Deep learning-based traffic jam intelligent identification method and system

Abstract

The invention provides a traffic jam intelligent identification method and system based on deep learning. The method comprises the steps of obtaining a multi-source traffic recognition optimized data set by obtaining and optimizing a multi-source traffic recognition data set, then constructing traffic scene label feature data, recognizing dynamic traffic scene category feature data, evaluating a traffic jam initial evaluation index, analyzing a traffic jam recognition influence factor, and optimizing to obtain a traffic jam correction index. And finally, performing threshold value comparison with a preset dynamic traffic jam identification threshold value, judging whether traffic jam exists or not according to a threshold value comparison result, analyzing and identifying a jam cause, and generating an intelligent traffic jam identification report at the same time. According to the invention, multi-source heterogeneous data is collected, data fusion and feature extraction are completed through the deep learning model, congestion identification is realized by combining dynamic scene classification and adaptive threshold determination, and congestion cause tracing analysis is completed by using the deep learning classification model, so that traffic congestion intelligent identification based on deep learning is realized.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technology, and more specifically, to a method and system for intelligent traffic congestion identification based on deep learning. Background Technology

[0002] Traffic congestion impacts urban operation and development efficiency, as well as travel experience, during urbanization. Traditional traffic congestion identification technologies are heavily reliant on data, primarily sourced from ETC gantries, traffic cameras, and navigation devices, resulting in high deployment costs and susceptibility to environmental interference. Furthermore, they neglect the impact of non-motorized vehicle and pedestrian traffic data analysis on accurate traffic congestion identification. Currently, traditional technologies use fixed algorithms and single fixed thresholds for judgment, making them prone to misjudgments in extreme weather and temporary events, failing to accurately reflect real-time traffic conditions. Some traditional technologies employ conventional data fusion and identification algorithms, which struggle to effectively extract deep features from multi-source data, hindering congestion root cause analysis and providing precise decision support for traffic management. Therefore, there is an urgent need for an intelligent traffic congestion identification method based on deep learning technology, integrating multi-source heterogeneous data, and adaptable to dynamic scenarios.

[0003] Effective technical solutions are urgently needed to address the above problems. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for intelligent traffic congestion identification based on deep learning. It can collect multi-source heterogeneous data, complete data fusion and feature extraction through a deep learning model, and achieve congestion identification by combining dynamic scene classification and adaptive threshold determination. At the same time, it uses a deep learning classification model to trace the causes of congestion, thereby realizing intelligent traffic congestion identification based on deep learning.

[0005] Firstly, this application provides a deep learning-based intelligent traffic congestion identification method, including the following steps: Obtain the multi-source traffic recognition dataset corresponding to the preset traffic segment ID, and perform data preprocessing to obtain the multi-source traffic recognition optimized dataset; Traffic scene label feature data is constructed based on the multi-source traffic identification optimization dataset, and then input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data. The multi-source traffic identification optimization dataset is used to extract features and input into a preset traffic congestion identification model for processing to obtain a preliminary traffic congestion index. The multi-source traffic identification optimization dataset is analyzed and processed to obtain the traffic congestion identification influencing factors. The traffic congestion identification influencing factors are used to correct the initial traffic congestion index to obtain the traffic congestion correction index. The traffic congestion correction index is compared with a preset dynamic traffic congestion identification threshold. If the traffic congestion correction index is less than the preset dynamic traffic congestion identification threshold, the preset traffic segment is determined to be non-congested. If the traffic congestion correction index is greater than or equal to the preset dynamic traffic congestion identification threshold, then the preset traffic segment is determined to be congested, and the cause of congestion is identified. A traffic congestion intelligent identification report is generated based on the preset traffic segment ID and congestion causes.

[0006] Optionally, in the deep learning-based intelligent traffic congestion recognition method described in this application, the step of obtaining a multi-source traffic recognition dataset corresponding to a preset traffic segment ID and performing data preprocessing to obtain an optimized multi-source traffic recognition dataset includes: The traffic segments within the preset urban area are divided into grids, and grid IDs are assigned to obtain the preset traffic segment IDs; Obtain the multi-source traffic recognition dataset corresponding to the preset traffic segment ID, including motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road segment associated public service scenario operation data, and environmental collection data. Among them, motor vehicle traffic operation data includes average speed of motor vehicles, average distance between motor vehicles, and proportion of freight vehicles; non-motor vehicle traffic operation data includes average speed of non-motor vehicles and non-motor vehicle driving trajectory data; pedestrian traffic operation data includes pedestrian flow data and average pedestrian dwell time; municipal traffic operation data includes road water accumulation data, municipal manhole cover operation status data, and event feature data; road segment associated public service scenario operation data includes public service peak hours, vehicle entry and exit frequency, and average vehicle dwell time; and environmental collection data includes real-time rainfall and visibility. The motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road segment associated public service scenario operation data, and environmental collection data are subjected to spatiotemporal alignment and outlier cleaning preprocessing to obtain a multi-source traffic recognition optimization dataset.

[0007] Optionally, in the deep learning-based intelligent traffic congestion recognition method described in this application, the step of constructing traffic scene label feature data based on the multi-source traffic recognition optimization dataset and inputting it into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data includes: Traffic scene label feature data is constructed based on the event feature data, real-time rainfall and visibility, and data collection timestamps. The traffic scene label feature data is input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data.

[0008] Optionally, in the deep learning-based intelligent traffic congestion identification method described in this application, the step of extracting features from the multi-source traffic identification optimization dataset and inputting it into a preset traffic congestion identification model for processing to obtain a preliminary traffic congestion index includes: Spatial feature extraction, temporal feature extraction, and congestion-induced feature extraction are performed on the motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, and road segment-related public service scenario operation data to obtain spatial feature data, temporal feature data, and congestion-induced feature data. Based on the dynamic traffic scenario category feature data, query the preset traffic scenario category and feature weight value mapping table to obtain the weight values ​​corresponding to spatial feature data, temporal feature data and congestion-induced feature data, including spatial weight values, temporal weight values ​​and congestion-induced weight values. The spatial feature data, temporal feature data, and congestion-induced feature data are combined with the spatial weight value, temporal weight value, and congestion-induced weight value for data fusion processing to obtain preliminary traffic congestion assessment feature data. The initial traffic congestion assessment feature data is input into a preset traffic congestion identification model for processing to obtain the initial traffic congestion assessment index.

[0009] Optionally, in the deep learning-based intelligent traffic congestion identification method described in this application, the step of analyzing and processing the multi-source traffic identification optimization dataset to obtain traffic congestion identification influencing factors includes: The road waterlogging data is compared with the preset waterlogging warning value to obtain the road waterlogging exceeding the warning rate; The road waterlogging exceeding the warning rate is combined with the municipal manhole cover operation status data and event characteristic data, and a weighted sum is performed to obtain the municipal traffic operation impact factor. The real-time rainfall is compared with the preset historical average rainfall for the same period to obtain the rainfall exceedance rate; The visibility is compared with a preset visibility warning value to obtain the visibility insufficiency rate; The environmental impact factor is obtained by weighted summation of the above-average rainfall rate and the visibility deficiency rate. The municipal traffic operation impact factors, environmental impact factors, and the proportion of freight vehicles are normalized and weighted and summed to obtain the traffic congestion identification impact factors.

