors of the reset gate and update gate of the t-th step of the r-th parallel branch of the first hidden layer respectively;

[0115] represents the candidate state generated in the tth step of the rth parallel branch of the first hidden layer;

[0116] and They represent the update states of the t-th step and the t-1-th step of the r-th parallel branch of the first hidden layer respectively;

[0117] J, K, M, F, P, Q represent the first, second, third, fourth, fifth, and sixth transformation matrices (trainable parameters), respectively;

[0118] e, c, and v represent the first, second, and third deviation parameters (trainable parameters), respectively;

[0119] r∈G, G represents the set of all objects, t∈{1, 2, 3, …, n}, t=1 ;

[0120] Represents the Sigmoid activation function;

[0121] represents the tanh activation function;

[0122] In one embodiment of the present invention, a second hidden layer for learning to identify association information related to traffic monitoring between objects in a first structural dimension is provided, which includes a plurality of serial layers, and the expression of the l+1th layer is as follows:

[0123]

[0124]

[0125] where i∈G, , represents a set of objects that have a relationship with the i-th object related to traffic monitoring;

[0126] Representation Object In the second hidden layer The feature representation of the layer.

[0127] Represents the object To Object Related information related to traffic monitoring;

[0128] Represents an aggregation function that aggregates messages from neighbors (e.g. sum, average, or maximum).

[0129] Represents an update function used to update object features (such as GRU or simple MLP).

[0130] represents the information transfer function, for example, a summation function with a weight parameter, as follows:

[0131] ,in Represents the weight parameter of the lth layer of the second hidden layer.

[0132] l∈{0, 1, 2, 3, ..., L-1}, L represents the total number of the second hidden layer, when l=0, , , and They represent the updated states generated in the tth step of the i-th parallel branch and the j-th parallel branch of the first hidden layer respectively;

[0133] The pre-training layer can adopt a general classification layer, for example, output a vector, the d-th component of the vector indicates whether the d-th truck has dangerous driving behavior.

[0134] The decoding layer can use a general fully connected layer. The decoding layer concatenates the feature representations of all objects in the Lth layer. The concatenation of the feature representations of the Lth layer is expressed as ,in It represents the concatenation of the feature representations of all objects at layer L. Represents the feature representation of the i-th object in the L-th layer of the second hidden layer;

[0135] The output of the decoding layer is a matrix representation. The first row and g column of the matrix representation represents the location where the g-th truck is stopped (if the stop is not performed, the value is 0), the second row and g column represent the ID of the replacement driver after the stop (if the driver is not replaced, the value is 0), and the third row and g column represent the distribution center where the g-th truck needs to be transferred after the stop (if the goods are not transferred, the value is 0).

[0136] Of course, other layers having the function of decoding the output of the hidden layer, such as MLP, are not excluded as decoding layers.

[0137] The following is an example of a specific application scenario of the above method:

[0138] This example focuses on the road network in Liaoning Province, China, especially the busy freight routes (limited to the province). Liaoning Province is an important province in Northeast China, with a developed economy, active logistics industry, and large truck traffic. Therefore, the incidence of dangerous driving behavior is relatively high. The main cities in the study area include Shenyang, Dalian, Anshan, Fushun, etc. These cities have close transportation links and are important logistics hubs.

[0139] Traffic conditions

[0140] The region has a well-developed highway network, with a total mileage of more than 3,000 kilometers and an average daily traffic volume of more than 60,000 vehicles. Among them, trucks account for about 35%, and can reach more than 45% during peak hours. The main freight routes connect major industrial areas and ports, transporting a wide variety of goods, including but not limited to steel, auto parts, agricultural products, etc.

[0141] Data preparation

[0142] Real-time vehicle data

[0143] Vehicle speed: recorded once per second by the vehicle's onboard GPS device.

[0144] Acceleration: recorded once per second by the onboard accelerometer.

[0145] Driver physiological data: recorded once a minute via heart rate monitor and breathing monitor.

[0146] Video data: A video is recorded every 5 minutes by the in-car camera.

[0147] Vehicle location: GPS coordinates are recorded once a second.

[0148] Historical dangerous driving incident data

[0149] Location of occurrence: GPS coordinates.

[0150] Time: The timestamp when the event occurred.

[0151] Cargo conditions: cargo type, weight, value, etc.

[0152] Traffic conditions: traffic flow, weather conditions, etc. at the time of the accident.

[0153] Processing results: the processing measures and effects at that time.

[0154] Traffic information

[0155] Road congestion: obtained through traffic surveillance cameras and traffic management centers.

[0156] Alternative roads: obtained through the map API.

[0157] Weather information: obtained through weather stations and online weather APIs.

[0158] Traffic Flow: Real-time traffic flow data, including the number and distribution of trucks and other vehicles.

[0159] Cargo Information

[0160] Type: Steel, auto parts, agricultural products, etc.

[0161] Data Annotation

[0162] Dangerous driving behavior labeling: Based on historical event data and expert knowledge, the data is labeled to mark dangerous driving behaviors.

