Intelligent traffic congestion recognition method and system based on deep learning

Abstract

The application provides a deep learning-based intelligent traffic congestion identification method and system. The method comprises: obtaining a multi-source traffic identification optimized data set by acquiring a multi-source traffic identification data set and optimizing processing, constructing traffic scene label feature data, identifying dynamic traffic scene category feature data, evaluating a traffic congestion preliminary evaluation index, analyzing traffic congestion identification influence factors, optimizing to obtain a traffic congestion correction index, and finally comparing the threshold value with a preset dynamic traffic congestion identification threshold value, determining whether it is traffic congestion according to the threshold value comparison result, analyzing the congestion identification cause, and generating an intelligent traffic congestion identification report; the application collects multi-source heterogeneous data, completes data fusion and feature extraction through a deep learning model, realizes congestion identification by combining dynamic scene classification and adaptive threshold determination, and completes congestion cause tracing analysis by using a deep learning classification model, thereby realizing deep learning-based intelligent traffic congestion identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent transportation, in particular to a traffic congestion intelligent identification method and system based on deep learning. BACKGROUND

[0002] Traffic congestion affects the efficiency of urban operation and development, as well as the travel experience in the process of urbanization. Traditional traffic congestion identification technology has strong data dependence, and the data mainly comes from ETC gantry, traffic cameras, navigation equipment, etc. The deployment cost is high, and it is easy to be affected by environmental interference. At the same time, it ignores the influence of non-motor vehicle and pedestrian traffic operation data analysis on accurate identification of traffic congestion. At present, the traditional technology adopts fixed algorithm and single fixed threshold for judgment. In the extreme weather and temporary event scene, it is easy to misjudge and cannot accurately reflect the traffic situation. Some traditional technologies use traditional data fusion and identification algorithm, which is difficult to effectively extract the deep features of multi-source data, and cannot realize the cause tracing analysis of congestion and provide accurate decision support for traffic relief. Therefore, there is an urgent need for a traffic congestion intelligent identification method based on deep learning technology, which can fuse multi-source heterogeneous data and adapt to dynamic scenes.

[0003] In view of the above problems, an effective technical solution is currently needed. SUMMARY

[0004] The purpose of the present application is to provide a traffic congestion intelligent identification method and system based on deep learning, which can collect multi-source heterogeneous data, complete data fusion and feature extraction through a deep learning model, realize congestion identification by combining dynamic scene classification and adaptive threshold judgment, and complete cause tracing of congestion by using a deep learning classification model, thereby realizing intelligent identification of traffic congestion based on deep learning.

[0005] In the first aspect, the present application provides a traffic congestion intelligent identification method based on deep learning, comprising the following steps:

[0006] Obtain a multi-source traffic identification data set corresponding to a preset traffic section ID, and perform data preprocessing to obtain a multi-source traffic identification optimization data set;

[0007] Construct traffic scene label feature data according to the multi-source traffic identification optimization data set, and input the data into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data;

[0008] Extract features from the multi-source traffic identification optimization data set, and input the data into a preset traffic congestion identification model for processing to obtain a traffic congestion preliminary evaluation index;

[0009] Analyze and process the multi-source traffic identification optimization data set to obtain a traffic congestion identification influence factor;

[0010] According to the traffic congestion identification influence factor, the traffic congestion preliminary evaluation index is corrected to obtain a traffic congestion correction index;

[0011] The traffic congestion correction index is compared with a preset dynamic traffic congestion identification threshold value;

[0012] If the traffic congestion correction index is less than the preset dynamic traffic congestion identification threshold value, it is determined that the preset traffic section is not congested;

[0013] If the traffic congestion correction index is greater than or equal to the preset dynamic traffic congestion identification threshold value, it is determined that the preset traffic section is congested, and the congestion cause is identified;

[0014] According to the preset traffic section ID and the congestion cause, a traffic congestion intelligent identification report is generated.

[0015] Optionally, in the deep learning-based traffic congestion intelligent identification method described in the present application, the multi-source traffic identification data set corresponding to the preset traffic section ID is obtained, and data preprocessing is performed to obtain a multi-source traffic identification optimization data set, including:

[0016] The traffic sections within the preset urban range are grid-divided, and a grid ID is assigned to obtain a preset traffic section ID;

[0017] The multi-source traffic identification data set corresponding to the preset traffic section ID is obtained, including motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road section associated public service scene operation data, and environment collection data, wherein the motor vehicle traffic operation data includes motor vehicle average speed, motor vehicle average spacing, and freight motor vehicle proportion; the non-motor vehicle traffic operation data includes non-motor vehicle average speed and non-motor vehicle trajectory data; the pedestrian traffic operation data includes pedestrian flow data and pedestrian stay duration average; the municipal traffic operation data includes road water accumulation data, municipal manhole cover operation state data, and event characteristic data; the road section associated public service scene operation data includes public service peak period, vehicle entry and exit frequency, and vehicle stay duration average; and the environment collection data includes real-time rainfall and visibility;

[0018] The motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road section associated public service scene operation data, and environment collection data are spatio-temporally aligned and abnormally value cleaned and preprocessed to obtain a multi-source traffic identification optimization data set.

[0019] Optionally, in the deep learning-based intelligent traffic congestion identification method described in the present application, the traffic scene label feature data is constructed according to the multi-source traffic identification optimization data set, and is input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data, including:

[0020] The traffic scene label feature data is constructed according to the event feature data and the real-time rainfall and visibility combination data collection time stamp;

[0021] 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.

[0022] Optionally, in the deep learning-based intelligent traffic congestion identification method described in the present application, the multi-source traffic identification optimization data set is subjected to feature extraction and input into a preset traffic congestion identification model for processing to obtain a traffic congestion preliminary evaluation index, including:

[0023] The motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, and road section associated public service scene operation data are subjected to spatial feature extraction, time sequence feature extraction, and congestion inducing feature extraction to obtain spatial feature data, time sequence feature data, and congestion inducing feature data;

[0024] According to the dynamic traffic scene category feature data, a preset traffic scene category and feature weight value mapping table is queried to obtain weight values corresponding to the spatial feature data, time sequence feature data, and congestion inducing feature data, including spatial weight values, time sequence weight values, and congestion inducing weight values;

[0025] The spatial feature data, time sequence feature data, and congestion inducing feature data are combined with the spatial weight values, time sequence weight values, and congestion inducing weight values for data fusion processing to obtain traffic congestion preliminary evaluation feature data;

[0026] The traffic congestion preliminary evaluation feature data is input into a preset traffic congestion identification model for processing to obtain a traffic congestion preliminary evaluation index.

[0027] Optionally, in the deep learning-based intelligent traffic congestion identification method described in the present application, the multi-source traffic identification optimization data set is subjected to analysis and processing to obtain a traffic congestion identification influence factor, including:

[0028] The road water accumulation data is compared with a preset water accumulation warning value to obtain a road water accumulation over-warning rate;

[0029] The road water accumulation over-warning rate is combined with the municipal manhole cover operation state data and event feature data for weighted summation processing to obtain a municipal traffic operation influence factor;

[0030] comparing the real-time rainfall with a preset historical same-period rainfall average to obtain a rainfall over-average rate;

[0031] comparing the visibility with a preset visibility warning value to obtain a visibility deficiency rate;

[0032] weighting and summing the rainfall over-average rate and the visibility deficiency rate to obtain an environmental impact factor;

[0033] normalizing and weighting and summing the municipal traffic operation impact factor, the environmental impact factor, and the freight motor vehicle proportion to obtain a traffic congestion identification impact factor.

