Artificial intelligence-based traffic congestion prediction management method, system and medium

Abstract

The application discloses a traffic congestion prediction management method and system based on artificial intelligence and a medium, relates to the technical field of intelligent transportation systems, and comprises the following steps: collecting multi-source traffic data to construct a space-time traffic data set, classifying traffic states based on the space-time traffic data set, and generating a multi-level congestion probability distribution map; combining the multi-level congestion probability distribution map with real-time traffic event data to predict congestion, and determining a congestion evolution path; managing traffic according to the congestion evolution path, generating a dynamic relief instruction set, and sending the dynamic relief instruction set to a traffic control device of a target road section to manage traffic congestion. The application solves the technical problem that traditional traffic management methods cannot adapt to complex and changeable traffic conditions and cannot accurately predict congestion and efficiently manage traffic, achieves accurate prediction of traffic congestion, efficiently generates and executes a dynamic relief strategy according to real-time traffic conditions, and further improves the technical effects of traffic management efficiency and road traffic capacity.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation systems, and in particular to a method, system, and medium for predicting and managing traffic congestion based on artificial intelligence. Background Technology

[0002] In traffic congestion management, accurate prediction and timely, effective mitigation are crucial for ensuring smooth urban traffic flow. Currently, traditional traffic congestion management methods largely rely on simple monitoring data and human experience. For example, based on traffic flow monitoring data from fixed points, people manually assess congestion and then implement mitigation measures. These methods can be effective when traffic conditions are relatively simple, but their limitations become apparent as urban traffic becomes increasingly complex. Urban traffic flow changes frequently, and unexpected events such as traffic accidents and road construction are increasing. Traditional methods cannot comprehensively collect traffic data, accurately predict congestion, or adjust mitigation strategies in a timely manner based on real-time conditions. This results in inefficient traffic congestion management, failing to meet the demands of efficient traffic management in modern cities. Summary of the Invention

[0003] This application provides a traffic congestion prediction and management method, system, and medium based on artificial intelligence, which is used to solve the technical problem that traditional traffic management methods cannot adapt to complex and ever-changing traffic conditions and are difficult to achieve accurate congestion prediction and efficient management.

[0004] The first aspect of this application provides an artificial intelligence-based traffic congestion prediction and management method, the method comprising: collecting multi-source traffic data to construct a spatiotemporal traffic dataset; classifying traffic states based on the spatiotemporal traffic dataset to generate a multi-level congestion probability distribution map; combining the multi-level congestion probability distribution map with real-time traffic event data to predict congestion and determine the congestion evolution path; performing traffic management according to the congestion evolution path, generating a dynamic diversion instruction set, and sending the dynamic diversion instruction set to the traffic control equipment of the target road segment for traffic congestion management.

[0005] A second aspect of this application provides an artificial intelligence-based traffic congestion prediction and management system, the system comprising: a distribution map generation module, used to collect multi-source traffic data to construct a spatiotemporal traffic dataset, classify traffic states based on the spatiotemporal traffic dataset, and generate a multi-level congestion probability distribution map; a congestion prediction module, used to combine the multi-level congestion probability distribution map with real-time traffic event data to predict congestion and determine the congestion evolution path; and a traffic management module, used to perform traffic management according to the congestion evolution path, generate a dynamic diversion instruction set, and send the dynamic diversion instruction set to the traffic control equipment of the target road segment for traffic congestion management.

[0006] A third aspect of this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the AI-based traffic congestion prediction and management method provided in this application.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] This application constructs a spatiotemporal traffic dataset by conducting roadside traffic sensing on target road sections, retrieving electronic map databases, and incorporating historical traffic fluctuation data. Next, the dataset is divided into grids and features are extracted to determine the probabilities of multiple congestion levels, generating a multi-level congestion probability distribution map. Then, real-time traffic event data is obtained by parsing road surveillance video streams and spatiotemporally correlated with the distribution map. A spatiotemporal graph convolutional network is used to predict and plot speed decay curves, determining the congestion evolution path. Finally, diffusion analysis is performed based on the path, and a dynamic diversion instruction set is generated by combining it with the real-time traffic pressure index. This set is then sent to traffic control equipment for congestion management, achieving accurate prediction of traffic congestion and efficient generation and execution of dynamic diversion strategies based on real-time traffic conditions, thereby improving traffic management efficiency and road capacity. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating the traffic congestion prediction and management method based on artificial intelligence provided in the embodiments of this application.

[0011] Figure 2 This is a schematic diagram of the structure of the traffic congestion prediction and management system based on artificial intelligence provided in the embodiments of this application.

[0012] Attached image labels: Distribution map generation module 1, congestion prediction module 2, traffic management module 3. Detailed Implementation

[0013] This application provides a traffic congestion prediction and management method, system, and medium based on artificial intelligence, which is used to solve the technical problem that traditional traffic management methods cannot adapt to complex and ever-changing traffic conditions and are difficult to achieve accurate congestion prediction and efficient management.

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0015] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, 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 such processes, methods, products, or devices.

[0016] Example 1, as Figure 1 As shown, an artificial intelligence-based traffic congestion prediction and management method includes:

[0017] Step A100: Collect multi-source traffic data to construct a spatiotemporal traffic dataset, classify traffic states based on the spatiotemporal traffic dataset, and generate a multi-level congestion probability distribution map.

