Traffic jam monitoring analysis method and platform adopting deep learning

Abstract

The invention discloses a traffic jam monitoring analysis method and platform adopting deep learning, and relates to the technical field of intelligent traffic systems, and the method comprises the steps: collecting the input and output flow data of different types of vehicles at an upstream road section, building a flow prediction model, and predicting the flow of a to-be-analyzed road section through the model. The flow data specifically comprise flow data of miniature vehicles, small vehicles, compact vehicles, medium-sized vehicles, medium-large vehicles and large vehicles. Whether a congestion threshold value is reached or not is judged by calculating the ratio (traffic congestion ratio) of the predicted flow to the road section capacity, traffic congestion is identified when the threshold value is reached, and finally congestion information is sent to a traffic monitoring client to be processed. The technical problems of low prediction accuracy and poor real-time performance caused by the fact that multi-source data cannot be fully utilized in the existing traffic monitoring technology are solved, and the technical effect of improving the accuracy and the real-time performance of traffic jam monitoring is achieved.

Description

Technical Field

[0001] This application relates to the technical field of intelligent transportation systems, and particularly to a traffic congestion monitoring and analysis method and platform using deep learning. Background Art

[0002] With the acceleration of the urbanization process, the traffic flow has increased, and the problem of traffic congestion has become increasingly serious. Especially during peak hours and under special weather conditions, traffic jams have brought great challenges to people's travel and urban traffic management. Traditional traffic monitoring methods usually rely on single traffic flow monitoring devices, unable to fully consider the influence of various factors such as weather, traffic incidents, and vehicle types, resulting in low prediction accuracy and poor real-time performance. In addition, traditional methods often rely on manually set thresholds and simple rules, lacking dynamic adaptability and unable to respond to traffic changes in a timely manner, further exacerbating the traffic congestion problem. Therefore, how to comprehensively utilize multi-source data and apply advanced analysis technologies to improve prediction accuracy has become an urgent problem to be solved in the field of traffic congestion management.

[0003] In the related technologies at the present stage, there are technical problems in traffic monitoring technologies that multi-source data cannot be fully utilized, resulting in low prediction accuracy and poor real-time performance. Summary of the Invention

[0004] This application solves the technical problems in the existing traffic monitoring technologies that multi-source data cannot be fully utilized, resulting in low prediction accuracy and poor real-time performance by providing a traffic congestion monitoring and analysis method and platform using deep learning.

[0005] This application provides a traffic congestion monitoring and analysis method using deep learning, including: Obtain the upstream section inflow of the section to be analyzed, where the upstream section inflow includes the upstream inflow information of mini cars, small cars, compact cars, medium-sized cars, large medium-sized cars, and large cars; obtain the upstream section outflow of the section to be analyzed, where the upstream section outflow includes the upstream outflow information of mini cars, small cars, compact cars, medium-sized cars, large medium-sized cars, and large cars; obtain the traffic flow prediction model of the section to be analyzed, where the topological structure of the traffic flow prediction model is the same as the traffic topology, and the input nodes of the traffic flow prediction model are the upstream traffic flow inlet and the upstream traffic flow outlet; input the upstream inflow information of mini cars, the upstream inflow information of small cars, the upstream inflow information of compact cars, the upstream inflow information of medium-sized cars, the upstream inflow information of large medium-sized cars, the upstream inflow information of large cars, the upstream outflow information of mini cars, the upstream outflow information of small cars, the upstream outflow information of compact cars, the upstream outflow information of medium-sized cars, the upstream outflow information of large medium-sized cars, and the upstream outflow information of large cars into the input nodes, and output the predicted traffic flow of the section to be analyzed; calculate the ratio of the predicted traffic flow of the section to be analyzed to the section capacity threshold, which is set as the traffic congestion ratio; when the traffic congestion ratio is greater than or equal to the traffic congestion ratio threshold, perform traffic congestion identification on the section to be analyzed and send it to the traffic congestion monitoring client.

[0006] The present application provides a traffic congestion monitoring and analysis platform using deep learning, including: Upstream section inflow acquisition module, the upstream section inflow acquisition module is used to obtain the upstream section inflow of the section to be analyzed. Among them, the upstream section inflow includes mini - car upstream inflow information, small - car upstream inflow information, compact - car upstream inflow information, medium - car upstream inflow information, medium - large - car upstream inflow information, and large - car upstream inflow information; upstream section outflow acquisition module, the upstream section outflow acquisition module is used to obtain the upstream section outflow of the section to be analyzed. Among them, the upstream section outflow includes mini - car upstream outflow information, small - car upstream outflow information, compact - car upstream outflow information, medium - car upstream outflow information, medium - large - car upstream outflow information, and large - car upstream outflow information; flow prediction model acquisition module, the flow prediction model acquisition module is used to obtain the flow prediction model of the section to be analyzed. Among them, the topological structure of the flow prediction model is the same as the traffic topology, and the input nodes of the flow prediction model are the upstream section flow inflow intersection and the upstream section flow outflow intersection; predicted flow output module of the section to be analyzed, the predicted flow output module of the section to be analyzed is used to input the mini - car upstream inflow information, the small - car upstream inflow information, the compact - car upstream inflow information, the medium - car upstream inflow information, the medium - large - car upstream inflow information, the large - car upstream inflow information, the mini - car upstream outflow information, the small - car upstream outflow information, the compact - car upstream outflow information, the medium - car upstream outflow information, the medium - large - car upstream outflow information, and the large - car upstream outflow information into the input nodes, and output the predicted flow of the section to be analyzed; ratio calculation module, the ratio calculation module is used to calculate the ratio of the predicted flow of the section to be analyzed to the section capacity threshold, which is set as the traffic congestion ratio; traffic congestion identification module, the traffic congestion identification module is used to identify traffic congestion for the section to be analyzed when the traffic congestion ratio is greater than or equal to the traffic congestion ratio threshold, and send it to the traffic congestion monitoring client.

[0007] It is intended to propose a traffic congestion monitoring and analysis method and platform using deep learning through this application. First, by collecting the inflow and outflow flow data of different types of vehicles in the upstream section, a flow prediction model is established, and this model is used to predict the flow of the section to be analyzed. Specifically, it includes the flow data of mini - cars, small - cars, compact - cars, medium - cars, medium - large - cars, and large - cars. By calculating the ratio of the predicted flow to the section capacity (traffic congestion ratio), it is judged whether the congestion threshold is reached. When the threshold is reached, traffic congestion is identified, and finally the congestion information is sent to the traffic monitoring client for processing, achieving the technical effect of improving the accuracy and real - time performance of traffic congestion monitoring. Brief Description of the Drawings

[0008] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the platform according to the embodiments of the present application. It should be understood that the operations above or below do not necessarily need to be executed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. At the same time, other operations can also be added to these processes, or one or more steps can be removed from these processes.

[0009] Figure 1 It is a schematic flowchart of a traffic congestion monitoring and analysis method using deep learning provided by an embodiment of the present application; Figure 2 It is a schematic structural diagram of a traffic congestion monitoring and analysis platform using deep learning provided by an embodiment of the present application.

[0010] Explanation of reference numerals: Upstream section inflow acquisition module 10, upstream section outflow acquisition module 20, traffic flow prediction model acquisition module 30, predicted traffic flow output module 40 for the section to be analyzed, ratio calculation module 50, traffic congestion identification module 60. Detailed implementation manners

[0011] The above description is only an overview of the technical solutions of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of this application more obvious and understandable, the following specifically gives the detailed implementation manners of this application.

[0012] In order to make the purpose, technical solutions and advantages of this application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limitations of the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0013] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments. However, it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. The terms "first" and "second" involved are only used to distinguish similar objects and do not represent a specific order for the objects. The terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, platform, product or server comprising a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or modules not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art of the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application.