[0034] Optionally, in the deep learning-based traffic congestion intelligent identification method described in the present application, if the traffic congestion correction index is greater than or equal to a preset traffic congestion identification threshold, it is determined that the preset traffic section is in traffic congestion, and the cause of the congestion is identified, including:

[0035] extracting data change characteristics from the motor vehicle traffic operation data in a preset time period to obtain speed change characteristic data, running track change characteristic data, and flow change data;

[0036] inputting the traffic congestion preliminary evaluation characteristic data, the speed change characteristic data, the running track change characteristic data, and the flow change data, and the municipal traffic operation impact factor, the environmental impact factor, and the freight motor vehicle proportion into a preset traffic congestion cause identification model for analysis and processing to obtain the cause of the congestion;

[0037] The cause of the congestion includes flow congestion, event congestion, or facility abnormality congestion.

[0038] In a second aspect, the present application provides a deep learning-based traffic congestion intelligent identification system, which comprises a memory and a processor, wherein the memory comprises a deep learning-based traffic congestion intelligent identification method program, and the deep learning-based traffic congestion intelligent identification method program is executed by the processor to implement the following steps:

[0039] obtaining a multi-source traffic identification data set corresponding to a preset traffic section ID, and performing data preprocessing to obtain a multi-source traffic identification optimization data set;

[0040] constructing traffic scene label feature data according to the multi-source traffic identification optimization data set, and inputting 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;

[0041] extracting features from the multi-source traffic identification optimization data set, and inputting the multi-source traffic identification optimization data set into a preset traffic congestion identification model for processing to obtain a traffic congestion preliminary evaluation index;

[0042] According to the multi-source traffic recognition optimization data set, an analysis processing is performed to obtain a traffic congestion recognition influence factor;

[0043] According to the traffic congestion recognition influence factor, the traffic congestion preliminary evaluation index is corrected to obtain a traffic congestion correction index;

[0044] The traffic congestion correction index is compared with a preset dynamic traffic congestion recognition threshold value;

[0045] If the traffic congestion correction index is less than the preset dynamic traffic congestion recognition threshold value, it is determined that the preset traffic section is not a traffic congestion;

[0046] If the traffic congestion correction index is greater than or equal to the preset dynamic traffic congestion recognition threshold value, it is determined that the preset traffic section is a traffic congestion, and a congestion cause is recognized;

[0047] According to the preset traffic section ID and the congestion cause, a traffic congestion intelligent recognition report is generated.

[0048] Optionally, in the deep learning-based traffic congestion intelligent recognition system described in the present application, the multi-source traffic recognition data set corresponding to the preset traffic section ID is obtained, and data preprocessing is performed to obtain a multi-source traffic recognition optimization data set, including:

[0049] The traffic sections within the preset range of the city are grid-divided, and a grid ID is allocated to obtain a preset traffic section ID;

[0050] The multi-source traffic recognition data set corresponding to the preset traffic section ID is obtained, including motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road section associated public service scene operation data, and environment collection data, wherein the motor vehicle traffic operation data includes motor vehicle average speed, motor vehicle average spacing, and freight motor vehicle proportion, the non-motor vehicle traffic operation data includes non-motor vehicle average speed and non-motor vehicle trajectory data, the pedestrian traffic operation data includes passenger flow data and pedestrian stay duration average, the municipal traffic operation data includes road water accumulation data, municipal manhole cover operation state data, and event characteristic data, the road section associated public service scene operation data includes public service peak period, vehicle entry and exit frequency, and vehicle stay duration average, and the environment collection data includes real-time rainfall and visibility;

[0051] The motor vehicle traffic operation data, the non-motor vehicle traffic operation data, the pedestrian traffic operation data, the municipal traffic operation data, the road section associated public service scene operation data, and the environment collection data are processed by time-space alignment and abnormal value cleaning preprocessing to obtain a multi-source traffic recognition optimization data set.

[0052] Optionally, in the deep learning-based intelligent traffic congestion identification system provided in the present application, the traffic scene label feature data is constructed according to the multi-source traffic identification optimization data set, and is input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data, including:

[0053] The traffic scene label feature data is constructed according to the event feature data and the real-time rainfall and visibility combination data collection time stamp;

[0054] 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.

[0055] Optionally, in the deep learning-based intelligent traffic congestion identification system provided in the present application, the multi-source traffic identification optimization data set is subjected to feature extraction, and is input into a preset traffic congestion identification model for processing to obtain a traffic congestion preliminary evaluation index, including:

[0056] The motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data and road section associated public service scene operation data are subjected to spatial feature extraction, time sequence feature extraction and congestion inducing feature extraction to obtain spatial feature data, time sequence feature data and congestion inducing feature data;

[0057] According to the dynamic traffic scene category feature data, a preset traffic scene category and feature weight value mapping table is queried to obtain weight values corresponding to the spatial feature data, time sequence feature data and congestion inducing feature data, including a spatial weight value, a time sequence weight value and a congestion inducing weight value;

[0058] The spatial feature data, time sequence feature data and congestion inducing feature data are combined with the spatial weight value, time sequence weight value and congestion inducing weight value for data fusion processing to obtain traffic congestion preliminary evaluation feature data;

[0059] The traffic congestion preliminary evaluation feature data is input into a preset traffic congestion identification model for processing to obtain a traffic congestion preliminary evaluation index.

[0060] As can be seen from the above, the deep learning-based intelligent traffic congestion identification method and system provided in the present application realizes intelligent traffic congestion identification based on deep learning by collecting multi-source heterogeneous data, completing data fusion and feature extraction through a deep learning model, combining dynamic scene classification and adaptive threshold determination to realize congestion identification, and simultaneously utilizing a deep learning classification model to complete congestion cause tracing.

[0061] Other features and advantages of the present application will be set forth in the following description, and in part will be apparent from the description, or can be learned by practice of the application. The objects and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims thereof as well as the appended drawings. BRIEF DESCRIPTION OF DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments of the present application will be briefly introduced as follows. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be considered as a limitation to the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0063] Figure 1 a flow chart of the intelligent traffic congestion recognition method based on deep learning provided by the embodiments of the present application;

[0064] Figure 2 a flow chart of obtaining a multi-source traffic recognition optimization data set of the intelligent traffic congestion recognition method based on deep learning provided by the embodiments of the present application;

[0065] Figure 3 a flow chart of obtaining dynamic traffic scene category feature data of the intelligent traffic congestion recognition method based on deep learning provided by the embodiments of the present application. DETAILED DESCRIPTION

[0066] The technical solutions in the embodiments of the present application will be described clearly and completely in the embodiments of the present application in combination with the drawings. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of the present application.

[0067] It should be noted that similar reference numerals and letters indicate similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. Meanwhile, in the description of the present application, the terms “first”, “second”, etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0068] Reference will be made to Figure 1 , Figure 1is a flowchart of a deep learning-based intelligent traffic congestion identification method in some embodiments of the present application. The deep learning-based intelligent traffic congestion identification method is used in terminal equipment such as computers, mobile phones, and the like. The deep learning-based intelligent traffic congestion identification method comprises the following steps:

[0069] S11, a plurality of source traffic identification data sets corresponding to a preset traffic section ID are obtained, and data preprocessing is performed to obtain a plurality of source traffic identification optimization data sets;

[0070] S121, traffic scene label feature data is constructed according to the plurality of source traffic identification optimization data sets, and is input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data;

[0071] S122, the plurality of source traffic identification optimization data sets are subjected to feature extraction and input into a preset traffic congestion identification model for processing to obtain a traffic congestion preliminary evaluation index;

[0072] S123, the plurality of source traffic identification optimization data sets are analyzed and processed to obtain a traffic congestion identification influence factor;

[0073] S13, the traffic congestion preliminary evaluation index is corrected according to the traffic congestion identification influence factor to obtain a traffic congestion correction index;

[0074] S14, the traffic congestion correction index is compared with a preset dynamic traffic congestion identification threshold value;

[0075] S151, if the traffic congestion correction index is less than the preset dynamic traffic congestion identification threshold value, it is determined that the preset traffic section is not congested;

[0076] S152, if the traffic congestion correction index is greater than or equal to the preset dynamic traffic congestion identification threshold value, it is determined that the preset traffic section is congested, and the congestion cause is identified;

[0077] S16, a traffic congestion intelligent identification report is generated according to the preset traffic section ID and the congestion cause.

