Intelligent road large model-driven traffic jam prediction method and system, and medium

Abstract

The invention discloses a traffic jam prediction method and system driven by an intelligent road large model and a medium, and relates to the related technical field of traffic jam prediction, and the method comprises the steps: dividing a target region according to multi-dimensional traffic road features; collecting historical traffic flow monitoring samples based on a plurality of traffic road types, and outputting a multi-class traffic flow monitoring sample set; performing time sequence cyclic convolution prediction network training, and constructing a plurality of traffic jam prediction models; reading traffic road types and real-time traffic flow monitoring data according to the intelligent road acquisition terminal; and calling the matched traffic jam prediction model to carry out traffic jam prediction on the real-time traffic flow monitoring data, and returning a traffic jam prediction result. The technical problem that in the prior art, dynamic changes of different road types are difficult to adapt, and consequently traffic jam prediction accuracy and real-time performance are insufficient is solved, and the technical effects of achieving efficient and real-time traffic jam prediction and improving traffic jam prediction precision are achieved.

Description

Technical Field

[0001] This application relates to the technical field of traffic congestion prediction, specifically to traffic congestion prediction methods, systems, and media driven by intelligent road models. Background Technology

[0002] With the continuous growth of motor vehicle ownership, traffic congestion has become one of the major challenges facing modern cities, leading to longer travel times, energy waste, and environmental pollution. However, traditional traffic congestion prediction mainly relies on statistical analysis of historical traffic flow data, such as time series models and regression models. While these models can reflect the periodic changes in traffic flow to some extent, they neglect the multidimensional characteristics of road types, number of lanes, surrounding facilities, and their complex impact on traffic flow. Furthermore, existing traffic congestion prediction methods are ill-suited to the dynamic changes and real-time requirements of different road types, resulting in limited prediction accuracy and generalization ability.

[0003] Therefore, current technologies have limitations in adapting to the dynamic changes of different road types, resulting in insufficient accuracy and real-time performance in traffic congestion prediction. Summary of the Invention

[0004] This application provides a traffic congestion prediction method, system, and medium driven by a smart road model, which solves the technical problem in the prior art that it is difficult to adapt to the dynamic changes of different road types, resulting in insufficient accuracy and real-time performance of traffic congestion prediction. It achieves the technical effect of realizing efficient real-time prediction of traffic congestion and improving the accuracy of traffic congestion prediction.

[0005] This application provides a traffic congestion prediction method driven by a smart road model. The method includes: dividing a target area according to multi-dimensional traffic road features to obtain multiple traffic road types; collecting historical traffic flow monitoring samples corresponding to each traffic road type based on the multiple traffic road types, and outputting a multi-class traffic flow monitoring sample set; training a temporal recurrent convolutional prediction network on the multi-class traffic flow monitoring sample set to construct multiple traffic congestion prediction models, and downloading the multiple traffic congestion prediction models to a model-driven center; reading the traffic road type and real-time traffic flow monitoring data of the current vehicle from the smart road acquisition terminal; the model-driven center determining a matching traffic congestion prediction model from the multiple traffic congestion prediction models based on the traffic road type, calling the matching traffic congestion prediction model to predict traffic congestion on the real-time traffic flow monitoring data, and returning the traffic congestion prediction result.

[0006] In one possible implementation, the traffic congestion prediction method driven by the intelligent road model further performs the following processing: constructing road analysis units, which are obtained by identifying the road proportion of the target area; performing initial road unit decomposition on the target area according to the road analysis units to obtain an initial set of road analysis units; dividing the multi-source area description dataset according to the initial set of road analysis units and outputting the divided multi-source area description data subset; performing feature similarity clustering on the divided multi-source area description data subset according to the defined multi-dimensional traffic road features to obtain similarity clustering results, which include multiple traffic road types after clustering.

[0007] In one possible implementation, the traffic congestion prediction method driven by the intelligent road model further performs the following processing: constructing road analysis units, which are obtained by identifying the road proportion of the target area; performing initial road unit decomposition on the target area according to the road analysis units to obtain an initial set of road analysis units; dividing the multi-source area description dataset according to the initial set of road analysis units and outputting the divided multi-source area description data subset; performing feature similarity clustering on the divided multi-source area description data subset according to the defined multi-dimensional traffic road features to obtain similarity clustering results, which include multiple traffic road types after clustering.