[0014] The embodiments of this application provide a traffic congestion monitoring and analysis method using deep learning, as Figure 1 shown, the method includes: Step S100, obtaining the upstream road section inflow of the road section to be analyzed, where the upstream road section inflow includes the upstream inflow information of mini cars, small cars, compact cars, medium-sized cars, medium and large-sized cars, and large cars.

[0015] Specifically, the road section to be analyzed refers to any road section in the road network. The user terminal can delineate the road section to be analyzed and its upstream road section according to the map software or traffic planning map. After comprehensively analyzing the road conditions, geomagnetic sensors, video monitoring, and electronic license plate recognition systems are installed at key positions.

[0016] Furthermore, when a vehicle passes by, the geomagnetic sensor obtains basic traffic flow data. The video monitoring distinguishes vehicle types based on image recognition technology and correlates them with the geomagnetic data. The electronic license plate recognition system calibrates and supplements vehicle information, so as to accurately obtain the upstream road section inflow information of each vehicle type. The inflow is the vehicle inflow at the intersection where vehicles enter the main road upstream of the road section to be analyzed. Specifically, the vehicle types at each intersection include mini cars, small cars, compact cars, medium-sized cars, medium and large-sized cars, and large cars. Preferably, the vehicle type classification standard in the embodiments of this application can be the conventional vehicle type classification standard, or the user terminal can classify vehicles according to the vehicle size by itself, and there is no limitation on this. Since different vehicle types will have different impacts on the traffic flow of the road section to be analyzed, the traffic flow in the embodiments of this application is subdivided into each vehicle type, thereby improving the refinement degree of road network traffic flow monitoring.

[0017] Step S200, obtain the upstream section outflow of the section to be analyzed, where the upstream section outflow includes the upstream outflow information of mini-cars, small cars, compact cars, medium-sized cars, large-medium-sized cars, and large cars.

[0018] Specifically, clarify the section to be analyzed and its upstream section. The former is a specific road section selected for analyzing traffic conditions, and the latter is the part before flowing into the section to be analyzed. Then, use the traffic geographic information system to plan the traffic flow monitoring area and select key nodes such as intersections and lane merging and diverging points. Next, bury geomagnetic sensors at key points of each lane in the monitoring area to record the number of vehicles passing by; install high-definition video cameras at high places around to capture video images; install an electronic license plate recognition system at the entrance or key nodes to obtain vehicle identity and type information, where the types cover the upstream outflow information of mini-cars, small cars, compact cars, medium-sized cars, large-medium-sized cars, and large cars. After that, the geomagnetic sensors send the data to the data processing center for preprocessing; the video server uses deep learning algorithms to detect and classify the vehicles in the video images and associate and fuse them with the geomagnetic data; the data processing center then corrects and supplements with the electronic license plate information, and finally accurately obtains the upstream outflow information of each vehicle type. The outflow is the vehicle outflow at the intersection where vehicles flow out to the upstream main road of the section to be analyzed.

[0019] Step S300, obtain the traffic flow prediction model of the section to be analyzed, where the topological structure of the traffic flow prediction model is the same as the traffic topology, and the input nodes of the traffic flow prediction model are the upstream section inflow intersection and the upstream section outflow intersection.

[0020] Specifically, use the geographic information system data, combine the historical data of traffic flow monitoring devices and on-site traffic surveys, and accurately draw the traffic topology map of the section to be analyzed and its upstream, determine the road network layout and the surrounding traffic facilities, and analyze the temporal and spatial change trends of traffic flow; further, based on the topology map, construct a traffic flow prediction model topological structure consistent with it, set the upstream section inflow intersection and the outflow intersection as input nodes, and at the same time analyze the directionality of traffic flow and lane usage rules to set the node connection relationship. Then, by arranging monitoring devices such as geomagnetic sensors and video cameras at intersections, collect the traffic flow parameters of each vehicle type as the parameter values of the input nodes, so that the model can accurately predict the traffic flow situation of the section to be analyzed according to the real-time traffic conditions and provide strong support for traffic management decisions.

[0021] In a possible implementation, a traffic flow prediction model for the road section to be analyzed is obtained. The topological structure of the traffic flow prediction model is the same as the traffic topology. The input nodes of the traffic flow prediction model are the upstream road section traffic inflow intersection and the upstream road section traffic outflow intersection. Step S300 further includes step S310 of configuring the upstream road section analysis distance.

[0022] Specifically, the upstream road section analysis distance refers to a preset road range extending upstream from the starting point of the road section to be analyzed. The traffic flow outside the upstream road section analysis distance has little impact on the traffic flow of the road section to be analyzed, so it does not need to be considered.

[0023] Step S320: Based on the upstream road section analysis distance, perform upstream road section topological segmentation on the road section to be analyzed to obtain the upstream road section traffic topology.

[0024] Specifically, the upstream road section analysis distance is a specific range extending upstream from the starting point of the road section to be analyzed. The upstream road section traffic topology refers to the road network structure within a specific range extending upstream from the starting point of the road section to be analyzed. Specifically, first extract the road map data within the upstream road section analysis distance through a geographic information system, extract the road distribution data, the inflow intersection distribution data, and the outflow intersection distribution data, and construct and generate the upstream road section traffic topology.

[0025] Step S330: Perform neural network topology simulation according to the upstream road section traffic topology to obtain the traffic flow prediction model topology. The input nodes of the traffic flow prediction model topology are the upstream road section traffic inflow intersection and the upstream road section traffic outflow intersection.

[0026] Specifically, neural network topology simulation is a process of simulating the upstream road section traffic topology with a neural network topology. The traffic flow prediction model topology is a graph neural network topology generated by simulating the upstream road section traffic topology through neural network topology simulation. At the same time, the upstream road section traffic inflow intersection and the upstream road section traffic outflow intersection are used as the input nodes of the traffic flow prediction model topology. Among them, the upstream road section traffic inflow intersection refers to the intersection in the upstream road section traffic topology where vehicles enter the main road of the road section to be analyzed, and the upstream road section traffic outflow intersection refers to the intersection in the upstream road section traffic topology where vehicles enter the main road of the road section to be analyzed.

[0027] By configuring the traffic flow prediction model topology as the graph neural topology structure for subsequent training of the traffic flow prediction model, the impact of upstream traffic flow on the traffic flow of the road section to be analyzed can be analyzed, and thus accurate monitoring and control of traffic flow can be achieved.

[0028] Step S340: Collect the traffic topology flow monitoring record data of the upstream section, where the flow monitoring record data includes the upstream inflow record information of minicars, small cars, compact cars, medium-sized cars, mid-large cars, and large cars, the upstream outflow record information of minicars, small cars, compact cars, medium-sized cars, mid-large cars, and large cars, and the summary flow record information of the section to be analyzed.

[0029] Specifically, the traffic topology of the upstream section represents the road network structure information of the upstream section; the flow monitoring record data is a set of flow data of various vehicles in different directions collected within a specific period. Key monitoring points such as intersections and merging / splitting points are determined based on the traffic topology of the upstream section, and geomagnetic sensors, high-definition cameras, and electronic license plate recognition systems are installed and debugged and calibrated to establish a data collection system. During data collection, each device operates in a predetermined manner. The geomagnetic sensor records the passing signals and quantities of vehicles, the camera identifies the vehicle types and passing situations through image recognition, and the electronic license plate recognition system obtains the detailed vehicle information, and then classifies and records them in the corresponding data tables according to vehicle types and flow directions.