[0078] It should be noted that, in order to realize accurate identification of traffic congestion in a dynamic traffic scene, first, multi-source heterogeneous data is collected, data preprocessing is performed, and a multi-source traffic identification optimization data set is obtained; then, dynamic traffic scene category feature data is determined, data analysis and evaluation are performed, a traffic congestion preliminary evaluation index and a traffic congestion identification influence factor are obtained, the traffic congestion preliminary evaluation index is optimized according to the traffic congestion identification influence factor, and a traffic congestion correction index is obtained, for example, the traffic congestion preliminary evaluation index is a, the traffic congestion identification influence factor is y, and (1+y)×a is the traffic congestion correction index; whether it is traffic congestion is determined through threshold comparison, wherein the preset dynamic traffic congestion identification threshold is determined according to the dynamic traffic scene category feature data querying the preset dynamic traffic scene category and weight value relationship mapping table determined by the person skilled in the art according to a large number of historical cases, and the preset dynamic traffic scene category and weight value relationship mapping table is obtained by the person skilled in the art according to a large number of historical cases, and can be dynamically adjusted, if it is traffic congestion, the congestion cause is further identified; finally, a traffic congestion intelligent identification report is generated according to the divided preset traffic road section ID and the congestion cause, and data support is provided for traffic relief and decision-making.

[0079] Please refer to Figure 2 , Figure 2 is the flowchart of obtaining a multi-source traffic identification optimization data set in the deep learning-based traffic congestion intelligent identification method in some embodiments of the present application. According to the embodiment of the present application, the multi-source traffic identification data set corresponding to the preset traffic road section ID is obtained, and data preprocessing is performed to obtain the multi-source traffic identification optimization data set, which comprises:

[0080] S21, grid division is performed on the traffic road sections in the city preset range, and a grid ID is allocated to obtain a preset traffic road section ID;

[0081] S22, a multi-source traffic identification data set corresponding to the preset traffic road section ID is obtained, including motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road section associated public service scene operation data and environment collection data, wherein the motor vehicle traffic operation data includes motor vehicle average speed, motor vehicle average spacing and freight motor vehicle proportion, the non-motor vehicle traffic operation data includes non-motor vehicle average speed and non-motor vehicle trajectory data, the pedestrian traffic operation data includes passenger flow data and pedestrian stay time average, the municipal traffic operation data includes road water accumulation data, municipal manhole cover operation state data and event feature data, the road section associated public service scene operation data includes public service peak period, vehicle entry and exit frequency and vehicle stay time average, and the environment collection data includes real-time rainfall and visibility;

[0082] S23, the motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road section associated public service scene operation data and environment collection data are spatio-temporally aligned and pre-processed by removing outliers, to obtain a multi-source traffic recognition optimization dataset.

[0083] It should be noted that, in order to overcome the single data source and strong dependence on special traffic equipment of the traditional technology, first, the traffic sections in the preset range of the city are divided into grids according to 500*500 meters, and a unique grid ID is assigned, to obtain a preset traffic section ID, then the data collection covers different traffic participants (such as motor vehicles, non-motor vehicles and pedestrians) and general civil and commercial facilities (such as street lamp poles, shop front houses and municipal manhole covers), public services (such as schools and hospitals) and multiple dimensions of collected environment, breaking through the focus on only motor vehicle data, at the same time, reducing the cost of equipment arrangement through civil facilities, effectively mining the relationship between traffic scene inducement and traffic congestion, and being practical for mixed traffic scenes in the city; finally, the collected multi-source data is spatio-temporally aligned through "500m*500m grid coding, 1 minute time slice", invalid data is filtered by combining the 3σ principle and the CNN anomaly detection model, and the attention interpolation model is used to fill in the gaps, to obtain a multi-source traffic recognition optimization dataset, wherein the road section associated public service scene operation data refers to the collection data of the public service scene in the preset range of the road section, the municipal manhole cover operation state includes normal or abnormal, and the municipal manhole cover operation state data is represented by different identifiers, and the event feature data is represented by different identifiers.

[0084] Please refer to Figure 3 , Figure 3 is a flowchart for obtaining dynamic traffic scene category feature data in the deep learning-based intelligent traffic congestion recognition method in some embodiments of the present application. According to the embodiment of the present application, the traffic scene label feature data is constructed according to the multi-source traffic recognition optimization dataset, and is input into a preset traffic scene category classification model for analysis and processing, to obtain dynamic traffic scene category feature data, which includes:

[0085] S31, the traffic scene label feature data is constructed according to the event feature data and the real-time rainfall and visibility combined data collection time stamp;

[0086] S32, 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.

[0087] It should be noted that in order to adapt to the congestion identification of the dynamic traffic scene, first, according to the collected multi-source traffic identification optimization data set, traffic scene label feature data covering 'time, environment and event' are constructed, specifically by constructing event feature data, real-time rainfall and visibility combined with data collection time stamp, then the constructed traffic scene label feature data is input into the preset traffic scene category classification model for analysis and processing, and dynamic traffic scene category feature data is obtained, the dynamic traffic scene category is like 'weekday morning peak, normal weather and no special event', and the dynamic traffic scene category feature data is represented by a unique identifier, wherein the preset traffic scene category classification model is obtained by training the initial Transformer scene classification model through a large number of historical sample traffic scene label feature data and corresponding dynamic traffic scene category feature data.

[0088] According to the embodiment of the application, the multi-source traffic identification optimization data set is extracted and input into the preset traffic congestion identification model for processing to obtain a traffic congestion preliminary evaluation index, which comprises:

[0089] The motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data and road section associated public service scene operation data are subjected to spatial feature extraction, time sequence feature extraction and congestion inducing feature extraction to obtain spatial feature data, time sequence feature data and congestion inducing feature data;

[0090] According to the dynamic traffic scene category feature data, a preset traffic scene category and feature weight value mapping table is queried to obtain the weight values corresponding to the spatial feature data, time sequence feature data and congestion inducing feature data, including spatial weight value, time sequence weight value and congestion inducing weight value;

[0091] The spatial feature data, time sequence feature data and congestion inducing feature data are combined with the spatial weight value, time sequence weight value and congestion inducing weight value for data fusion processing to obtain traffic congestion preliminary evaluation feature data;

[0092] The traffic congestion preliminary evaluation feature data is input into the preset traffic congestion identification model for processing to obtain a traffic congestion preliminary evaluation index.

