An induction and control system for ensuring all-weather traffic of intelligent expressway

Abstract

The present application relates to the technical field of intelligent transportation, and more particularly to an induction and control system for guaranteeing all-weather traffic of intelligent expressway, comprising a perception layer, a network transmission layer, an edge computing layer, a central decision layer and a terminal interaction layer which are sequentially connected in communication. The perception layer is responsible for collecting road environment sensing data and road condition sensing data; the network transmission layer completes the reception and bidirectional transmission of various data. The edge computing layer implements multiple local processing on the collected data and uploads the results, and the central decision layer combines the data, uses digital twin modeling and reinforcement learning algorithm to optimize and generate global control strategy. The edge computing layer fuses local data and global strategy to generate regular control instructions, and synchronously generates emergency control instructions when identifying emergency working conditions, and finally the terminal interaction layer uniformly executes the corresponding control instructions. The present application remedies the defects of traditional expressway control through the five-layer architecture, realizes all-scene low-latency adaptive control, and improves traffic safety and efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to a guidance and control system that ensures all-weather passage on intelligent highways. Background Technology

[0002] In recent years, with the continuous growth of my country's expressway mileage and the significant increase in traffic flow, expressway operation and management are rapidly evolving from traditional manual inspections and static signage guidance towards informatization and intelligentization. On the one hand, roadside sensing equipment such as video surveillance, microwave radar, and weather detectors have been widely deployed, enabling real-time collection of data on traffic flow, speed, visibility, and road conditions. On the other hand, the gradual maturation of emerging technologies such as vehicle-to-everything (V2X), 5G communication, and edge computing provides a technological foundation for real-time data interaction and rapid response on expressways. Against this backdrop, how to utilize these technologies to ensure safe passage on expressways in all weather conditions such as rain, fog, snow, and nighttime has become a research hotspot and engineering challenge in the field of intelligent transportation. The all-weather traffic guidance and control system aims to achieve real-time monitoring, dynamic control, and precise guidance of the expressway operating environment and traffic conditions through an integrated design of perception, communication, decision-making, and execution, thereby reducing the risk of accidents and improving traffic efficiency in severe weather.

[0003] In existing technologies, all-weather traffic guidance and control systems typically employ a centralized cloud platform architecture: data collected by roadside sensors and some vehicle-mounted terminals is transmitted to a central server via a communication network. The server generates control strategies such as speed limits, lane closures, and traffic light activation based on preset threshold rules (e.g., a speed limit of 20 km / h when visibility is below 50 meters, or an icing warning when the road surface temperature is below 0°C) or offline-trained traffic flow models. These strategies are then distributed to drivers via variable message signs, roadside broadcasts, mobile apps, or vehicle-to-infrastructure (V2I) messages. Some advanced research attempts to introduce digital twin technology to construct virtual road networks for situation simulation or to use reinforcement learning algorithms to dynamically optimize control parameters, but these efforts mostly remain at the stage of offline simulation or closed test tracks. However, the aforementioned existing technologies still have significant shortcomings in actual 24 / 7 operation: First, the fusion level of perception data is relatively shallow, and multi-source heterogeneous data (weather, video, radar, vehicle status) lacks localized calibration and hierarchical processing. Furthermore, central decision-making is highly dependent on communication links; once the network is interrupted or the central platform fails, edge road sections completely lose their independent control capabilities, making it impossible to respond quickly to emergencies such as sudden fog, instantaneous road icing, or traffic accidents. Second, control strategies based on fixed threshold rules cannot adaptively adjust according to dynamic parameters such as real-time traffic flow density and vehicle speed distribution, making it difficult to balance traffic efficiency and safety risks. Intelligent algorithms such as reinforcement learning suffer from excessive computational latency and are impractical if trained online in real time, while offline implementation cannot adapt to real-time changes in operating conditions. Third, existing systems do not distinguish between routine control and emergency control scenarios, lacking a collaborative emergency mechanism between the edge computing layer and the central decision-making layer. In the event of communication anomalies or central policy timeouts, they cannot automatically switch to local emergency control mode, severely impacting the reliability and response timeliness of 24 / 7 traffic. Summary of the Invention