[0030] Step S350: Use the summary flow record information of the section to be analyzed as the supervised data, and use the upstream inflow record information of minicars, small cars, compact cars, medium-sized cars, mid-large cars, and large cars, the upstream outflow record information of minicars, small cars, compact cars, medium-sized cars, mid-large cars, and large cars as the input data of the input nodes to train the traffic prediction model topology and obtain the traffic prediction model.

[0031] Specifically, the summarized traffic flow record information of the road section to be analyzed is the overall vehicle traffic historical data of this road section, which is the summarized traffic flow record data of the road section to be analyzed affected by the upstream traffic flow record information of each vehicle type. It is set as the supervised data. The upstream traffic flow record information of each vehicle type details the traffic flow of different vehicle types at each intersection. Preferably, the input data of each intersection is a multi-dimensional array, and the dimension of each array is the number of vehicle type categories, serving as the input node data. The summarized traffic flow record information of the road section to be analyzed and the upstream traffic flow record information of each vehicle type are set as the data for constructing the traffic flow prediction model. The data for constructing the traffic flow prediction model is divided into a training set, a validation set, and a test set according to the ratio of 8:1:1. Then, the training set data is used to train the topology of the traffic flow prediction model, and at the same time, the error between the predicted value and the true value is calculated. When in 1000 consecutive trainings, at least 990 times the error between the predicted value and the true value is less than or equal to the error threshold preset by the user, the test set is used to evaluate the model performance. If in 100 consecutive tests, at least 95 times the error between the predicted value and the true value is less than or equal to the error threshold preset by the user, it is considered convergent, and a traffic flow prediction model is generated. The trained model can fully explore the relationship between traffic flow data, improve the prediction accuracy and refinement degree, and can quickly predict the traffic flow of the road section to be analyzed based on the upstream real-time monitoring data, providing a basis for traffic management decisions.

[0032] In a possible implementation manner, when configuring the analysis distance of the upstream road section, step S310 further includes: Step S311, perform the configuration of the analysis distance without an outlet intersection to obtain the initial analysis distance of the upstream road section.

[0033] Step S312, based on the road section to be analyzed, collect the number of traffic flow outlet intersections within the initial analysis distance of the upstream road section.

[0034] Step S313, perform the analysis distance configuration according to the number of traffic flow outlet intersections to obtain the analysis distance of the upstream road section.

[0035] Specifically, the analysis distance configuration without an outflow intersection refers to the process of analyzing the maximum distance that affects the traffic flow of the road section to be analyzed when there is no outflow intersection. The initial upstream road section analysis distance is the maximum distance that affects the traffic flow of the road section to be analyzed when there is no outflow intersection obtained by performing the analysis distance configuration without an outflow intersection. Further, the road section to be analyzed refers to the road area that needs to be monitored and analyzed for road traffic flow, which can be any road range on the traffic road. The number of flow outflow intersections refers to the total number of intersections where vehicles drive out and flow to other areas within the initial upstream road section analysis distance starting from the starting point of the road section to be analyzed. The upstream road section analysis distance is a specific length range extending upstream from the starting point of the road section to be analyzed. When the upstream road traffic flow beyond the upstream road section analysis distance has little impact on the road traffic flow of the road section to be analyzed, only the road traffic flow data within the range of the upstream road section analysis distance extending upstream from the starting point of the road section to be analyzed is statistically analyzed for the traffic flow of the road section to be analyzed.

[0036] Furthermore, the above process involves two execution steps. One is the analysis distance configuration without an outflow intersection, and the other is the analysis distance configuration based on the number of flow outflow intersections. The two execution algorithm processes are exactly the same, only the data screening conditions are different. The analysis distance configuration without an outflow intersection screens data samples under the condition of no outflow intersection, while the analysis distance configuration based on the number of flow outflow intersections collects data samples with the same number of flow outflow intersections within the initial upstream road section analysis distance. The other execution steps are exactly the same. The embodiments of the present application take the analysis distance configuration without an outflow intersection as an example for elaboration, as follows: In a possible implementation manner, when performing the analysis distance configuration without an outflow intersection to obtain the initial upstream road section analysis distance, step S311 further includes step S3111 of obtaining the road section traffic flow monitoring record data without an outflow intersection. Among them, the road section traffic flow monitoring record data includes the aggregated road section traffic flow record data and the upstream road section traffic flow record data. The upstream road section traffic flow record data has a record distance label, a record speed average label, and a first record time label. The aggregated road section traffic flow record data has a second record time label. The second record time label is later than the first record time label, and the time deviation between the second record time label and the first record time label is equal to the ratio of the record distance label to the record speed average label.

[0037] Specifically, collect the road segment flow monitoring data of the road segment to be analyzed without an upstream outlet, and store it as road segment flow monitoring record data. The road segment flow monitoring record data includes the summary road segment flow record data and the upstream road segment flow record data of the road segment to be analyzed. Preferably, the upstream road segment flow data has a record distance tag representing the distance from the starting point of the road segment to be analyzed, a speed mean tag of the vehicle at the corresponding position, and a first record time tag at the collection time. The summary road segment flow data has a second record time tag later than the former, and the deviation between the first record time tag and the second record time tag is equal to the ratio of the record distance to the speed mean. In this way, it can be considered that the summary road segment flow record data is affected by the upstream road segment flow record data. By collecting the upstream road segment flow record data at different positions, it can provide basic data for analyzing the distance threshold that affects the summary road segment flow record data of the road segment to be analyzed later, so as to obtain the initial upstream road segment analysis distance.

[0038] Step S3112: According to the upstream road segment flow record data, based on the flow deviation threshold, perform clustering analysis on the road segment flow monitoring record data to obtain multiple clusters of road segment flow monitoring record data.

[0039] Specifically, the flow deviation threshold refers to a preset flow deviation threshold value. When the upstream road segment flow deviation is greater than the flow deviation threshold, it is considered to have a flow deviation; otherwise, it is considered not to have a flow deviation. When there is no flow deviation, the corresponding two road segment flow monitoring record data can be added as the same cluster of road segment flow monitoring record data. After repeated analysis, when all the road segment flow monitoring record data have been pairwise compared, output multiple clusters of road segment flow monitoring record data. In each cluster of road segment flow monitoring record data, the upstream road segment flow is considered to have no deviation, which is equivalent to a quantitative value. At this time, the variables are only the summary road segment flow record data and the record distance tag. Therefore, the relationship between the summary road segment flow record data and the record distance tag can be analyzed.

[0040] Step S3113: Traverse the multiple clusters of road segment flow monitoring record data, and perform correlation analysis on the summary road segment flow record data and the record distance tag to obtain the first record distance correlation degree, the second record distance correlation degree until the Qth record distance correlation degree.

[0041] Specifically, the record distance correlation represents the correlation between the aggregated road segment traffic record data and the record distance tags. Q represents the total number of record distance tags triggered by the multi-cluster road segment traffic monitoring record data. Traverse the multi-cluster road segment traffic monitoring record data, analyze the correlation between the aggregated road segment traffic record data in each cluster of road segment traffic monitoring record data and the record distance tags, and obtain the first record distance correlation, the second record distance correlation up to the Qth record distance correlation. The analysis and acquisition processes of the first record distance correlation, the second record distance correlation up to the Qth record distance correlation are the same. Taking the first record distance correlation as an example, the details are as follows: In a possible implementation, traverse the multi-cluster road segment traffic monitoring record data, perform a correlation analysis on the aggregated road segment traffic record data and the record distance tags, and obtain the first record distance correlation, the second record distance correlation up to the Qth record distance correlation. Step S3113 further includes step S31131, according to the multi-cluster road segment traffic monitoring record data, extract the first cluster of road segment traffic monitoring record data.