[0010] Optionally, in the deep learning-based intelligent traffic congestion identification method described in this application, the step of determining a preset traffic segment as congested and identifying the causes of congestion if the traffic congestion correction index is greater than or equal to a preset traffic congestion identification threshold includes: Based on the motor vehicle traffic operation data within a preset time period, data change features are extracted to obtain vehicle speed change feature data, operation trajectory change feature data, and traffic flow change data; The traffic congestion preliminary assessment feature data, vehicle speed change feature data, operation trajectory change feature data, and flow change data, as well as the municipal traffic operation influencing factors, environmental influencing factors, and the proportion of freight vehicles, are input into a preset traffic congestion cause identification model for analysis and processing to obtain the causes of congestion. The causes of congestion include traffic congestion, event-related congestion, or facility-related congestion.

[0011] Secondly, this application provides a deep learning-based intelligent traffic congestion recognition system, which includes: a memory and a processor. The memory includes a program for a deep learning-based intelligent traffic congestion recognition method. When the program for the deep learning-based intelligent traffic congestion recognition method is executed by the processor, it performs the following steps: Obtain the multi-source traffic recognition dataset corresponding to the preset traffic segment ID, and perform data preprocessing to obtain the multi-source traffic recognition optimized dataset; Traffic scene label feature data is constructed based on the multi-source traffic identification optimization dataset, and then input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data. The multi-source traffic identification optimization dataset is used to extract features and input into a preset traffic congestion identification model for processing to obtain a preliminary traffic congestion index. The multi-source traffic identification optimization dataset is analyzed and processed to obtain the traffic congestion identification influencing factors. The traffic congestion identification influencing factors are used to correct the initial traffic congestion index to obtain the traffic congestion correction index. The traffic congestion correction index is compared with a preset dynamic traffic congestion identification threshold. If the traffic congestion correction index is less than the preset dynamic traffic congestion identification threshold, the preset traffic segment is determined to be non-congested. If the traffic congestion correction index is greater than or equal to the preset dynamic traffic congestion identification threshold, then the preset traffic segment is determined to be congested, and the cause of congestion is identified. A traffic congestion intelligent identification report is generated based on the preset traffic segment ID and congestion causes.

[0012] Optionally, in the deep learning-based intelligent traffic congestion recognition system described in this application, the step of obtaining a multi-source traffic recognition dataset corresponding to a preset traffic segment ID and performing data preprocessing to obtain an optimized multi-source traffic recognition dataset includes: The traffic segments within the preset urban area are divided into grids, and grid IDs are assigned to obtain the preset traffic segment IDs; Obtain the multi-source traffic recognition dataset corresponding to the preset traffic segment ID, including motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road segment associated public service scenario operation data, and environmental collection data. Among them, motor vehicle traffic operation data includes average speed of motor vehicles, average distance between motor vehicles, and proportion of freight vehicles; non-motor vehicle traffic operation data includes average speed of non-motor vehicles and non-motor vehicle driving trajectory data; pedestrian traffic operation data includes pedestrian flow data and average pedestrian dwell time; municipal traffic operation data includes road water accumulation data, municipal manhole cover operation status data, and event feature data; road segment associated public service scenario operation data includes public service peak hours, vehicle entry and exit frequency, and average vehicle dwell time; and environmental collection data includes real-time rainfall and visibility. The motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road segment associated public service scenario operation data, and environmental collection data are subjected to spatiotemporal alignment and outlier cleaning preprocessing to obtain a multi-source traffic recognition optimization dataset.

[0013] Optionally, in the deep learning-based intelligent traffic congestion recognition system described in this application, the step of constructing traffic scene label feature data based on the multi-source traffic recognition optimization dataset and inputting it into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data includes: Traffic scene label feature data is constructed based on the event feature data, real-time rainfall and visibility, and data collection timestamps. The traffic scene label feature data is input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data.

[0014] Optionally, in the deep learning-based intelligent traffic congestion recognition system described in this application, the step of extracting features from the multi-source traffic recognition optimization dataset and inputting it into a preset traffic congestion recognition model for processing to obtain a preliminary traffic congestion index includes: Spatial feature extraction, temporal feature extraction, and congestion-induced feature extraction are performed on the motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, and road segment-related public service scenario operation data to obtain spatial feature data, temporal feature data, and congestion-induced feature data. Based on the dynamic traffic scenario category feature data, query the preset traffic scenario category and feature weight value mapping table to obtain the weight values ​​corresponding to spatial feature data, temporal feature data and congestion-induced feature data, including spatial weight values, temporal weight values ​​and congestion-induced weight values. The spatial feature data, temporal feature data, and congestion-induced feature data are combined with the spatial weight value, temporal weight value, and congestion-induced weight value for data fusion processing to obtain preliminary traffic congestion assessment feature data. The initial traffic congestion assessment feature data is input into a preset traffic congestion identification model for processing to obtain the initial traffic congestion assessment index.

[0015] As can be seen from the above, the traffic congestion intelligent identification method and system based on deep learning provided in this application collects multi-source heterogeneous data, completes data fusion and feature extraction through a deep learning model, and achieves congestion identification by combining dynamic scene classification and adaptive threshold judgment. At the same time, it uses a deep learning classification model to complete the source tracing of congestion causes, thereby realizing traffic congestion intelligent identification based on deep learning.

[0016] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing embodiments of this application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of a deep learning-based intelligent traffic congestion recognition method provided in an embodiment of this application; Figure 2 A flowchart illustrating the process of obtaining a multi-source traffic recognition optimization dataset using a deep learning-based intelligent traffic congestion recognition method provided in this application embodiment; Figure 3 This is a flowchart illustrating the process of obtaining dynamic traffic scene category feature data using a deep learning-based intelligent traffic congestion identification method provided in this application embodiment. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] Please refer to Figure 1 , Figure 1 This is a flowchart of a deep learning-based intelligent traffic congestion recognition method according to some embodiments of this application. This deep learning-based intelligent traffic congestion recognition method is used in terminal devices, such as computers and mobile phones. The deep learning-based intelligent traffic congestion recognition method includes the following steps: S11. Obtain the multi-source traffic recognition dataset corresponding to the preset traffic segment ID, and perform data preprocessing to obtain the multi-source traffic recognition optimized dataset. S121. Construct traffic scene label feature data based on the multi-source traffic identification optimization dataset, and input it into the preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data. S122. Extract features from the multi-source traffic identification optimization dataset and input it into a preset traffic congestion identification model for processing to obtain a preliminary traffic congestion index. S123. Analyze and process the multi-source traffic identification optimization dataset to obtain traffic congestion identification influencing factors; S13. Correct the initial traffic congestion index based on the traffic congestion identification influencing factors to obtain the traffic congestion correction index. S14. Compare the traffic congestion correction index with the preset dynamic traffic congestion identification threshold. S151. If the traffic congestion correction index is less than the preset dynamic traffic congestion identification threshold, the preset traffic segment is determined to be non-congested. S152. If the traffic congestion correction index is greater than or equal to the preset dynamic traffic congestion identification threshold, the preset traffic segment is determined to be congested, and the cause of congestion is identified. S16. Generate a traffic congestion intelligent identification report based on the preset traffic segment ID and congestion causes.