[0018] In this embodiment, the spatiotemporal traffic dataset is a collection of data constructed by integrating multi-source traffic data. The congestion probability distribution map is a graphical representation of the likelihood of traffic congestion, constructed based on the spatiotemporal traffic dataset.

[0019] Specifically, the process involves collecting multi-source traffic data to construct a spatiotemporal traffic dataset, conducting road tests on the target road segment to obtain real-time dynamic information and generate the first subset, retrieving the road segment topology from the electronic map database to generate the second subset, introducing historical traffic fluctuation data for correlation analysis to generate the third subset, and finally aligning these three subsets spatiotemporally to construct the spatiotemporal traffic dataset. The specific steps are explained in detail in A110-A140.

[0020] The process of generating a multi-level congestion probability distribution map based on a spatiotemporal traffic dataset involves first dividing the dataset into grids to construct a spatiotemporal grid matrix, then traversing the matrix to extract traffic feature vectors, classifying the dataset according to traffic conditions based on these vectors to determine the probabilities of multiple congestion levels, and finally mapping these probabilities onto the spatiotemporal grid matrix to complete the construction of the multi-level congestion probability distribution map. The specific steps are explained in detail in pages A150-A180.

[0021] Step A200: Combine the multi-level congestion probability distribution map with real-time traffic event data to predict congestion and determine the congestion evolution path.

[0022] In this embodiment, the congestion evolution path is a congestion development trajectory determined by combining multi-level congestion probability distribution maps and real-time traffic event data after a series of analyses and processing.

[0023] Optionally, by combining multi-level congestion probability distribution maps and real-time traffic event data, the congestion evolution path can be determined. First, the road monitoring video stream of the target road segment is analyzed to obtain real-time traffic event data. This data is then spatiotemporally correlated with the distribution map to generate a dynamic event impact weight matrix. A spatiotemporal graph convolutional network is used to perform multi-step iterative prediction on the matrix to draw the road traffic speed decay curve. By traversing the curve, congestion propagation nodes are identified. These nodes are then connected according to the road segment topology to obtain the congestion evolution path. The specific steps are explained in detail in A210-A250.

[0024] Step A300: Perform traffic management based on the congestion evolution path, generate a dynamic diversion instruction set, and send the dynamic diversion instruction set to the traffic control equipment of the target road segment for traffic congestion management.

[0025] In this embodiment, the dynamic traffic management instruction set is a series of instructions generated based on the congestion evolution path and used for traffic congestion management.

[0026] In one embodiment of this application, the step of generating a dynamic diversion instruction set based on the congestion evolution path is as follows: performing diffusion analysis on the path based on the congestion propagation nodes to determine the diffusion direction vector; combining the vector and nodes to perform traffic diversion analysis to obtain multiple candidate diversion strategies; prioritizing the strategies according to the real-time traffic pressure index to determine the diversion trigger sequence; and constructing a dynamic diversion instruction set based on the sequence. The specific steps are described in detail in A310-A330.

[0027] Then, the dynamic traffic management instruction set is sent to the traffic control equipment on the target road segment for traffic congestion management. Traffic control equipment includes intelligent traffic lights, variable message signs, and electronic traffic enforcement systems. When the dynamic traffic management instruction set is received, for intelligent traffic lights, if the instruction requests an extension of the green light duration in a certain direction, the traffic light control system will automatically adjust the time settings to increase the passage time for vehicles in that direction, alleviating congestion. Variable message signs will guide vehicles to change routes based on the displayed instructions (such as "Congestion ahead, please detour"), balancing traffic flow across the road network. Electronic traffic enforcement systems, based on the instructions, will capture and record vehicles violating temporary traffic control measures (such as restricting passage on specific road sections), ensuring the effective implementation of traffic management measures.

[0028] By generating dynamic traffic management instruction sets and sending them to traffic control equipment, the system can accurately predict congestion development based on real-time traffic conditions, thereby efficiently managing traffic congestion and improving road traffic efficiency.

[0029] Furthermore, step A100 in the method provided in this application embodiment includes:

[0030] A110: Based on the target road segment, perform road test traffic sensing to generate the first real-time dynamic subset dataset.

[0031] A120: Retrieve the electronic map database of the target road segment to extract the road segment topology and generate a second static road network subset.

[0032] A130: Introduce historical traffic fluctuation data of the target road segment for correlation analysis to generate a third correlation factor dataset.

[0033] A140: Spatiotemporally align the first real-time dynamic subset, the second static road network subset, and the third correlation factor dataset to construct the spatiotemporal traffic dataset.

[0034] In this embodiment, the road segment topology refers to the collection of information such as the connection relationship between the target road segment and surrounding roads in geographic space, road orientation, intersection layout, number of lanes, and road grade. Traffic fluctuation data is a record of the changes in traffic parameters such as traffic flow, vehicle speed, and vehicle density of the target road segment over a period of time.

[0035] Specifically, firstly, road testing traffic sensing is conducted on the target road segment. Various traffic sensors are deployed on the target road segment, such as geomagnetic sensors, radar sensors, and video monitoring equipment. Geomagnetic sensors can monitor real-time changes in the geomagnetic field as vehicles pass, thereby acquiring data such as vehicle transit time and speed; radar sensors can accurately measure vehicle speed and distance; and video monitoring equipment can visually observe information such as the number of vehicles, vehicle types, and driving trajectories on the road segment. The data collected by these sensors, after integration, constitutes the first real-time dynamic subset dataset, which reflects the real-time traffic dynamics of the target road segment at the current moment. For example, at a certain moment, the traffic flow on this road segment is 50 vehicles per minute, and the average vehicle speed is 30 kilometers per hour.