[0042] Specifically, the first cluster of road segment traffic monitoring record data refers to any cluster of data belonging to the multi-cluster road segment traffic monitoring record data set, which is an exemplary cluster used to illustrate the correlation analysis and does not represent a specific cluster.

[0043] Step S31132, according to the first cluster of road segment traffic monitoring record data, extract the first aggregated road segment traffic record data, perform a dimensionless processing, and generate a reference data sequence.

[0044] Specifically, from the selected first cluster of road segment traffic monitoring record data, identify and extract the first aggregated road segment traffic record data therein; in order to enable data of different magnitudes to be compared and analyzed on the same scale, use a standardized dimensionless method, preferably Min-Max standardization (map the data value to the interval from 0 to 1), process the aggregated road segment traffic data, so as to obtain a dimensionless reference data sequence that can highlight the relative change trend of the traffic, so as to perform correlation analysis with other data subsequently.

[0045] Step S31133, according to the first aggregated road segment traffic record data, extract the corresponding first record distance tags from the first cluster of road segment traffic monitoring record data, perform a dimensionless processing, and generate a comparison data sequence.

[0046] Specifically, since the first recorded distance tags correspond one-to-one between the first cluster of road segment traffic monitoring record data and the first summarized road segment traffic record data, the first recorded distance tags corresponding one-to-one are extracted from the first cluster of road segment traffic monitoring record data according to the first summarized road segment traffic record data, and a dimensionless processing is performed to generate a comparison data sequence. Using the same method as the dimensionless processing of traffic data, the first recorded distance tags are dimensionless processed to convert them into a dimensionless comparison data sequence with a unified comparison scale, enabling effective correlation analysis with the previously generated reference data sequence.

[0047] Step S31134, perform grey relational analysis according to the comparison data sequence and the reference data sequence to obtain the first recorded distance correlation degree.

[0048] Specifically, the generated comparison data sequence (i.e., the dimensionless first recorded distance tag sequence) and the reference data sequence (i.e., the dimensionless first summarized road segment traffic record data sequence) are used as inputs, and the grey relational analysis method is applied. The grey relational analysis method calculates the correlation coefficients between the two sequences at each time point, and then combines the correlation coefficients to obtain the overall correlation degree between the two sequences, that is, the first recorded distance correlation degree. The correlation degree value will quantitatively reflect the tightness and mutual influence relationship between the summarized road segment traffic and the recorded distance in the first cluster of road segment traffic monitoring record data, providing important data support for subsequent further analysis and decision-making.

[0049] Step S3114, according to the first recorded distance correlation degree, the second recorded distance correlation degree until the Qth recorded distance correlation degree, extract the maximum distance greater than or equal to the correlation degree threshold, and set it as the initial upstream road segment analysis distance.

[0050] Specifically, the correlation degree threshold refers to the grey relational analysis threshold regarded as non-correlated, which is user-defined. If not configured, the default grey relational analysis threshold is equal to 0.2. Therefore, extracting the maximum distance of the recorded distance greater than or equal to the correlation degree threshold is regarded as the maximum influence distance without an inflow intersection and is stored as the initial upstream road segment analysis distance. In the same way, the upstream road segment analysis distance can be obtained.

[0051] By determining the upstream road segment analysis distance, the subsequent road traffic collection range is determined, avoiding the collection of redundant road traffic data, thus ensuring the calculation efficiency.

[0052] In a possible implementation, the aggregated traffic record information of the road section to be analyzed is used as the supervision data, and the upstream inflow record information of the minicar, the upstream inflow record information of the small car, the upstream inflow record information of the compact car, the upstream inflow record information of the medium-sized car, the upstream inflow record information of the mid-large car, and the upstream inflow record information of the large car, the upstream outflow record information of the minicar, the upstream outflow record information of the small car, the upstream outflow record information of the compact car, the upstream outflow record information of the medium-sized car, the upstream outflow record information of the mid-large car, and the upstream outflow record information of the large car are used as the input data of the input nodes to train the topology of the traffic prediction model, and the traffic prediction model is obtained. Step S350 further includes step S351 of performing weight assignment on the upstream inflow of the minicar, the upstream inflow of the small car, the upstream inflow of the compact car, the upstream inflow of the medium-sized car, the upstream inflow of the mid-large car, and the upstream inflow of the large car, the upstream outflow of the minicar, the upstream outflow of the small car, the upstream outflow of the compact car, the upstream outflow of the medium-sized car, the upstream outflow of the mid-large car, and the upstream outflow of the large car to obtain a weight distribution result, wherein the weight distribution result is weighted based on the Delphi weighting method.

[0053] Specifically, invite experts, scholars in the traffic field, and traffic engineers with rich practical experience to form an expert panel, ensuring that the panel members have an in-depth understanding of the traffic flow characteristics of different vehicle types and can make professional judgments on the importance of the upstream inflow and outflow of various vehicle types. Design a special questionnaire to ask the expert panel about their views on the relative importance of the upstream inflow and the corresponding outflow of the minicar, the small car, the compact car, the medium-sized car, the mid-large car, and the large car. After each round of investigation, summarize the opinions of the experts and feedback them to them, allowing the experts to refer to the opinions of other members in the subsequent rounds, re-examine and adjust their judgments. After multiple rounds of repeated consultations until the opinions of the experts tend to be stable and consistent. Organize and statistically analyze the importance scores of the upstream and downstream traffic of various vehicle types finally determined by the experts. Through the arithmetic mean method, etc., convert the scores of the experts into the weight values of the upstream and downstream traffic of each vehicle type, so as to obtain a weight distribution result based on the Delphi weighting method, clarify the relative importance of the traffic of different vehicle types in the overall traffic flow analysis, and provide key weight parameters for subsequent model training.

[0054] In step S352, after performing input weight configuration on the input nodes according to the weight distribution result, using the aggregated traffic flow record information of the road section to be analyzed as the supervision data, and using the upstream inflow record information of mini-cars, the upstream inflow record information of small cars, the upstream inflow record information of compact cars, the upstream inflow record information of medium-sized cars, the upstream inflow record information of large-medium-sized cars, the upstream inflow record information of large cars, the upstream outflow record information of mini-cars, the upstream outflow record information of small cars, the upstream outflow record information of compact cars, the upstream outflow record information of medium-sized cars, the upstream outflow record information of large-medium-sized cars, and the upstream outflow record information of large cars as the input data of the input nodes, train the traffic flow prediction model topology to obtain the traffic flow prediction model.

[0055] Specifically, based on the previously obtained weight distribution result, perform corresponding input weight configuration on the input nodes. Set the upstream inflow record information of different vehicle types (including mini-cars, small cars, compact cars, medium-sized cars, large-medium-sized cars, and large cars) and the upstream outflow record information according to the weights assigned to each, so that during the model training process, the traffic flow data of different vehicle types can have different degrees of influence on the model according to their importance levels. Use the aggregated traffic flow record information of the road section to be analyzed as the supervision data. The supervision data represents the total traffic flow situation actually occurring on the road section to be analyzed and is the target value used to measure the prediction accuracy during the model training process. At the same time, use the upstream inflow and outflow record information of various vehicle types as the input data of the input nodes to provide detailed traffic flow characteristic information for the model, enabling the model to learn the change relationship of traffic flow of different vehicle types between upstream and downstream and their comprehensive influence mechanism on the aggregated traffic flow of the road section to be analyzed. Adopt a neural network algorithm to construct the traffic flow prediction model topology. Input the input data with configured weights into the model, and through continuously adjusting the internal parameters of the model (such as the weights and biases of the neural network, etc.), with the goal of minimizing the error between the model prediction result and the supervision data (the aggregated traffic flow record of the road section to be analyzed), perform multiple iterative trainings. During the training process, continuously monitor the performance indicators of the model, such as the mean square error, mean absolute error, etc., and optimize and adjust the model according to the indicators until the model reaches better prediction performance, and finally obtain a traffic flow prediction model that can accurately predict the traffic flow of the road section to be analyzed, providing strong tool support for the real-time monitoring and future trend prediction of traffic flow, helping the traffic management department formulate countermeasures in advance, optimize traffic resource allocation, and alleviate traffic congestion.

