AI-based intelligent optimization management system for urban traffic congestion

Abstract

This application relates to the field of artificial intelligence, specifically disclosing an AI-based intelligent optimization management system for urban traffic congestion. It aims to address the congestion and inefficiency problems caused by sluggish response, localized optimization, and low data utilization in existing traffic management systems. The system comprises five modules: data acquisition and fusion, traffic state perception and prediction, decision optimization, instruction issuance and execution, and human-computer interaction and visualization. Through multi-source data fusion, graph neural network prediction, and multi-agent reinforcement learning, it achieves real-time traffic flow perception, accurate prediction, and adaptive control, effectively alleviating congestion and improving the overall operational efficiency and resilience of the road network.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence, specifically relating to an intelligent optimization management system for urban traffic congestion based on artificial intelligence. Background Technology

[0002] Artificial intelligence (AI), as a core frontier in the field of information technology, has been widely applied in various fields such as data analysis, pattern recognition, and intelligent decision-making, profoundly changing social production and lifestyles. Its powerful learning, reasoning, and optimization capabilities provide innovative approaches to solving dynamic management problems in complex mega-systems. Particularly in the field of urban governance, AI technology has demonstrated enormous potential and value in improving resource allocation efficiency and optimizing public services. Urban transportation systems are a crucial component of the complex mega-systems of modern society, and their efficient operation is directly related to urban economic development and the quality of life of residents.

[0003] Urban traffic congestion management, a core challenge in modern urban planning and operation, aims to optimize the allocation of road resources through effective guidance and scheduling of traffic flow, thereby alleviating traffic pressure and improving traffic efficiency. Research and practice in this field are crucial for building efficient and sustainable urban transportation systems.

[0004] Existing technologies face numerous challenges in urban traffic congestion management. Traditional traffic management models rely heavily on historical experience and fixed rules, lacking the ability to dynamically perceive, accurately predict, and rapidly respond to real-time traffic flow. This results in delayed and inefficient responses to sudden congestion or abnormal traffic conditions. Although Intelligent Transportation Systems (ITS) introduce a large number of data collection devices, they suffer from severe data silos and difficulties in effectively integrating heterogeneous data, failing to form a comprehensive and accurate understanding of the urban traffic situation. Furthermore, existing optimization algorithms are mostly based on local information for decision-making, making it difficult to achieve global collaborative optimization of the urban traffic network. This leads to uneven traffic flow distribution and the transfer of congestion between different areas, and path planning and traffic guidance lack real-time adaptive adjustment mechanisms. These shortcomings collectively make it difficult for existing technologies to effectively cope with the complex and ever-changing urban traffic environment, urgently requiring a management system capable of intelligent prediction, accurate decision-making, and global optimization. Summary of the Invention

[0005] This invention provides an AI-based intelligent optimization management system for urban traffic congestion, aiming to address the common technical contradictions in existing urban traffic management systems when dealing with complex and dynamic traffic conditions, such as slow response, localized optimization leading to global suboptimal performance, and low data utilization. Existing technologies typically rely on preset traffic rules, historical data analysis, or isolated intersection control strategies, making it difficult to perceive the global state of the urban traffic network in real time and accurately predict its evolution trends. Furthermore, they are unable to deeply integrate and intelligently decide on massive amounts of heterogeneous data, resulting in frequent traffic congestion and low traffic efficiency. This invention constructs a multi-layered, collaborative intelligent management framework to achieve real-time perception, accurate prediction, and adaptive optimization control of urban traffic flow, thereby significantly improving the overall operational efficiency and resilience of the urban traffic network.

[0006] To achieve the above objectives, the present invention provides an intelligent optimization management system for urban traffic congestion based on artificial intelligence, which includes: a data acquisition and fusion module, a traffic state perception and prediction module, a decision optimization module, an instruction issuance and execution module, and a human-computer interaction and visualization module.

[0007] In one embodiment of the present invention, a data acquisition and fusion module is used to continuously acquire and preprocess various real-time traffic data in the urban transportation network, and to standardize and fuse heterogeneous data. This data acquisition and fusion module includes:

[0008] The traffic sensor data acquisition unit is used to acquire data from various physical sensors distributed in the road network, including but not limited to induction coil detectors, geomagnetic detectors, video detectors, millimeter-wave radar detectors, and RFID readers. This unit acquires real-time information on traffic flow, vehicle speed, lane occupancy, vehicle queue length, and license plate recognition.

[0009] The mobile terminal data acquisition unit is used to obtain anonymized location, speed, direction, and travel trajectory data from GPS devices, mobile communication base stations, and smartphone applications of connected vehicles.

[0010] The environmental and event data acquisition unit is used to acquire real-time weather data, road construction information, traffic accident reports, large-scale event forecasts, and information on emergencies.

[0011] The data preprocessing and fusion unit is used to clean, denoise, convert formats, align time, and spatially map the collected raw data. It employs multi-source data fusion algorithms, such as Kalman filtering or deep learning-based fusion methods, to integrate heterogeneous data from different sources, with different sampling frequencies, and different precisions into a unified, high-dimensional representation of urban traffic conditions. This fusion process generates a comprehensive dataset containing spatiotemporal traffic flow parameters, event parameters, and environmental parameters.

[0012] In one embodiment of the present invention, a traffic condition perception and prediction module is used to accurately perceive the real-time operating status of an urban traffic network based on a comprehensive dataset provided by a data acquisition and fusion module, and to predict short- to medium-term traffic flow trends. This traffic condition perception and prediction module includes:

[0013] The real-time traffic status perception unit analyzes the fused data and calculates and updates key traffic indicators such as vehicle density, average speed, capacity, congestion level, and queue length for each road segment and intersection on the road grid in real time. This unit accurately identifies and locates traffic congestion areas and assesses the severity and impact of congestion by setting dynamic thresholds and using pattern recognition algorithms.