[0004] To address the technical deficiencies in the background technology, this invention proposes a guidance and control system for ensuring all-weather passage on intelligent highways, solving the aforementioned technical problems and meeting practical needs. The specific technical solution is as follows: A guidance and control system for ensuring all-weather passage on smart highways includes a perception layer, a network transmission layer, an edge computing layer, a central decision-making layer, and a terminal interaction layer that are sequentially connected in communication. The sensing layer is used to collect environmental sensing data and road condition sensing data; The network transport layer is used to receive and transmit data; The edge computing layer is used to perform multiple localization processes based on environmental sensing data and road condition sensing data, obtain localization processing results, and upload them to the central decision-making layer. The central decision-making layer is used to generate global management and control strategy data based on localized processing results, through digital twin modeling and reinforcement learning algorithm iterative optimization, and then distribute it to the edge computing layer. The edge computing layer combines localized processing results and global management and control strategy data to generate regular management and control instructions and output them to the terminal interaction layer. When an emergency occurs, it generates emergency control instructions. The terminal interaction layer is used to execute routine control commands and emergency control commands.

[0005] Specifically, the perception layer includes roadside sensing components and vehicle-mounted sensing components. The roadside sensing components include millimeter-wave radar, a meteorological sensor array, a video analysis component, and a road surface condition sensor. The vehicle-mounted sensing components are vehicle-mounted OBU terminals. The roadside sensing components are used to collect environmental sensing data and some road condition sensing data. The vehicle-mounted components are used to collect vehicle driving status data. The environmental sensing data includes visibility, precipitation intensity, road surface temperature, and icing risk parameters. The road condition sensing data includes vehicle location, driving speed, vehicle driving status, traffic accidents, and road debris parameters.

[0006] Specifically, the edge computing layer performs the following steps for multiple localization processes: Based on environmental and road condition sensor data, multi-source data fusion calibration is performed to remove error and redundant data and generate calibrated sensor data. Traffic incident detection is performed based on calibrated sensor data to identify abnormal events and generate incident detection results; Based on the calibrated sensor data and event detection results, a warning level determination is performed, the warning level is matched, and a warning determination result is generated. The warning levels include Level 1 regular warning, Level 2 attention warning, and Level 3 emergency warning. Based on the warning determination result, the calibrated sensor data is classified and graded. The calibrated sensor data corresponding to Level 3 emergency warning is marked as emergency event data, and the remaining calibrated sensor data is marked as regular data, generating a data classification and grading result. The calibrated sensor data, event detection results, early warning judgment results, and data classification results are integrated to form localized processing results, which are then uploaded to the central decision-making level.

[0007] Specifically, an emergency condition is determined when any one of the following criteria is met: The traffic incident types corresponding to the emergency incident data have a correspondence with the items in the preset emergency incident type list; The network transport layer communication link was abnormally interrupted; The global control strategy data was not received for more than a preset time period.

[0008] Specifically, the network transport layer uses 5G slicing technology to build a dedicated communication network, pre-dividing low-latency emergency data transmission channels and regular bandwidth data transmission channels; When an emergency occurs, the edge computing layer generates an emergency control command based on the localized processing results and sends a channel scheduling command to the network transport layer to start a low-latency emergency data transmission channel, pushing the emergency control command to the terminal interaction layer for immediate execution in real time.

[0009] Specifically, the steps for the central decision-making layer to generate global management and control strategy data are as follows: Accept the results of localization processing and simultaneously retrieve the static road network ledger of the expressway, the historical traffic control database and the shared big data of the whole region's meteorology as the basic dataset; A high-precision real-scene digital twin model is built based on the basic dataset. The localized processing results are then mounted and bound to the corresponding road segment units of the digital twin model in real time, generating a dynamic synchronous simulation data stream and real-time simulation and judgment results of the digital twin of the entire road segment. Construct a reinforcement learning algorithm operating architecture, define a safe passage reward function and a road network congestion penalty coefficient, and delineate the environmental state space and the control action space; Using historical control samples, offline iterative training and simulation are carried out to optimize the speed limit threshold, lane release ratio and induced start-stop parameters cycle by cycle until the model converges to the preset control accuracy and outputs the optimized and finalized algorithm parameter set. In actual operation, the dynamic synchronous simulation data stream is input, the optimized and finalized algorithm parameter set is called to perform fast strategy calculation, and combined with the real-time simulation and judgment results of the digital twin of the whole road segment, the hierarchical control logic is matched according to the road segment partition to generate the original global strategy set including hierarchical speed limit adjustment, dynamic lane closure, long-distance traffic flow detour guidance and multi-mode guidance strategy switching. The original set of global strategies is standardized in format and validated for rationality, and then solidified into standardized global control strategy data.