[0093] It should be noted that in order to quantitatively express the traffic congestion, firstly, different types of data collected are extracted by using a preset special deep learning sub-model, such as the motor vehicle traffic operation data collected by the millimeter wave radar, the spatial feature data including speed and distance are extracted by using a preset 3D-CNN model, the time sequence feature data of the non-motor vehicle traffic operation data and the pedestrian traffic operation data are extracted by using a preset GR model, and the congestion inducing feature data including peak period and vehicle in-out frequency characteristics are extracted by using a preset MLP model; then, according to the determined dynamic traffic scene category feature data, a preset traffic scene category and feature weight value mapping table is queried to determine the corresponding weight value, and the spatial feature data, the time sequence feature data and the congestion inducing feature data are weighted and fused to obtain traffic congestion preliminary evaluation feature data; finally, the traffic congestion preliminary evaluation feature data is input into a preset traffic congestion recognition model for processing to obtain a traffic congestion preliminary evaluation index, wherein the preset traffic scene category and feature weight value mapping table is analyzed and constructed by the person skilled in the art according to historical data, and can be dynamically adjusted, the traffic congestion preliminary evaluation feature data is represented as a feature vector, the preset traffic congestion recognition model is obtained by training a large number of historical sample traffic congestion preliminary evaluation feature data and corresponding traffic congestion preliminary evaluation indexes, and the preset 3D-CNN model, the preset GR model and the preset MLP model are obtained by pre-training a large number of historical sample cases by the person skilled in the art.

[0094] According to the embodiment of the present application, the analysis and processing according to the multi-source traffic recognition optimization data set are performed to obtain a traffic congestion recognition influence factor, which includes:

[0095] The road water accumulation data is compared with a preset water accumulation warning value to obtain a road water accumulation over-warning rate;

[0096] The road water accumulation over-warning rate is combined with the municipal well cover operation state data and the event feature data to perform weighted summation processing to obtain a municipal traffic operation influence factor;

[0097] The real-time rainfall is compared with a preset historical same period rainfall average value to obtain a rainfall over-average rate;

[0098] The visibility is compared with a preset visibility warning value to obtain a visibility deficiency rate;

[0099] The rainfall over-average rate and the visibility deficiency rate are weighted and summed to obtain an environmental influence factor;

[0100] The municipal traffic operation influence factor and the environmental influence factor are normalized and weighted and summed with the freight motor vehicle proportion to obtain a traffic congestion recognition influence factor.

[0101] It should be noted that, in order to adapt to the dynamically changing traffic scene and improve the accuracy of traffic congestion identification, the traffic congestion identification influence factors are obtained by analyzing and evaluating municipal traffic operation data and environmental collection data, and the traffic congestion preliminary evaluation index obtained by analyzing motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data and road section associated public service scene operation data is optimized and corrected, wherein the road water over-warning rate is the ratio of the difference between the road water data and the preset water warning value to the preset water warning value, if it is positive, it is recorded normally, otherwise it is recorded as 0, that is, the road water data is less than or equal to the preset water warning value, the municipal well cover operation state data is used to indicate that the municipal well cover is normal or abnormal, and the event characteristic data is used to indicate whether there is a temporary event (such as road construction), the municipal well cover operation state data and the event characteristic data are identified by different identifiers, such as 0 for normal municipal well cover and 1 for abnormal municipal well cover, and the municipal traffic operation influence factor is determined by weighted summation of the road water over-warning rate, the municipal well cover operation state data and the event characteristic data; the rainfall over-average rate is the ratio of the difference between the real-time rainfall and the preset historical same period rainfall average to the preset historical same period rainfall average, if it is positive, it is recorded normally, otherwise it is recorded as 0, that is, the real-time rainfall is less than or equal to the preset historical same period rainfall average, and the insufficient visibility rate is the ratio of the absolute value of the difference between the visibility and the preset visibility warning value to the preset visibility warning value, if the visibility is greater than the preset visibility warning value, it is recorded as 0, and the environmental influence factor is determined by weighted summation of the rainfall over-average rate and the insufficient visibility rate; finally, the municipal traffic operation influence factor and the environmental influence factor and the freight motor vehicle proportion are normalized and mapped to the interval [0, 1], and then weighted summation processing is performed to finally determine the traffic congestion identification influence factor.

[0102] According to the embodiment of the present application, if the traffic congestion correction index is greater than or equal to the preset traffic congestion identification threshold, it is determined that the preset traffic section is in traffic congestion, and the congestion cause is identified, including:

[0103] According to the motor vehicle traffic operation data in the preset time period, data change characteristic data is extracted to obtain speed change characteristic data, running track change characteristic data and flow change data;

[0104] The traffic congestion preliminary evaluation characteristic data, the speed change characteristic data, the running track change characteristic data and the flow change data, and the municipal traffic operation influence factor, the environmental influence factor and the freight motor vehicle proportion are input into a preset traffic congestion cause identification model for analysis and processing to obtain the congestion cause;

[0105] The congestion cause includes flow congestion, event congestion or facility abnormality congestion.

[0106] It should be noted that, based on the deep learning classification model of multi-task learning, the traffic congestion preliminary evaluation feature data fused by a large number of historical samples, the vehicle speed change feature data, the running track change feature data and the flow change data reflecting the data change characteristics of the congestion period data, the municipal traffic operation influence factor, the environmental influence factor and the proportion of freight motor vehicles are combined as inputs, and the preset traffic congestion cause identification model is trained, which is used for identifying three types of congestion causes, then, based on real-time data, the congestion cause tracing analysis is carried out, and data support is provided for traffic relief and decision-making, wherein the flow congestion refers to the congestion caused by too large traffic flow, the event congestion refers to the traffic congestion caused by temporary events, and the facility abnormal congestion refers to the traffic congestion caused by municipal facility abnormalities.

[0107] It is worth mentioning that, according to the embodiment of the application, further comprising:

[0108] If it is flow congestion, the motor vehicle driving track data, public service scene feature data and road network structure feature data are acquired;

[0109] The motor vehicle driving track data, public service scene feature data and road network structure feature data are input into a preset flow congestion source analysis model for analysis and processing, and congestion flow source feature data is obtained;

[0110] Traffic flow data of a preset time period is acquired, and flow time sequence feature data is constructed according to the traffic flow data;

[0111] The flow time sequence feature data and preset historical same period flow time sequence feature data are input into a preset flow peak feature analysis model for analysis and processing, and flow peak parameter data is obtained;

[0112] The congestion flow source and flow peak parameter data are sent to the operation and maintenance end for display.

[0113] It should be noted that after determining the traffic congestion, in order to further accurately determine the traffic congestion source and the traffic peak characteristics, first, the corresponding motor vehicle driving trajectory data, public service scene characteristic data and road network structure characteristic data are obtained, wherein the motor vehicle driving trajectory data includes data collection timestamp, motor vehicle position coordinates and motor vehicle instantaneous speed, the public service scene characteristic data refers to the number of permanent residents, peak travel rate and travel direction within a preset range around the road section, and the road network structure characteristic data refers to abstracting the city road network into a node and edge graph structure, the node is an intersection or a road section, and the edge is a connection relationship between nodes, then the preset traffic congestion source analysis model is analyzed and processed to obtain congestion traffic source characteristic data, such as 80% of the congestion vehicles come from a certain residential area around the road section, and 20% come from other places, at the same time, traffic flow data in a preset time period (such as the past 4h) is collected, time granularity is divided by 5min, and the number of vehicles of each time granularity, the average number of vehicles of the previous 3 time granularities and the vehicle growth rate (i.e. the difference between the number of vehicles of the next time granularity and the number of vehicles of the previous time granularity, and the ratio of the number of vehicles of the previous time granularity) of a certain road section are counted, so as to construct traffic time sequence characteristic data, and then the preset historical same period traffic time sequence characteristic data is input into the preset traffic peak characteristic analysis model for analysis and processing to obtain traffic peak parameter data, including peak traffic starting time, peak traffic duration and peak deviation rate (i.e. the absolute value of the difference between the current peak traffic and the historical same period peak traffic, and the ratio of the historical same period peak traffic), finally, the obtained congestion traffic source and traffic peak parameter data are sent to the operation end for display, providing accurate data support for traffic relief, wherein the preset traffic congestion source analysis model is obtained by training a large number of historical samples of motor vehicle driving trajectory data, public service scene characteristic data and road network structure characteristic data and corresponding congestion traffic source characteristic data, and the preset traffic peak characteristic analysis model is obtained by training a large number of historical samples of traffic time sequence characteristic data and preset historical same period traffic time sequence characteristic data and corresponding traffic peak parameter data.