[0036] Next, the electronic map database of the target road segment is retrieved to extract its topology. The electronic map database stores rich road information, including road direction, length, number of lanes, and intersection connections. This information is extracted from the electronic map database to generate a second static road network subset. For example, the electronic map shows that the target road segment is a two-way four-lane road, 2 kilometers long, and connected to five surrounding intersections. This data provides the basic framework for subsequent analysis of traffic flow.

[0037] Then, historical traffic fluctuation data of the target road segment is used for correlation analysis. This historical traffic fluctuation data can be obtained from traffic management department databases, traffic research institution monitoring data, and other sources. This data includes information such as traffic flow and speed changes of the target road segment over different time periods. Those skilled in the art can perform correlation analysis on this historical data, for example, comparing traffic flow patterns during weekly weekday morning rush hours or analyzing the impact of seasonal factors on the road segment's traffic, thereby generating a third correlation factor dataset. This dataset can reveal the long-term trends and potential patterns of traffic changes on the target road segment.

[0038] Finally, the three datasets were spatiotemporally aligned. Since the three datasets have different data sources and collection times, they needed to be integrated into a unified spatiotemporal framework through temporal calibration and spatial matching. For example, data collected at a specific moment in the real-time dynamic dataset was matched and integrated with the corresponding spatial location on the electronic map and related data from the same time period in historical data. Through this spatiotemporal alignment, a spatiotemporal traffic dataset was constructed.

[0039] Through the above steps of collecting data from multiple sources and performing spatiotemporal alignment and integration, traffic data was collected comprehensively from multiple dimensions, constructing a spatiotemporal traffic dataset that can more accurately and comprehensively reflect the traffic conditions of the target road segment, providing a solid data foundation for subsequent accurate traffic congestion prediction and management.

[0040] Furthermore, step A100 in the method provided in this application embodiment includes:

[0041] A150: Divide the spatiotemporal traffic dataset into grids and construct a spatiotemporal grid matrix.

[0042] A160: Traverse the spatiotemporal grid matrix to extract features and obtain multiple traffic feature vectors, which correspond to the spatiotemporal grid matrix.

[0043] A170: Based on the multiple traffic feature vectors, classify the traffic state of the spatiotemporal traffic dataset and determine the probability of multiple congestion levels.

[0044] A180: Map the probabilities of the multiple congestion levels to the spatiotemporal grid matrix to construct the multi-level congestion probability distribution map.

[0045] Optionally, firstly, a spatiotemporal grid matrix is ​​constructed by dividing the spatiotemporal traffic dataset into grids. The spatiotemporal traffic dataset contains rich traffic information, such as real-time vehicle speed, traffic flow, road segment topology, and historical traffic fluctuation data. The traffic space of the target area is divided into small grid units according to certain rules. For example, based on the length of road segments and the distribution of intersections, each 500-meter interval or the area between each intersection can be considered a grid unit. Simultaneously, combining this with the time dimension, with time intervals of 15 minutes or 30 minutes, a three-dimensional spatiotemporal grid matrix is ​​constructed. Each grid has clear spatial and temporal boundaries, facilitating subsequent precise location and analysis of traffic data.

[0046] Next, feature extraction is performed by traversing the spatiotemporal grid matrix to obtain multiple traffic feature vectors. Within each grid cell, various feature data related to traffic conditions are extracted. For example, the average vehicle speed and traffic flow for that grid within a specific time period are obtained from real-time dynamic data; the number of lanes and road grade of the road segment where the grid is located are extracted from static road network data; and combined with historical traffic fluctuation data, information such as the past traffic flow trend and congestion frequency for that time period and region is obtained. These data are integrated and processed to form a traffic feature vector containing multi-dimensional information. For example, the traffic feature vector of a certain grid may include: an average vehicle speed of 30 km / h, a traffic flow of 200 vehicles / 15 minutes, 4 lanes, a road grade of arterial road, and a congestion frequency of 30% in the same time period over the past week. Furthermore, each traffic feature vector corresponds one-to-one with a grid cell in the corresponding spatiotemporal grid matrix.

[0047] Then, based on multiple traffic feature vectors, the spatiotemporal traffic dataset is classified to determine the probability of multiple congestion levels. These traffic feature vectors are analyzed and trained using a neural network algorithm. First, a multi-layer neural network model is constructed, containing an input layer, multiple hidden layers, and an output layer. The previously extracted traffic feature vectors are used as node data in the input layer, such as average vehicle speed, traffic flow, number of lanes, road grade, and historical congestion frequency. The hidden layers consist of many neurons connected by weights. During the training phase, a large number of traffic data samples labeled with congestion states (mild congestion, moderate congestion, severe congestion, or free flow, etc.) are input into the neural network. During data propagation, the input layer data undergoes weighted calculations and activation function processing by the hidden layer neurons, gradually extracting higher-level features. With each hidden layer, the feature representation of the data becomes more abstract. When the data propagates to the output layer, a predicted congestion level probability distribution is obtained. This result is compared with the actual congestion level of the sample to calculate the loss value, such as cross-entropy loss. Next, the weights between neurons are adjusted based on the loss value using the backpropagation algorithm, continuously reducing the loss value. After multiple iterations of training with a large number of data samples, the neural network gradually learns the complex intrinsic relationship between traffic features and congestion status, ultimately establishing a stable relationship model. At this point, by inputting the current traffic feature vector into the trained model, the model can accurately output the probability that the grid cell is in different congestion levels.