[0014] The short-to-medium-term traffic flow prediction unit is used to predict traffic flow, average speed, and congestion index within a specified future time window (e.g., 5 minutes, 15 minutes, or 30 minutes) based on historical traffic data, real-time traffic conditions, and environmental and event data, using a deep learning model. This prediction unit employs a graph neural network-based prediction model, capable of capturing the complex spatiotemporal correlations within the traffic network. For example, the prediction model can be represented as:

[0015]

[0016] in, Indicates time The traffic state vector for each time moment includes parameters such as flow rate, speed, and density for each road segment; This represents a nonlinear mapping function implemented by a deep neural network (e.g., a spatiotemporal graph network that combines graph convolutional networks and long short-term memory networks); A graph structure representing the urban transportation network, where nodes represent intersections or road segments, and edges represent the connections between road segments; Indicates from the current time to the past Traffic state sequence at each time step Indicates time Observed environmental and event data vectors; This represents the model's learning parameters. The model learns the spatiotemporal evolution of traffic flow to provide a reliable predictive basis for subsequent decision optimization.

[0017] In one embodiment of the present invention, a decision optimization module is used to generate globally optimal traffic signal control strategies, dynamic path planning schemes, and emergency response measures based on real-time perception results and prediction information provided by the traffic state perception and prediction module. This decision optimization module employs a multi-agent reinforcement learning framework to achieve adaptive collaborative optimization of the traffic network. The decision optimization module includes:

[0018] The multi-agent cooperative control unit treats the signal controllers at various intersections in the urban traffic network as independent agents. Each agent learns the optimal action strategy through interaction with the environment. The state space of each agent is defined as the traffic state of the current intersection and adjacent intersections (including traffic flow, queue length, average speed, etc.), the predicted future traffic state, and the current signal phase and duration. The action space is defined as the switching sequence of signal light phases, their duration, and the adjustment of the green light ratio. Agents learn the optimal strategy by maximizing cumulative rewards. The reward function is designed to reflect a comprehensive objective, including overall network efficiency, congestion mitigation, and travel comfort. For example, the cumulative reward function can be defined as:

[0019]

[0020] in Indicates at time step Total network rewards; Indicates the number of intersections in the network; Indicates an intersection The average vehicle delay time; Indicates an intersection The average queue length; Indicates an intersection Vehicle throughput; This indicates the number of times a vehicle stops on the network. , , , All are adjustable positive weight coefficients used to balance different optimization objectives. Each agent uses depth... The network is trained using deep reinforcement learning algorithms such as Direct Quality Network (DQN) or Proximal Policy Optimization (PPO). The value update rule can be expressed as:

[0021]

[0022] in Indicates the state Take action below The expected cumulative reward, It is the learning rate. From state Take action Later arrival state The instant rewards received; It is a discount factor, indicating the importance of future rewards; Indicates the next state The maximum obtainable Value. Through this collaborative learning mechanism, each agent can discover and execute signal control strategies that satisfy its own local optimization while also promoting the efficiency of the global network.

[0023] The dynamic route planning and guidance unit provides real-time optimal route suggestions to vehicles or navigation services connected to the system based on predicted traffic congestion. This unit comprehensively considers the actual travel time of the current road segment, the predicted future travel time, the congestion level of the road segment, and user preferences (such as shortest time, shortest distance, avoiding congestion, etc.) to calculate the optimal route from the origin to the destination. The path cost function can be expressed as:

[0024]

[0025] in Representing a path Total cost; It is a path A section of the road; It is a section of road The current actual travel time; It is a section of road Predicting future travel time, It is a section of road The level of congestion; , , These are weighting coefficients used to balance various indicators. This unit updates the toll costs of the road network in real time and dynamically generates and pushes optimal routes to users using heuristic search algorithms (such as the A* algorithm or variants of Dijkstra's algorithm) to achieve a balanced distribution of traffic flow in the road network.

[0026] The traffic incident emergency response unit is used to immediately activate emergency plans upon detecting sudden events such as traffic accidents, road construction, and natural disasters. Based on the nature, location, and severity of the event, this unit automatically adjusts signal control strategies in the affected area and route guidance plans for surrounding road sections. For example, it may extend the green light time for the obstructed direction, guide vehicles to alternative routes, or temporarily close some ramps. Through rapid response, the unit minimizes the impact of the event on traffic flow and avoids secondary congestion.

[0027] In one embodiment of the present invention, the instruction issuance and execution module is used to convert the various decision instructions generated by the decision optimization module into specific control signals and send them to the corresponding execution devices. The instruction issuance and execution module includes:

[0028] The signal control command sending unit is used to send optimized signal phase, duration, and green light ratio parameters to the traffic signal controllers at each intersection via a standard communication protocol (such as NTCIP) to achieve real-time adjustment of the traffic lights. This unit ensures the accuracy and low latency of the commands.

[0029] The information dissemination and guidance instruction sending unit is used to disseminate dynamic route planning results, congestion information, event warnings, and other information to drivers and the public in real time through variable information signs (VMS), broadcasts, and interfaces with third-party navigation applications, guiding them to choose appropriate travel routes and modes of transportation.

[0030] The execution status feedback unit is used to receive execution status feedback information from devices such as traffic signal controllers, variable information signs, and vehicle navigation systems, and send it back to the data acquisition and fusion module to form closed-loop control, which is used to evaluate the decision-making effect and further optimize the model.

[0031] As one embodiment of the present invention, the human-computer interaction and visualization module is used to provide traffic management personnel with an intuitive and comprehensive monitoring, analysis, and management platform. The human-computer interaction and visualization module includes:

[0032] The real-time situation display unit is used to display the operational status of the entire urban traffic network on a digital map in real time, including information such as vehicle speed, traffic flow, congestion level, queue length and traffic light phase of each road segment, and supports the overlay display of multi-dimensional information.

[0033] The decision-making scheme review and intervention unit allows traffic management personnel to view the traffic optimization decision-making schemes generated by the system and provides the function of manually reviewing, approving, modifying or enforcing specific instructions to deal with special situations or to conduct expert intervention.

[0034] The historical data analysis and reporting unit provides query, statistical, and analytical functions for historical traffic data, system operation logs, and optimization effect evaluation reports, providing data support for traffic planning and management decisions.

[0035] The simulation and evaluation unit provides offline simulation capabilities, allowing managers to input different traffic scenarios or events to simulate the system's response under different strategies, thus facilitating strategy pre-evaluation and optimization.

[0036] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0037] This invention constructs a closed-loop intelligent urban traffic management system integrating data acquisition, status perception, accurate prediction, intelligent decision-making, and command execution. Firstly, this invention overcomes the problems of data fragmentation and poor consistency in traditional systems through deep fusion technology of multi-source heterogeneous data, achieving real-time, comprehensive, and accurate situational awareness of the urban traffic network.