[0008] In one possible implementation, the traffic congestion prediction method driven by the intelligent road model further performs the following processing: obtaining initial road analysis units by identifying the road proportion of the target area; calculating the fitness of the initial road analysis units to obtain fitness evaluation indicators, including computing resource adaptation evaluation indicators and multi-dimensional traffic road feature performance evaluation indicators; when the fitness evaluation indicators are less than a preset fitness threshold, adjusting the granularity of the initial road analysis units, re-obtaining road analysis units, until outputting road analysis units that meet the preset fitness threshold.

[0009] In one possible implementation, the traffic congestion prediction method driven by the intelligent road big model further performs the following processing: performing time-series processing on the multi-class traffic flow monitoring sample set respectively; constructing multi-class learning samples and known labels supervising the degree of traffic congestion based on the time-series processed multi-class traffic flow monitoring sample set; initializing a time-series recurrent convolutional prediction network; training the initialized time-series recurrent convolutional prediction network with multi-class learning samples and known labels supervising the degree of traffic congestion; learning model parameters with the goal of minimizing the mean square error of traffic congestion prediction; and obtaining multiple converged traffic congestion prediction models.

[0010] In one possible implementation, the traffic congestion prediction method driven by the intelligent road big model further performs the following processing: obtaining a meta-training dataset based on the prediction samples output by the multiple converged traffic congestion prediction models; training the meta-training dataset using a meta-learner to obtain a fused traffic congestion prediction model obtained by fusing the multiple converged traffic congestion prediction models; and updating the model parameters of the multiple converged traffic congestion prediction models using the model parameters of the fused traffic congestion prediction model to optimize the multiple traffic congestion prediction models.

[0011] This application also provides a traffic congestion prediction system driven by a smart road model. The system includes: a region segmentation module for segmenting a target region according to multi-dimensional traffic road features and obtaining multiple traffic road types; a historical data collection module for collecting historical traffic flow monitoring samples corresponding to each of the multiple traffic road types and outputting a set of multiple types of traffic flow monitoring samples; a prediction model construction module for training a temporal recurrent convolutional prediction network on the set of multiple types of traffic flow monitoring samples to construct multiple traffic congestion prediction models and downloading the multiple traffic congestion prediction models to a model-driven center; a real-time data acquisition module for reading the traffic road type and real-time traffic flow monitoring data of the current vehicle from the smart road acquisition terminal; and a traffic congestion prediction module for the model-driven center to determine a matching traffic congestion prediction model from the multiple traffic congestion prediction models based on the traffic road type, call the matching traffic congestion prediction model to predict traffic congestion on the real-time traffic flow monitoring data, and return the traffic congestion prediction result.

[0012] This application also provides a computer-readable storage medium, including: a computer program stored thereon, which, when executed by a processor, implements a traffic congestion prediction method driven by a smart road big model.

[0013] This application proposes a traffic congestion prediction method, system, and medium driven by a smart road model. The method involves dividing the target area according to multi-dimensional traffic road characteristics; collecting historical traffic flow monitoring samples based on multiple traffic road types to output multi-class traffic flow monitoring sample sets; training a temporal recurrent convolutional prediction network to construct multiple traffic congestion prediction models; reading traffic road types and real-time traffic flow monitoring data from a smart road acquisition terminal; calling the matching traffic congestion prediction model to predict traffic congestion based on the real-time traffic flow monitoring data; and returning the traffic congestion prediction results. This addresses the technical problem in existing technologies where the prediction is difficult to adapt to the dynamic changes of different road types, leading to insufficient accuracy and real-time performance in traffic congestion prediction. It achieves efficient and real-time traffic congestion prediction, improving the accuracy of traffic congestion prediction. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0015] Figure 1 This is a schematic diagram of the traffic congestion prediction method driven by the intelligent road big model provided in the embodiments of this application.

[0016] Figure 2 A schematic diagram of the traffic congestion prediction system driven by the intelligent road big model provided in this application embodiment.