[0048] Finally, the determined probabilities of multiple congestion levels are mapped to a spatiotemporal grid matrix to construct a multi-level congestion probability distribution map. The probabilities of different congestion levels corresponding to each grid cell are visually labeled on the previously constructed spatiotemporal grid matrix. Different colors or shades of color can be used to represent different congestion level probabilities; for example, green indicates a high probability of smooth traffic flow, yellow indicates a relatively high probability of mild congestion, and red indicates a relatively high probability of severe congestion. This results in an intuitive multi-level congestion probability distribution map, clearly showing the distribution of the probability of traffic congestion occurring in different time and spatial ranges.

[0049] By following the steps described above—from grid partitioning, feature extraction, traffic state classification to probability mapping to construct a distribution map—the complex traffic situation is presented in an intuitive and accurate multi-level congestion probability distribution map, providing strong decision support for subsequent traffic congestion prediction and management.

[0050] Furthermore, step A200 in the method provided in this application embodiment includes:

[0051] A210: Analyze the road surveillance video stream of the target road segment to obtain real-time traffic event data.

[0052] A220: Spatiotemporally correlate the real-time traffic event data with the multi-level congestion probability distribution map to generate a dynamic event impact weight matrix.

[0053] A230: A spatiotemporal graph convolutional network is used to perform multi-step iterative prediction of the influence weight matrix of the dynamic event, and a road traffic speed decay curve is plotted.

[0054] A240: Traverse the road traffic speed decay curve to identify congestion and determine multiple congestion propagation nodes.

[0055] A250: Connect the multiple congestion propagation nodes according to the road segment topology to generate the congestion evolution path.

[0056] In this embodiment, the spatiotemporal graph convolutional network is a neural network model that integrates graph convolution and time series analysis capabilities, specifically designed to process data with spatiotemporal characteristics. The congestion propagation node is a key location determined based on the road speed decay curve during the analysis of traffic congestion development.

[0057] Specifically, firstly, the road surveillance video stream of the target road segment is analyzed to obtain real-time traffic event data. Surveillance cameras are then widely deployed on the road, continuously collecting video stream data. Video analytics techniques are then used to process the images in the video frame by frame.

[0058] The algorithm identifies target objects such as vehicles, pedestrians, traffic accident scenes, and road construction areas using object detection algorithms. First, the images from road surveillance videos are preprocessed to a format suitable for the algorithm. Next, the algorithm uses pre-trained models (such as convolutional neural network-based models, constructed in the same way as step A170) to extract features from the images. These models have been trained on a large number of images containing various types of target objects and are able to identify the characteristic patterns of different objects. Then, a region proposal network generates candidate regions that may contain target objects. These candidate regions are then classified and their locations regressed to ultimately determine the category and location of target objects such as vehicles, pedestrians, traffic accident scenes, and road construction areas.

[0059] Behavioral analysis algorithms are used to determine if abnormal traffic behavior exists, such as sudden braking or driving in the wrong direction. Based on the target detection results, the trajectory of the target object is first followed. By analyzing parameters such as speed, direction, and acceleration of the trajectory, combined with preset normal behavior pattern rules, anomalies are determined. For example, if a vehicle's speed drops sharply beyond the normal deceleration range in a short period of time, it is identified as sudden braking; if the vehicle's direction of movement is opposite to the prescribed direction of the road, it is identified as driving in the wrong direction.

[0060] Next, the real-time traffic event data is spatiotemporally correlated with the multi-level congestion probability distribution map to generate a dynamic event impact weight matrix. The specific steps are explained in detail in A221-A225.

[0061] Then, a spatiotemporal graph convolutional network is used to perform multi-step iterative prediction of the dynamic event impact weight matrix, and to plot the road traffic speed decay curve. A spatiotemporal graph convolutional network is a neural network model specifically designed for processing spatiotemporal data. Taking the dynamic event impact weight matrix as input, the model considers the topology of the traffic network (i.e., the connectivity between road segments) and its temporal changes. Internally, convolutional operations extract features from the weights of spatially adjacent grid regions, capturing the spatial propagation characteristics of event impact; a recurrent neural network structure models the temporal changes in weights, learning the evolution of event impact over time. After multi-step iterative calculations, the model predicts the changes in traffic speed at different time points and on different road segments. Plotting these predicted speed data in chronological order yields the road traffic speed decay curve, which visually demonstrates the decreasing trend of road traffic speed over time, reflecting the development of congestion.

[0062] Next, the road speed decay curve is traversed to identify congestion and determine multiple congestion propagation nodes. Several congestion judgment thresholds are set (which can be set by those skilled in the art based on actual traffic conditions). For example, if the road speed is below a certain value (e.g., 20 km / h) for a sustained period (e.g., 15 minutes), the road segment is judged to be congested. Along the speed decay curve, the time points when the speed drops below the threshold and meets the duration condition, and the corresponding road segment locations, are identified. These locations are the congestion propagation nodes. For example, if the speed decay curve shows that a certain road segment's speed is consistently below 20 km / h between 10:00 and 10:15, then the location corresponding to that road segment is identified as a congestion propagation node. These nodes mark key areas where congestion may begin to spread or worsen.