[0010] Specifically, the steps for the edge computing layer to generate regular control instructions are as follows: The global control strategy data and the localized processing results are normalized and standardized respectively to unify the timestamps, road segment coordinates and parameter units, and achieve accurate alignment and matching in the spatiotemporal dimensions. Extract the core feature parameters of the localized processing results and compare them item by item with the built-in standard thresholds of the global control strategy to identify abnormal points where the global strategy does not match the local field conditions and where parameters conflict. Adaptive correction and compensation are performed for abnormal locations, and the working condition adaptation similarity score is calculated. Only matching parameter combinations with adaptation scores higher than a preset threshold are retained, and the fusion parameter package after adaptation and correction is generated. Based on the modified and adapted fusion parameter combination, standardized routine control instructions are generated, including dynamic roadside lighting guidance, variable information hierarchical text warnings, and human-vehicle collaborative interactive broadcasts.

[0011] Specifically, the terminal interaction layer includes a roadside terminal unit, an in-vehicle terminal unit, and a management platform terminal unit; The roadside terminal unit is used to perform intelligent guidance light dimming, variable message sign graphic and text display, and roadside directional voice warning operations. The vehicle-mounted terminal unit is used to perform operations such as vehicle-road cooperative information push, AR real-scene navigation guidance, and vehicle assisted driving strategy reminders; The management platform terminal unit is used to perform operations such as visual display of the overall control status, sound and light alarms for on-site abnormal events, and receiving and issuing manual intervention instructions.

[0012] Specifically, the vehicle-mounted terminal unit is an OBU device that integrates a Beidou-3 positioning module, a 5G dual-mode communication module, and an inertial navigation unit.

[0013] Specifically, the central decision-making layer is also used to receive on-site execution data of instructions from the terminal interaction layer, and to iteratively optimize the parameters of the digital twin 3D model and the configuration of the reinforcement learning algorithm based on the feedback execution data.

[0014] Compared with existing technologies, the guidance and control system for ensuring all-weather smart highway traffic provided by this invention has the following beneficial effects: This invention establishes a perception layer, network transmission layer, edge computing layer, central decision-making layer, and terminal interaction layer connected in sequence. The edge computing layer performs multiple localized processing on multi-source sensor data. Under normal operating conditions, it receives the global optimization strategy generated by the central decision-making layer based on digital twins and reinforcement learning offline training, adaptively corrects it, and outputs standardized instructions. In emergency situations, the edge computing layer autonomously generates emergency control instructions and pushes them to the terminal interaction layer for execution in real time via a dedicated low-latency 5G slicing channel, thus achieving dual protection of global optimization and local emergency response. The central decision-making layer uses historical data to train a reinforcement learning model offline and combines it with real-time digital twin simulation to generate strategies such as graded speed limits, dynamic lane closures, and long-distance detour guidance. It also supports iterative optimization based on on-site execution data feedback, significantly improving the system's adaptability to different weather and traffic flow conditions. The terminal interaction layer covers roadside guidance lights, variable message signs, vehicle-mounted OBUs, and the management platform, achieving comprehensive guidance and early warning. This effectively ensures the safe passage efficiency of smart highways under all-weather conditions such as rain, fog, snow, and nighttime, reducing the risk of traffic accidents and possessing high reliability and low-latency response capabilities. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a module of an intelligent highway guidance and control system that ensures all-weather passage. Detailed Implementation

[0016] In the description of this invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "middle," and "inner," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, it should be noted that unless otherwise explicitly specified and limited, the terms "installed," "connected," and "joined" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention through specific circumstances.