[0017] Figure labeling: Area division module 10, historical data collection module 20, prediction model construction module 30, real-time data acquisition module 40, traffic congestion prediction module 50. Detailed Implementation

[0018] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below.

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" can be the same or different subsets of all possible embodiments and can be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly 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 one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0021] This application provides a traffic congestion prediction method driven by a smart road large model, such as... Figure 1 As shown, the method includes:

[0022] Step S100: Divide the target area according to the multi-dimensional traffic road characteristics to obtain multiple traffic road types.

[0023] Step S100 further includes step S110, collecting a multi-source regional description dataset of the target area, the multi-source regional description dataset including basic geographic information, traffic control data, traffic flow data, and auxiliary data; step S120, defining multi-dimensional traffic road features, wherein the multi-dimensional traffic road features include a physical topology dimension describing the road structure, a control method dimension describing the road control rules, a functional scenario dimension describing the scene in which the road is located, and a peak time period dimension for dynamic regulation; step S130, analyzing the multi-source regional description dataset according to the defined multi-dimensional traffic road features to obtain multiple traffic road types of the target area.

[0024] Preferably, basic geographic information, traffic control data, traffic flow data, and auxiliary data of the target area are collected from different dimensions and combined to form a multi-source regional description dataset. Specifically, basic geographic information refers to the static physical attribute data of the road network, including the number of lanes, road width, road grade, slope, curvature, material, and the geometric structure of intersections, where the road grade may be highway, urban arterial road, or ramp; traffic control data refers to the rules and facilities data for managing traffic flow, including traffic light timing schemes, signs and markings, tidal lane switching times, and the setting of guardrails and isolation facilities; traffic flow data refers to dynamic monitoring data reflecting the vehicle operating status, including historical and real-time data such as the number of vehicles, average vehicle speed, time occupancy, and headway collected by sensors such as geomagnetic coils, cameras, and radar; auxiliary data refers to the surrounding environmental data that can indirectly affect traffic flow, including weather conditions, date type, and the distribution and type of points of interest along the road, such as schools, shopping malls, office buildings, and hospitals.

[0025] Preferably, the roads in the target area are analyzed from the perspectives of spatial structure and connectivity, traffic management rules, social functions and usage purposes, and temporal variation patterns to obtain multi-dimensional traffic road characteristics. Specifically, the physical topology dimension describing the road structure is defined to quantify the physical form of the road, such as the number of intersections, the number of lanes, and whether it is a grade-separated interchange; the control method dimension describing the road management rules is defined to quantify the constraints of traffic management, such as signal cycle duration, speed limit, and whether parking is allowed; the functional scenario dimension describing the road's location is defined to quantify the road's service targets, such as the proportion of commuter traffic, the intensity of commercial activities, and the area surrounding schools; and the peak period dimension of dynamic regulation is defined to quantify the dynamic patterns of road traffic flow in different time periods, such as the start time of the morning peak, the duration of the evening peak, and the weekend traffic curve.

[0026] Preferably, the multi-source regional description dataset is analyzed according to the defined multi-dimensional traffic road features. That is, the multi-source regional description dataset is mapped to multi-dimensional traffic road features according to its attributes and multiple corresponding feature vectors are calculated. Then, K-Means clustering or DBSCAN clustering is used to learn the multiple feature vectors, including calculating similarity and classifying similar roads into the same category. Finally, multiple traffic road types are output, such as commuter main roads with tidal congestion on weekday mornings, roads around shopping districts that are congested on weekend afternoons, and logistics channels in industrial areas that are open all day but have frequent large trucks.

[0027] Furthermore, step S130 also includes step S131, constructing road analysis units, wherein the road analysis units are obtained by identifying the road proportion of the target area; step S132, performing initial road unit decomposition on the target area according to the road analysis units to obtain an initial set of road analysis units; step S133, dividing the multi-source area description dataset according to the initial set of road analysis units and outputting the divided multi-source area description data subset; step S134, performing feature similarity clustering on the divided multi-source area description data subset according to the defined multi-dimensional traffic road features to obtain similarity clustering results, wherein the similarity clustering results include multiple traffic road types after clustering.