[0063] Finally, multiple congestion propagation nodes are connected according to the road segment topology to generate a congestion evolution path. The road segment topology describes the connection methods and directions between roads. Based on the previously determined locations of the congestion propagation nodes, combined with road segment topology information such as which road segments are interconnected and intersection turning rules, congestion propagation nodes at adjacent time points are connected by actual roads in chronological order to form a continuous path, i.e., the congestion evolution path. This path shows which road segments congestion might propagate and spread from its starting point, providing crucial information for traffic management departments to take proactive measures to alleviate congestion.

[0064] Through the steps described above—from acquiring real-time data, correlation analysis, model prediction, congestion identification to path generation—the system achieves accurate prediction of congestion evolution paths, providing strong support for the early development of effective traffic management strategies.

[0065] Furthermore, step A220 in the method provided in this application embodiment includes:

[0066] A221: Based on the real-time traffic event data, location identification is performed to determine the geographical coordinates of multiple events.

[0067] A222: Construct an event-road network mapping network, transform the geographic coordinates of the multiple events to the event-road network mapping network, and extract upstream and downstream related nodes.

[0068] A223: Based on the upstream and downstream related nodes, conduct an event impact assessment to determine the scope of the event's impact.

[0069] A224: Divide the multi-level congestion probability distribution map into regions according to the event impact range data, and determine multiple impact level regions.

[0070] A225: Calculate the weight increment based on the multiple influence level regions to construct the dynamic event influence weight matrix.

[0071] In this embodiment, the event-road network mapping network is a model built based on a geographic information system and road network topology, primarily used to establish the correspondence between the geographic coordinates of traffic events and the actual road network. Upstream and downstream related nodes refer to the road nodes located upstream and downstream of the location where the traffic event occurred in the event-road network mapping network. Event impact range data refers to the specific range of impact information on the road network traffic conditions determined after evaluating various factors related to the traffic event.

[0072] Specifically, firstly, location identification is performed based on real-time traffic incident data to determine the geographic coordinates of multiple incidents. After obtaining real-time traffic incident data through road surveillance video analysis, technologies such as the Global Positioning System (GPS) and Geographic Information System (GIS) are used to accurately locate each traffic incident. For example, when a traffic accident is detected, the latitude and longitude coordinates of the accident site are determined through the vehicle's built-in GPS device or nearby base stations. If multiple locations are involved (such as the area where accident vehicles are scattered), multiple geographic coordinates are obtained, providing an accurate spatial reference for subsequent analysis.

[0073] Next, an event-road network mapping network is constructed to transform the geographic coordinates of multiple events into this network and extract upstream and downstream related nodes. During the construction of the event-road network mapping network, road topology data is first extracted from the electronic map database. This data includes detailed information such as road length, direction, number of lanes, intersection connection methods, and road grade. Roads are then abstracted into a graph structure composed of nodes and edges, with each intersection and road junction as a node and road segments as edges connecting nodes. Subsequently, attribute information is assigned to nodes and edges, such as the geographic coordinates of nodes, the direction of travel of edges, and speed limits. Next, geocoding technology is used to accurately match the actual geographic coordinates of roads with the nodes and edges in the graph structure, establishing a correspondence between geospatial data and the road network structure. Simultaneously, the flow direction of the road network is analyzed and marked to clarify the vehicle travel direction of each road segment. Finally, a data indexing and query mechanism is constructed to quickly map the input event geographic coordinates to the corresponding nodes and road segments, thus completing the construction of the event-road network mapping network and laying the foundation for subsequent operations such as extracting upstream and downstream related nodes.

[0074] Furthermore, the event-road network mapping network is constructed based on the road topology structure of electronic maps, establishing a correspondence between actual geographic coordinates and nodes and road segments in the road network. When the geographic coordinates of an event are input into the mapping network, the road segment where those coordinates are located can be automatically identified, and the upstream and downstream related nodes of the event occurrence point can be determined based on road flow relationships. For example, if an accident occurs on a main road, the mapping network can identify three intersection nodes upstream and two intersection nodes downstream of the accident point; these nodes are key areas where traffic flow is affected by the event.

[0075] Then, an impact assessment is conducted based on upstream and downstream related nodes to determine the scope of the event's impact. The assessment process comprehensively considers factors such as event type (traffic accidents, road construction, etc.), severity (number of vehicles involved, scale of construction), time of occurrence (morning / evening rush hour or off-peak), and traffic flow on the road segment. For example, a multi-vehicle rear-end collision on a main road during morning rush hour may affect multiple intersections upstream and downstream due to high traffic volume and limited road capacity; while a minor collision during off-peak hours may only affect the area near the accident site. By establishing an assessment model, the specific scope of the event's impact is calculated, such as a 1-kilometer stretch upstream and downstream of the accident site.