[0017] The embodiments of the present invention will be described below with reference to the accompanying drawings and related examples. The embodiments of the present invention are not limited to the following examples, and the present invention relates to the relevant necessary components in this technical field, which should be regarded as well-known technology in this technical field and can be known and mastered by those skilled in this technical field.

[0018] See Figure 1 The present invention provides an guidance and control system for ensuring all-weather passage on smart highways, comprising a perception layer, a network transmission layer, an edge computing layer, a central decision-making layer and a terminal interaction layer that are connected in sequence. The sensing layer is used to collect environmental sensing data and road condition sensing data; Specifically, the perception layer is the data acquisition entry point of the intelligent highway system. It belongs to the front-end sensing unit and is responsible for acquiring raw environmental and road condition data of the entire highway area. It is the data foundation of the entire system, but has no data processing or decision-making functions.

[0019] The network transport layer is used to receive and transmit data; Specifically, the network transport layer is the data communication hub of the system, undertaking the data forwarding and transmission tasks between various layers. It is only responsible for the efficient and stable transmission of data and does not participate in data calculation and instruction generation.

[0020] The edge computing layer is used to perform multiple localization processes based on environmental sensing data and road condition sensing data, obtain localization processing results, and upload them to the central decision-making layer. Specifically, the edge computing layer is the system's local core processing unit, deployed on-site along the road. It is responsible for localized data processing, autonomous emergency situation assessment, and generation of routine / emergency commands, balancing real-time performance and autonomy to address cloud-based decision-making delays and communication interruptions. Multiple localized processing involves the edge computing layer performing multi-step on-site processing on raw sensor data, including data fusion calibration, anomaly detection, warning level determination, and data classification. This local data preprocessing can be completed without uploading to the cloud. The localized processing result is a standardized dataset output by the edge computing layer after completing multiple localized processing. It includes calibrated data, event detection results, warning levels, and data classification labels, serving as the common basis for both local and cloud-based decision-making.

[0021] The central decision-making layer is used to generate global management and control strategy data based on localized processing results, through digital twin modeling and reinforcement learning algorithm iterative optimization, and then distribute it to the edge computing layer. Specifically, the central decision-making layer is the system's global optimization hub, belonging to the cloud / platform-level computing unit. Based on massive amounts of data and intelligent algorithms, it generates comprehensive control strategies, responsible for the overall planning and optimization of the road network, ensuring the comprehensiveness and rationality of the control strategies. The global control strategy data is a comprehensive road network control plan generated by the central decision-making layer based on digital twin modeling and iterative optimization using reinforcement learning algorithms, covering global rules such as speed limits, lane control, detour guidance, and guidance modes.

[0022] The edge computing layer combines localized processing results and global management and control strategy data to generate regular management and control instructions and output them to the terminal interaction layer. When an emergency occurs, it generates emergency control instructions. Specifically, routine control commands are standardized commands generated by the edge computing layer based on local data and global policies under normal road conditions, normal communication, and effective cloud policies, balancing global optimization with local adaptation. Emergency control commands are instantaneous response commands autonomously generated by the edge computing layer in emergency situations such as sudden accidents, extreme weather, communication interruptions, and cloud policy timeouts, without waiting for cloud decisions, ensuring timely emergency response. Emergency situations are sudden scenarios that threaten highway traffic safety and require millisecond-level response, including traffic accidents, road debris, extreme low visibility, network interruptions, and cloud policy timeouts.

[0023] The terminal interaction layer is used to execute routine control commands and emergency control commands.

[0024] Specifically, the terminal interaction layer is the end point of the system's instruction execution, directly facing vehicles, roads, and management terminals. It is responsible for implementing control instructions and completing functions such as traffic guidance, early warning prompts, and status feedback.

[0025] This invention employs a five-layer collaborative system architecture comprising a perception layer, a network transmission layer, an edge computing layer, a central decision-making layer, and a terminal interaction layer. Relying on the edge computing layer, it performs multiple localized processing of environmental and road condition sensor data. This architecture can both coordinate with the central decision-making layer to generate global control strategy data based on digital twin modeling and reinforcement learning algorithms, producing routine control instructions tailored to actual road conditions, and autonomously generate emergency control instructions in emergency situations. Finally, the terminal interaction layer executes both types of instructions, achieving an organic combination of local real-time control and global optimization decision-making. This effectively improves the response speed and operational stability of intelligent highway traffic guidance and control, ensuring the reliability and safety of all-weather highway traffic management.