[0114] It is worth mentioning that, according to the embodiment of the application, further comprising:

[0115] If it is an event congestion, the road section traffic monitoring video data is acquired;

[0116] The road section traffic monitoring video data is input into a preset target detection model for analysis and processing to obtain congestion event position coordinates and event type characteristic data;

[0117] According to the congestion event position coordinates of the preset continuous video frame and the motor vehicle driving trajectory data, a preset traffic congestion influence semantic segmentation model is marked to obtain congestion influence range data;

[0118] Send the congestion event position coordinate and event type feature data and the congestion influence range data to the operation and maintenance end display.

[0119] It should be noted that, in order to further determine the congestion position and the congestion influence range of the event congestion, and to provide accurate data support for traffic relief, first, the road traffic monitoring video data and the motor vehicle driving trajectory data of the event congestion road section are collected, the adaptive image enhancement processing is performed on the road traffic monitoring video data, the Kalman filter is adopted to complete the short-time missing trajectory data, the road traffic monitoring video data is input into the preset target detection model for analysis and processing, the congestion event position coordinate and the event type feature data are obtained, such as the event type is vehicle rear-end, the congestion event position coordinate is 106.05°E, 38.02°N, at the same time, the congestion event position coordinate and the motor vehicle driving trajectory data of the preset continuous video frames (such as 5 continuous frames) are labeled through the preset traffic congestion influence semantic segmentation model, the congestion influence range data is obtained, such as the influence range is 500 meters behind, finally, the obtained congestion event position coordinate and event type feature data and the congestion influence range data are sent to the operation and maintenance end display, wherein, the preset target detection model is obtained by training a large number of historical sample road traffic monitoring video data and the corresponding congestion event position coordinate and event type feature data, and the preset traffic congestion influence semantic segmentation model is obtained by training a large number of historical sample congestion event position coordinate and motor vehicle driving trajectory data and the corresponding congestion influence range data.

[0120] The application further discloses a traffic congestion intelligent recognition system based on deep learning, which comprises a memory and a processor, the memory comprises a traffic congestion intelligent recognition method program based on deep learning, and the traffic congestion intelligent recognition method program based on deep learning realizes the following steps when being executed by the processor:

[0121] A plurality of source traffic recognition data sets corresponding to a preset traffic road section ID are acquired, and data preprocessing is performed to obtain a plurality of source traffic recognition optimization data sets;

[0122] Traffic scene label feature data are constructed according to the plurality of source traffic recognition optimization data sets, and are input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data;

[0123] The plurality of source traffic recognition optimization data sets are subjected to feature extraction and input into a preset traffic congestion recognition model for processing to obtain a traffic congestion preliminary evaluation index;

[0124] The plurality of source traffic recognition optimization data sets are analyzed and processed to obtain a traffic congestion recognition influence factor;

[0125] According to the traffic congestion identification influence factor, the traffic congestion preliminary evaluation index is corrected to obtain a traffic congestion correction index;

[0126] The traffic congestion correction index is compared with a preset dynamic traffic congestion identification threshold value in threshold comparison;

[0127] If the traffic congestion correction index is less than the preset dynamic traffic congestion identification threshold value, it is determined that the preset traffic section is not congested;

[0128] If the traffic congestion correction index is greater than or equal to the preset dynamic traffic congestion identification threshold value, it is determined that the preset traffic section is congested, and the congestion cause is identified;

[0129] A traffic congestion intelligent identification report is generated according to the preset traffic section ID and the congestion cause.

[0130] It should be noted that, in order to realize accurate identification of traffic congestion in a dynamic traffic scene, first, multi-source heterogeneous data is collected, data preprocessing is performed, and a multi-source traffic identification optimization data set is obtained; then, dynamic traffic scene category feature data is determined, data analysis and evaluation are performed, a traffic congestion preliminary evaluation index and a traffic congestion identification influence factor are obtained, the traffic congestion preliminary evaluation index is optimized according to the traffic congestion identification influence factor, and a traffic congestion correction index is obtained, for example, if the traffic congestion preliminary evaluation index is a and the traffic congestion identification influence factor is y, then (1+y)×a is the traffic congestion correction index; whether it is traffic congestion is determined through threshold comparison, wherein the preset dynamic traffic congestion identification threshold value is determined according to the dynamic traffic scene category feature data by querying a preset dynamic traffic scene category and weight value relationship mapping table, the preset dynamic traffic scene category and weight value relationship mapping table is obtained by a person skilled in the art according to a large number of historical cases, and can be dynamically adjusted, if it is traffic congestion, the congestion cause is further identified; finally, a traffic congestion intelligent identification report is generated according to the divided preset traffic section ID and the congestion cause, providing data support for traffic relief and decision-making.

[0131] According to the embodiment of the application, the multi-source traffic identification data set corresponding to the preset traffic section ID is obtained, and data preprocessing is performed to obtain a multi-source traffic identification optimization data set, which comprises:

[0132] The traffic sections in the city preset range are grid divided, and a grid ID is allocated to obtain a preset traffic section ID;

[0133] Obtaining a multi-source traffic identification data set corresponding to a preset traffic section ID, including motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road section associated public service scene operation data and environment collection data, wherein the motor vehicle traffic operation data includes motor vehicle average speed, motor vehicle average spacing and freight motor vehicle proportion, the non-motor vehicle traffic operation data includes non-motor vehicle average speed and non-motor vehicle trajectory data, the pedestrian traffic operation data includes passenger flow data and pedestrian stay time average, the municipal traffic operation data includes road water accumulation data, municipal manhole cover operation state data and event feature data, the road section associated public service scene operation data includes public service peak period, vehicle entry and exit frequency and vehicle stay time average, and the environment collection data includes real-time rainfall and visibility.

[0134] The motor vehicle traffic operation data, the non-motor vehicle traffic operation data, the pedestrian traffic operation data, the municipal traffic operation data, the road section associated public service scene operation data and the environment collection data are subjected to spatio-temporal alignment and abnormal value cleaning preprocessing to obtain a multi-source traffic identification optimized data set.

[0135] It should be noted that, in order to overcome the single data source and strong dependence on special traffic equipment of the traditional technology, first, the traffic sections in a preset range of the city are divided into grids according to 500*500 meters, and a unique grid ID is assigned to obtain a preset traffic section ID, then the data collection covers different traffic participants (such as motor vehicles, non-motor vehicles and pedestrians) and ordinary civil and commercial facilities (such as street lamp poles, shop front houses and municipal manhole covers), public services (such as schools and hospitals) and multiple dimensions of collected environment, breaking through the focus on only motor vehicle data, at the same time, reducing the cost of device arrangement through civil facilities, effectively mining the relationship between traffic scene inducement and traffic congestion, and being practical for urban mixed traffic scenes; finally, the collected multi-source data is subjected to spatio-temporal alignment through "500m*500m grid coding, 1-minute time slice", invalid data is filtered in combination with the 3σ principle and the CNN abnormal detection model, and the attention interpolation model is used to fill in the gaps to obtain a multi-source traffic identification optimized data set, wherein the road section associated public service scene operation data refers to the collection data of the public service scene in the preset range of the road section, the municipal manhole cover operation state includes normal or abnormal, and the municipal manhole cover operation state data is represented by different identifiers, and the event feature is, for example, construction, and the event feature data is represented by different identifiers.