[0076] When establishing the assessment model, a large amount of historical data on different types of traffic events was first collected, covering various events such as traffic accidents and road construction. The data included information such as event type, severity (e.g., damage to vehicles involved in the accident, number of construction equipment and personnel), time of occurrence, traffic flow on the affected road segment, road grade, and changes in traffic conditions before and after the event. Next, statistical methods were used to preprocess the data, removing noise and normalizing it to convert various factors into a unified and calculable numerical form. Then, a deep learning model (such as a convolutional neural network, following the same process as step A170) was used to train the model, using event type, severity, time of occurrence, and traffic flow on the affected road segment as input features, and the actual impact range of the event as the output label. During training, the model parameters were continuously adjusted to enable the model to accurately learn the complex relationship between various factors and the impact range of the event. After multiple rounds of iterative training and validation, an assessment model that could accurately calculate the impact range of an event based on the input factors was finally established, thus obtaining the event impact range data.

[0077] Next, the multi-level congestion probability distribution map is divided into regions based on the event's impact range data, identifying multiple impact levels. According to the event's impact range, different areas are divided on the multi-level congestion probability distribution map and assigned different levels based on the degree of impact. Generally, these can be divided into core impact areas, secondary impact areas, and peripheral impact areas. The core impact area is the location where the event occurred and the directly affected road sections, experiencing the greatest traffic disruption; the secondary impact area consists of road sections adjacent to the core area and indirectly affected; the peripheral impact area is an area that may be slightly affected by factors such as traffic detours. For example, the core impact area is assigned the highest level, level 3, the secondary impact area is level 2, and the peripheral impact area is level 1.

[0078] Finally, a dynamic event impact weight matrix is ​​constructed by calculating weight increments based on multiple impact level regions. A corresponding weight increment is assigned to each impact level region, with higher levels receiving larger increments. For example, the weight increment for the core impact region is set to 0.8, for the secondary impact region 0.5, and for the peripheral impact region 0.2. These weight increments are mapped to a grid matrix of a multi-level congestion probability distribution map. Each grid receives a corresponding weight adjustment based on its region, ultimately forming the dynamic event impact weight matrix. This matrix clearly demonstrates the differences in the degree of impact of real-time traffic events on various regions of the traffic network.

[0079] Through the steps described above, from event location, network mapping, impact assessment, regional division to matrix construction, and by comprehensively utilizing geographic information technology, road topology analysis, and multi-factor evaluation models, real-time traffic event data is deeply integrated with multi-level congestion probability distribution maps to generate a dynamic event impact weight matrix. This allows for precise quantification of the impact of real-time traffic events on the traffic network, providing more accurate data support for traffic congestion prediction.

[0080] Furthermore, step A300 in the method provided in this application embodiment includes:

[0081] A310: Based on the multiple congestion propagation nodes, perform diffusion analysis on the congestion evolution path to determine the diffusion direction vector.

[0082] A320: Based on the diffusion direction vector and the multiple congestion propagation nodes, conduct traffic management analysis to determine multiple candidate traffic management strategies.

[0083] A330: Prioritize the multiple candidate traffic diversion strategies according to the real-time traffic pressure index, determine the diversion trigger sequence, and construct the dynamic traffic diversion instruction set based on the diversion trigger sequence.

[0084] In this embodiment, the diffusion direction vector is a vector used to describe the direction and extent of traffic congestion diffusion in the road network. The real-time traffic pressure index is a quantitative indicator that comprehensively reflects the current traffic congestion level and traffic pressure on a road segment.

[0085] Specifically, firstly, a diffusion analysis of the congestion evolution path is performed based on multiple congestion propagation nodes to determine the diffusion direction vector. Congestion propagation nodes are key locations for congestion spread determined by road speed decay curves. By analyzing the temporal and spatial distribution of these nodes, vector calculation methods are used, combined with road topology and traffic flow direction, to determine the potential direction of congestion spread. For example, if multiple consecutive congestion propagation nodes show a distribution trend from east to west, and the traffic flow direction of the corresponding road segment is also from east to west, then the congestion diffusion direction vector can be determined to be westward. Simultaneously, factors such as the carrying capacity of roads in each direction and the connectivity of the surrounding road network are comprehensively considered to further refine the diffusion direction vector, making it more closely reflect the actual congestion diffusion trend.

[0086] Next, based on the direction of congestion propagation, traffic management strategies are developed for different congestion propagation nodes, taking into account the characteristics of their respective road segments. For nodes near intersections, traffic light timings can be adjusted to increase the green light duration for non-congested directions, guiding vehicles to pass quickly and alleviating congestion pressure. For nodes on main roads, detour prompts can be issued through variable message signs to guide vehicles to surrounding side roads. If there are parking lots near the node, temporary parking resources can be coordinated to reduce the number of vehicles parked on the road. For example, at a congestion propagation node on a main road, considering the large number of surrounding side roads and their relatively good traffic capacity, a candidate strategy of guiding vehicles to detour to side roads is developed. At a congestion propagation node at an intersection, a candidate strategy of extending the green light time is developed, ultimately forming multiple different types of candidate traffic management strategies applicable to different scenarios.

[0087] Then, multiple candidate traffic management strategies are prioritized according to the real-time traffic pressure index to determine the traffic management trigger sequence. A dynamic traffic management instruction set is then constructed based on this sequence. The real-time traffic pressure index is a quantitative indicator calculated by integrating real-time traffic data such as road segment traffic flow, average vehicle speed, and vehicle queue length. First, real-time traffic data such as road segment traffic flow, average vehicle speed, and vehicle queue length are collected. Then, each data point is assigned a weight based on its importance to traffic pressure. Generally, higher traffic flow has a greater impact on traffic pressure and can therefore have a relatively higher weight. The weights for average vehicle speed and vehicle queue length are reasonably allocated by those skilled in the art based on the actual situation. Finally, these data are weighted and summed to obtain a specific value, the real-time traffic pressure index, which directly and quantitatively reflects the current congestion level and traffic pressure situation of the road segment.