[0028] Preferably, the smallest analysis unit is defined by analyzing the area proportion of the road network in the entire target area. This unit is used as the initial road unit decomposition of the target area, including dividing the roads in the entire target area into grids to obtain multiple initial road analysis units and a set of initial road analysis units. Then, a multi-source region description dataset is divided based on the set of initial road analysis units. That is, based on the spatial coordinates of each multi-source region description data, it is associated and matched to the corresponding initial road analysis unit to obtain multiple corresponding data subsets. This outputs the divided multi-source region description data subsets, each data subset uniquely corresponding to a grid unit and containing all relevant data occurring within the initial road analysis unit. Finally, feature similarity clustering is performed on the divided multi-source region description data subsets according to the defined multi-dimensional traffic road features. This includes using K-Means clustering or DBSCAN clustering to analyze the feature vectors corresponding to all initial road analysis units, calculating the similarity between feature vectors of different units, and grouping the initial road analysis units corresponding to feature vectors with similarity that meet a preset similarity threshold into one category. Each cluster corresponds to one traffic road type, and the similarity clustering result is obtained. The similarity clustering result includes multiple traffic road types after clustering.

[0029] Furthermore, step S131 also includes step A, obtaining an initial road analysis unit by identifying the road proportion of the target area; step B, performing fitness calculation on the initial road analysis unit to obtain fitness evaluation indicators, including computing resource adaptation evaluation indicators and multi-dimensional traffic road feature performance evaluation indicators; step C, when the fitness evaluation indicators are less than a preset fitness threshold, adjusting the granularity of the initial road analysis unit, re-obtaining road analysis units, until a road analysis unit that meets the preset fitness threshold is output.

[0030] Preferably, the ratio of road area to total area within the target area is calculated to generate initial road analysis units. For example, in a city center with a high road ratio, the initial road analysis unit might be set to 300m × 300m, while in a suburban area with a low road ratio, it might be set to 1000m × 1000m. Then, the fitness of the initial road analysis units is calculated. This involves quantitatively evaluating the quality of the initial road analysis units based on computational resource adaptation assessment indicators and multidimensional traffic road feature performance assessment indicators. The computational resource adaptation assessment indicator measures computational efficiency, evaluating how many units are generated based on the current initial road analysis unit division, thereby estimating the memory usage and computation time required for the entire data processing and model training process. A higher computational resource adaptation assessment indicator value indicates a heavier computational burden. The multidimensional traffic road feature performance assessment indicator measures data quality, evaluating whether the traffic characteristics within each unit are consistent based on the current initial road analysis unit division. This is achieved by calculating the variance or entropy of the features within the unit. A higher multidimensional traffic road feature performance assessment indicator value indicates more chaotic and inconsistent features within the unit, resulting in lower quality.

[0031] Preferably, a preset fitness threshold is used, which includes high computational resource efficiency and good feature quality performance. If the fitness evaluation index is less than the preset fitness threshold, the granularity of the initial road analysis unit is adjusted. Specifically, if the computational resource index exceeds the threshold, the size of the initial road analysis unit is increased; otherwise, the size of the initial road analysis unit is decreased. Then, road analysis units are re-acquired, and evaluation and iterative comparison are performed until road analysis units that meet the preset fitness threshold are output, ensuring the high intelligence and practicality of the current partitioning scheme and guaranteeing prediction accuracy.

[0032] Step S200: Collect historical traffic flow monitoring samples corresponding to each of the multiple traffic road types, and output a set of multiple types of traffic flow monitoring samples.

[0033] Preferably, based on multiple road types, historical traffic flow monitoring samples corresponding to each road type are extracted and collected from past monitoring records. These samples may include traffic data monitored within historical time periods, such as average vehicle speed, traffic volume, road occupancy rate, and average vehicle speed and traffic conditions after a preset prediction time, such as smooth flow, slow flow, or congestion. This results in the output of multiple traffic flow monitoring sample sets, which may include traffic flow monitoring sample sets for commuter arteries, traffic flow monitoring sample sets for roads surrounding commercial districts, and traffic flow monitoring sample sets for highway connectors.

[0034] Step S300: Train a temporal recurrent convolutional prediction network on the multi-type traffic flow monitoring sample set to construct multiple traffic congestion prediction models, and download the multiple traffic congestion prediction models to the model driving center.