[0038] Secondly, this invention introduces a spatiotemporal traffic flow prediction model based on graph neural networks, which can effectively capture the complex nonlinear characteristics and spatial correlation of traffic flow, provide high-precision short- and medium-term traffic condition predictions, transform passive response into proactive early warning and early intervention, and thus provide a solid predictive foundation for intelligent decision-making.

[0039] Furthermore, this invention employs a multi-agent reinforcement learning framework, treating the signal controllers and path guidance systems at various intersections in the urban traffic network as collaborative learning agents. Through a sophisticated shared reward mechanism and learning algorithm, globally optimal collaboration between traffic signal control and path guidance is achieved. This framework overcomes the bottleneck of local optimization leading to suboptimal global outcomes in traditional traffic management systems. It can adaptively adjust strategies based on real-time traffic conditions and predictive information, effectively alleviating urban traffic congestion and improving road network capacity and vehicle operating efficiency.

[0040] Furthermore, this invention, through dynamic path planning and guidance units combined with real-time congestion prediction, provides users with personalized optimal travel routes, effectively guiding the balanced distribution of traffic flow in the road network, reducing traffic load on specific road sections, and decreasing vehicle travel time. Simultaneously, the traffic incident emergency response unit can rapidly generate and execute emergency plans for sudden events, minimizing the impact of emergencies on traffic operations and enhancing the resilience of the transportation system.

[0041] Finally, the human-computer interaction and visualization module provided by this invention enables traffic management personnel to intuitively monitor traffic conditions, review and intervene in system decisions, and conduct historical data analysis and simulation, greatly improving the intelligence, precision, and scientific level of traffic management. This invention significantly reduces urban traffic congestion rates, reduces vehicle fuel consumption and exhaust emissions, and improves residents' travel efficiency and satisfaction, demonstrating significant social and economic benefits. Attached Figure Description

[0042] Figure 1 This is a diagram of the overall technical architecture of the system of the present invention. Detailed Implementation

[0043] Please refer to Figure 1This embodiment discloses an intelligent optimization management system for urban traffic congestion based on artificial intelligence, aiming to comprehensively improve the operational efficiency of urban transportation networks and their resilience in dealing with complex and dynamic traffic conditions. This system constructs a closed-loop intelligent traffic management framework by integrating multi-source heterogeneous data acquisition and fusion, advanced traffic condition perception and accurate prediction, multi-agent collaborative decision-making optimization, and real-time command issuance and execution functions.

[0044] This AI-based intelligent optimization management system for urban traffic congestion includes the following core functional modules: data acquisition and fusion, traffic condition perception and prediction, decision optimization, command issuance and execution, and human-computer interaction and visualization. These modules work together to achieve in-depth insight into urban traffic flow, intelligent decision-making, and precise control.

[0045] The data acquisition and fusion module is responsible for continuously acquiring various real-time traffic data from the urban transportation network, and preprocessing, standardizing, and deeply fusing this raw data to provide a high-quality, unified data foundation for subsequent traffic condition perception and prediction. The internal components of this module work collaboratively to ensure the integrity and real-time nature of the data stream.

[0046] Specifically, the data acquisition and fusion module includes:

[0047] The traffic sensor data acquisition unit is dedicated to acquiring data from various physical sensors widely deployed in the urban road network. These sensor types include induction coil detectors, geomagnetic detectors, video detectors, millimeter-wave radar detectors, and RFID readers. This unit acquires diverse traffic parameters in real time by establishing communication interfaces with different types of sensor devices. For example, induction coil detectors provide data on traffic flow, average vehicle speed, and lane occupancy for a specific lane within a given time window by detecting changes in electromagnetic induction as vehicles pass. Geomagnetic detectors utilize the principle of Earth's magnetic field disturbance to obtain information on vehicle presence, direction of travel, and speed, and are particularly suitable for detecting the length of queues formed by parked vehicles. Video detectors, through image recognition and computer vision algorithms, can not only identify license plate information but also statistically analyze traffic flow, average speed, vehicle type, queue length, and the precise extent of congestion areas. Millimeter-wave radar detectors, with their high precision and all-weather operation, are used to measure vehicle speed, distance, and trajectory, maintaining reliable data acquisition even under adverse weather conditions. Radio frequency identification (RFID) readers are primarily used to acquire unique identification information and transit times of vehicles bearing RFID tags (e.g., public transportation vehicles or specially managed vehicles), assisting in journey time estimation. This sensor data is transmitted to the receiving end of the data acquisition and fusion module via wired or wireless communication networks (e.g., fiber optic networks, 4G, 5G). At the receiving end, the data undergoes preliminary verification, such as checking the integrity of data packets, the accuracy of timestamps, and filtering out abnormal data caused by sensor malfunctions or communication interruptions. Typical data acquisition frequencies can range from multiple times per second (e.g., video and radar) to once per minute (e.g., induction coils) to accommodate the characteristics of different data sources.

[0048] The mobile terminal data acquisition unit focuses on acquiring anonymized vehicle location, speed, direction, and complete travel trajectory data from the GPS devices, mobile communication base stations, and smartphone applications of connected vehicles. This unit establishes connections with the vehicle manufacturer's telematics system, the mobile network operator's base station data interface, and the application programming interface (API) of third-party navigation or travel service providers to acquire massive amounts of vehicle movement data. The GPS device provides the vehicle's precise latitude and longitude coordinates, altitude, instantaneous speed, and direction of travel. The mobile communication base station provides approximate location information of the vehicle within its coverage area through triangulation or signal strength analysis, serving as an important auxiliary positioning method, especially in densely populated urban areas or when GPS signals are limited. The smartphone application, through its integrated location services, provides the user's vehicle's real-time location, speed, and travel trajectory. This data undergoes strict anonymization to ensure user privacy is not violated. During data acquisition, the system implements multiple privacy protection measures, such as hash encryption of vehicle identification numbers and user identity information, differential privacy enhancement, and retaining only statistically significant macroscopic traffic flow information. The data is typically updated every 3 to 10 seconds, ensuring real-time capture of dynamic changes in traffic flow. This data is crucial for understanding macro-level traffic patterns, analyzing origin-destination (OD) travel demand, and identifying potential congestion points.