[0136] According to the embodiment of the application, the multi-source traffic identification optimized data set is used to construct traffic scene label feature data, and 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, including:

[0137] According to the event feature data and the real-time rainfall and visibility combined data collection timestamp, traffic scene label feature data is constructed;

[0138] The traffic scene label feature data is input into a preset traffic scene category classification model for analysis and processing, and dynamic traffic scene category feature data is obtained.

[0139] It should be noted that, in order to adapt to the congestion identification of dynamic traffic scenes, first, traffic scene label feature data covering "time, environment and event" is constructed according to the collected multi-source traffic identification optimization data set, specifically by event feature data, real-time rainfall and visibility combined data collection timestamp, then the constructed traffic scene label feature data is input into a preset traffic scene category classification model for analysis and processing, and dynamic traffic scene category feature data is obtained, the dynamic traffic scene category is like "weekday morning peak, normal weather and no special event", and the dynamic traffic scene category feature data is represented by a unique identifier, wherein the preset traffic scene category classification model is obtained by training an initial Transformer scene classification model through a large number of historical sample traffic scene label feature data and corresponding dynamic traffic scene category feature data.

[0140] According to the embodiment of the application, the multi-source traffic identification optimization data set is feature extracted and input into a preset traffic congestion identification model for processing to obtain a traffic congestion preliminary evaluation index, which includes:

[0141] The motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data and road section associated public service scene operation data are subjected to spatial feature extraction, time sequence feature extraction and congestion inducing feature extraction to obtain spatial feature data, time sequence feature data and congestion inducing feature data;

[0142] According to the dynamic traffic scene category feature data, a preset traffic scene category and feature weight value mapping table is queried to obtain the weight values corresponding to the spatial feature data, time sequence feature data and congestion inducing feature data, including spatial weight values, time sequence weight values and congestion inducing weight values;

[0143] The spatial feature data, time sequence feature data and congestion inducing feature data are combined with the spatial weight values, time sequence weight values and congestion inducing weight values for data fusion processing to obtain traffic congestion preliminary evaluation feature data;

[0144] The traffic congestion preliminary evaluation feature data is input into a preset traffic congestion identification model for processing to obtain a traffic congestion preliminary evaluation index.

[0145] It should be noted that in order to quantitatively express the traffic congestion, firstly, different types of data collected are extracted by using a preset special deep learning sub-model, such as the motor vehicle traffic operation data collected by the millimeter wave radar, the spatial feature data including speed and distance are extracted by using a preset 3D-CNN model, the time sequence feature data of the non-motor vehicle traffic operation data and the pedestrian traffic operation data are extracted by using a preset GR model, and the congestion inducing feature data including peak period and vehicle in-out frequency characteristics are extracted by using a preset MLP model; then, according to the determined dynamic traffic scene category feature data, a preset traffic scene category and feature weight value mapping table is queried to determine the corresponding weight value, and the spatial feature data, the time sequence feature data and the congestion inducing feature data are weighted and fused to obtain traffic congestion preliminary evaluation feature data; finally, the traffic congestion preliminary evaluation feature data is input into a preset traffic congestion recognition model for processing to obtain a traffic congestion preliminary evaluation index, wherein the preset traffic scene category and feature weight value mapping table is analyzed and constructed by the person skilled in the art according to historical data, and can be dynamically adjusted, the traffic congestion preliminary evaluation feature data is represented as a feature vector, the preset traffic congestion recognition model is obtained by training a large number of historical sample traffic congestion preliminary evaluation feature data and corresponding traffic congestion preliminary evaluation indexes, and the preset 3D-CNN model, the preset GR model and the preset MLP model are obtained by pre-training a large number of historical sample cases by the person skilled in the art.

[0146] According to the embodiment of the present application, the analysis and processing according to the multi-source traffic recognition optimization data set are performed to obtain a traffic congestion recognition influence factor, which includes:

[0147] The road water accumulation data is compared with a preset water accumulation warning value to obtain a road water accumulation over-warning rate;

[0148] The road water accumulation over-warning rate is combined with the municipal well cover operation state data and the event feature data to perform weighted summation processing to obtain a municipal traffic operation influence factor;

[0149] The real-time rainfall is compared with a preset historical same period rainfall average value to obtain a rainfall over-average rate;

[0150] The visibility is compared with a preset visibility warning value to obtain a visibility deficiency rate;

[0151] The rainfall over-average rate and the visibility deficiency rate are weighted and summed to obtain an environmental influence factor;

[0152] The municipal traffic operation influence factor and the environmental influence factor are normalized and weighted and summed with the freight motor vehicle proportion to obtain a traffic congestion recognition influence factor.

[0153] It should be noted that, in order to adapt to the dynamically changing traffic scene and improve the accuracy of traffic congestion identification, the traffic congestion identification influence factors are obtained by analyzing and evaluating municipal traffic operation data and environmental collection data, and the traffic congestion preliminary evaluation index obtained by analyzing motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data and road section associated public service scene operation data is optimized and corrected, wherein the road water over-warning rate is the ratio of the difference between the road water data and the preset water warning value to the preset water warning value, if it is positive, it is recorded normally, otherwise it is recorded as 0, that is, the road water data is less than or equal to the preset water warning value, the municipal well cover operation state data is used to indicate that the municipal well cover is normal or abnormal, and the event feature data is used to indicate whether there is a temporary event (such as road construction), the municipal well cover operation state data and the event feature data are identified by different identifiers, such as 0 for normal municipal well cover and 1 for abnormal municipal well cover, and the municipal traffic operation influence factor is determined by weighted summation of the road water over-warning rate, the municipal well cover operation state data and the event feature data; the rainfall over-average rate is the ratio of the difference between the real-time rainfall and the preset historical same period rainfall average to the preset historical same period rainfall average, if it is positive, it is recorded normally, otherwise it is recorded as 0, that is, the real-time rainfall is less than or equal to the preset historical same period rainfall average, and the insufficient visibility rate is the ratio of the absolute value of the difference between the visibility and the preset visibility warning value to the preset visibility warning value, if the visibility is greater than the preset visibility warning value, it is recorded as 0, and the environmental influence factor is determined by weighted summation of the rainfall over-average rate and the insufficient visibility rate; finally, the municipal traffic operation influence factor and the environmental influence factor and the freight motor vehicle proportion are normalized and mapped to the interval [0, 1], and then weighted summation processing is performed to finally determine the traffic congestion identification influence factor.

[0154] According to the embodiment of the present application, if the traffic congestion correction index is greater than or equal to the preset traffic congestion identification threshold, it is determined that the preset traffic section is in traffic congestion, and the congestion cause is identified, including:

[0155] According to the motor vehicle traffic operation data in the preset time period, data change characteristic data, running track change characteristic data and flow change data are extracted;

[0156] The traffic congestion preliminary evaluation characteristic data, the speed change characteristic data, the running track change characteristic data and the flow change data, and the municipal traffic operation influence factor, the environmental influence factor and the freight motor vehicle proportion are input into a preset traffic congestion cause identification model for analysis and processing to obtain the congestion cause;

[0157] The congestion cause includes flow congestion, event congestion or facility abnormality congestion.

[0158] It should be noted that, based on the deep learning classification model of multi-task learning, the traffic congestion preliminary evaluation feature data fused by a large number of historical samples, the vehicle speed change feature data, the running track change feature data and the flow change data reflecting the data change characteristics of the congestion period data, the municipal traffic operation influence factor, the environmental influence factor and the proportion of freight motor vehicles are combined as inputs, and the preset traffic congestion cause identification model is trained, which is used for identifying three types of congestion causes, then, based on real-time data, the congestion cause tracing analysis is carried out, and data support is provided for traffic relief and decision-making, wherein the flow congestion refers to the congestion caused by too large traffic flow, the event congestion refers to the traffic congestion caused by temporary events, and the facility abnormal congestion refers to the traffic congestion caused by municipal facility abnormalities.