[0056] Step S400: Input the upstream inflow information of the minicar, the upstream inflow information of the small car, the upstream inflow information of the compact car, the upstream inflow information of the medium-sized car, the upstream inflow information of the mid-large car, and the upstream inflow information of the large car, the upstream outflow information of the minicar, the upstream outflow information of the small car, the upstream outflow information of the compact car, the upstream outflow information of the medium-sized car, the upstream outflow information of the mid-large car, and the upstream outflow information of the large car into the input node, and output the predicted flow of the section to be analyzed.

[0057] Specifically, first, collect the upstream inflow and outflow information of minicars, small cars, compact cars, medium-sized cars, mid-large cars, and large cars covering different time periods from the traffic monitoring system; construct the traffic flow data of various vehicles at each intersection into a multi-dimensional array and input it into the input node of the traffic flow prediction model. Process the input data through the topology of the traffic flow prediction model, comprehensively analyze the mutual relationship of the traffic flows of various vehicle types and their impact on the downstream section, and finally output the predicted flow value of the section to be analyzed. The predicted flow value is presented in digital form or visually displayed.

[0058] Step S500: Calculate the ratio of the predicted flow of the section to be analyzed to the section capacity threshold, and set it as the traffic congestion ratio.

[0059] Specifically, divide the predicted flow of the section to be analyzed by the section capacity threshold to obtain the traffic congestion ratio. The traffic congestion ratio reflects the traffic congestion degree of the section, providing a quantitative index and decision-making basis for the traffic management department to evaluate congestion, formulate traffic diversion strategies, and take preventive measures.

[0060] Step S600: When the traffic congestion ratio is greater than or equal to the traffic congestion ratio threshold, mark the section to be analyzed as congested and send it to the traffic congestion monitoring client.

[0061] Specifically, an expert presets a traffic congestion ratio threshold for each section of the road. For example, the threshold for the urban arterial road is set to 0.8. When the traffic congestion ratio is greater than or equal to the threshold, the system automatically marks the section as a congested section in the database, records information such as the start time of congestion and the congestion ratio, and sends it to the traffic congestion monitoring client through communication protocols and interfaces, such as the command center display screen, the traffic police mobile law enforcement terminal, and the intelligent transportation APP, providing a basis and reference for traffic management and public travel.

[0062] In a possible implementation manner, when the traffic congestion ratio is greater than or equal to the traffic congestion ratio threshold, mark the section to be analyzed as congested and send it to the traffic congestion monitoring client. Step S600 further includes Step S610: When the traffic congestion ratio is less than the traffic congestion ratio threshold, obtain the monitoring image information of the section to be analyzed.

[0063] Specifically, the intelligent transportation system set up in the traffic management center continuously calculates the traffic congestion ratio calculation results for each road section. Once it is found that the traffic congestion ratio of a road section to be analyzed is lower than the preset traffic congestion ratio threshold, the system will automatically start the image acquisition program. By establishing a connection with the high-definition monitoring cameras arranged along the road section, the current monitoring image information is extracted from the storage device or real-time video stream of the cameras. The layout of the cameras should ensure that key areas of the road section can be covered in all directions and from multiple angles, such as intersections, lane merging and diverging points, curves and slopes prone to accidents, etc., so as to obtain image data that can accurately reflect the details of the road traffic conditions and provide intuitive visual materials for subsequent traffic bottleneck analysis.

[0064] Step S620, perform traffic bottleneck analysis based on the monitoring image information to obtain a traffic bottleneck binary value.

[0065] Specifically, then, the obtained monitoring image information is transmitted to a dedicated image analysis and processing module. The image analysis and processing module uses the algorithm model of image recognition technology to identify and analyze traffic elements in the image. Detect the driving state of vehicles (whether queuing, whether the vehicle speed is extremely slow), road conditions (whether there are obstacles, whether the lanes are normally passable), traffic signal states (whether the signal lights are working properly, whether the timing is reasonable), and other factors that may affect traffic flow (such as roadside parking, pedestrians jaywalking, etc.). According to the detection results, through preset rules and scoring systems, the traffic bottleneck binary value is calculated. This value simplifies the traffic conditions into two states: there is a bottleneck situation that may lead to a reduction in traffic flow or a decrease in flow velocity, and a normal state where there is no such situation, so as to facilitate subsequent rapid judgment and decision-making.

[0066] Step S630, where when the monitoring image information shows factors that cause a reduction in traffic flow or a decrease in traffic flow velocity, the traffic bottleneck binary value is in the bottleneck trigger state.

[0067] Specifically, during the traffic bottleneck analysis process, if the image analysis module identifies obvious factors that will cause a reduction in traffic flow or a decrease in flow velocity, such as road construction areas (construction vehicles and warning signs occupying lanes), traffic accident scenes (vehicle collisions or malfunctions causing lane blockages), illegally parked vehicles hindering normal passage, traffic signal failures causing chaos at intersections, etc., according to the established assignment rules, the traffic bottleneck binary value is set to the bottleneck trigger state, indicating that although the current congestion ratio of this road section does not exceed the standard, there are already potential traffic congestion risks and attention needs to be paid and corresponding measures should be taken to prevent the further deterioration of the traffic conditions.

[0068] Step S640, otherwise, the traffic bottleneck binary value is in the non-bottleneck state.

[0069] Specifically, conversely, if no abnormal factors that may lead to traffic congestion are found through the analysis of the monitored images, that is, the vehicles are driving orderly, the road is unobstructed, and the traffic facilities are operating normally, then the traffic bottleneck binary value is determined to be in a non-bottleneck state, indicating that the current traffic condition of this section is good, there is no obvious potential traffic bottleneck hazard, the traffic flow can maintain a relatively stable and smooth operation state, and no special traffic intervention measures are required, and only regular traffic monitoring needs to be continued.

[0070] Step S650, when the traffic bottleneck binary value is in a bottleneck trigger state, perform a traffic congestion label on the section to be analyzed and send it to the traffic congestion monitoring client.

[0071] Specifically, finally, when the system determines that the traffic bottleneck binary value is in a bottleneck trigger state, the traffic management system immediately performs a traffic congestion label on the section to be analyzed in its database, records detailed information such as the label time and possible bottleneck reasons (based on the image analysis results), etc., for subsequent statistical analysis and traffic diversion plan formulation. At the same time, a data packet containing the traffic congestion label of the section and related bottleneck information is sent to various traffic congestion monitoring clients through a wireless communication network, such as the large-screen display system in the traffic command center, so that the on-duty traffic police can intuitively see the potential congestion risks of the section and allocate police forces in time for on-site diversion; it will also be sent to the mobile law enforcement terminals of the traffic police to facilitate them to understand the section situation in advance and make responses when patrolling or handling other affairs; it will also be pushed to the intelligent transportation APP to provide real-time road condition warnings for the public to travel, help them reasonably plan travel routes, avoid sections that may be congested, thereby improving the operation efficiency and service quality of the entire traffic system and effectively preventing and alleviating traffic congestion.

[0072] In the above text, reference is made to Figure 1 The traffic congestion monitoring and analysis method using deep learning according to the embodiments of the present invention is described in detail. Next, reference will be made to Figure 2 Describe a traffic congestion monitoring and analysis platform using deep learning according to the embodiments of the present invention.