[0022] It should be noted that, in order to achieve accurate identification of traffic congestion in dynamic traffic scenarios, firstly, multi-source heterogeneous data is collected and preprocessed to obtain a multi-source traffic identification optimization dataset; then, the category feature data of dynamic traffic scenarios is determined, and data analysis and evaluation are performed to obtain the initial traffic congestion index and the traffic congestion identification impact factor. Based on the traffic congestion identification impact factor, the initial traffic congestion index is optimized to obtain the traffic congestion correction index. For example, if the initial traffic congestion index is a and the traffic congestion identification impact factor is y, then (1+y)×a is the traffic congestion correction index. Then, by comparing thresholds, it is determined whether there is traffic congestion. The preset dynamic traffic congestion identification threshold is determined by querying a pre-constructed preset dynamic traffic scene category and weight value relationship mapping table based on the determined dynamic traffic scene category feature data. The preset dynamic traffic scene category and weight value relationship mapping table is obtained by those skilled in the art based on the analysis of a large number of historical cases and can be dynamically adjusted. If there is traffic congestion, the causes of congestion are further identified. Finally, a traffic congestion intelligent identification report is generated based on the preset traffic segment IDs and congestion causes to provide data support for traffic management and decision-making.

[0023] Please refer to Figure 2 , Figure 2 This is a flowchart illustrating the process of obtaining a multi-source traffic identification optimization dataset using a deep learning-based intelligent traffic congestion identification method as described in some embodiments of this application. According to embodiments of the present invention, obtaining the multi-source traffic identification dataset corresponding to a preset traffic segment ID and performing data preprocessing to obtain the multi-source traffic identification optimization dataset includes: S21. Divide the traffic segments within the preset urban area into grids and assign grid IDs to obtain the preset traffic segment IDs; S22. Obtain the multi-source traffic recognition dataset corresponding to the preset traffic segment ID, including motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road segment associated public service scenario operation data, and environmental collection data. Among them, motor vehicle traffic operation data includes the average speed of motor vehicles, the average distance between motor vehicles, and the proportion of freight motor vehicles; non-motor vehicle traffic operation data includes the average speed of non-motor vehicles and non-motor vehicle driving trajectory data; pedestrian traffic operation data includes pedestrian flow data and the average pedestrian dwell time; municipal traffic operation data includes road water accumulation data, municipal manhole cover operation status data, and event feature data; road segment associated public service scenario operation data includes public service peak hours, vehicle entry and exit frequency, and average vehicle dwell time; and environmental collection data includes real-time rainfall and visibility. S23. Perform spatiotemporal alignment and outlier cleaning preprocessing on the motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road segment associated public service scenario operation data, and environmental collection data to obtain a multi-source traffic recognition optimization dataset.

[0024] It should be noted that, in order to overcome the limitations of traditional technologies that rely on a single data source and strong dependence on dedicated traffic equipment, the city's pre-defined traffic segments are first divided into 500×500 meter grids, each assigned a unique grid ID to obtain the pre-defined traffic segment IDs. Then, data collection covers multiple dimensions, including different traffic participants (such as motor vehicles, non-motor vehicles, and pedestrians), ordinary civilian and commercial facilities (such as streetlights, shop fronts, and municipal manhole covers), public services (such as schools and hospitals), and the data collection environment. This approach breaks away from focusing solely on motor vehicle data. Simultaneously, by utilizing civilian facilities, it reduces equipment deployment costs and effectively uncovers the causes of traffic scenarios and traffic congestion. The relationship is highly practical for urban mixed traffic scenarios. Finally, the collected multi-source data is spatiotemporally aligned by using a "500m×500m grid coding and 1-minute time slice". Invalid data is filtered by combining the 3σ principle and a CNN anomaly detection model. Then, an attention interpolation model is used to fill the gaps to obtain a multi-source traffic recognition optimization dataset. Among them, the road segment associated public service scenario operation data refers to the collected data of public service scenarios within the preset range of the road segment. The operation status of municipal manhole covers includes normal or abnormal. The operation status data of municipal manhole covers is represented by different identifiers. Event features, such as construction, are represented by different identifiers.

[0025] Please refer to Figure 3 , Figure 3This is a flowchart illustrating the process of obtaining dynamic traffic scene category feature data using a deep learning-based intelligent traffic congestion recognition method according to some embodiments of this application. According to an embodiment of the present invention, the step of constructing traffic scene label feature data based on the multi-source traffic recognition optimization dataset and inputting it into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data includes: S31. Construct traffic scene label feature data based on the event feature data, the real-time rainfall and visibility, and the data collection timestamp; S32. Input the traffic scene label feature data into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data.

[0026] It should be noted that, in order to adapt to congestion identification in dynamic traffic scenarios, firstly, traffic scenario label feature data covering "time, environment, and event" is constructed based on the collected multi-source traffic identification optimization dataset. Specifically, this is constructed by combining event feature data, real-time rainfall, and visibility with data collection timestamps. Then, the constructed traffic scenario label feature data is input into a preset traffic scenario category classification model for analysis and processing to obtain dynamic traffic scenario category feature data. Dynamic traffic scenario categories include "weekday morning rush hour, normal weather, and no special events." The dynamic traffic scenario category feature data is represented by a unique identifier. The preset traffic scenario category classification model is obtained by training the initial Transformer scenario classification model with a large amount of historical sample traffic scenario label feature data and the corresponding dynamic traffic scenario category feature data.

[0027] According to an embodiment of the present invention, the step of extracting features from the multi-source traffic identification optimization dataset and inputting it into a preset traffic congestion identification model for processing to obtain a preliminary traffic congestion index includes: Spatial feature extraction, temporal feature extraction, and congestion-induced feature extraction are performed on the motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, and road segment-related public service scenario operation data to obtain spatial feature data, temporal feature data, and congestion-induced feature data. Based on the dynamic traffic scenario category feature data, query the preset traffic scenario category and feature weight value mapping table to obtain the weight values ​​corresponding to spatial feature data, temporal feature data and congestion-induced feature data, including spatial weight values, temporal weight values ​​and congestion-induced weight values. The spatial feature data, temporal feature data, and congestion-induced feature data are combined with the spatial weight value, temporal weight value, and congestion-induced weight value for data fusion processing to obtain preliminary traffic congestion assessment feature data. The initial traffic congestion assessment feature data is input into a preset traffic congestion identification model for processing to obtain the initial traffic congestion assessment index.

[0028] It should be noted that, in order to quantify traffic congestion, firstly, pre-defined dedicated deep learning sub-models are used to extract features from different types of collected data. For example, for motor vehicle traffic data collected by millimeter-wave radar, a pre-defined 3D-CNN model is used to extract spatial feature data, including speed and spacing; for non-motorized vehicle traffic data and pedestrian traffic data, a pre-defined GR model is used to extract temporal feature data; for road segment-related public service scenario operation data, a pre-defined MLP model is used to extract congestion-inducing feature data, including peak hours and vehicle entry / exit frequency features. Then, based on the determined dynamic traffic scenario category feature data, a pre-defined traffic scenario category and feature weight value mapping table is queried to determine the corresponding weight value. The extracted spatial features are then weighted according to the weight value. The traffic congestion preliminary assessment feature data is obtained by weighted fusion of the feature data, time-series feature data, and congestion-induced feature data. Finally, the traffic congestion preliminary assessment feature data is input into the preset traffic congestion identification model for processing to obtain the traffic congestion preliminary assessment index. The preset traffic scenario category and feature weight value mapping table is constructed by those skilled in the art based on historical data analysis and can be dynamically adjusted. The traffic congestion preliminary assessment feature data is represented by feature vectors. The preset traffic congestion identification model is obtained by training with a large number of historical samples of traffic congestion preliminary assessment feature data and corresponding traffic congestion preliminary assessment indices. The preset 3D-CNN model, preset GR model, and preset MLP model are obtained by those skilled in the art through pre-training based on a large number of historical sample cases.