[0049] The Environmental and Event Data Acquisition Unit is responsible for acquiring various environmental and event information affecting urban traffic operations. This information includes, but is not limited to, real-time weather data, road construction information, traffic accident reports, large-scale event announcements, and emergency notifications. Weather data is acquired through connections with meteorological service agencies' weather station networks, including temperature, rainfall, snowfall, visibility, and wind speed and direction. This data is crucial for predicting the impact of slippery roads, icy conditions, or low visibility on traffic. Road construction information is acquired through integration with the information systems of urban construction management departments, including construction sections, construction time, road occupancy status, and estimated completion time, used to adjust traffic organization in advance. Traffic accident reports are acquired through integration with the accident handling system or event reporting platform of public security traffic management departments, including accident type, location, number of affected lanes, duration, and rescue progress, to achieve rapid response and traffic management. Large-scale event announcements are acquired through integration with the event management systems of cultural and sports institutions, convention centers, and other organizations, including event location, time, expected number of participants, and traffic organization plans, used to formulate traffic safety measures in advance. Information on emergencies, such as fires, emergency rescues, and mass incidents, is obtained through information sharing with the city's emergency command center, ensuring the system can detect and respond immediately. This event data is typically acquired in the form of text, structured tables, or geospatial information and transmitted to the system via a unified interface.

[0050] The data preprocessing and fusion unit is the core of the data acquisition and fusion module, responsible for the refined processing and deep fusion of the aforementioned traffic sensor data, mobile terminal data, and environmental and event data. First, multi-stage preprocessing operations are performed on the collected raw data. Data cleaning includes removing duplicate records, filling in missing values ​​(e.g., using interpolation algorithms or predictions based on historical data), and correcting erroneous inputs (e.g., through rule matching or outlier detection). Denoising techniques such as Gaussian filtering and median filtering are used to eliminate sensor noise or communication interference, improving data quality. Format conversion unifies the raw formats of different data sources into a standard format (e.g., JSON, XML, or proprietary binary format) for subsequent processing. Time alignment is a critical step. Since the sampling frequencies and timestamps of different data sources may differ, this unit uses a high-precision time synchronization protocol (e.g., Network Time Protocol) for global time calibration and resampling or interpolating the data to unify all data to a preset sampling frequency (e.g., every 30 seconds or every minute). Spatial mapping unifies data from different coordinate systems or geocoding onto a standard geographic information system base map of the urban transportation network, ensuring the consistency of all transportation elements in geospatial space.

[0051] Building upon this foundation, this unit employs multi-source data fusion algorithms, such as Kalman filtering-based or deep learning-based fusion methods, to integrate heterogeneous data from diverse sources with varying sampling frequencies and accuracies into a unified, high-dimensional representation of urban traffic conditions. Kalman filtering-based fusion methods are suitable for linear or near-linear dynamic data models and can fuse sensor data with measurement and system noise through prediction and correction loops to obtain more accurate traffic condition estimates. Deep learning fusion methods, such as multimodal deep learning models (e.g., autoencoders, convolutional neural networks, or recurrent neural networks), can automatically learn and extract deep features from heterogeneous data and achieve feature fusion and abstraction through the hierarchical structure of neural networks. For example, image features from video detectors, traffic sequences from geomagnetic detectors, GPS trajectory data, and weather conditions can be used as multimodal inputs and mapped to the same high-dimensional feature space through a shared encoder, thereby achieving deep semantic fusion.

[0052] This fusion process ultimately generates a comprehensive dataset containing spatiotemporal traffic flow parameters, event parameters, and environmental parameters. Spatiotemporal traffic flow parameters include real-time traffic volume, average speed, lane occupancy, queue length, travel time, and traffic density distribution across the road network for each road segment and intersection. Event parameters include the type, location, impact range, and duration of various emergencies or planned events. Environmental parameters include real-time weather conditions and air quality index. This comprehensive dataset is stored in a structured manner, for example, using a time-series database or a graph database, and is provided to the traffic condition perception and prediction module via a high-bandwidth interface as the underlying data support for its operation.

[0053] The traffic condition perception and prediction module, based on the comprehensive dataset provided by the data acquisition and fusion module, accurately perceives the real-time operating status of the urban traffic network and makes high-precision predictions of short- and medium-term traffic flow trends within a specified future time window, providing real-time and forward-looking traffic information for the decision optimization module.

[0054] Specifically, the traffic condition perception and prediction module includes:

[0055] The real-time traffic status perception unit is responsible for deep analysis of the fused comprehensive dataset, calculating and updating key traffic indicators for each road segment and intersection on the urban road grid in real time. These indicators include, but are not limited to: vehicle density, representing the number of vehicles per unit length of road segment, reflecting the degree of traffic congestion; average speed, representing the average speed of all vehicles on the road segment; capacity, representing the maximum number of vehicles that can pass through a road segment or intersection per unit time; congestion level, which classifies traffic conditions into different levels such as smooth, basically smooth, light congestion, moderate congestion, and severe congestion by setting multiple dynamic thresholds; and vehicle queue length, indicating the length of the vehicle queue waiting to pass at intersections or bottleneck road segments.

[0056] This unit accurately identifies and locates traffic congestion areas and assesses the severity and impact of congestion by setting dynamic thresholds and pattern recognition algorithms. The dynamic thresholds are adaptively adjusted based on historical data, time periods (e.g., morning and evening rush hours versus off-peak hours), weather conditions, and special events to ensure accurate congestion assessment. The pattern recognition algorithm analyzes abnormal changes in traffic flow parameters, such as sudden drops in vehicle speed, sustained high traffic volume, and sharp increases in queue length, to identify the occurrence, development, and dissipation of congestion. Once congestion is identified, the system immediately locates its geographical position and calculates its impact range, such as the length of affected road segments, the number of intersections affected, and the estimated duration. These perception results are updated in real-time within the system in structured data formats (e.g., congestion maps, congestion reports) and then passed to the decision optimization module.