[0073] The traffic congestion monitoring and analysis platform using deep learning according to the embodiments of the present invention is used to solve the technical problems in the existing traffic monitoring technology that multi-source data cannot be fully utilized, resulting in low prediction accuracy and poor real-time performance, and achieves the technical effect of improving the accuracy and real-time performance of traffic congestion monitoring. The traffic congestion monitoring and analysis platform using deep learning includes: an upstream section inflow acquisition module 10, an upstream section outflow acquisition module 20, a traffic flow prediction model acquisition module 30, a predicted traffic flow output module 40 for the section to be analyzed, a ratio calculation module 50, and a traffic congestion label module 60.

[0074] The upstream section inflow acquisition module 10 is used to obtain the upstream section inflow of the section to be analyzed. Among them, the upstream section inflow includes the upstream inflow information of mini cars, the upstream inflow information of small cars, the upstream inflow information of compact cars, the upstream inflow information of medium-sized cars, the upstream inflow information of medium and large-sized cars, and the upstream inflow information of large cars.

[0075] The upstream section outflow acquisition module 20 is used to obtain the upstream section outflow of the section to be analyzed. Among them, the upstream section outflow includes the upstream outflow information of mini cars, the upstream outflow information of small cars, the upstream outflow information of compact cars, the upstream outflow information of medium-sized cars, the upstream outflow information of medium and large-sized cars, and the upstream outflow information of large cars.

[0076] The traffic flow prediction model acquisition module 30 is used to obtain the traffic flow prediction model of the section to be analyzed. Among them, the topological structure of the traffic flow prediction model is the same as the traffic topology, and the input nodes of the traffic flow prediction model are the upstream traffic flow merging intersections and the upstream traffic flow outflow intersections.

[0077] The predicted traffic flow output module 40 of the section to be analyzed is used to input the upstream inflow information of mini cars, the upstream inflow information of small cars, the upstream inflow information of compact cars, the upstream inflow information of medium-sized cars, the upstream inflow information of medium and large-sized cars, the upstream inflow information of large cars, the upstream outflow information of mini cars, the upstream outflow information of small cars, the upstream outflow information of compact cars, the upstream outflow information of medium-sized cars, the upstream outflow information of medium and large-sized cars, and the upstream outflow information of large cars into the input nodes, and output the predicted traffic flow of the section to be analyzed.

[0078] The ratio calculation module 50 is used to calculate the ratio of the predicted traffic flow of the section to be analyzed to the section capacity threshold, which is set as the traffic congestion ratio.

[0079] The traffic congestion identification module 60 is used to identify the traffic congestion of the section to be analyzed when the traffic congestion ratio is greater than or equal to the traffic congestion ratio threshold, and send it to the traffic congestion monitoring client.

[0080] Next, the specific configuration of the traffic flow prediction model acquisition module 30 will be described in detail. As described above, a traffic flow prediction model for the road section to be analyzed is obtained, wherein the topological structure of the traffic flow prediction model is the same as the traffic topology, and the input nodes of the traffic flow prediction model are the upstream road section traffic inflow intersection and the upstream road section traffic outflow intersection. The traffic flow prediction model acquisition module 30 further includes: an analysis distance configuration unit for configuring the upstream road section analysis distance; a topology segmentation unit for performing upstream road section topology segmentation on the road section to be analyzed based on the upstream road section analysis distance to obtain the upstream road section traffic topology; a traffic flow prediction model topology unit for performing neural network topology simulation according to the upstream road section traffic topology to obtain the traffic flow prediction model topology, and the input nodes of the traffic flow prediction model topology are the upstream road section traffic inflow intersection and the upstream road section traffic outflow intersection; a traffic flow monitoring record data acquisition unit for acquiring the traffic flow monitoring record data of the upstream road section traffic topology, wherein the traffic flow monitoring record data includes the upstream inflow traffic record information of mini-cars, the upstream inflow traffic record information of small cars, the upstream inflow traffic record information of compact cars, the upstream inflow traffic record information of medium-sized cars, the upstream inflow traffic record information of medium and large-sized cars, and the upstream inflow traffic record information of large cars, the upstream outflow traffic record information of mini-cars, the upstream outflow traffic record information of small cars, the upstream outflow traffic record information of compact cars, the upstream outflow traffic record information of medium-sized cars, the upstream outflow traffic record information of medium and large-sized cars, and the upstream outflow traffic record information of large cars, as well as the summary traffic flow record information of the road section to be analyzed; a traffic flow prediction model acquisition unit for using the summary traffic flow record information of the road section to be analyzed as the supervision data, and using the upstream inflow traffic record information of mini-cars, the upstream inflow traffic record information of small cars, the upstream inflow traffic record information of compact cars, the upstream inflow traffic record information of medium-sized cars, the upstream inflow traffic record information of medium and large-sized cars, and the upstream inflow traffic record information of large cars, the upstream outflow traffic record information of mini-cars, the upstream outflow traffic record information of small cars, the upstream outflow traffic record information of compact cars, the upstream outflow traffic record information of medium-sized cars, the upstream outflow traffic record information of medium and large-sized cars, and the upstream outflow traffic record information of large cars as the input data of the input nodes to train the traffic flow prediction model topology to obtain the traffic flow prediction model.

[0081] Among them, for configuring the analysis distance of the upstream section, the analysis distance configuration unit further includes: an analysis distance configuration execution subunit, which is used to execute the analysis distance configuration without an outflow intersection to obtain the initial upstream section analysis distance; a traffic outflow intersection quantity analysis subunit, which is used to collect the quantity of traffic outflow intersections within the initial upstream section analysis distance based on the section to be analyzed; and an upstream section analysis distance acquisition subunit, which is used to perform analysis distance configuration according to the quantity of traffic outflow intersections to obtain the upstream section analysis distance.

[0082] Among them, for executing the analysis distance configuration without an outflow intersection to obtain the initial upstream section analysis distance, the analysis distance configuration execution subunit further includes: a monitoring record data acquisition micro-unit, which is used to obtain the section traffic monitoring record data of the section without an outflow intersection. Among them, the section traffic monitoring record data includes the aggregated section traffic record data and the upstream section traffic record data. The upstream section traffic record data has a record distance label, a record speed mean label, and a first record time label. The aggregated section traffic record data has a second record time label, and the second record time label is later than the first record time label, and the time deviation between the second record time label and the first record time label is equal to the ratio of the record distance label to the record speed mean label; a clustering analysis micro-unit, which is used to perform clustering analysis on the section traffic monitoring record data according to the upstream section traffic record data based on a traffic deviation threshold to obtain multiple clusters of section traffic monitoring record data; a relevance analysis micro-unit, which is used to traverse the multiple clusters of section traffic monitoring record data and perform relevance analysis on the aggregated section traffic record data and the record distance label to obtain the first record distance correlation degree, the second record distance correlation degree until the Qth record distance correlation degree; and a maximum distance extraction micro-unit, which is used to extract the maximum distance greater than or equal to the correlation degree threshold according to the first record distance correlation degree, the second record distance correlation degree until the Qth record distance correlation degree, and set it as the initial upstream section analysis distance.

[0083] Among them, traverse the multi-cluster road section traffic monitoring record data, perform correlation analysis on the summarized road section traffic record data and the record distance label to obtain the first record distance correlation degree, the second record distance correlation degree until the Qth record distance correlation degree. The correlation analysis micro-unit further includes: a first cluster road section traffic monitoring record data extraction unit, which is used to extract the first cluster road section traffic monitoring record data according to the multi-cluster road section traffic monitoring record data; a reference data sequence generation unit, which is used to extract the first summarized road section traffic record data according to the first cluster road section traffic monitoring record data, perform dimensionless processing, and generate a reference data sequence; a dimensionless processing unit, which is used to extract the corresponding first record distance label from the first cluster road section traffic monitoring record data according to the first summarized road section traffic record data, perform dimensionless processing, and generate a comparison data sequence; a grey correlation degree analysis unit, which is used to perform grey correlation degree analysis according to the comparison data sequence and the reference data sequence to obtain the first record distance correlation degree.