[0026] It should be noted that the perception layer includes roadside sensing components and vehicle-mounted sensing components. The roadside sensing components include millimeter-wave radar, a meteorological sensor array, a video analysis component, and a road surface condition sensor. The vehicle-mounted sensing components are vehicle-mounted OBU terminals. The roadside sensing components are used to collect environmental sensing data and some road condition sensing data. The vehicle-mounted components are used to collect vehicle driving status data. The environmental sensing data includes visibility, precipitation intensity, road surface temperature, and icing risk parameters. The road condition sensing data includes vehicle position, driving speed, vehicle driving status, traffic accidents, and road debris parameters.

[0027] Specifically, the roadside sensing components are fixed hardware units deployed on the roadside, including four types of equipment: millimeter-wave radar, meteorological sensor array, video analysis components, and road surface condition sensors. Among them, millimeter-wave radar is used to collect basic road condition data such as vehicle position and speed without being affected by rain, fog, or sunlight. The meteorological sensor array collects meteorological environmental data such as visibility and precipitation intensity. The video analysis components identify abnormal road condition events such as traffic accidents and road debris. The road surface condition sensors monitor road safety parameters such as road surface temperature and icing risk, and are mainly responsible for collecting public environmental data and road condition data across the entire area. The vehicle-mounted sensing components are vehicle-mounted OBU terminals, which serve as vehicle-side mobile sensing units and are specifically designed to collect vehicle's own driving status data, filling the blind spots in microscopic driving information of individual vehicles that cannot be covered by roadside sensing.

[0028] In one embodiment of the present invention, the specific steps of the edge computing layer performing multiple localization processes are as follows: Step S101: Based on environmental sensor data and road condition sensor data, multi-source data fusion calibration is performed to remove error and redundant data and generate calibrated sensor data. Specifically, after the edge computing layer receives the multi-source heterogeneous raw data from roadside sensing components (millimeter-wave radar, weather sensors, video analytics, and road surface sensors) and vehicle-mounted OBU terminals transmitted from the perception layer, the first step is to perform spatiotemporal synchronization and alignment: all data are uniformly calibrated to UTC standard timestamps and matched with spatial coordinates such as highway kilometer markers and lane numbers to achieve accurate spatiotemporal matching between roadside data and vehicle-mounted data; the second step is to perform data normalization processing: data with different dimensions such as vehicle speed, temperature, visibility, and traffic density are converted into 0-1 standardized values ​​to unify the data format; the third step is to perform filtering, noise reduction, and anomaly removal: the Kalman filter algorithm is used to filter random noise collected by sensors, and abnormal data exceeding reasonable ranges (such as invalid values ​​such as vehicle speed > 180 km / h and negative visibility) are removed by threshold judgment, redundant data collected repeatedly is deleted, and key data missing in a short period of time is supplemented; finally, calibrated sensing data that is accurate in time, spatially matched, error-free, and redundant is generated.

[0029] In this embodiment, the specific implementation steps of multi-source data fusion calibration are as follows: First, all sensor data are linearly interpolated and aligned based on GPS standard timestamps, with a unified data sampling frequency of 10Hz, and the standard spatial coordinates of highway kilometer markers + lane numbers are matched to achieve a one-to-one correspondence between the spatiotemporal dimensions of roadside and vehicle-mounted data. Second, all sensor data of different dimensions are uniformly mapped to the 0-1 interval through min-max linear transformation to complete the normalization process. Finally, the Kalman filter algorithm is used to filter the random noise of the sensor data, and abnormal data exceeding the reasonable range is removed by the 3σ criterion. Duplicate and redundant data are deleted and short-term missing data is filled in to generate calibrated sensor data.