[0029] According to an embodiment of the present invention, the step of analyzing and processing the multi-source traffic identification optimization dataset to obtain traffic congestion identification influencing factors includes: The road waterlogging data is compared with the preset waterlogging warning value to obtain the road waterlogging exceeding the warning rate; The road waterlogging exceeding the warning rate is combined with the municipal manhole cover operation status data and event characteristic data, and a weighted sum is performed to obtain the municipal traffic operation impact factor. The real-time rainfall is compared with the preset historical average rainfall for the same period to obtain the rainfall exceedance rate; The visibility is compared with a preset visibility warning value to obtain the visibility insufficiency rate; The environmental impact factor is obtained by weighted summation of the above-average rainfall rate and the visibility deficiency rate. The municipal traffic operation impact factors, environmental impact factors, and the proportion of freight vehicles are normalized and weighted and summed to obtain the traffic congestion identification impact factors.

[0030] It should be noted that, in order to adapt to dynamically changing traffic scenarios and improve the accuracy of traffic congestion identification, analysis is conducted based on municipal traffic operation data and environmental data collection to assess and obtain influencing factors for traffic congestion identification. The initial traffic congestion assessment index, obtained from the analysis of motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, and road segment-related public service scenario operation data, is then optimized and corrected. Specifically, the road waterlogging exceeding warning rate refers to the ratio of the difference between the road waterlogging data and the preset waterlogging warning value to the preset waterlogging warning value. If it is positive, it is recorded normally; otherwise, it is recorded as 0, meaning the road waterlogging data is less than or equal to the preset waterlogging warning value. Municipal manhole cover operation status data is used to indicate whether the manhole covers are normal or abnormal, and event characteristic data is used to indicate whether temporary events (such as road construction) exist. Municipal manhole cover operation status data and event characteristic data are identified using different identifiers, such as municipal... The impact factor of municipal traffic operation is determined by weighted summation of the following: 0 for normal manhole cover conditions and 1 for abnormal conditions. The impact factor is determined by weighted summation of the road waterlogging exceeding warning rate, municipal manhole cover operation status data, and event characteristic data. The rainfall exceeding average rate is the ratio of the difference between real-time rainfall and the preset historical average rainfall for the same period to the preset historical average rainfall for the same period. If positive, it is recorded normally; otherwise, it is recorded as 0, indicating that the real-time rainfall is less than or equal to the preset historical average rainfall for the same period. The visibility deficiency rate is the ratio of the absolute value of the difference between visibility and the preset visibility warning value to the preset visibility warning value. If visibility is greater than the preset visibility warning value, it is recorded as 0. The environmental impact factor is determined by weighted summation of the rainfall exceeding average rate and the visibility deficiency rate. Finally, the impact factors of municipal traffic operation, the environmental impact factor, and the proportion of freight vehicles are normalized and mapped to the [0, 1] interval, then weighted summation is performed to finally determine the traffic congestion identification impact factor.

[0031] According to an embodiment of the present invention, if the traffic congestion correction index is greater than or equal to a preset traffic congestion identification threshold, then determining that a preset traffic segment is congested and identifying the cause of congestion includes: Based on the motor vehicle traffic operation data within a preset time period, data change features are extracted to obtain vehicle speed change feature data, operation trajectory change feature data, and traffic flow change data; The traffic congestion preliminary assessment feature data, vehicle speed change feature data, operation trajectory change feature data, and flow change data, as well as the municipal traffic operation influencing factors, environmental influencing factors, and the proportion of freight vehicles, are input into a preset traffic congestion cause identification model for analysis and processing to obtain the causes of congestion. The causes of congestion include traffic congestion, event-related congestion, or facility-related congestion.

[0032] It should be noted that the deep learning classification model based on multi-task learning uses a large amount of historical samples to fuse preliminary traffic congestion assessment feature data. It also uses vehicle speed change feature data, trajectory change feature data, and flow change data reflecting congestion time-period data changes, combined with municipal traffic operation influencing factors, environmental influencing factors, the proportion of freight vehicles, and corresponding congestion causes as inputs to train a pre-defined traffic congestion cause identification model. This model identifies three types of congestion causes. Then, based on real-time data, it performs congestion cause tracing analysis to improve data support for traffic management and decision-making. Specifically, flow congestion refers to congestion caused by excessive vehicle flow, event congestion refers to traffic congestion caused by temporary events, and facility anomaly congestion refers to traffic congestion caused by malfunctions in municipal facilities.

[0033] It is worth mentioning that, according to embodiments of the present invention, it further includes: If there is traffic congestion, then obtain vehicle driving trajectory data, public service scenario feature data, and road network structure feature data; The vehicle driving trajectory data, public service scenario feature data and road network structure feature data are input into a preset traffic congestion source analysis model for analysis and processing to obtain congestion traffic source feature data. Obtain traffic flow data for a preset time period, and construct traffic flow time-series feature data based on the traffic flow data; The traffic flow time series characteristic data and the preset historical traffic flow time series characteristic data are input into the preset traffic peak characteristic analysis model for analysis and processing to obtain traffic peak parameter data. The congestion traffic source and peak traffic parameter data are sent to the operation and maintenance terminal for display.

[0034] It should be noted that after traffic congestion is identified, to further accurately determine the source of congestion and the characteristics of peak traffic flow, the following steps are taken: First, acquire corresponding vehicle driving trajectory data, public service scenario characteristic data, and road network structure characteristic data. Vehicle driving trajectory data includes data collection timestamps, vehicle location coordinates, and instantaneous vehicle speed. Public service scenario characteristic data refers to the number of permanent residents, peak travel rate, and travel direction within a preset range around the road segment. Road network structure characteristic data refers to abstracting the urban road network into a node-edge graph structure, where nodes are intersections or road segments, and edges represent the connections between nodes. Then, analyze and process the data using a preset traffic congestion source analysis model to obtain congestion traffic source characteristic data, such as 80% of congested vehicles originating from a residential area around the road segment, and 20% from other areas. Simultaneously, collect traffic flow data for a preset time period (e.g., the past 4 hours), break it down into 5-minute time granularities, and calculate the number of vehicles on a specific road segment at each time granularity, the average number of vehicles and the vehicle growth rate (i.e., the subsequent...) of the three time granularities preceding the current time. The traffic flow time-series characteristic data is constructed by comparing the difference between the number of vehicles at the current time granularity and the number of vehicles at the previous time granularity with the number of vehicles at the previous time granularity. This data is then combined with preset historical traffic flow time-series characteristic data and input into a preset traffic flow peak characteristic analysis model for analysis and processing. This yields traffic flow peak parameter data, including the peak traffic start time, peak traffic duration, and peak deviation rate (i.e., the absolute value of the difference between the current peak traffic and the historical peak traffic and the ratio of the historical peak traffic). Finally, the obtained congestion traffic source and traffic flow peak parameter data are sent to the operation and maintenance terminal for display, providing accurate data support for traffic management. The preset traffic congestion source analysis model is trained by acquiring a large amount of historical sample vehicle driving trajectory data, public service scenario characteristic data, road network structure characteristic data, and corresponding congestion traffic source characteristic data. The preset traffic flow peak characteristic analysis model is trained by acquiring a large amount of historical sample traffic flow time-series characteristic data, preset historical traffic flow time-series characteristic data, and corresponding traffic flow peak parameter data.