[0035] Step S300 further includes step S310, performing time-series processing on the multi-class traffic flow monitoring sample set, and constructing multi-class learning samples and known labels for supervising the degree of traffic congestion based on the time-series processed multi-class traffic flow monitoring sample set; step S320, initializing a time-series recurrent convolutional prediction network, training the initialized time-series recurrent convolutional prediction network with multi-class learning samples and known labels for supervising the degree of traffic congestion, and learning model parameters with the goal of minimizing the mean square error of traffic congestion prediction to obtain multiple converged traffic congestion prediction models.

[0036] Preferably, a temporal recurrent convolutional prediction network is trained on multiple traffic flow monitoring sample sets. Specifically, the multiple traffic flow monitoring sample sets are processed temporally, i.e., a sliding window sampling is performed on the traffic flow monitoring sample set of each type of road, where the sliding window may be one hour, thus obtaining the time-processed multi-type traffic flow monitoring sample sets. These are then used to construct corresponding learning samples to describe the evolution of historical traffic conditions. Known labels for monitoring traffic congestion levels are set, such as average vehicle speed, road occupancy, or traffic condition category. Then, based on convolutional neural networks and recurrent neural networks, a temporal recurrent convolutional prediction network is initialized, while simultaneously capturing the spatial dependencies between different monitoring points within the target area, such as how congestion at upstream intersections affects downstream intersections, and how traffic flow changes with... The system learns to understand the long-term dependence and dynamic patterns of temporal changes, such as the onset, duration, and dissipation of traffic congestion during the morning rush hour. Then, it inputs multiple types of training samples and known labels supervising traffic congestion levels into an initialized temporal recurrent convolutional prediction network for training. This involves calculating predicted values ​​based on the training samples and comparing them with the known labels of traffic congestion levels, aiming to minimize the mean squared error of traffic congestion predictions. The model parameters are learned through backpropagation and a gradient descent optimizer to adjust the parameters within the temporal recurrent convolutional prediction network. After multiple iterations of training, several converged traffic congestion prediction models are obtained. These models are then downloaded to the model-driven center for deployment. One prediction model is trained for each type of road to handle the corresponding traffic mode, ensuring accurate traffic congestion predictions.

[0037] Furthermore, step S320 also includes step S321, obtaining a meta-training dataset based on the prediction samples output by the multiple converged traffic congestion prediction models; step S322, training the meta-training dataset using a meta-learner to obtain a fused traffic congestion prediction model obtained by fusing the multiple converged traffic congestion prediction models; and step S323, updating the model parameters of the multiple converged traffic congestion prediction models using the model parameters of the fused traffic congestion prediction model to optimize the multiple traffic congestion prediction models.

[0038] Preferably, multiple converged traffic congestion prediction models are input using a validation dataset. The output prediction samples from each model are used to form new feature vectors, which are then used to construct a meta-training dataset. This dataset includes the prediction results of multiple traffic congestion prediction models for the same input data and the corresponding actual congestion values. A meta-learner is then trained on the meta-training dataset. This meta-learner may be a linear regression model, a fully connected neural network, or a gradient boosting tree. It learns how to optimally fuse the prediction results of multiple traffic congestion prediction models, resulting in a fused traffic congestion prediction model. This fused model is used to coordinate and integrate the multiple traffic congestion prediction models. Finally, based on knowledge distillation, the model parameters of the fused traffic congestion prediction model are used to update the model parameters of the converged multiple traffic congestion prediction models, thereby improving the prediction level of the traffic congestion prediction models. For each data sample, the prediction of the traffic congestion prediction model is required to be close to the true label and also close to the final prediction of the fused traffic congestion prediction model. Finally, optimized multiple traffic congestion prediction models are output, thereby improving the overall accuracy and reliability of traffic congestion prediction.

[0039] Step S400: Read the current road type and real-time traffic flow monitoring data of the vehicle based on the intelligent road data acquisition terminal.