[0084] Among them, using the aggregated flow record information of the road section to be analyzed as the supervised data, and using the upstream inflow record information of mini-cars, the upstream inflow record information of small cars, the upstream inflow record information of compact cars, the upstream inflow record information of medium-sized cars, the upstream inflow record information of mid-large cars, and the upstream inflow record information of large cars, the upstream outflow record information of mini-cars, the upstream outflow record information of small cars, the upstream outflow record information of compact cars, the upstream outflow record information of medium-sized cars, the upstream outflow record information of mid-large cars, and the upstream outflow record information of large cars as the input data of the input nodes, training the topology of the flow prediction model to obtain the flow prediction model. The flow prediction model acquisition unit further includes: a weight distribution result acquisition subunit, which is used to perform weight assignment on the upstream inflow of mini-cars, the upstream inflow of small cars, the upstream inflow of compact cars, the upstream inflow of medium-sized cars, the upstream inflow of mid-large cars, and the upstream inflow of large cars, the upstream outflow of mini-cars, the upstream outflow of small cars, the upstream outflow of compact cars, the upstream outflow of medium-sized cars, the upstream outflow of mid-large cars, and the upstream outflow of large cars to obtain a weight distribution result, where the weight distribution result is weighted based on the Delphi weighting method; a flow prediction model acquisition subunit, which is used to configure the input weights of the input nodes according to the weight distribution result, and then use the aggregated flow record information of the road section to be analyzed as the supervised data, and use the upstream inflow record information of mini-cars, the upstream inflow record information of small cars, the upstream inflow record information of compact cars, the upstream inflow record information of medium-sized cars, the upstream inflow record information of mid-large cars, and the upstream inflow record information of large cars, the upstream outflow record information of mini-cars, the upstream outflow record information of small cars, the upstream outflow record information of compact cars, the upstream outflow record information of medium-sized cars, the upstream outflow record information of mid-large cars, and the upstream outflow record information of large cars as the input data of the input nodes, and train the topology of the flow prediction model to obtain the flow prediction model.

[0085] Next, the specific configuration of the traffic congestion identification module 60 will be described in detail. As described above, when the traffic congestion ratio is greater than or equal to the traffic congestion ratio threshold, traffic congestion identification is performed on the section to be analyzed and sent to the traffic congestion monitoring client. The traffic congestion identification module 60 further includes: a monitoring image information acquisition unit for the section to be analyzed, which is used to obtain the monitoring image information of the section to be analyzed when the traffic congestion ratio is less than the traffic congestion ratio threshold; a traffic bottleneck binary value acquisition unit, which is used to perform traffic bottleneck analysis based on the monitoring image information to obtain a traffic bottleneck binary value; a trigger status judgment unit, where when the monitoring image information shows factors that cause a decrease in traffic flow or a decrease in traffic velocity, the traffic bottleneck binary value is in a bottleneck trigger state; a non-bottleneck status judgment unit, which is used to otherwise, the traffic bottleneck binary value is in a non-bottleneck state; a traffic congestion monitoring client sending unit, which is used to perform traffic congestion identification on the section to be analyzed and send it to the traffic congestion monitoring client when the traffic bottleneck binary value is in a bottleneck trigger state.

[0086] The traffic congestion monitoring and analysis platform using deep learning provided by the embodiments of the present invention can execute the traffic congestion monitoring and analysis method using deep learning provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0087] Although the present application makes various references to certain modules in the platform according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server. The included individual units and modules are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0088] The above specific implementation manners do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present application shall be included within the protection scope of the present application. In some cases, the actions or steps recorded in the present application can be executed in a different order from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A traffic congestion monitoring and analysis method using deep learning, characterized in that: include: Obtaining the upstream section incoming flow of the section to be analyzed, wherein the upstream section incoming flow includes upstream incoming flow information of mini vehicles, upstream incoming flow information of small vehicles, upstream incoming flow information of compact vehicles, upstream incoming flow information of medium-sized vehicles, upstream incoming flow information of medium-to-large vehicles, and upstream incoming flow information of large vehicles; Obtaining the upstream road segment outflow flow of the road segment to be analyzed, wherein the upstream road segment outflow flow includes upstream outflow flow information of mini vehicles, upstream outflow flow information of small vehicles, upstream outflow flow information of compact vehicles, upstream outflow flow information of medium vehicles, upstream outflow flow information of medium and large vehicles, and upstream outflow flow information of large vehicles; Obtaining a flow prediction model for the road section to be analyzed, wherein the topological structure of the flow prediction model is the same as the traffic topology, and the input nodes of the flow prediction model are the flow inlet intersection of the upstream road section and the flow outlet intersection of the upstream road section; Input the upstream inflow flow information of the micro-car, the upstream inflow flow information of the small car, the upstream inflow flow information of the compact car, the upstream inflow flow information of the medium-sized car, the upstream inflow flow information of the medium-to-large car, and the upstream inflow flow information of the large car, the upstream outflow flow information of the micro-car, the upstream outflow flow information of the small car, the upstream outflow flow information of the compact car, the upstream outflow flow information of the medium-sized car, the upstream outflow flow information of the medium-to-large car, and the upstream outflow flow information of the large car into the input node, and output the predicted flow of the road section to be analyzed; Calculating the ratio of the predicted flow rate of the road section to be analyzed to the road section capacity threshold, and setting it as the traffic congestion ratio; When the traffic congestion ratio is greater than or equal to the traffic congestion ratio threshold, the road section to be analyzed is marked with traffic congestion and sent to the traffic congestion monitoring client.

2. The method according to claim 1, characterized in that Obtaining a traffic prediction model for the road section to be analyzed, including: Configure upstream segment analysis distance; Based on the upstream road section analysis distance, performing upstream road section topological segmentation on the road section to be analyzed to obtain the upstream road section traffic topology; Performing neural network topology simulation according to the upstream road section traffic topology to obtain a traffic prediction model topology, wherein the input nodes of the traffic prediction model topology are the upstream road section traffic inflow intersection and the upstream road section traffic outflow intersection; Collecting flow monitoring record data of the traffic topology of the upstream road section, wherein the flow monitoring record data includes upstream incoming flow record information of mini-cars, upstream incoming flow record information of small-cars, upstream incoming flow record information of compact-cars, upstream incoming flow record information of medium-sized-cars, upstream incoming flow record information of medium-to-large-sized-cars, and upstream incoming flow record information of large-sized-cars, upstream outgoing flow record information of mini-cars, upstream outgoing flow record information of small-cars, upstream outgoing flow record information of compact-cars, upstream outgoing flow record information of medium-sized-cars, upstream outgoing flow record information of medium-to-large-sized-cars, and upstream outgoing flow record information of large-sized-cars, as well as summary flow record information of the road section to be analyzed; The summarized traffic record information of the road section to be analyzed is used as the supervision data, and the upstream incoming traffic record information of the micro-car, the upstream incoming traffic record information of the small car, the upstream incoming traffic record information of the compact car, the upstream incoming traffic record information of the medium-sized car, the upstream incoming traffic record information of the medium-to-large-sized car, and the upstream incoming traffic record information of the large car, the upstream outgoing traffic record information of the micro-car, the upstream outgoing traffic record information of the small car, the upstream outgoing traffic record information of the compact car, the upstream outgoing traffic record information of the medium-sized car, the upstream outgoing traffic record information of the medium-to-large-sized car, and the upstream outgoing traffic record information of the large car are used as the input data of the input nodes to train the traffic prediction model topology and obtain the traffic prediction model.