[0057] The short-to-medium-term traffic flow prediction unit uses advanced deep learning models based on historical traffic data, real-time traffic conditions, and environmental and event data to predict traffic flow, average speed, and congestion index within specified future time windows (e.g., 5 minutes, 15 minutes, and 30 minutes). Historical traffic data includes traffic flow, speed, and congestion patterns accumulated over the past few years or even decades, used by the model to learn the periodicity, trends, and randomness of traffic flow. Real-time traffic condition data provides a snapshot of traffic at the current moment, serving as the starting point for prediction. Environmental and event data (e.g., weather forecasts, known construction plans, and information on large-scale events) are incorporated into the prediction model as external influencing factors to improve prediction accuracy.

[0058] This prediction unit employs a graph neural network-based prediction model, capable of capturing the complex spatiotemporal correlations within traffic networks. Graph neural networks abstract urban traffic networks into a graph structure, where nodes represent intersections or road segments, and edges represent the connections between road segments, thus effectively modeling the spatial propagation effects and temporal evolution of traffic flow.

[0059] For example, the prediction model can be expressed as:

[0060]

[0061] in, Indicates time A time-predicted traffic state vector containing key parameters such as future traffic flow, average speed, and traffic density for each road segment in the urban traffic network. (Symbol) This represents a nonlinear mapping function implemented by a deep neural network, specifically a spatiotemporal graph network combining graph convolutional networks and long short-term memory networks or spatiotemporal graph attention networks. This network can learn nonlinear transformation relationships from complex spatiotemporal data. (Symbol) This represents the inherent graph structure of an urban transportation network. In this graph, nodes represent important traffic intersections or road segments, while edges explicitly represent the physical connections between these segments or intersections. (Symbols) This constitutes an observed traffic state sequence, representing the sequence from the current time. Going back to the past Traffic condition data observed at each time step. This data includes historical information on flow rate, speed, and density. (Symbols) Indicates the current time The observed environmental and event data vector, which includes external influencing factors such as weather conditions, road construction information, or emergencies. (Symbol) This represents the set of all trainable parameters learned by the deep learning model during the training process through optimization algorithms.

[0062] This model enables the system to learn the spatiotemporal evolution of traffic flow, such as how congestion on upstream road segments propagates to downstream segments within a specific time period, or how traffic accidents in a certain area affect the flow distribution of the surrounding road network. These high-precision short- and medium-term predictions, such as the traffic flow trends on major arterial roads within the next 15 minutes and the congestion index predictions for specific intersections, provide reliable predictive basis for subsequent decision optimization, realizing a shift from passive response to proactive early warning and intervention. The prediction results are stored in vector form and continuously updated to ensure that the decision optimization module always makes decisions based on the latest predictive information.

[0063] The decision optimization module generates globally optimal traffic signal control strategies, dynamic route planning schemes, and traffic incident emergency response measures based on real-time perception results and prediction information provided by the traffic state perception and prediction module. The core advantage of this module lies in its use of a multi-agent reinforcement learning framework to achieve adaptive collaborative optimization of the entire urban traffic network, overcoming the limitations of traditional local optimization.

[0064] Specifically, the decision optimization module includes:

[0065] The multi-agent cooperative control unit treats traffic signal controllers at various intersections in the urban transportation network as independent agents. Each agent learns and executes the optimal action strategy that maximizes global cumulative rewards through continuous interaction with its environment (i.e., surrounding traffic conditions). Each agent's state space is precisely defined, including the traffic conditions at the current intersection and its adjacent intersections, such as real-time traffic flow, average queue length, average vehicle speed, lane occupancy, and predicted future traffic conditions (provided by the short-to-medium-term traffic flow prediction unit), as well as the current signal phase and the duration of each phase. The action space is defined as the switching sequence of signal phases, the duration of each phase, and the adjustment of the green light ratio.

[0066] The agent learns the optimal policy by maximizing cumulative rewards. The reward function is carefully designed to reflect a combination of objectives, including overall network efficiency, congestion mitigation, and travel comfort. For example, the cumulative reward function can be defined as:

[0067]

[0068] in, Indicates at time step At any given moment, this represents the total reward received by the entire city's transportation network. This reward value is a key indicator for evaluating the overall effectiveness of the current signal control strategy. (Symbol) This indicates the total number of intersections in the city's transportation network. (Symbol) Indicates at time step intersection The average delay time for vehicles. A smaller value indicates smoother traffic flow. (Symbol) Indicates at time step intersection The average queue length at this location. A smaller value indicates less congestion. (Symbol) Indicates at time step intersection The number of vehicles passing through during the current time period. A higher value indicates a higher intersection throughput. (Symbol) Indicates at time step Total number of stops for all vehicles across the network. Fewer stops result in greater travel comfort and reduced fuel consumption. (Symbol) , , , All of these are adjustable positive weighting coefficients, used to balance different optimization objectives; for example, they can be increased during peak periods. Prioritize vehicle throughput during peak hours, and increase it during off-peak hours. and To reduce delays and queues.

[0069] Each agent uses depth The system is trained using deep reinforcement learning algorithms such as network or near-end policy optimization to learn the optimal approach. Value function or policy function. The value update rule can be expressed as:

[0070]

[0071] in, Indicates the current state Take action below The expected cumulative reward is the basis for the agent's decision-making. (Symbol) The learning rate controls the step size of each update, affecting the convergence speed and stability of the learning process. (Symbol) From state Take action Afterwards, the agent reaches the new state. An instant reward received immediately. (Symbol) This is a discount factor, ranging from 0 to 1, used to measure the importance of future rewards to the current decision; the closer to 1, the more important future rewards are. (Symbol) Indicates the next state The agent then evaluates all possible actions. The maximum that can be obtained This value is used to guide the agent in choosing the optimal action for the future.

[0072] Through this collaborative learning mechanism, each agent can discover and execute signal control strategies that can both satisfy its own local optimization and promote the improvement of global network efficiency, thereby effectively alleviating traffic congestion and improving the overall traffic capacity and vehicle operation efficiency of the road network.

[0073] The dynamic route planning and guidance unit provides real-time optimal route suggestions to vehicles connected to the system or third-party navigation services based on real-time congestion information and future forecasts provided by the traffic condition perception and prediction module. This unit comprehensively considers multiple factors to calculate the best route from origin to destination. These factors include the actual travel time of the current road segment (calculated based on real-time sensor data and vehicle trajectory data), the predicted future travel time (provided by the short-to-medium-term traffic flow prediction unit), the congestion level of the road segment (assessed by the real-time traffic condition perception unit), and the user's preferences set in the navigation system (e.g., shortest time, shortest distance, avoiding congested areas, or selecting specific road segments).