[0040] Preferably, the intelligent road data acquisition terminal refers to intelligent sensing devices installed on road infrastructure, which may include high-definition cameras for identifying the number of vehicles, vehicle type, vehicle speed, and lane occupancy; millimeter-wave radar / LiDAR for accurately detecting vehicle position, speed, and trajectory; geomagnetic coils for detecting vehicle passage and presence; and roadside units for bidirectional data exchange with intelligent connected vehicles. Specifically, the intelligent road data acquisition terminal reads the current road type and real-time traffic flow monitoring data of the vehicle. Specifically, the intelligent road data acquisition terminal identifies the vehicle and obtains its precise geographical location through its own GPS, then matches it with multiple road types obtained through cluster analysis to determine the current road type of the vehicle. Then, the intelligent road data acquisition terminal collects real-time traffic flow monitoring data such as vehicle flow, speed, road occupancy, and queue length.

[0041] In step S500, the model-driven center determines a matching traffic congestion prediction model from the multiple traffic congestion prediction models based on the type of traffic road it is on, calls the matching traffic congestion prediction model to perform traffic congestion prediction on the real-time traffic flow monitoring data, and returns the traffic congestion prediction result.

[0042] Preferably, the model-driven center is typically deployed on a cloud server, storing multiple traffic congestion prediction models and receiving real-time requests from intelligent road data collection terminals. Specifically, the model-driven center determines a matching traffic congestion prediction model from among the multiple models based on the type of traffic road it is located on. Then, it calls the matching traffic congestion prediction model, using real-time traffic flow monitoring data as input features to predict traffic congestion. For example, if it finds that the current speed and occupancy combination pattern is highly similar to the pattern before severe congestion occurs 15 minutes later in historical data, it outputs a traffic congestion prediction result, which may be a prediction of the average vehicle speed in 15 minutes, a prediction of congestion in 15 minutes, or a prediction of an 85% probability of future congestion, etc. This allows the center to plan new routes for users, provide travel time predictions, or dynamically adjust the traffic light timings for the next cycle using the traffic signal control system to alleviate impending traffic congestion.

[0043] In the above text, refer to Figure 1 This paper describes in detail a traffic congestion prediction method driven by a large intelligent road model according to an embodiment of the present invention. Next, reference will be made to... Figure 2 This invention describes a traffic congestion prediction system driven by a large-scale intelligent road model according to an embodiment of the present invention.

[0044] The traffic congestion prediction system driven by the intelligent road large model according to embodiments of the present invention is used to solve the technical problems existing in the prior art, which are difficult to adapt to the dynamic changes of different road types, resulting in insufficient accuracy and real-time performance of traffic congestion prediction. It achieves the technical effect of realizing efficient real-time traffic congestion prediction and improving the accuracy of traffic congestion prediction. Figure 2 As shown, the traffic congestion prediction system driven by the intelligent road big model includes: a regional division module 10, a historical data collection module 20, a prediction model construction module 30, a real-time data acquisition module 40, and a traffic congestion prediction module 50.

[0045] The system comprises the following modules: a region segmentation module 10, which divides the target region according to multi-dimensional traffic road characteristics and obtains multiple traffic road types; a historical data collection module 20, which collects historical traffic flow monitoring samples corresponding to each of the multiple traffic road types and outputs a set of multiple types of traffic flow monitoring samples; a prediction model construction module 30, which trains a temporal recurrent convolutional prediction network on the set of multiple types of traffic flow monitoring samples to construct multiple traffic congestion prediction models and downloads the multiple traffic congestion prediction models to the model-driven center; a real-time data acquisition module 40, which reads the traffic road type and real-time traffic flow monitoring data of the current vehicle from the intelligent road acquisition terminal; and a traffic congestion prediction module 50, which is used by the model-driven center to determine a matching traffic congestion prediction model from the multiple traffic congestion prediction models based on the traffic road type, call the matching traffic congestion prediction model to predict traffic congestion on the real-time traffic flow monitoring data, and return the traffic congestion prediction result.

[0046] The specific configuration of the region segmentation module 10 will be described in detail below. The region segmentation module 10 further includes: collecting a multi-source region description dataset of the target region, the multi-source region description dataset including basic geographic information, traffic control data, traffic flow data, and auxiliary data; defining multi-dimensional traffic road features, wherein the multi-dimensional traffic road features include a physical topology dimension describing the road structure, a control method dimension describing the road control rules, a functional scenario dimension describing the road's location, and a peak time period dimension for dynamic regulation; and analyzing the multi-source region description dataset according to the defined multi-dimensional traffic road features to obtain multiple traffic road types for the target region.