[0030] Step S102: Perform traffic incident detection based on the calibrated sensor data, identify abnormal events, and generate incident detection results; Specifically, the edge computing layer retrieves calibrated sensor data and performs event detection across multiple dimensions: 1. For road condition events, it uses vehicle trajectory analysis from millimeter-wave radar and target recognition algorithms from video analysis components to detect events such as traffic accidents, road debris, vehicles driving in the wrong direction / illegally parked, and traffic congestion, marking the event type, location, affected lanes, and duration; 2. For environmental events, it uses meteorological sensor arrays and road condition sensor data to detect abnormal environmental events such as extreme low visibility (fog patches), heavy precipitation, road icing, and low-temperature frost; all detected events are labeled with attributes according to their impact, forming standardized event detection results that include event type, location, range, and severity.

[0031] Step S103: Based on the calibrated sensor data and event detection results, perform a warning level determination, match the warning levels, and generate a warning determination result. The warning levels include Level 1 regular warning, Level 2 attention warning, and Level 3 emergency warning. Based on the warning determination result, classify the calibrated sensor data. Mark the calibrated sensor data corresponding to Level 3 emergency warning as emergency event data, and mark the remaining calibrated sensor data as regular data to generate a data classification result. Specifically, the edge computing layer combines calibrated sensor data with event detection results to determine the warning level according to preset rules: Level 1 Regular Warning: No abnormal events, visibility ≥ 500m, no road icing, smooth traffic flow, no accidents / discharge; Level 2 Caution Warning: Minor abnormal events, visibility 200-500m, road surface temperature 0-4℃ (risk of icing), slow traffic flow, no major safety hazards; Level 3 Emergency Warning: Major abnormal events, visibility < 200m (fog), road icing, traffic accidents / discharge on the road, vehicles driving in the wrong direction / illegally parked; After completing the level determination, data classification is performed: the calibrated sensor data corresponding to the Level 3 Emergency Warning is marked as emergency event data, and the calibrated sensor data corresponding to the Level 1 and Level 2 Warnings are uniformly marked as regular data, generating data classification results.

[0032] Step S104: Integrate the calibrated sensor data, event detection results, early warning judgment results, and data classification results to form a localized processing result, and upload it to the central decision-making level.

[0033] Specifically, the edge computing layer encapsulates four types of data—calibrated sensor data, event detection results, early warning judgment results, and data classification results—in a structured manner according to the highway management and control standard format, forming a complete localized processing result. On the one hand, the result is cached locally for later use in emergency situation judgment and routine instruction generation; on the other hand, it is uploaded to the central decision-making layer in real time through the network transmission layer, serving as the core input data for digital twin modeling, reinforcement learning algorithm iterative optimization, and global management and control strategy generation.

[0034] It should be noted that an emergency situation is determined when any one of the following criteria is met: The traffic incident types corresponding to the emergency incident data have a correspondence with the items in the preset emergency incident type list; Specifically, the edge computing layer retrieves the generated data classification results, extracts the traffic / environmental event types corresponding to the data marked as emergency events, and compares them precisely with a locally pre-stored list of preset emergency event types. If the event type is in the list (e.g., traffic accidents, road icing, fog with visibility <200m), it is immediately identified as an emergency situation. The preset emergency event type list is a high-risk event standard list that conforms to highway operation specifications and is pre-stored locally by the edge computing layer, covering event types that require immediate handling, such as traffic accidents, road debris, extreme fog, road icing, and vehicles driving in the wrong direction / illegally parked.

[0035] The network transport layer communication link was abnormally interrupted; Specifically, the edge computing layer continuously sends heartbeat detection packets to the network transport layer to monitor link connectivity, transmission rate and packet loss rate in real time. If there is no feedback for more than three consecutive heartbeat packets, the packet loss rate exceeds 90% or the signal is completely lost, the communication link is determined to be abnormally interrupted, triggering an emergency condition.

[0036] The global control strategy data was not received for more than a preset time period.

[0037] Specifically, after the edge computing layer uploads the localized processing results, it starts a policy waiting timer (the preset duration is the maximum response latency calibrated by the system). If no valid global control policy data is received from the central decision-making layer before the timer expires, it is determined that the emergency condition is met. The preset duration is set to 5 seconds by default when the system first starts. After the central decision-making layer has received no less than 100 valid policy responses, it automatically switches to 1.5 times the historical response latency P95 value, and is recalculated every 24 hours to achieve dynamic adaptive adjustment.