[0163] Cargo information labeling: label the cargo type, timeliness and other information.

[0164] Traffic condition marking: marking traffic flow, weather conditions, etc.

[0165] A set of desensitized data samples such as Figure 2 and 3 shown.

[0166] Model training (pre-training)

[0167] Backpropagation of the general binary classification loss is performed through the annotations of the data samples.

[0168] Model training (reinforcement learning)

[0169] Reinforcement Learning Framework

[0170] Environment: Highway network and freight system data, including vehicles, roads, traffic conditions, cargo information, etc.

[0171] Agent: Traffic monitoring system responsible for detecting and handling dangerous driving behavior.

[0172] Status: traffic monitoring data and freight system data, including real-time truck onboard data (such as speed, acceleration, navigation path, driver physiological data, driver personal information, etc.);

[0173] Historical dangerous driving incident data (location, time, cargo conditions, traffic conditions, etc.);

[0174] Traffic information (road congestion, alternative roads, etc.);

[0175] Cargo information (type, timeliness, emergency response measures, etc.).

[0176] Action: Traffic management measures.

[0177] Rewards: Rewards are given based on the effectiveness of the handling, such as successfully avoiding accidents, reducing traffic congestion, etc.

[0178] Reinforcement Learning Algorithms

[0179] Q-learning: Use the Q-learning algorithm to learn the optimal strategy.

[0180] The reward function is designed as follows:

[0181]

[0182] in, Represents the reward value obtained by executing the ath action in the bth state;

[0183] It means accident prevention reward;

[0184] If no accident or emergency occurs after the action is performed, In the event of an accident or emergency, ;

[0185] Indicates time-limited reward;

[0186] If the goods arrive at the destination on time, ; If the goods arrive late, , is the penalty coefficient of delay time, Indicates the delay time for all shipments, Indicates the scheduled transportation time for all shipments;

[0187] Indicates the value of the goods reward;

[0188] If the goods are intact during transportation, If the goods are damaged in any way, , is the penalty coefficient of the damage degree, Indicates the loss of the value of all goods. Indicates the value of all goods;

[0189] These are the first, second, and third hyperparameters used to adjust the influence of different reward factors on the total reward. The default values ​​are 0.4, 0.3, and 0.3.

[0190] In a simulated environment, the environment and the agent are modeled, and actions are performed to train and verify the model. The training results are as follows: Figure 4 As shown in the figure, the horizontal axis represents the training rounds, and the vertical axis represents the average Q value. The average Q value reaches a stable state when the rounds are close to 80. Since the time step of reinforcement learning is long, the Q value is normalized.

[0191] At this point, the model can basically prevent accidents caused by dangerous driving behaviors, basically avoid the loss of the value of the goods, and better control the delay time of the goods.

[0192] The above example is only an example of how to implement it, and does not represent all the possible implementations. For example, the reward function can include the impact of truck stopping on other vehicles, or more objectives. Of course, this will increase the difficulty of model convergence.

[0193] The data samples in the examples are simplified, but it should be understood that the information of the data samples should include all the information indicated, and of course more information that is helpful for analysis can be added.

[0194] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. An operation and maintenance method of an intelligent traffic monitoring system, characterized in that: The following steps are involved: Step 1, collect traffic monitoring data and freight system data; Step 2, constructing the first structured data based on the collected data. The first structured data includes two structural dimensions. The basic structural elements included in the first structural dimension are objects. The objects of the first structured data include: drivers, vehicles, goods, roads, and historical dangerous driving events; the basic structural elements included in the second structural dimension are time series units; Step 3, build a model, which includes a first hidden layer, a second hidden layer, a pre-training layer and a decoding layer. The first hidden layer is used to learn and identify the time series information in the second structural dimension to obtain the update status of all objects to the second hidden layer. The second hidden layer is used to learn and identify the correlation information related to traffic monitoring between objects in the first structural dimension to obtain the feature representation of all objects to the decoding layer. The decoding layer concatenates the feature representations of all objects and maps them to the policy space to output the truck control policy. Step 4, execute the truck control strategy obtained by inputting the traffic monitoring data collected at the current moment into the model; the truck control strategy is a matrix representation, the first row and the gth column of the matrix representation represent the location where the gth truck is stopped, the second row and the gth column represent the ID of the driver who is replaced after the stop, and the third row and the gth column represent the distribution center where the gth truck needs to be transferred after the stop; The model is trained by reinforcement learning, and the state of reinforcement learning is defined as traffic monitoring data and freight system data; the action of reinforcement learning is traffic management measures, and the reward function is: ； in, Represents the reward value obtained by executing the ath action in the bth state; represents an accident prevention reward, which is inversely proportional to the number of accidents or emergencies that occur; It represents the timeliness reward, which is inversely proportional to the time of the goods’ arrival delay; It represents the reward for the value of the goods, which is inversely proportional to the loss of the value of the goods during transportation; , , are the first, second, and third hyperparameters.