[0047] The specific configuration of the region segmentation module 10 will be described in detail below. The region segmentation module 10 further includes: constructing road analysis units, which are obtained by identifying the road proportion of the target region; performing initial road unit decomposition on the target region according to the road analysis units to obtain an initial set of road analysis units; dividing the multi-source region description dataset according to the initial set of road analysis units, and outputting the divided multi-source region description data subset; performing feature similarity clustering on the divided multi-source region description data subset according to defined multi-dimensional traffic road features to obtain a similarity clustering result, wherein the similarity clustering result includes multiple traffic road types after clustering.

[0048] The specific configuration of the region division module 10 will be described in detail below. The region division module 10 further includes: obtaining initial road analysis units by identifying the road proportion of the target region; performing fitness calculations on the initial road analysis units to obtain fitness evaluation indicators, including computing resource adaptation evaluation indicators and multi-dimensional traffic road feature performance evaluation indicators; adjusting the granularity of the initial road analysis units when the fitness evaluation indicators are less than a preset fitness threshold, and re-obtaining road analysis units until road analysis units that meet the preset fitness threshold are output.

[0049] The specific configuration of the prediction model construction module 30 will be described in detail below. The prediction model construction module 30 further includes: performing time-series processing on the multi-class traffic flow monitoring sample sets respectively; constructing multi-class learning samples and known labels supervising the degree of traffic congestion based on the time-series processed multi-class traffic flow monitoring sample sets; initializing a temporal recurrent convolutional prediction network; training the initialized temporal recurrent convolutional prediction network using the multi-class learning samples and known labels supervising the degree of traffic congestion; learning model parameters with the goal of minimizing the mean square error of traffic congestion prediction; and obtaining multiple converged traffic congestion prediction models.

[0050] The specific configuration of the prediction model building module 30 will be described in detail below. The prediction model building module 30 further includes: obtaining a meta-training dataset based on the prediction samples output by the multiple converged traffic congestion prediction models; training the meta-training dataset using a meta-learner to obtain a fused traffic congestion prediction model obtained by fusing the multiple converged traffic congestion prediction models; and updating the model parameters of the multiple converged traffic congestion prediction models using the model parameters of the fused traffic congestion prediction model to optimize the multiple traffic congestion prediction models.

[0051] The traffic congestion prediction system driven by the intelligent road model provided in this embodiment of the invention can execute the traffic congestion prediction method driven by the intelligent road model provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0052] Based on the foregoing embodiments, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the traffic congestion prediction method driven by the intelligent road big model as described in any of the preceding embodiments.

[0053] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included 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 each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0054] The specific embodiments described above do not constitute a limitation on the scope of protection of this 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 principles of this application should be included within the scope of protection of this application.

Claims

1. A traffic congestion prediction method driven by a smart road large model, characterized in that, The method includes: The target area is divided according to multi-dimensional traffic road characteristics to obtain multiple traffic road types; Based on the multiple road types, historical traffic flow monitoring samples corresponding to each road type are collected, and a set of multiple types of traffic flow monitoring samples is output. Train a temporal recurrent convolutional prediction network on the multi-type traffic flow monitoring sample set to construct multiple traffic congestion prediction models, and download the multiple traffic congestion prediction models to the model driving center; The intelligent road data collection terminal reads the current road type and real-time traffic flow monitoring data of the vehicle. The model-driven center determines a matching traffic congestion prediction model from among the multiple traffic congestion prediction models based on the type of traffic road it is on, calls the matching traffic congestion prediction model to perform traffic congestion prediction on the real-time traffic flow monitoring data, and returns the traffic congestion prediction result.