[0159] It is worth mentioning that, according to the embodiment of the application, further comprising:

[0160] If it is flow congestion, the motor vehicle driving track data, public service scene feature data and road network structure feature data are acquired;

[0161] The motor vehicle driving track data, public service scene feature data and road network structure feature data are input into a preset flow congestion source analysis model for analysis and processing, and congestion flow source feature data is obtained;

[0162] Traffic flow data of a preset time period is acquired, and flow time sequence feature data is constructed according to the traffic flow data;

[0163] The flow time sequence feature data and preset historical same period flow time sequence feature data are input into a preset flow peak feature analysis model for analysis and processing, and flow peak parameter data is obtained;

[0164] The congestion flow source and flow peak parameter data are sent to the operation and maintenance end for display.

[0165] It should be noted that after determining the traffic congestion, in order to further accurately determine the traffic congestion source and the traffic peak characteristics, first, the corresponding motor vehicle driving trajectory data, public service scene characteristic data and road network structure characteristic data are obtained, wherein the motor vehicle driving trajectory data includes data collection timestamp, motor vehicle position coordinates and motor vehicle instantaneous speed, the public service scene characteristic data refers to the number of permanent residents, peak travel rate and travel direction within a preset range around the road section, and the road network structure characteristic data refers to that the city road network is abstracted as a node and edge graph structure, the node is an intersection or a road section, and the edge is a connection relationship between nodes, then the preset traffic congestion source analysis model is analyzed and processed to obtain congestion traffic source characteristic data, for example, 80% of the congestion vehicles come from a certain residential area around the road section, and 20% come from other places, at the same time, traffic flow data in a preset time period (such as the past 4h) is collected, the time granularity is divided by 5min, and the number of vehicles of each time granularity, the average number of vehicles of the past 3 time granularities before the current time and the vehicle growth rate (that is, the difference between the number of vehicles of the next time granularity and the number of vehicles of the previous time granularity, and the ratio of the number of vehicles of the previous time granularity) of a certain road section are counted, so as to construct traffic time sequence characteristic data, and then the preset historical same period traffic time sequence characteristic data is input into the preset traffic peak characteristic analysis model for analysis and processing to obtain traffic peak parameter data, including peak traffic starting time, peak traffic duration and peak deviation rate (that is, the absolute value of the difference between the current peak traffic and the historical same period peak traffic, and the ratio of the historical same period peak traffic), finally, the obtained congestion traffic source and traffic peak parameter data are sent to the operation end for display, providing accurate data support for traffic relief, wherein the preset traffic congestion source analysis model is obtained by training a large number of historical samples of motor vehicle driving trajectory data, public service scene characteristic data and road network structure characteristic data and corresponding congestion traffic source characteristic data, and the preset traffic peak characteristic analysis model is obtained by training a large number of historical samples of traffic time sequence characteristic data and preset historical same period traffic time sequence characteristic data and corresponding traffic peak parameter data.

[0166] It is worth mentioning that, according to the embodiment of the application, further comprising:

[0167] If it is an event congestion, the road section traffic monitoring video data is acquired;

[0168] The road section traffic monitoring video data is input into a preset target detection model for analysis and processing to obtain congestion event position coordinates and event type characteristic data;

[0169] According to the congestion event position coordinates of the preset continuous video frame and the motor vehicle driving trajectory data, a preset traffic congestion influence semantic segmentation model is marked to obtain congestion influence range data;

[0170] Send the congestion event position coordinate and event type feature data and the congestion influence range data to the operation and maintenance end display.

[0171] It should be noted that, in order to further determine the congestion position and the congestion influence range of the event congestion, and to provide accurate data support for traffic relief, first, the road traffic monitoring video data and the motor vehicle driving trajectory data of the event congestion road section are collected, the adaptive image enhancement processing is performed on the road traffic monitoring video data, the Kalman filter is adopted to complete the short-time missing trajectory data, the road traffic monitoring video data is input into the preset target detection model for analysis and processing, the congestion event position coordinate and the event type feature data are obtained, such as the event type is vehicle rear-end, the congestion event position coordinate is 106.05°E, 38.02°N, at the same time, the congestion event position coordinate and the motor vehicle driving trajectory data of the preset continuous video frames (such as 5 continuous frames) are labeled through the preset traffic congestion influence semantic segmentation model, the congestion influence range data is obtained, such as the influence range is 500 meters behind, finally, the obtained congestion event position coordinate and event type feature data and the congestion influence range data are sent to the operation and maintenance end display, wherein, the preset target detection model is obtained by training a large number of historical sample road traffic monitoring video data and corresponding congestion event position coordinate and event type feature data, and the preset traffic congestion influence semantic segmentation model is obtained by training a large number of historical sample congestion event position coordinate and motor vehicle driving trajectory data and corresponding congestion influence range data.

[0172] The disclosed traffic congestion intelligent recognition method and system based on deep learning realize congestion recognition by collecting multi-source heterogeneous data, completing data fusion and feature extraction through a deep learning model, combining dynamic scene classification and adaptive threshold determination, and completing congestion cause tracing through a deep learning classification model, so as to realize intelligent recognition of traffic congestion based on deep learning.

[0173] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only illustrative, for example, the division of the units is only a logical function division, and actual implementation can have another division mode, for example, multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed components can be through some interfaces, indirect coupling or communication connection of the devices or units, which can be electrical, mechanical or other forms.

[0174] The units described as separate components above can or can not be physically separate, and the components shown as units can or can not be physical units; they can be located in one place or distributed on multiple network units; and part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0175] In addition, each functional unit in each embodiment of the present application 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 the form of hardware or in the form of hardware plus software functional units.

[0176] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction-related hardware, and the foregoing program can be stored in a readable storage medium, and when the program is executed, the steps of the method embodiments are executed; and the foregoing storage medium includes: a mobile storage device, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various storage medium that can store program codes.

[0177] Alternatively, the integrated units of the present application, if implemented in the form of software functional modules and sold or used as independent products, can also be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present application can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the embodiments of the present application. The foregoing storage medium includes: a mobile storage device, a ROM, a RAM, a magnetic disk or an optical disk, and various storage medium that can store program codes.