[0035] It is worth mentioning that, according to embodiments of the present invention, it further includes: If the congestion is due to an incident, obtain traffic monitoring video data for the affected road segment; The traffic monitoring video data of the road section is input into a preset target detection model for analysis and processing to obtain the location coordinates of the congestion event and the event type feature data. Based on the location coordinates of the congestion event and the vehicle trajectory data of the preset continuous video frames, the data of the congestion impact range is obtained by labeling them using a preset traffic congestion impact semantic segmentation model. The location coordinates of the congestion event, the event type characteristic data, and the congestion impact range data are sent to the operation and maintenance terminal for display.

[0036] It should be noted that, to further determine the location and impact range of the congestion, and to provide accurate data support for traffic management, the following steps were taken: First, traffic monitoring video data and vehicle trajectory data of the congested road segment were collected. Adaptive image enhancement processing was performed on the traffic monitoring video data, and Kalman filtering was used to complete short-term missing trajectory data for the vehicle trajectory data. The traffic monitoring video data was then input into a pre-defined target detection model for analysis and processing to obtain the location coordinates and event type feature data of the congestion event. For example, if the event type is a rear-end collision, the location coordinates of the congestion event are 106.05°E. At 38.02°N, the location coordinates of congestion events and vehicle trajectory data of preset continuous video frames (e.g., 5 consecutive frames) are labeled using a preset traffic congestion impact semantic segmentation model to obtain congestion impact range data, such as an impact range of 500 meters behind. Finally, the obtained congestion event location coordinates, event type feature data, and congestion impact range data are sent to the operation and maintenance terminal for display. The preset target detection model is trained by acquiring a large number of historical samples of road segment traffic monitoring video data and corresponding congestion event location coordinates and event type feature data. The preset traffic congestion impact semantic segmentation model is trained by acquiring a large number of historical samples of congestion event location coordinates, vehicle trajectory data, and corresponding congestion impact range data.

[0037] This invention also discloses a deep learning-based intelligent traffic congestion recognition system, including a memory and a processor. The memory includes a deep learning-based intelligent traffic congestion recognition method program, which, when executed by the processor, performs the following steps: Obtain the multi-source traffic recognition dataset corresponding to the preset traffic segment ID, and perform data preprocessing to obtain the multi-source traffic recognition optimized dataset; Traffic scene label feature data is constructed based on the multi-source traffic identification optimization dataset, and then input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data. The multi-source traffic identification optimization dataset is used to extract features and input into a preset traffic congestion identification model for processing to obtain a preliminary traffic congestion index. The multi-source traffic identification optimization dataset is analyzed and processed to obtain the traffic congestion identification influencing factors. The traffic congestion identification influencing factors are used to correct the initial traffic congestion index to obtain the traffic congestion correction index. The traffic congestion correction index is compared with a preset dynamic traffic congestion identification threshold. If the traffic congestion correction index is less than the preset dynamic traffic congestion identification threshold, the preset traffic segment is determined to be non-congested. If the traffic congestion correction index is greater than or equal to the preset dynamic traffic congestion identification threshold, then the preset traffic segment is determined to be congested, and the cause of congestion is identified. A traffic congestion intelligent identification report is generated based on the preset traffic segment ID and congestion causes.

[0038] It should be noted that, in order to achieve accurate identification of traffic congestion in dynamic traffic scenarios, firstly, multi-source heterogeneous data is collected and preprocessed to obtain a multi-source traffic identification optimization dataset; then, the category feature data of dynamic traffic scenarios is determined, and data analysis and evaluation are performed to obtain the initial traffic congestion index and the traffic congestion identification impact factor. Based on the traffic congestion identification impact factor, the initial traffic congestion index is optimized to obtain the traffic congestion correction index. For example, if the initial traffic congestion index is a and the traffic congestion identification impact factor is y, then (1+y)×a is the traffic congestion correction index. Then, by comparing thresholds, it is determined whether there is traffic congestion. The preset dynamic traffic congestion identification threshold is determined by querying a pre-constructed preset dynamic traffic scene category and weight value relationship mapping table based on the determined dynamic traffic scene category feature data. The preset dynamic traffic scene category and weight value relationship mapping table is obtained by those skilled in the art based on the analysis of a large number of historical cases and can be dynamically adjusted. If there is traffic congestion, the causes of congestion are further identified. Finally, a traffic congestion intelligent identification report is generated based on the preset traffic segment IDs and congestion causes to provide data support for traffic management and decision-making.

[0039] According to an embodiment of the present invention, the step of obtaining a multi-source traffic identification dataset corresponding to a preset traffic segment ID and performing data preprocessing to obtain an optimized multi-source traffic identification dataset includes: The traffic segments within the preset urban area are divided into grids, and grid IDs are assigned to obtain the preset traffic segment IDs; Obtain the multi-source traffic recognition dataset corresponding to the preset traffic segment ID, including motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road segment associated public service scenario operation data, and environmental collection data. Among them, motor vehicle traffic operation data includes average speed of motor vehicles, average distance between motor vehicles, and proportion of freight vehicles; non-motor vehicle traffic operation data includes average speed of non-motor vehicles and non-motor vehicle driving trajectory data; pedestrian traffic operation data includes pedestrian flow data and average pedestrian dwell time; municipal traffic operation data includes road water accumulation data, municipal manhole cover operation status data, and event feature data; road segment associated public service scenario operation data includes public service peak hours, vehicle entry and exit frequency, and average vehicle dwell time; and environmental collection data includes real-time rainfall and visibility. The motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road segment associated public service scenario operation data, and environmental collection data are subjected to spatiotemporal alignment and outlier cleaning preprocessing to obtain a multi-source traffic recognition optimization dataset.

[0040] It should be noted that, in order to overcome the limitations of traditional technologies that rely on a single data source and strong dependence on dedicated traffic equipment, the city's pre-defined traffic segments are first divided into 500×500 meter grids, each assigned a unique grid ID to obtain the pre-defined traffic segment IDs. Then, data collection covers multiple dimensions, including different traffic participants (such as motor vehicles, non-motor vehicles, and pedestrians), ordinary civilian and commercial facilities (such as streetlights, shop fronts, and municipal manhole covers), public services (such as schools and hospitals), and the data collection environment. This approach breaks away from focusing solely on motor vehicle data. Simultaneously, by utilizing civilian facilities, it reduces equipment deployment costs and effectively uncovers the causes of traffic scenarios and traffic congestion. The relationship is highly practical for urban mixed traffic scenarios. Finally, the collected multi-source data is spatiotemporally aligned by using a "500m×500m grid coding and 1-minute time slice". Invalid data is filtered by combining the 3σ principle and a CNN anomaly detection model. Then, an attention interpolation model is used to fill the gaps to obtain a multi-source traffic recognition optimization dataset. Among them, the road segment associated public service scenario operation data refers to the collected data of public service scenarios within the preset range of the road segment. The operation status of municipal manhole covers includes normal or abnormal. The operation status data of municipal manhole covers is represented by different identifiers. Event features, such as construction, are represented by different identifiers.

[0041] According to an embodiment of the present invention, the step of constructing traffic scene label feature data based on the multi-source traffic recognition optimization dataset and inputting it into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data includes: Traffic scene label feature data is constructed based on the event feature data, real-time rainfall and visibility, and data collection timestamps. The traffic scene label feature data is input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data.