2. The operation and maintenance method of an intelligent traffic monitoring system according to claim 1, characterized in that: The data collected in Step 1 include: real-time truck onboard data, historical dangerous driving incident data, road condition information, and cargo information.

3. The operation and maintenance method of an intelligent traffic monitoring system according to claim 1, characterized in that: The method of constructing the first structured data includes: extracting a first structured data object; Extract specific features associated with the object, and divide the extracted specific features into multiple parts according to the collection period, and map one part to a time sequence unit. The order of the time sequence units is consistent with the time order of the collection period of the specific features; Establishing a mapping between the extracted objects and their associated specific features; For any two objects, if there is a correlation related to traffic monitoring between the associated specific features or there is a correlation related to traffic monitoring between the objects, an object relationship is established between the two objects; The first structured data objects include: Driver object: represents different truck drivers; Vehicle object: represents a truck; Cargo object: represents the goods being transported; Road object: represents a road or road section; Historical dangerous driving event object: represents historical dangerous driving events; Specific characteristics associated with the driver: driver personal information, driver physiological data, continuous driving time; Specific characteristics associated with the vehicle: load, speed, acceleration, vehicle type; Specific characteristics associated with the goods: type of goods, shelf life, unit value, unit transportation price; Specific features associated with roads: road monitoring images, road traffic, width, number of lanes, speed limit; Specific features associated with historical dangerous driving events: location, time, traffic conditions, vehicle, cargo in the vehicle, and description text of the result of the congestion; Relevance to traffic monitoring includes: The driver and the vehicle being driven; The vehicle and the cargo it carries; The vehicle and the road on which it is located; the vehicle’s association with historical dangerous driving incidents; Driver associations with historical risky driving events.

4. The operation and maintenance method of an intelligent traffic monitoring system according to claim 1, characterized in that: The first hidden layer contains multiple parallel branches. One parallel branch inputs the sequential unit of an object. Each parallel branch performs step-by-step calculations. The expression for the t-th step calculation of the r-th parallel branch is as follows: ； ； ； ； and They represent the activation vectors of the reset gate and update gate of the t-th step of the r-th parallel branch of the first hidden layer respectively; represents the candidate state generated in the tth step of the rth parallel branch of the first hidden layer; and They represent the update states of the t-th step and the t-1-th step of the r-th parallel branch of the first hidden layer respectively; J, K, M, F, P, and Q represent the first, second, third, fourth, fifth, and sixth conversion matrices, respectively; e, c, and v represent the first, second, and third deviation parameters, respectively; r∈G, G represents the set of all objects, t∈{1, 2, 3, …, n}, t=1 ; Represents the Sigmoid activation function; Represents the tanh activation function.

5. The operation and maintenance method of an intelligent traffic monitoring system according to claim 4, characterized in that: The second hidden layer contains multiple serial layers, and the expression of the l+1th layer is as follows: ； ； where i∈G, , represents a set of objects that have a relationship with the i-th object related to traffic monitoring; Representation Object In the second hidden layer Feature representation of the layer; Represents the object To Object Related information related to traffic monitoring; Represents an aggregate function; represents the update function; represents the information transfer function; l∈{0, 1, 2, 3, ..., L-1}, L represents the total number of the second hidden layer, when l=0, , , and They represent the updated states generated by the tth step of the i-th parallel branch and the j-th parallel branch of the first hidden layer respectively.

6. The operation and maintenance method of an intelligent traffic monitoring system according to claim 5, characterized in that: ,in Represents the weight parameter of the lth layer of the second hidden layer.

7. The operation and maintenance method of an intelligent traffic monitoring system according to claim 1, characterized in that: The decoding layer concatenates the feature representations of all objects in the Lth layer. The concatenation of the feature representations of the Lth layer is expressed as ,in It represents the concatenation of the feature representations of all objects at layer L. represents the feature representation of the i-th object in the L-th layer of the second hidden layer, The fully connected layer of the input decoding layer is mapped to the policy space and the truck control policy is output.

8. The operation and maintenance method of an intelligent traffic monitoring system according to claim 1, characterized in that: The output of the second hidden layer is connected to the pre-training layer, and the pre-training layer outputs the result indicating whether the driver's object has dangerous behavior; the model is pre-trained through the loss value of the output of the pre-training layer; and then the model is retrained through the loss of the output of the decoding layer.

9. The operation and maintenance method of an intelligent traffic monitoring system according to claim 1, characterized in that: If no accident or emergency occurs after the action is performed, In the event of an accident or emergency, ; If the goods arrive at the destination on time, ; If the shipment is delayed, , is the penalty coefficient of delay time, Indicates the delay time for all shipments, Indicates the scheduled transportation time for all shipments; If the goods are intact during transportation, If the goods are damaged in any way, , is the penalty coefficient of the damage degree, Indicates the loss of the value of all goods. Indicates the value of all goods.

10. A computer storage medium, characterized in that: It is used to store instructions that can be read by a computer, and when the instructions stored therein are read, an operation and maintenance method of an intelligent traffic monitoring system as described in any one of claims 1-9 can be executed.

Images