[0088] Finally, each candidate traffic diversion strategy is matched and evaluated against the real-time traffic pressure index, and strategies that can alleviate congestion on high-pressure road sections are given higher priority. For example, when a road section is severely saturated with traffic flow and the average vehicle speed is extremely low, detour strategies that can quickly divert vehicles have higher priority than strategies that adjust traffic light timings. Based on the priority, a traffic diversion trigger sequence is determined, and these strategies are converted into specific instructions according to the trigger sequence, such as traffic light control instructions, information dissemination instructions, and traffic control instructions, ultimately constructing a dynamic traffic diversion instruction set.

[0089] By following a series of steps—from analyzing the direction of congestion spread to formulating candidate diversion strategies, and then prioritizing and constructing an instruction set based on real-time traffic pressure indices—it is possible to accurately generate dynamic diversion instruction sets based on congestion evolution paths, thereby effectively improving the targeting and efficiency of traffic congestion management and ensuring smooth road traffic.

[0090] In summary, the AI-based traffic congestion prediction and management method provided in this application has the following technical effects:

[0091] This application constructs a spatiotemporal traffic dataset by collecting multi-source traffic data. Based on this dataset, traffic state classification is performed to generate a multi-level congestion probability distribution map. Combined with real-time traffic event data, congestion prediction is performed to determine the congestion evolution path. Finally, a dynamic traffic management instruction set is generated based on the congestion evolution path and sent to traffic control equipment. Through the integrated analysis of multi-source data, model construction, and algorithm application, accurate prediction of traffic congestion is achieved, and dynamic traffic management strategies are efficiently generated and executed based on real-time traffic conditions, thereby improving traffic management efficiency and road capacity.

[0092] Example 2, as Figure 2As shown, based on the same inventive concept as in Embodiment 1 above, this application provides an artificial intelligence-based traffic congestion prediction and management system, the system comprising:

[0093] Distribution map generation module 1 is used to collect multi-source traffic data to construct a spatiotemporal traffic dataset, classify traffic states based on the spatiotemporal traffic dataset, and generate a multi-level congestion probability distribution map.

[0094] Congestion prediction module 2 is used to combine the multi-level congestion probability distribution map with real-time traffic event data to predict congestion and determine the congestion evolution path.

[0095] Traffic management module 3 is used to manage traffic according to the congestion evolution path, generate a dynamic diversion instruction set, and send the dynamic diversion instruction set to the traffic control equipment of the target road segment for traffic congestion management.

[0096] Furthermore, the distribution map generation module 1 is used to perform the following steps:

[0097] Based on the target road segment, road test traffic sensing is performed to generate a first real-time dynamic subset; the road segment topology is extracted from the electronic map database of the target road segment to generate a second static road network subset; historical traffic fluctuation data of the target road segment is introduced for correlation analysis to generate a third correlation factor dataset; the first real-time dynamic subset, the second static road network subset, and the third correlation factor dataset are spatiotemporally aligned to construct the spatiotemporal traffic dataset.

[0098] Furthermore, the distribution map generation module 1 is used to perform the following steps:

[0099] The spatiotemporal traffic dataset is divided into grids to construct a spatiotemporal grid matrix. Features are extracted by traversing the spatiotemporal grid matrix to obtain multiple traffic feature vectors, which correspond to the spatiotemporal grid matrix. Based on the multiple traffic feature vectors, the spatiotemporal traffic dataset is classified into traffic states to determine multiple congestion level probabilities. The multiple congestion level probabilities are mapped to the spatiotemporal grid matrix to construct the multi-level congestion probability distribution map.

[0100] Furthermore, the congestion prediction module 2 is used to perform the following steps:

[0101] The road surveillance video stream of the target road segment is parsed to obtain real-time traffic event data; the real-time traffic event data is spatiotemporally correlated with the multi-level congestion probability distribution map to generate a dynamic event influence weight matrix; a spatiotemporal graph convolutional network is used to perform multi-step iterative prediction on the dynamic event influence weight matrix to draw a road traffic speed decay curve; the road traffic speed decay curve is traversed to identify congestion and determine multiple congestion propagation nodes; the multiple congestion propagation nodes are connected according to the road segment topology to generate the congestion evolution path.

[0102] Furthermore, the congestion prediction module 2 is used to perform the following steps:

[0103] Based on the real-time traffic event data, location identification is performed to determine the geographic coordinates of multiple events; an event-road network mapping network is constructed to transform the geographic coordinates of the multiple events into the event-road network mapping network, and upstream and downstream related nodes are extracted; based on the upstream and downstream related nodes, event impact assessment is performed to determine the event impact range data; according to the event impact range data, the multi-level congestion probability distribution map is divided into regions to determine multiple impact level regions; based on the multiple impact level regions, weight increments are calculated to construct the dynamic event impact weight matrix.