[0042] It should be noted that, in order to adapt to congestion identification in dynamic traffic scenarios, firstly, traffic scenario label feature data covering "time, environment, and event" is constructed based on the collected multi-source traffic identification optimization dataset. Specifically, this is constructed by combining event feature data, real-time rainfall, and visibility with data collection timestamps. Then, the constructed traffic scenario label feature data is input into a preset traffic scenario category classification model for analysis and processing to obtain dynamic traffic scenario category feature data. Dynamic traffic scenario categories include "weekday morning rush hour, normal weather, and no special events." The dynamic traffic scenario category feature data is represented by a unique identifier. The preset traffic scenario category classification model is obtained by training the initial Transformer scenario classification model with a large amount of historical sample traffic scenario label feature data and the corresponding dynamic traffic scenario category feature data.

[0043] According to an embodiment of the present invention, the step of extracting features from the multi-source traffic identification optimization dataset and inputting it into a preset traffic congestion identification model for processing to obtain a preliminary traffic congestion index includes: Spatial feature extraction, temporal feature extraction, and congestion-induced feature extraction are performed on the motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, and road segment-related public service scenario operation data to obtain spatial feature data, temporal feature data, and congestion-induced feature data. Based on the dynamic traffic scenario category feature data, query the preset traffic scenario category and feature weight value mapping table to obtain the weight values ​​corresponding to spatial feature data, temporal feature data and congestion-induced feature data, including spatial weight values, temporal weight values ​​and congestion-induced weight values. The spatial feature data, temporal feature data, and congestion-induced feature data are combined with the spatial weight value, temporal weight value, and congestion-induced weight value for data fusion processing to obtain preliminary traffic congestion assessment feature data. The initial traffic congestion assessment feature data is input into a preset traffic congestion identification model for processing to obtain the initial traffic congestion assessment index.

[0044] It should be noted that, in order to quantify traffic congestion, firstly, pre-defined dedicated deep learning sub-models are used to extract features from different types of collected data. For example, for motor vehicle traffic data collected by millimeter-wave radar, a pre-defined 3D-CNN model is used to extract spatial feature data, including speed and spacing; for non-motorized vehicle traffic data and pedestrian traffic data, a pre-defined GR model is used to extract temporal feature data; for road segment-related public service scenario operation data, a pre-defined MLP model is used to extract congestion-inducing feature data, including peak hours and vehicle entry / exit frequency features. Then, based on the determined dynamic traffic scenario category feature data, a pre-defined traffic scenario category and feature weight value mapping table is queried to determine the corresponding weight value. The extracted spatial features are then weighted according to the weight value. The traffic congestion preliminary assessment feature data is obtained by weighted fusion of the feature data, time-series feature data, and congestion-induced feature data. Finally, the traffic congestion preliminary assessment feature data is input into the preset traffic congestion identification model for processing to obtain the traffic congestion preliminary assessment index. The preset traffic scenario category and feature weight value mapping table is constructed by those skilled in the art based on historical data analysis and can be dynamically adjusted. The traffic congestion preliminary assessment feature data is represented by feature vectors. The preset traffic congestion identification model is obtained by training with a large number of historical samples of traffic congestion preliminary assessment feature data and corresponding traffic congestion preliminary assessment indices. The preset 3D-CNN model, preset GR model, and preset MLP model are obtained by those skilled in the art through pre-training based on a large number of historical sample cases.

[0045] According to an embodiment of the present invention, the step of analyzing and processing the multi-source traffic identification optimization dataset to obtain traffic congestion identification influencing factors includes: The road waterlogging data is compared with the preset waterlogging warning value to obtain the road waterlogging exceeding the warning rate; The road waterlogging exceeding the warning rate is combined with the municipal manhole cover operation status data and event characteristic data, and a weighted sum is performed to obtain the municipal traffic operation impact factor. The real-time rainfall is compared with the preset historical average rainfall for the same period to obtain the rainfall exceedance rate; The visibility is compared with a preset visibility warning value to obtain the visibility insufficiency rate; The environmental impact factor is obtained by weighted summation of the above-average rainfall rate and the visibility deficiency rate. The municipal traffic operation impact factors, environmental impact factors, and the proportion of freight vehicles are normalized and weighted and summed to obtain the traffic congestion identification impact factors.

[0046] It should be noted that, in order to adapt to dynamically changing traffic scenarios and improve the accuracy of traffic congestion identification, analysis is conducted based on municipal traffic operation data and environmental data collection to assess and obtain influencing factors for traffic congestion identification. The initial traffic congestion assessment index, obtained from the analysis of motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, and road segment-related public service scenario operation data, is then optimized and corrected. Specifically, the road waterlogging exceeding warning rate refers to the ratio of the difference between the road waterlogging data and the preset waterlogging warning value to the preset waterlogging warning value. If it is positive, it is recorded normally; otherwise, it is recorded as 0, meaning the road waterlogging data is less than or equal to the preset waterlogging warning value. Municipal manhole cover operation status data is used to indicate whether the manhole covers are normal or abnormal, and event characteristic data is used to indicate whether temporary events (such as road construction) exist. Municipal manhole cover operation status data and event characteristic data are identified using different identifiers, such as municipal... The impact factor of municipal traffic operation is determined by weighted summation of the following: 0 for normal manhole cover conditions and 1 for abnormal conditions. The impact factor is determined by weighted summation of the road waterlogging exceeding warning rate, municipal manhole cover operation status data, and event characteristic data. The rainfall exceeding average rate is the ratio of the difference between real-time rainfall and the preset historical average rainfall for the same period to the preset historical average rainfall for the same period. If positive, it is recorded normally; otherwise, it is recorded as 0, indicating that the real-time rainfall is less than or equal to the preset historical average rainfall for the same period. The visibility deficiency rate is the ratio of the absolute value of the difference between visibility and the preset visibility warning value to the preset visibility warning value. If visibility is greater than the preset visibility warning value, it is recorded as 0. The environmental impact factor is determined by weighted summation of the rainfall exceeding average rate and the visibility deficiency rate. Finally, the impact factors of municipal traffic operation, the environmental impact factor, and the proportion of freight vehicles are normalized and mapped to the [0, 1] interval, then weighted summation is performed to finally determine the traffic congestion identification impact factor.

[0047] According to an embodiment of the present invention, if the traffic congestion correction index is greater than or equal to a preset traffic congestion identification threshold, then determining that a preset traffic segment is congested and identifying the cause of congestion includes: Based on the motor vehicle traffic operation data within a preset time period, data change features are extracted to obtain vehicle speed change feature data, operation trajectory change feature data, and traffic flow change data; The traffic congestion preliminary assessment feature data, vehicle speed change feature data, operation trajectory change feature data, and flow change data, as well as the municipal traffic operation influencing factors, environmental influencing factors, and the proportion of freight vehicles, are input into a preset traffic congestion cause identification model for analysis and processing to obtain the causes of congestion. The causes of congestion include traffic congestion, event-related congestion, or facility-related congestion.

[0048] It should be noted that the deep learning classification model based on multi-task learning uses a large amount of historical samples to fuse preliminary traffic congestion assessment feature data. It also uses vehicle speed change feature data, trajectory change feature data, and flow change data reflecting congestion time-period data changes, combined with municipal traffic operation influencing factors, environmental influencing factors, the proportion of freight vehicles, and corresponding congestion causes as inputs to train a pre-defined traffic congestion cause identification model. This model identifies three types of congestion causes. Then, based on real-time data, it performs congestion cause tracing analysis to improve data support for traffic management and decision-making. Specifically, flow congestion refers to congestion caused by excessive vehicle flow, event congestion refers to traffic congestion caused by temporary events, and facility anomaly congestion refers to traffic congestion caused by malfunctions in municipal facilities.