[0074] The path cost function can be expressed as:

[0075]

[0076] in, Representing a path The total cost. This cost value is used to quantify the overall merits of a path; the system aims to find the path with the minimum cost. (Symbol) It is a path A section of road within the route. It consists of a series of interconnected road segments. (Symbol) It is a section of road The current actual travel time is dynamically calculated based on real-time traffic flow, vehicle speed, and other data, reflecting the current traffic efficiency of the road segment. (Symbol) It is a section of road The predicted future travel time, provided by the short-to-medium-term traffic flow forecasting unit, estimates the traffic conditions of a road segment over a future period, enabling forward-looking route planning. (Symbol) It is a section of road The congestion level is usually expressed numerically, from 0 to 5, where 5 represents severe congestion. This indicator visually reflects the degree of congestion on a road segment. (Symbol) , , These are adjustable weighting coefficients used to balance various indicators. By adjusting these weights, the system can change the emphasis of path planning according to actual needs or user preferences; for example, it can increase the weighting in emergency situations. It places greater emphasis on predicting future travel times.

[0077] This unit updates the toll costs of each road segment in the road network in real time and dynamically generates optimal paths through real-time queries and calculations using heuristic search algorithms (such as the A algorithm or a variant of Dijkstra's algorithm, combined with spatiotemporal network extensions). For example, the A algorithm combines the shortest path finding capability of Dijkstra's algorithm with the efficiency advantages of heuristic functions, enabling it to quickly find the globally optimal solution in complex traffic networks. The generated path information is pushed to vehicle navigation systems or smartphone applications connected to the system in the form of navigation instructions, map highlights, or text prompts to achieve a balanced distribution of traffic flow in the urban road network, effectively reduce the traffic load on specific road segments, and reduce vehicle travel time.

[0078] The traffic incident emergency response unit immediately activates pre-set emergency plans and generates corresponding emergency response strategies upon detecting emergencies such as traffic accidents, road construction, and natural disasters. Based on the nature of the event (e.g., car crash, fire, road collapse), its precise geographical location, and its severity (e.g., number of lanes affected, degree of traffic disruption), the unit automatically adjusts traffic signal control strategies and dynamic route guidance schemes for the affected area and surrounding road sections. For example, in the event of a traffic accident, the system may extend the green light time in the blocked direction to expedite the passage of stranded vehicles or guide surrounding vehicles to alternative routes. For road construction or natural disasters, the system can temporarily close some ramps or road sections to prevent vehicles from entering and exacerbating congestion.

[0079] The generation process of emergency response strategies is highly dynamic and adaptive. First, the system matches corresponding contingency plan templates based on the event type. These templates contain a series of standardized emergency measures. Next, by combining real-time traffic perception data and predictive information, the contingency plan templates are parametrically adjusted, such as determining the specific number of seconds the green light is extended and the specific planning of detour routes. During execution, the system continuously monitors the traffic conditions in the event area and dynamically adjusts the emergency strategy according to changes in traffic flow to ensure that the impact of the event on traffic flow is minimized, secondary congestion is avoided, and the accessibility of emergency lanes is guaranteed.

[0080] The instruction issuance and execution module is responsible for converting the various optimization decision instructions generated by the decision optimization module into specific control signals that can be directly recognized and executed by physical devices, and sending them to the corresponding execution devices in the urban traffic management system. Simultaneously, this module also receives feedback information from the execution devices, forming a closed-loop control system and continuously evaluating the decision-making effectiveness.

[0081] Specifically, the instruction issuance and execution module includes:

[0082] The signal control command sending unit is responsible for sending the optimized signal phase sequence, duration of each phase, and green light ratio in each direction—key parameters such as the green light ratio in each direction—to the traffic signal controllers at each intersection via a standard communication protocol (e.g., the National Traffic Information Communication Protocol NTCIP or a locally customized protocol). The communication interface employs a highly reliable, high-bandwidth connection method, such as a dedicated fiber optic network or industrial Ethernet, to ensure the accuracy and low latency of command transmission. Before sending commands, the system performs multiple checks, such as checksum verification and command format validation, to prevent the transmission of erroneous commands. Upon receiving a command, the signal controller immediately parses and executes it, thereby achieving real-time adjustment of the traffic lights. This unit is also responsible for monitoring the command transmission status and initiating a retransmission mechanism or alarm mechanism in case of transmission failure.

[0083] The Information Dissemination and Guidance Instruction Sending Unit is responsible for disseminating real-time optimal route suggestions, traffic congestion information, event warnings, and other important traffic information generated by the dynamic route planning and guidance unit to drivers and the public through diverse channels. Dissemination channels include Variable Message Systems (VMS), which are typically deployed on highways, urban expressways, and major roads, visually displaying traffic information in graphic or textual form. Simultaneously, information is pushed to in-vehicle navigation systems and smartphone applications via digital broadcasting systems, urban traffic information platforms, and interfaces with third-party navigation applications (e.g., through open application programming interfaces). This information content is highly abstracted and refined, aiming to convey key information in the most concise way, such as "Road construction ahead, please detour to XX Road" or "Severe congestion on XX Road, estimated delay of 20 minutes." This unit is also responsible for the timeliness management of information, ensuring that outdated information is promptly removed or updated.

[0084] The execution status feedback unit is responsible for receiving and summarizing execution status feedback information from various execution devices. This feedback information primarily originates from traffic signal controllers (e.g., current signal light phase, actual green light duration, controller operating status), variable information flags (e.g., display content confirmation, device operating status), and some vehicle navigation systems integrated into this system (e.g., whether the vehicle adopted the recommended route, actual travel time). The feedback data is transmitted back to this unit in real time via the corresponding communication interface for preliminary analysis and verification. For example, the system compares the parameters sent in the command with the parameters actually executed to assess the accuracy of the execution. This feedback information is crucial for forming closed-loop control. It is immediately sent back to the data acquisition and fusion module as part of the next round of data fusion to evaluate the effectiveness of the decisions generated by the decision optimization module. For example, by comparing the predicted congestion relief with actual congestion data, the effectiveness of the signal control strategy is evaluated, and valuable training and validation data are provided for further model optimization.