[0038] In this embodiment, the edge computing layer and the central decision-making layer achieve bidirectional policy synchronization through a preset heartbeat synchronization protocol. The edge computing layer uploads the localized processing result to the central decision-making layer every 100ms, carrying a unique incrementing version number of the local cached policy. When the central decision-making layer issues global management policy data, it carries a unique incrementing policy version number and policy validity period. The edge computing layer updates the local policy and resets the policy waiting timer only when the received policy version number is higher than the local cached version. When the system is powered on for the first time, the edge computing layer waits for the global policy for a maximum of 30 seconds. If it does not receive the policy within the time limit, it directly enters the local emergency control mode. When an emergency is triggered, the edge computing layer automatically calls the latest valid policy in the local cache as a backup and generates an emergency control command in combination with the localized processing result. The maximum validity period of the cached policy is 72 hours. If the validity period is exceeded, it will automatically downgrade to the basic security management mode.

[0039] It should be noted that the network transport layer uses 5G slicing technology to build a dedicated communication network, pre-dividing low-latency emergency data transmission channels and regular bandwidth data transmission channels; The low-latency emergency data transmission channel is configured with the highest transmission priority, ≤10ms ultra-low latency, and dedicated fixed bandwidth. It is only authorized to transmit emergency control instructions and prohibits regular data access. The regular bandwidth data transmission channel is configured with normal transmission priority and general shared bandwidth. It is used to transmit raw sensing data, localized processing results, global management strategy data, and regular management instructions.

[0040] When an emergency occurs, the edge computing layer generates an emergency control command based on the localized processing results and sends a channel scheduling command to the network transport layer to start a low-latency emergency data transmission channel, pushing the emergency control command to the terminal interaction layer for immediate execution in real time.

[0041] Specifically, the edge computing layer directly calls the locally cached localized processing results, without waiting for the central decision-making layer to issue global policies, and quickly generates emergency control instructions that match the current emergency scenario; the edge computing layer sends a dedicated channel scheduling instruction to the network transport layer, the instruction carrying an emergency channel activation identifier, transmission priority marker, and target terminal address; after receiving the scheduling instruction, the network transport layer immediately activates the low-latency emergency data transmission channel, prioritizes the allocation of core network resources to this channel, suspends non-critical data transmission on regular channels, and avoids resource preemption; the emergency control instruction is transmitted to the terminal interaction layer through the emergency channel with ultra-low latency and no packet loss, and the terminal skips the regular verification process and immediately executes the emergency control operation.

[0042] The emergency control command shall include at least one or more of the following: Dynamic speed limit instructions: Set speed limits for road sections based on the type and scope of the emergency (e.g., 20 km / h for foggy sections and 10 km / h for accident sections). Lane control instructions: clearly specify the closed lane number, closed section, and conditions for reopening; Instructions for guiding light mode: Set the color, flashing frequency, and guiding direction of the guiding light; Strong reminder command from the vehicle terminal: Forcefully trigger the vehicle OBU voice alarm and AR navigation pop-up; Management platform alarm command: Trigger the management platform's audible and visual alarms and automatically locate the event point; Emergency control commands adopt a "one command, one code" approach, with non-overriding priority. Once received by the terminal interaction layer, they are executed before regular control commands.

[0043] In one embodiment of the present invention, the specific steps for the central decision-making layer to generate global management and control strategy data are as follows: Step S201: Accept the localization processing results and simultaneously retrieve the highway static road network ledger, historical traffic control database and whole-area meteorological shared big data as the basic dataset; Specifically, the static road network ledger for expressways is a fixed basic information database for expressways, containing static parameters such as road network topology, number of lanes, lane width, kilometer markers, bridge and tunnel locations, roadside guidance equipment / sensor locations, and speed limit sign locations. The historical traffic control database stores structured historical data from the past 1-3 years, including traffic flow data, speed distribution, control strategy execution records, traffic accident records, severe weather response plans, and emergency control effectiveness. The province-wide shared meteorological big data comprises comprehensive meteorological data issued by meteorological departments, including real-time weather conditions for road sections, historical weather patterns, fog / rain / snow / icing warnings, and temperature / humidity / wind monitoring.