[0049] It is worth mentioning that, according to embodiments of the present invention, it further includes: If there is traffic congestion, then obtain vehicle driving trajectory data, public service scenario feature data, and road network structure feature data; The vehicle driving trajectory data, public service scenario feature data and road network structure feature data are input into a preset traffic congestion source analysis model for analysis and processing to obtain congestion traffic source feature data. Obtain traffic flow data for a preset time period, and construct traffic flow time-series feature data based on the traffic flow data; The traffic flow time series characteristic data and the preset historical traffic flow time series characteristic data are input into the preset traffic peak characteristic analysis model for analysis and processing to obtain traffic peak parameter data. The congestion traffic source and peak traffic parameter data are sent to the operation and maintenance terminal for display.

[0050] It should be noted that after traffic congestion is identified, to further accurately determine the source of congestion and the characteristics of peak traffic flow, the following steps are taken: First, acquire corresponding vehicle driving trajectory data, public service scenario characteristic data, and road network structure characteristic data. Vehicle driving trajectory data includes data collection timestamps, vehicle location coordinates, and instantaneous vehicle speed. Public service scenario characteristic data refers to the number of permanent residents, peak travel rate, and travel direction within a preset range around the road segment. Road network structure characteristic data refers to abstracting the urban road network into a node-edge graph structure, where nodes are intersections or road segments, and edges represent the connections between nodes. Then, analyze and process the data using a preset traffic congestion source analysis model to obtain congestion traffic source characteristic data, such as 80% of congested vehicles originating from a residential area around the road segment, and 20% from other areas. Simultaneously, collect traffic flow data for a preset time period (e.g., the past 4 hours), break it down into 5-minute time granularities, and calculate the number of vehicles on a specific road segment at each time granularity, the average number of vehicles and the vehicle growth rate (i.e., the subsequent...) of the three time granularities preceding the current time. The traffic flow time-series characteristic data is constructed by comparing the difference between the number of vehicles at the current time granularity and the number of vehicles at the previous time granularity with the number of vehicles at the previous time granularity. This data is then combined with preset historical traffic flow time-series characteristic data and input into a preset traffic flow peak characteristic analysis model for analysis and processing. This yields traffic flow peak parameter data, including the peak traffic start time, peak traffic duration, and peak deviation rate (i.e., the absolute value of the difference between the current peak traffic and the historical peak traffic and the ratio of the historical peak traffic). Finally, the obtained congestion traffic source and traffic flow peak parameter data are sent to the operation and maintenance terminal for display, providing accurate data support for traffic management. The preset traffic congestion source analysis model is trained by acquiring a large amount of historical sample vehicle driving trajectory data, public service scenario characteristic data, road network structure characteristic data, and corresponding congestion traffic source characteristic data. The preset traffic flow peak characteristic analysis model is trained by acquiring a large amount of historical sample traffic flow time-series characteristic data, preset historical traffic flow time-series characteristic data, and corresponding traffic flow peak parameter data.

[0051] It is worth mentioning that, according to embodiments of the present invention, it further includes: If the congestion is due to an incident, obtain traffic monitoring video data for the affected road segment; The traffic monitoring video data of the road section is input into a preset target detection model for analysis and processing to obtain the location coordinates of the congestion event and the event type feature data. Based on the location coordinates of the congestion event and the vehicle trajectory data of the preset continuous video frames, the data of the congestion impact range is obtained by labeling them using a preset traffic congestion impact semantic segmentation model. The location coordinates of the congestion event, the event type characteristic data, and the congestion impact range data are sent to the operation and maintenance terminal for display.

[0052] It should be noted that, to further determine the location and impact range of the congestion, and to provide accurate data support for traffic management, the following steps were taken: First, traffic monitoring video data and vehicle trajectory data of the congested road segment were collected. Adaptive image enhancement processing was performed on the traffic monitoring video data, and Kalman filtering was used to complete short-term missing trajectory data for the vehicle trajectory data. The traffic monitoring video data was then input into a pre-defined target detection model for analysis and processing to obtain the location coordinates and event type feature data of the congestion event. For example, if the event type is a rear-end collision, the location coordinates of the congestion event are 106.05°E. At 38.02°N, the location coordinates of congestion events and vehicle trajectory data of preset continuous video frames (e.g., 5 consecutive frames) are labeled using a preset traffic congestion impact semantic segmentation model to obtain congestion impact range data, such as an impact range of 500 meters behind. Finally, the obtained congestion event location coordinates, event type feature data, and congestion impact range data are sent to the operation and maintenance terminal for display. The preset target detection model is trained by acquiring a large number of historical samples of road segment traffic monitoring video data and corresponding congestion event location coordinates and event type feature data. The preset traffic congestion impact semantic segmentation model is trained by acquiring a large number of historical samples of congestion event location coordinates, vehicle trajectory data, and corresponding congestion impact range data.

[0053] The present invention discloses a deep learning-based intelligent traffic congestion identification method and system. By collecting multi-source heterogeneous data, the system completes data fusion and feature extraction through a deep learning model, and combines dynamic scene classification and adaptive threshold determination to achieve congestion identification. At the same time, the system uses a deep learning classification model to trace the causes of congestion, thereby realizing intelligent traffic congestion identification based on deep learning.

[0054] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0055] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0056] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0057] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0058] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

Claims

1. A deep learning-based intelligent traffic congestion identification method, characterized in that, Includes the following steps: Obtain the multi-source traffic recognition dataset corresponding to the preset traffic segment ID, and perform data preprocessing to obtain the multi-source traffic recognition optimized dataset; Traffic scene label feature data is constructed based on the multi-source traffic identification optimization dataset and input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data. The multi-source traffic identification optimization dataset is used to extract features and input into a preset traffic congestion identification model for processing to obtain a preliminary traffic congestion index. The multi-source traffic identification optimization dataset is analyzed and processed to obtain the traffic congestion identification influencing factors. The traffic congestion identification influencing factors are used to correct the initial traffic congestion index to obtain the traffic congestion correction index. The traffic congestion correction index is compared with a preset dynamic traffic congestion identification threshold. If the traffic congestion correction index is less than the preset dynamic traffic congestion identification threshold, the preset traffic segment is determined to be non-congested. If the traffic congestion correction index is greater than or equal to the preset dynamic traffic congestion identification threshold, then the preset traffic segment is determined to be congested, and the cause of congestion is identified. A traffic congestion intelligent identification report is generated based on the preset traffic segment ID and congestion causes.

2. The traffic congestion intelligent identification method based on deep learning according to claim 1, characterized in that, The step of obtaining the multi-source traffic recognition dataset corresponding to the preset traffic segment ID and performing data preprocessing to obtain the multi-source traffic recognition optimized dataset includes: The traffic segments within the preset urban area are divided into grids, and grid IDs are assigned to obtain the preset traffic segment IDs; Obtain the multi-source traffic recognition dataset corresponding to the preset traffic segment ID, including motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road segment associated public service scenario operation data, and environmental collection data. Among them, motor vehicle traffic operation data includes average speed of motor vehicles, average distance between motor vehicles, and proportion of freight vehicles; non-motor vehicle traffic operation data includes average speed of non-motor vehicles and non-motor vehicle driving trajectory data; pedestrian traffic operation data includes pedestrian flow data and average pedestrian dwell time; municipal traffic operation data includes road water accumulation data, municipal manhole cover operation status data, and event feature data; road segment associated public service scenario operation data includes public service peak hours, vehicle entry and exit frequency, and average vehicle dwell time; and environmental collection data includes real-time rainfall and visibility. The motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road segment associated public service scenario operation data, and environmental collection data are subjected to spatiotemporal alignment and outlier cleaning preprocessing to obtain a multi-source traffic recognition optimization dataset.