[0104] Furthermore, the traffic management module 3 is used to perform the following steps:

[0105] Based on the multiple congestion propagation nodes, a diffusion analysis is performed on the congestion evolution path to determine the diffusion direction vector; traffic diversion analysis is performed according to the diffusion direction vector and the multiple congestion propagation nodes to determine multiple candidate diversion strategies; the multiple candidate diversion strategies are prioritized according to the real-time traffic pressure index to determine the diversion trigger sequence, and the dynamic diversion instruction set is constructed based on the diversion trigger sequence.

[0106] In embodiment three, this application also provides a computer-readable storage medium that can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the AI-based traffic congestion prediction and management method in the embodiments of this application, thereby realizing the above-mentioned AI-based traffic congestion prediction and management method.

[0107] It should be understood that the embodiments disclosed in this application and the above description enable those skilled in the art to implement this application. However, this application is not limited to the embodiments mentioned above. It should be understood that those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this invention, and should all be included within the protection scope of this application.

Claims

1. A traffic congestion prediction and management method based on artificial intelligence, characterized in that, The method includes: Collect multi-source traffic data to construct a spatiotemporal traffic dataset, classify traffic conditions based on the spatiotemporal traffic dataset, and generate a multi-level congestion probability distribution map. By combining the multi-level congestion probability distribution map with real-time traffic event data, congestion prediction is performed to determine the congestion evolution path, including: The road surveillance video stream of the target road segment is analyzed to obtain real-time traffic event data; The real-time traffic event data is spatiotemporally correlated with the multi-level congestion probability distribution map to generate a dynamic event impact weight matrix, including: identifying the location based on the real-time traffic event data and determining the geographic coordinates of multiple events; Construct an event-road network mapping network, transform the geographic coordinates of the multiple events to the event-road network mapping network, and extract upstream and downstream related nodes; Based on the upstream and downstream related nodes, an event impact assessment is conducted to determine the scope of the event's impact. Based on the event impact range data, the multi-level congestion probability distribution map is divided into regions to determine multiple impact level regions; The weight increment is calculated based on the multiple influence level regions, and the dynamic event influence weight matrix is ​​constructed. A spatiotemporal graph convolutional network is used to perform multi-step iterative prediction of the weight matrix of the dynamic event influence, and a road traffic speed decay curve is plotted. The congestion is identified by traversing the road traffic speed decay curves to determine multiple congestion propagation nodes. The multiple congestion propagation nodes are connected according to the road segment topology to generate the congestion evolution path; Traffic management is carried out based on the congestion evolution path, a dynamic diversion instruction set is generated, and the dynamic diversion instruction set is sent to the traffic control equipment of the target road segment for traffic congestion management. Collect multi-source traffic data to construct a spatiotemporal traffic dataset, including: Based on the target road segment, road test traffic sensing is performed to generate the first real-time dynamic subset dataset; The electronic map database of the target road segment is retrieved to extract the road segment topology and generate a second static road network subset. Historical traffic fluctuation data of the target road segment are introduced for correlation analysis to generate a third correlation factor dataset. The first real-time dynamic sub-dataset, the second static road network sub-dataset, and the third correlation factor dataset are spatiotemporally aligned to construct the spatiotemporal traffic dataset.

2. The traffic congestion prediction and management method based on artificial intelligence as described in claim 1, characterized in that, Based on the aforementioned spatiotemporal traffic dataset, traffic state classification is performed to generate a multi-level congestion probability distribution map, including: The spatiotemporal traffic dataset is divided into grids to construct a spatiotemporal grid matrix. Feature extraction is performed by traversing the spatiotemporal grid matrix to obtain multiple traffic feature vectors, and the multiple traffic feature vectors have a corresponding relationship with the spatiotemporal grid matrix; Based on the multiple traffic feature vectors, the spatiotemporal traffic dataset is classified into traffic states to determine the probabilities of multiple congestion levels. The probabilities of the multiple congestion levels are mapped to the spatiotemporal grid matrix to construct the multi-level congestion probability distribution map.

3. The traffic congestion prediction and management method based on artificial intelligence as described in claim 1, characterized in that, Traffic management is performed based on the congestion evolution path, generating a dynamic traffic management instruction set, including: Based on the multiple congestion propagation nodes, a diffusion analysis is performed on the congestion evolution path to determine the diffusion direction vector; Traffic management analysis is performed based on the diffusion direction vector and the multiple congestion propagation nodes to determine multiple candidate traffic management strategies. The multiple candidate traffic diversion strategies are prioritized according to the real-time traffic pressure index to determine the diversion trigger sequence, and the dynamic diversion instruction set is constructed based on the diversion trigger sequence.

4. A traffic congestion prediction and management system based on artificial intelligence, characterized in that: The system is used to implement the AI-based traffic congestion prediction and management method according to any one of claims 1-3, the system comprising: The distribution map generation module is used to collect multi-source traffic data to construct a spatiotemporal traffic dataset, classify traffic states based on the spatiotemporal traffic dataset, and generate a multi-level congestion probability distribution map. The congestion prediction module is used to combine the multi-level congestion probability distribution map with real-time traffic event data to predict congestion and determine the congestion evolution path. The traffic management module is used to manage traffic based on the congestion evolution path, generate a dynamic diversion instruction set, and send the dynamic diversion instruction set to the traffic control equipment of the target road segment for traffic congestion management.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the AI-based traffic congestion prediction and management method as described in any one of claims 1 to 3.

Images