[0085] The Human-Computer Interaction and Visualization module provides an intuitive, comprehensive, and easy-to-use monitoring, analysis, and management platform. This platform enables traffic management personnel to monitor the real-time operational status of the urban traffic network, review system-generated intelligent decisions, make manual interventions, and conduct in-depth analysis and strategy simulation of historical data.

[0086] Specifically, the human-computer interaction and visualization module includes:

[0087] The real-time situation display unit centers on a high-resolution digital map, dynamically displaying the operational status of the entire urban traffic network in real time. The map clearly marks real-time vehicle speeds (e.g., represented by color coding), real-time traffic flow, congestion levels (e.g., represented by different shades of red), queue lengths, and the current traffic light phases and remaining green light time at each intersection. This unit supports multi-dimensional information overlay display; for example, managers can simultaneously display weather information, road construction areas, accident locations, and even the air quality index for specific areas. All data is presented in an intuitive graphical interface, such as heatmaps, traffic flow animations, and status charts, helping managers quickly locate problem areas, understand the dynamic evolution of traffic flow, and evaluate the immediate effects of various decisions. The display interface features zoom, pan, and layered display functions to adapt to different management granularities.

[0088] The Decision-Making Review and Intervention Unit provides traffic management personnel with a crucial decision-making interface. When the decision optimization module generates new traffic optimization decision-making schemes (e.g., a new set of signal timing schemes or a wide-ranging route guidance strategy), this unit presents the details of these schemes in a clear and understandable manner, including the expected effects, potential impacts, and the affected areas. Managers can review these schemes and, based on practical experience or specific instructions, choose to manually review and approve them for immediate implementation, or modify them to adapt to special circumstances. In emergency or special situations, such as the passage of a large convoy or a major natural disaster, managers can also enforce specific instructions, overriding the system's automatically generated strategies, ensuring the traffic management system possesses the highest level of flexibility and human intervention capability. All review, approval, modification, and enforcement operations are meticulously logged for subsequent auditing and accountability.

[0089] The Historical Data Analysis and Reporting unit provides powerful historical data query, statistics, and analysis functions. Managers can query historical traffic data for any time period (e.g., average daily traffic flow and monthly congestion duration for specific road segments), system operation logs (e.g., system decision generation time, execution results, and abnormal alarm records), and optimization effect evaluation reports. The evaluation reports include various key performance indicators (KPIs), such as the percentage increase in average network speed, reduction in average delay time, change in the number of stops per unit mileage, and carbon emission estimates. This unit supports custom report generation and displays data in various formats such as charts and reports, providing scientific data support and basis for urban traffic planning, long-term management decisions, and policy formulation. In-depth mining of historical data can reveal long-term traffic congestion patterns, assess the long-term effectiveness of specific management strategies, and provide data-driven insights for future system upgrades.

[0090] The simulation and evaluation unit provides an offline simulation environment that allows traffic managers to input different traffic scenarios or emergency situations without affecting actual traffic operations. For example, it can simulate a large-scale event, sudden severe weather, or a large-scale traffic accident that may occur in the future, and then test the system's response to different optimization strategies (e.g., different signal timing schemes, different route guidance rules) under these scenarios. This simulation environment can accurately simulate the dynamic changes in traffic flow and evaluate the performance of various traffic indicators (e.g., congestion level, average delay, capacity) under different strategies. Through simulation, managers can pre-evaluate and optimize new traffic management strategies, identify potential problems, and fully verify and adjust them before actual deployment, thereby reducing decision-making risks and improving the scientific rigor and effectiveness of management strategies.

[0091] In summary, the AI-based intelligent optimization management system for urban traffic congestion in this embodiment achieves real-time, comprehensive, and accurate situational awareness of the urban traffic network through deep fusion technology of multi-source heterogeneous data. Secondly, this system introduces a spatiotemporal traffic flow prediction model based on graph neural networks, providing high-precision short- and medium-term traffic condition predictions to achieve proactive early warning and intervention. Furthermore, this system employs a multi-agent reinforcement learning framework, treating the signal controllers and path guidance systems at various intersections in the urban traffic network as collaborative learning agents. Through a cleverly designed shared reward mechanism and learning algorithm, it achieves globally optimal coordination between traffic signal control and path guidance. This framework overcomes the bottleneck of local optimization leading to suboptimal global performance in traditional traffic management systems, adaptively adjusting strategies based on real-time traffic conditions and prediction information, effectively alleviating urban traffic congestion and improving road network capacity and vehicle operating efficiency. Moreover, this system, through dynamic path planning and guidance units combined with real-time congestion prediction, provides users with personalized optimal travel routes, effectively guiding the balanced distribution of traffic flow in the road network, reducing traffic load on specific road sections, and minimizing vehicle travel time. Meanwhile, the traffic incident emergency response unit can quickly generate and execute emergency plans for sudden events, minimizing the impact of such events on traffic operations and enhancing the resilience of the transportation system. Finally, the system's human-computer interaction and visualization modules enable traffic management personnel to intuitively monitor traffic conditions, review and intervene in system decisions, and conduct historical data analysis and simulations, greatly improving the intelligence, precision, and scientific level of traffic management. This system significantly reduces urban traffic congestion rates, decreases vehicle fuel consumption and emissions, and improves residents' travel efficiency and satisfaction, demonstrating significant social and economic benefits.