3. The traffic congestion intelligent identification method based on deep learning according to claim 2, characterized in that, The step of constructing traffic scene label feature data based on the multi-source traffic recognition optimization dataset and inputting it into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data includes: Traffic scene label feature data is constructed based on the event feature data, real-time rainfall and visibility, and data collection timestamps. The traffic scene label feature data is input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data.

4. The traffic congestion intelligent identification method based on deep learning according to claim 3, characterized in that, The step of extracting features from the multi-source traffic identification and optimization dataset and inputting it into a preset traffic congestion identification model for processing to obtain a preliminary traffic congestion index includes: Spatial feature extraction, temporal feature extraction, and congestion-induced feature extraction are performed on the motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, and road segment-related public service scenario operation data to obtain spatial feature data, temporal feature data, and congestion-induced feature data. Based on the dynamic traffic scenario category feature data, query the preset traffic scenario category and feature weight value mapping table to obtain the weight values ​​corresponding to spatial feature data, temporal feature data and congestion-induced feature data, including spatial weight values, temporal weight values ​​and congestion-induced weight values. The spatial feature data, temporal feature data, and congestion-induced feature data are combined with the spatial weight value, temporal weight value, and congestion-induced weight value for data fusion processing to obtain preliminary traffic congestion assessment feature data. The initial traffic congestion assessment feature data is input into a preset traffic congestion identification model for processing to obtain the initial traffic congestion assessment index.

5. The traffic congestion intelligent identification method based on deep learning according to claim 4, characterized in that, The step of analyzing and processing the multi-source traffic identification optimization dataset to obtain traffic congestion identification influencing factors includes: The road waterlogging data is compared with the preset waterlogging warning value to obtain the road waterlogging exceeding the warning rate; The road waterlogging exceeding the warning rate is combined with the municipal manhole cover operation status data and event characteristic data, and a weighted sum is performed to obtain the municipal traffic operation impact factor. The real-time rainfall is compared with the preset historical average rainfall for the same period to obtain the rainfall exceedance rate; The visibility is compared with a preset visibility warning value to obtain the visibility insufficiency rate; The environmental impact factor is obtained by weighted summation of the above-average rainfall rate and the visibility deficiency rate. The municipal traffic operation impact factors, environmental impact factors, and the proportion of freight vehicles are normalized and weighted and summed to obtain the traffic congestion identification impact factors.

6. The traffic congestion intelligent identification method based on deep learning according to claim 5, characterized in that, If the traffic congestion correction index is greater than or equal to a preset traffic congestion identification threshold, then the preset traffic segment is determined to be congested, and the causes of congestion are identified, including: Based on the motor vehicle traffic operation data within a preset time period, data change features are extracted to obtain vehicle speed change feature data, operation trajectory change feature data, and traffic flow change data; The traffic congestion preliminary assessment feature data, vehicle speed change feature data, operation trajectory change feature data, and flow change data, as well as the municipal traffic operation influencing factors, environmental influencing factors, and the proportion of freight vehicles, are input into a preset traffic congestion cause identification model for analysis and processing to obtain the causes of congestion. The causes of congestion include traffic congestion, event-related congestion, or facility-related congestion.

7. A traffic congestion intelligent recognition system based on deep learning, characterized in that, The system includes a memory and a processor. The memory contains a program for a deep learning-based intelligent traffic congestion recognition method. When the processor executes the program for the deep learning-based intelligent traffic congestion recognition method, it performs the following steps: Obtain the multi-source traffic recognition dataset corresponding to the preset traffic segment ID, and perform data preprocessing to obtain the multi-source traffic recognition optimized dataset; Traffic scene label feature data is constructed based on the multi-source traffic identification optimization dataset and input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data. The multi-source traffic identification optimization dataset is used to extract features and input into a preset traffic congestion identification model for processing to obtain a preliminary traffic congestion index. The multi-source traffic identification optimization dataset is analyzed and processed to obtain the traffic congestion identification influencing factors. The traffic congestion identification influencing factors are used to correct the initial traffic congestion index to obtain the traffic congestion correction index. The traffic congestion correction index is compared with a preset dynamic traffic congestion identification threshold. If the traffic congestion correction index is less than the preset dynamic traffic congestion identification threshold, the preset traffic segment is determined to be non-congested. If the traffic congestion correction index is greater than or equal to the preset dynamic traffic congestion identification threshold, then the preset traffic segment is determined to be congested, and the cause of congestion is identified. A traffic congestion intelligent identification report is generated based on the preset traffic segment ID and congestion causes.

8. The traffic congestion intelligent recognition system based on deep learning according to claim 7, characterized in that, The step of obtaining the multi-source traffic recognition dataset corresponding to the preset traffic segment ID and performing data preprocessing to obtain the multi-source traffic recognition optimized dataset includes: The traffic segments within the preset urban area are divided into grids, and grid IDs are assigned to obtain the preset traffic segment IDs; Obtain the multi-source traffic recognition dataset corresponding to the preset traffic segment ID, including motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road segment associated public service scenario operation data, and environmental collection data. Among them, motor vehicle traffic operation data includes average speed of motor vehicles, average distance between motor vehicles, and proportion of freight vehicles; non-motor vehicle traffic operation data includes average speed of non-motor vehicles and non-motor vehicle driving trajectory data; pedestrian traffic operation data includes pedestrian flow data and average pedestrian dwell time; municipal traffic operation data includes road water accumulation data, municipal manhole cover operation status data, and event feature data; road segment associated public service scenario operation data includes public service peak hours, vehicle entry and exit frequency, and average vehicle dwell time; and environmental collection data includes real-time rainfall and visibility. The motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road segment associated public service scenario operation data, and environmental collection data are subjected to spatiotemporal alignment and outlier cleaning preprocessing to obtain a multi-source traffic recognition optimization dataset.

9. The traffic congestion intelligent recognition system based on deep learning according to claim 8, characterized in that, The step of constructing traffic scene label feature data based on the multi-source traffic recognition optimization dataset and inputting it into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data includes: Traffic scene label feature data is constructed based on the event feature data, real-time rainfall and visibility, and data collection timestamps. The traffic scene label feature data is input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data.

10. The traffic congestion intelligent recognition system based on deep learning according to claim 9, characterized in that, The step of extracting features from the multi-source traffic identification and optimization dataset and inputting it into a preset traffic congestion identification model for processing to obtain a preliminary traffic congestion index includes: Spatial feature extraction, temporal feature extraction, and congestion-induced feature extraction are performed on the motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, and road segment-related public service scenario operation data to obtain spatial feature data, temporal feature data, and congestion-induced feature data. Based on the dynamic traffic scenario category feature data, query the preset traffic scenario category and feature weight value mapping table to obtain the weight values ​​corresponding to spatial feature data, temporal feature data and congestion-induced feature data, including spatial weight values, temporal weight values ​​and congestion-induced weight values. The spatial feature data, temporal feature data, and congestion-induced feature data are combined with the spatial weight value, temporal weight value, and congestion-induced weight value for data fusion processing to obtain preliminary traffic congestion assessment feature data. The initial traffic congestion assessment feature data is input into a preset traffic congestion identification model for processing to obtain the initial traffic congestion assessment index.

Images