3. The method according to claim 2, characterized in that Configure upstream segment analysis distance, including: Execute the non-outlet intersection analysis distance configuration to obtain the initial upstream section analysis distance; Based on the road section to be analyzed, the number of flow confluence intersections within the analysis distance of the initial upstream road section is collected; The analysis distance is configured according to the number of the flow outlet intersections to obtain the analysis distance of the upstream section.

4. The method according to claim 3, characterized in that Execute the non-outlet analysis distance configuration to obtain the initial upstream segment analysis distance, including: Obtaining traffic monitoring record data of a section without an outgoing junction, wherein the traffic monitoring record data of the section includes summary traffic record data of the section and upstream traffic record data of the section, the upstream traffic record data of the section has a record distance label, a record speed average label and a first record time label, the summary traffic record data of the section has a second record time label, the second record time label is later than the first record time label, and the time deviation between the second record time label and the first record time label is equal to the ratio of the record distance label to the record speed average label; According to the upstream section flow record data, based on the flow deviation threshold, cluster analysis is performed on the section flow monitoring record data to obtain multiple clusters of section flow monitoring record data; Traversing the multiple clusters of road section flow monitoring record data, performing correlation analysis on the aggregated road section flow record data and the record distance label, and obtaining a first record distance correlation degree, a second record distance correlation degree, and finally a Qth record distance correlation degree; According to the first recorded distance association degree, the second recorded distance association degree, and up to the Qth recorded distance association degree, a maximum distance greater than or equal to the association degree threshold is extracted and set as the initial upstream section analysis distance.

5. The method according to claim 4, characterized in that Traversing the multiple clusters of road section flow monitoring record data, performing correlation analysis on the aggregated road section flow record data and the record distance label, and obtaining a first record distance correlation degree, a second record distance correlation degree, and up to a Qth record distance correlation degree, including: Extracting a first cluster of road section flow monitoring record data according to the multiple clusters of road section flow monitoring record data; Extracting first summary section flow record data according to the first cluster of section flow monitoring record data, performing dedimensionalization processing, and generating a reference data sequence; Extracting one-to-one corresponding first record distance labels from the first cluster of road section flow monitoring record data according to the first summary road section flow record data, performing dedimensionalization processing, and generating a comparison data sequence; A grey correlation analysis is performed on the comparison data sequence and the reference data sequence to obtain the first record distance correlation.

6. The method according to claim 2, characterized in that The traffic flow record information of the road section to be analyzed is used as the supervision data, and the upstream inflow flow record information of the micro-car, the upstream inflow flow record information of the small car, the upstream inflow flow record information of the compact car, the upstream inflow flow record information of the medium-sized car, the upstream inflow flow record information of the medium-to-large car, and the upstream inflow flow record information of the large car, the upstream outflow flow record information of the micro-car, the upstream outflow flow record information of the small car, the upstream outflow flow record information of the compact car, the upstream outflow flow record information of the medium-sized car, the upstream outflow flow record information of the medium-to-large car, and the upstream outflow flow record information of the large car are used as the input data of the input node to train the traffic prediction model topology and obtain the traffic prediction model, including: Weight distribution is performed on the upstream inflow of mini vehicles, the upstream inflow of small vehicles, the upstream inflow of compact vehicles, the upstream inflow of medium vehicles, the upstream inflow of medium-to-large vehicles, and the upstream inflow of large vehicles, the upstream outflow of mini vehicles, the upstream outflow of small vehicles, the upstream outflow of compact vehicles, the upstream outflow of medium vehicles, the upstream outflow of medium-to-large vehicles, and the upstream outflow of large vehicles to obtain a weight distribution result, wherein the weight distribution result is weighted based on the Delphi weighting method; After configuring the input weights of the input nodes according to the weight distribution results, the traffic record information of the road section to be analyzed is summarized as the supervision data, and the upstream incoming traffic record information of the micro-car, the upstream incoming traffic record information of the small car, the upstream incoming traffic record information of the compact car, the upstream incoming traffic record information of the medium-sized car, the upstream incoming traffic record information of the medium-to-large-sized car, and the upstream incoming traffic record information of the large car, the upstream outgoing traffic record information of the micro-car, the upstream outgoing traffic record information of the small car, the upstream outgoing traffic record information of the compact car, the upstream outgoing traffic record information of the medium-sized car, the upstream outgoing traffic record information of the medium-to-large-sized car, and the upstream outgoing traffic record information of the large car are used as input data of the input nodes to train the traffic prediction model topology and obtain the traffic prediction model.

7. The method according to claim 1, characterized in that Also includes: When the traffic congestion ratio is less than the traffic congestion ratio threshold, obtaining monitoring image information of the road section to be analyzed; Performing traffic bottleneck analysis based on the monitoring image information to obtain a traffic bottleneck binary value; Wherein, when the monitoring image information shows factors that lead to a reduction in traffic flow or a decrease in traffic flow speed, the traffic bottleneck binary value is a bottleneck triggering state; Otherwise, the traffic bottleneck binary value is a no-bottleneck state; When the traffic bottleneck binary value is in a bottleneck triggering state, a traffic congestion mark is made for the road section to be analyzed and sent to a traffic congestion monitoring client.

8. The traffic congestion monitoring and analysis platform using deep learning is characterized by: The platform is used to implement the traffic congestion monitoring and analysis method using deep learning according to any one of claims 1 to 7, and the platform includes: An upstream section incoming flow acquisition module, the upstream section incoming flow acquisition module is used to obtain the upstream section incoming flow of the section to be analyzed, wherein the upstream section incoming flow includes the upstream incoming flow information of mini vehicles, the upstream incoming flow information of small vehicles, the upstream incoming flow information of compact vehicles, the upstream incoming flow information of medium-sized vehicles, the upstream incoming flow information of medium-to-large vehicles, and the upstream incoming flow information of large vehicles; An upstream section outflow flow acquisition module, the upstream section outflow flow acquisition module is used to obtain the upstream section outflow flow of the section to be analyzed, wherein the upstream section outflow flow includes micro-car upstream outflow flow information, small car upstream outflow flow information, compact car upstream outflow flow information, medium-sized car upstream outflow flow information, medium-to-large car upstream outflow flow information and large car upstream outflow flow information; A flow prediction model acquisition module, the flow prediction model acquisition module is used to obtain the flow prediction model of the section to be analyzed, wherein the topological structure of the flow prediction model is the same as the traffic topology, and the input nodes of the flow prediction model are the flow inlet intersection of the upstream section and the flow outlet intersection of the upstream section; The predicted flow output module of the road section to be analyzed is used to input the upstream inflow flow information of the micro-car, the upstream inflow flow information of the small car, the upstream inflow flow information of the compact car, the upstream inflow flow information of the medium-sized car, the upstream inflow flow information of the medium-to-large car and the upstream inflow flow information of the large car, the upstream outflow flow information of the micro-car, the upstream outflow flow information of the small car, the upstream outflow flow information of the compact car, the upstream outflow flow information of the medium-sized car, the upstream outflow flow information of the medium-to-large car and the upstream outflow flow information of the large car into the input node, and output the predicted flow of the road section to be analyzed; A ratio calculation module, the ratio calculation module is used to calculate the ratio of the predicted flow of the road section to be analyzed to the road section capacity threshold, which is set as a traffic congestion ratio; A traffic congestion identification module is used to identify the traffic congestion of the road section to be analyzed and send it to the traffic congestion monitoring client when the traffic congestion ratio is greater than or equal to the traffic congestion ratio threshold.

Images