Claims

1. An intelligent optimization management system for urban traffic congestion based on artificial intelligence, characterized in that, include: The data acquisition and fusion module is used to acquire various real-time heterogeneous traffic data in the urban traffic network, and to preprocess, standardize and fuse them from multiple sources to generate a comprehensive dataset containing spatiotemporal traffic flow parameters, event parameters and environmental parameters. The traffic condition perception and prediction module is communicatively connected to the data acquisition and fusion module. It is used to accurately perceive the real-time operating status of the urban traffic network based on the comprehensive dataset and to make high-precision predictions of the short- and medium-term traffic flow trends. The prediction module is a spatiotemporal graph network prediction model based on graph neural networks. The decision optimization module is communicatively connected to the traffic state perception and prediction module. It is used to generate globally optimal traffic signal control strategies, dynamic path planning schemes, and traffic incident emergency response measures based on real-time perception results and prediction information, using a multi-agent reinforcement learning framework. The instruction issuance and execution module, communicatively connected to the decision optimization module, is used to send the decision instructions generated by the decision optimization module to the corresponding execution device and receive the execution status feedback information from the execution device; and The human-computer interaction and visualization module is used to provide real-time monitoring of the operation status of the traffic network, review and intervention of decision-making schemes, historical data analysis, and simulation evaluation functions; The traffic condition perception and prediction module includes: The real-time traffic status perception unit is used to analyze the comprehensive dataset and calculate and update key traffic indicators such as vehicle density, average speed, traffic capacity, congestion level, and queue length of each road segment and intersection on the road grid in real time. The real-time traffic status sensing unit, through setting dynamic thresholds and pattern recognition algorithms, accurately identifies and locates traffic congestion areas, and assesses the severity and scope of congestion; and The short-to-medium-term traffic flow prediction unit is used to predict traffic flow, average speed and congestion index within a specified time window in the future, based on historical traffic data, real-time traffic conditions and environmental and event data, using the spatiotemporal graph network prediction model based on graph neural networks. The prediction model can capture the complex spatiotemporal correlations in the traffic network. The short-to-medium-term traffic flow prediction unit is specifically used for: The urban transportation network is abstracted as a graph structure, where nodes represent intersections or road segments, and edges represent the connection relationships between road segments. The traffic state vector is used as the output of the prediction model. The traffic state vector contains the future traffic flow, average speed and traffic density parameters of each road segment in the urban traffic network. The observed traffic state sequence, environmental and event data vectors, and the graph structure of the urban traffic network are used as inputs to the prediction model. The observed traffic state sequence represents observed traffic state data from the current time back to multiple past time steps; and The spatiotemporal evolution of traffic flow is learned through the prediction model. The decision optimization module includes: The multi-agent cooperative control unit is used to treat the signal controllers at each intersection in the urban traffic network as independent agents. The state space of each agent is defined as the traffic state of the current intersection and adjacent intersections, the predicted future traffic state, and the current signal phase and duration. The action space is defined as the switching sequence of the signal light phase, the duration, and the green ratio adjustment. The multi-agent collaborative control unit guides the agents to learn the optimal strategy by maximizing the cumulative reward function that reflects the overall network efficiency, congestion relief, and travel comfort. The agents are trained using deep networks or proximal policy optimization deep reinforcement learning algorithms to learn the optimal value function or policy function.

2. The intelligent optimization management system for urban traffic congestion based on artificial intelligence according to claim 1, characterized in that, The data acquisition and fusion module includes: The traffic sensor data acquisition unit is used to acquire traffic flow, vehicle speed, lane occupancy, vehicle queue length, and license plate recognition information from physical sensors distributed in the road network. A mobile terminal data acquisition unit is used to acquire anonymized location, speed, direction, and travel trajectory data from the GPS equipment, mobile communication base stations, and smartphone applications of connected vehicles; and The environmental and event data acquisition unit is used to acquire real-time weather data, road construction information, traffic accident reports, large-scale event forecasts, and emergency information.

3. The intelligent optimization management system for urban traffic congestion based on artificial intelligence according to claim 2, characterized in that, The data acquisition and fusion module also includes: The data preprocessing and fusion unit is used to clean, denoise, convert formats, align time, and spatially map the raw data collected by the traffic sensor data acquisition unit, the mobile terminal data acquisition unit, and the environment and event data acquisition unit. It also uses a multi-source data fusion algorithm to integrate the heterogeneous data from different sources, with different sampling frequencies, and different precisions into a unified, high-dimensional representation of urban traffic status. The multi-source data fusion algorithm includes fusion methods based on Kalman filtering or deep learning.

4. The intelligent optimization management system for urban traffic congestion based on artificial intelligence according to claim 1, characterized in that, The decision optimization module further includes: The dynamic path planning and guidance unit is used to provide real-time optimal path suggestions for vehicles or navigation services connected to the system based on the real-time perception results and prediction information. The dynamic path planning and guidance unit comprehensively considers the actual travel time of the current road segment, the predicted future travel time, the congestion level of the road segment, and user preferences to calculate the optimal path from the origin to the destination. It also dynamically generates and pushes the optimal path by updating the road network travel cost in real time and using a heuristic search algorithm to achieve a balanced distribution of traffic flow in the road network.

5. The intelligent optimization management system for urban traffic congestion based on artificial intelligence according to claim 4, characterized in that, The decision optimization module further includes: The traffic incident emergency response unit is used to immediately activate the emergency plan when a traffic accident, road construction, or natural disaster is detected. The traffic incident emergency response unit automatically adjusts the signal control strategy for the affected area and the route guidance scheme for surrounding road sections based on the nature, location, and severity of the incident, in order to minimize the impact of the incident on traffic flow.

6. The intelligent optimization management system for urban traffic congestion based on artificial intelligence according to claim 1, characterized in that, The instruction issuance and execution module includes: The signal control command sending unit is used to send the optimized signal phase, duration and green ratio parameters to the traffic signal controllers at each intersection through a standard communication protocol, so as to realize the real-time adjustment of the traffic lights; The information dissemination and guidance instruction sending unit is used to disseminate dynamic route planning results, congestion information, and event warning information to drivers and the public in real time through variable message signs, broadcasts, and interfaces with third-party navigation applications; and The execution status feedback unit is used to receive execution status feedback information from traffic signal controllers, variable information signs, and vehicle navigation system devices, and send it back to the data acquisition and fusion module for evaluating the decision-making effect and further optimizing the model.

7. The intelligent optimization management system for urban traffic congestion based on artificial intelligence according to claim 1, characterized in that, The human-computer interaction and visualization module includes: The real-time situation display unit is used to display the operating status of the entire urban traffic network on the digital map in real time, including vehicle speed, traffic flow, congestion level, queue length and traffic light phase information of each road segment, and supports multi-dimensional information overlay display. The decision-making scheme review and intervention unit allows traffic management personnel to view the traffic optimization decision-making schemes generated by the system and provides the function of manually reviewing, approving, modifying or enforcing specific instructions. The historical data analysis and reporting unit provides query, statistical, and analytical functions for historical traffic data, system operation logs, and optimization effect evaluation reports; and The simulation and evaluation unit provides offline simulation capabilities, allowing managers to input different traffic scenarios or events and simulate the system's response under different strategies.

Images