2. The traffic congestion prediction method driven by the intelligent road large model as described in claim 1, characterized in that, The target area is divided according to multi-dimensional traffic road characteristics to obtain multiple traffic road types. The methods include: Collect a multi-source regional description dataset for the target area, which includes basic geographic information, traffic control data, traffic flow data, and auxiliary data; Define multidimensional traffic road features, wherein the multidimensional traffic road features include a physical topology dimension describing the road structure, a control method dimension describing the road management rules, a functional scenario dimension describing the road's location, and a peak time period dimension for dynamic regulation; By analyzing the multi-source region description dataset according to the defined multi-dimensional traffic road features, multiple traffic road types of the target region are obtained.

3. The traffic congestion prediction method driven by the intelligent road large model as described in claim 2, characterized in that, The method involves analyzing the multi-source region description dataset according to the defined multi-dimensional traffic road features to obtain multiple traffic road types for the target region, including: A road analysis unit is constructed, which obtains information by identifying the road ratio of the target area; The target area is initially decomposed into road units according to the road analysis units to obtain an initial set of road analysis units; The multi-source region description dataset is divided according to the initial road analysis unit set, and the divided multi-source region description data subset is output. Based on the defined multidimensional traffic road features, feature similarity clustering is performed on the subsets of multi-source regional description data to obtain similarity clustering results, which include multiple traffic road types after clustering.

4. The traffic congestion prediction method driven by the intelligent road large model as described in claim 3, characterized in that, Methods for constructing road analysis units include: The initial road analysis unit is obtained by identifying the road proportion of the target area; The fitness of the initialized road analysis unit is calculated to obtain fitness evaluation indicators, including computing resource adaptation evaluation indicators and multi-dimensional traffic road feature performance evaluation indicators. When the fitness evaluation index is less than the preset fitness threshold, the granularity of the initial road analysis unit is adjusted, and a new road analysis unit is obtained until a road analysis unit that meets the preset fitness threshold is output.

5. The traffic congestion prediction method driven by the intelligent road large model as described in claim 1, characterized in that, Training a temporal recurrent convolutional prediction network on the aforementioned multi-class traffic flow monitoring sample set to construct multiple traffic congestion prediction models, the method including: The various traffic flow monitoring sample sets are subjected to time-series processing, and based on the time-series processed various traffic flow monitoring sample sets, various learning samples and known labels for supervising the degree of traffic congestion are constructed. A temporal recurrent convolutional prediction network is initialized. The network is then trained using multiple training samples and known labels that supervise the degree of traffic congestion. The model parameters are learned with the goal of minimizing the mean square error of traffic congestion prediction, resulting in multiple convergent traffic congestion prediction models.

6. The traffic congestion prediction method driven by the intelligent road large model as described in claim 5, characterized in that, After obtaining multiple convergent traffic congestion prediction models, the method also includes: The meta-training dataset is obtained by analyzing the prediction samples output by multiple converged traffic congestion prediction models. The meta-training dataset is trained using a meta-learner to obtain a fused traffic congestion prediction model, which is obtained by fusing and learning multiple converged traffic congestion prediction models. The model parameters of the fused traffic congestion prediction model are used to update the model parameters of multiple converged traffic congestion prediction models, thereby optimizing the multiple traffic congestion prediction models.

7. A traffic congestion prediction system driven by a smart road big model, characterized in that, The system is used to implement the traffic congestion prediction method driven by the intelligent road model as described in any one of claims 1 to 6, and the system comprises: The region segmentation module is used to segment the target region according to multi-dimensional traffic road characteristics and obtain multiple traffic road types; The historical data collection module is used to collect historical traffic flow monitoring samples corresponding to each of the multiple traffic road types, and output a set of multiple types of traffic flow monitoring samples. The prediction model building module is used to train a temporal recurrent convolutional prediction network on the multi-class traffic flow monitoring sample set, build multiple traffic congestion prediction models, and download the multiple traffic congestion prediction models to the model driving center. The real-time data acquisition module is used to read the current road type and real-time traffic flow monitoring data of the vehicle based on the intelligent road acquisition terminal; The traffic congestion prediction module is used by the model-driven center to determine a matching traffic congestion prediction model from the multiple traffic congestion prediction models based on the type of traffic road, call the matching traffic congestion prediction model to predict traffic congestion on the real-time traffic flow monitoring data, and return the traffic congestion prediction result.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the traffic congestion prediction method driven by the intelligent road large model as described in any one of claims 1-6.

Images