Claims

1. A deep learning-based intelligent traffic congestion recognition method, characterized in that, The method comprises the following steps: obtaining a multi-source traffic identification data set corresponding to a preset traffic section ID, and performing data preprocessing to obtain a multi-source traffic identification optimization data set, including motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, road section associated public service scene operation data, and environment collection data, wherein the municipal traffic operation data includes road water accumulation data, municipal manhole cover operation state data, and event feature data, and the road section associated public service scene operation data includes public service peak period, vehicle entry and exit frequency, and vehicle stay time average; constructing traffic scene label feature data according to the multi-source traffic identification optimization data set, and inputting 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; performing spatial feature extraction, time sequence feature extraction, and congestion inducing feature extraction on the motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, and road section associated public service scene operation data to obtain spatial feature data, time sequence feature data, and congestion inducing feature data; querying a preset traffic scene category and feature weight value mapping table according to the dynamic traffic scene category feature data to obtain weight values corresponding to the spatial feature data, time sequence feature data, and congestion inducing feature data, including spatial weight values, time sequence weight values, and congestion inducing weight values; performing data fusion processing on the spatial feature data, time sequence feature data, and congestion inducing feature data in combination with the spatial weight values, time sequence weight values, and congestion inducing weight values to obtain traffic congestion preliminary evaluation feature data; inputting the traffic congestion preliminary evaluation feature data into a preset traffic congestion identification model for processing to obtain a traffic congestion preliminary evaluation index; performing analysis and processing on the multi-source traffic identification optimization data set to obtain traffic congestion identification influence factors; correcting the traffic congestion preliminary evaluation index according to the traffic congestion identification influence factors to obtain a traffic congestion correction index; comparing the traffic congestion correction index with a preset dynamic traffic congestion identification threshold value; if the traffic congestion correction index is less than the preset dynamic traffic congestion identification threshold value, determining that the preset traffic section is not congested; if the traffic congestion correction index is greater than or equal to the preset dynamic traffic congestion identification threshold value, determining that the preset traffic section is congested, and identifying the congestion cause; generating a traffic congestion intelligent identification report according to the preset traffic section ID and the congestion cause. 2.The deep learning based intelligent traffic congestion recognition method according to claim 1, characterized in that, The method for obtaining a multi-source traffic identification data set corresponding to a preset traffic section ID and performing data preprocessing to obtain a multi-source traffic identification optimization data set comprises: dividing traffic sections within a preset city range into grids, and assigning grid IDs to obtain a preset traffic section ID; Obtain a multi-source traffic identification data set corresponding to a preset traffic section ID, wherein the motor vehicle traffic operation data includes motor vehicle average speed, motor vehicle average spacing, and freight motor vehicle proportion, the non-motor vehicle traffic operation data includes non-motor vehicle average speed and non-motor vehicle trajectory data, the pedestrian traffic operation data includes pedestrian flow data and pedestrian stay time average, and the environment 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 section associated public service scene operation data, and environment collection data are subjected to spatio-temporal alignment and abnormal value cleaning preprocessing to obtain a multi-source traffic identification optimization data set. 3.The deep learning based intelligent traffic congestion recognition method according to claim 2, characterized in that, The traffic scene label feature data is constructed according to the multi-source traffic identification optimization data set, and is input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data, including: The traffic scene label feature data is constructed according to the event feature data and the real-time rainfall and visibility combined with the data collection timestamp; 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 deep learning based intelligent traffic congestion recognition method according to claim 3, characterized in that, The traffic congestion identification influence factor is obtained by analyzing and processing the multi-source traffic identification optimization data set, including: The road water accumulation data is compared with a preset water accumulation warning value to obtain a road water accumulation over-warning rate; The road water accumulation over-warning rate is combined with the municipal manhole cover operation state data and event feature data for weighted summation processing to obtain a municipal traffic operation influence factor; The real-time rainfall is compared with a preset historical same-period rainfall average to obtain a rainfall over-average rate; The visibility is compared with a preset visibility warning value to obtain a visibility deficiency rate; The rainfall over-average rate and the visibility deficiency rate are weighted and summed to obtain an environmental influence factor; The municipal traffic operation influence factor and the environmental influence factor are normalized and weighted and summed with the freight motor vehicle proportion to obtain a traffic congestion identification influence factor. 5.The deep learning based intelligent traffic congestion recognition method according to claim 4, characterized in that, If the traffic congestion correction index is greater than or equal to a preset traffic congestion identification threshold, it is determined that the preset traffic section is congested, and the congestion cause is identified, including: Data change feature data, running track change feature data, and flow change data are obtained by extracting the data change feature according to the motor vehicle traffic operation data in a preset time period; The traffic congestion preliminary evaluation feature data, the speed change feature data, the running track change feature data, and the flow change data are input into a preset traffic congestion cause identification model for analysis and processing together with the municipal traffic operation influence factor, the environmental influence factor, and the freight motor vehicle proportion to obtain the congestion cause; The congestion cause includes flow congestion, event congestion, or facility abnormality congestion.

6. The intelligent traffic congestion recognition system based on deep learning, characterized in that, A memory and a processor are included, the memory includes a deep learning-based traffic congestion intelligent identification method program, and the deep learning-based traffic congestion intelligent identification method program is implemented when the processor is executed, including the following steps: The multi-source traffic identification data set corresponding to the preset traffic section ID is acquired, and data preprocessing is performed to obtain a multi-source traffic identification optimization data set, including motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, municipal traffic operation data, section-related public service scene operation data, and environment collection data, wherein the municipal traffic operation data includes road water accumulation data, municipal manhole cover operation state data, and event feature data, and the section-related public service scene operation data includes public service peak period, vehicle entry and exit frequency, and vehicle stay time average value; Traffic scene label feature data is constructed according to the multi-source traffic identification optimization data set, and is input into a preset traffic scene category classification model for analysis and processing to obtain dynamic traffic scene category feature data; The motor vehicle traffic operation data, non-motor vehicle traffic operation data, pedestrian traffic operation data, and section-related public service scene operation data are subjected to spatial feature extraction, time sequence feature extraction, and congestion inducing feature extraction to obtain spatial feature data, time sequence feature data, and congestion inducing feature data; According to the dynamic traffic scene category feature data, a preset traffic scene category and feature weight value mapping table is queried to obtain weight values corresponding to the spatial feature data, time sequence feature data, and congestion inducing feature data, including spatial weight values, time sequence weight values, and congestion inducing weight values; The spatial feature data, time sequence feature data, and congestion inducing feature data are combined with the spatial weight values, time sequence weight values, and congestion inducing weight values for data fusion processing to obtain traffic congestion preliminary evaluation feature data; The traffic congestion preliminary evaluation feature data is input into a preset traffic congestion identification model for processing to obtain a traffic congestion preliminary evaluation index; According to the multi-source traffic identification optimization data set, analysis and processing are performed to obtain traffic congestion identification influence factors; According to the traffic congestion identification influence factors, the traffic congestion preliminary evaluation index is corrected to obtain a traffic congestion correction index; The traffic congestion correction index is compared with a preset dynamic traffic congestion identification threshold value; If the traffic congestion correction index is less than the preset dynamic traffic congestion identification threshold value, it is determined that the preset traffic section is not congested; If the traffic congestion correction index is greater than or equal to the preset dynamic traffic congestion identification threshold value, it is determined that the preset traffic section is congested, and the congestion cause is identified; A traffic congestion intelligent identification report is generated according to the preset traffic section ID and the congestion cause. 7.The deep learning based intelligent traffic congestion recognition system according to claim 6, characterized in that, The multi-source traffic identification data set corresponding to the preset traffic section ID is acquired, and data preprocessing is performed to obtain a multi-source traffic identification optimization data set, including: The traffic sections within a preset city range are subjected to grid division, and grid IDs are assigned to obtain a preset traffic section ID; The multi-source traffic identification data set corresponding to the preset traffic section ID is acquired, wherein the motor vehicle traffic operation data includes motor vehicle average speed, motor vehicle average distance, and freight motor vehicle proportion, the non-motor vehicle traffic operation data includes non-motor vehicle average speed and non-motor vehicle trajectory data, the pedestrian traffic operation data includes passenger flow data and pedestrian stay time average value, and the environment collection data includes real-time rainfall and visibility; The motor vehicle traffic operation data, the non-motor vehicle traffic operation data, the pedestrian traffic operation data, the municipal traffic operation data, the road section associated public service scene operation data and the environment collection data are spatio-temporally aligned and abnormally value cleaned for preprocessing, to obtain a multi-source traffic recognition optimization dataset. 8.The deep learning based intelligent traffic congestion recognition system according to claim 7, characterized in that, The traffic scene label feature data is constructed according to the multi-source traffic recognition optimization dataset, and is input into a preset traffic scene category classification model for analysis and processing, to obtain dynamic traffic scene category feature data, including: The traffic scene label feature data is constructed according to the event feature data and the real-time rainfall and visibility combined data collection time stamp; 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.

Images