[0044] Step S202: Build a high-precision real-scene digital twin model based on the basic dataset, and bind the localized processing results to the corresponding road segment units of the digital twin model in real time to generate dynamic synchronous simulation data stream and real-time simulation and judgment results of the digital twin of the entire road segment. Specifically, the high-precision real-scene digital twin model is a 1:1 virtual simulation model of highways built based on GIS geographic information and 3D modeling technology, which can map real-time road conditions, environment, traffic flow, and equipment status. The dynamic synchronous simulation data stream is a real-time data stream of road network traffic flow, environmental status, and equipment status updated every second after the digital twin model is bound to real-time localized processing results. The real-time simulation assessment results of the entire road segment are conclusions drawn from the twin model simulation, including the congestion level, risk points, traffic efficiency, control priority, and optimal control scheme for the entire road segment.

[0045] Based on a basic dataset, a 1:1 high-precision digital twin model of a highway is built using a GIS geographic information interface and a 3D modeling engine. This model fully recreates the real-world structure of the road network, lanes, bridges, tunnels, roadside facilities, and slopes. Vehicle locations, speeds, traffic density, visibility, road surface conditions, and traffic events from the localized processing results are accurately mapped and bound to the corresponding units in the twin model in real time, using precise road segment coordinates and lane positions. This achieves millisecond-level dynamic synchronization between the model and the real-world environment, generating a dynamically synchronized simulation data stream. The twin model is used to simulate and extrapolate traffic flow, environmental risks, and control effects for the entire road segment over the next 5-15 minutes. It calculates congestion levels, accident risk values, and traffic efficiency for each segment, outputting real-time simulation and assessment results for the entire road segment, including risk points, control priorities, optimal control directions, and expected traffic effects.

[0046] Step S203: Construct the reinforcement learning algorithm operating architecture, define the safe passage reward function and the road network congestion penalty coefficient, and delineate the environmental state space and the control action space; Specifically, reinforcement learning algorithms are intelligent algorithms that interact with the road network environment and autonomously learn to optimize decisions. They are used to dynamically optimize parameters such as highway speed limits, lane control, and guidance strategies. The safe passage reward function is an algorithm evaluation function that positively rewards no traffic accidents, reasonable speeds, smooth traffic flow, and driving safety; the reward score is positively correlated with safe passage duration and smoothness. The road network congestion penalty coefficient is an algorithm evaluation coefficient that negatively punishes road network congestion, traffic delays, excessively low speeds, and inappropriate control; the penalty score is positively correlated with congestion duration and congestion severity. The environmental state space is the input dimension of reinforcement learning, including visibility, road surface temperature, icing risk, traffic density, average speed, and traffic event type / location / level. The control action space is the output dimension of reinforcement learning, including graded speed limit adjustments, dynamic lane closure / release, long-distance traffic detour guidance, guidance light mode switching, and information board text push.

[0047] The formula for the safe passage reward function is: Safe passage reward score = safe passage time × 0.4 + average speed compliance rate × 0.3 + reasonable traffic density rate × 0.2 + duration without abnormal events × 0.1; The above formula takes safety first, efficiency as a balance, order guarantee, and stability as its core design principles, and quantifies the four core objectives of highway management into weighted indicators. The reward score is obtained by weighted summation. The higher the score, the higher the level of road network safety and the better the traffic efficiency. The algorithm will automatically tend to output the optimal management strategy corresponding to the score.

[0048] Among them, safe passage time refers to the continuous running time within a road segment without safety risks such as traffic accidents, wrong-way driving / illegal parking, or malicious lane changes; 0.4 refers to the weight of safe passage time, which is determined through artificial scientific calculation using the Analytic Hierarchy Process (AHP), fitting experiments with 3 years of historical control data, and digital twin simulation experiments; safety is the core objective of highway control, and its negative correlation coefficient with the accident rate reaches 0.85, hence it is given the highest weight. Average speed compliance rate is the matching ratio between the actual average speed of a road segment and the theoretical safe speed limit (ratio = actual average speed / theoretical safe speed limit × 100%). 0.3 is the weight of the average speed compliance rate, calculated through AHP and fitting experiments with historical data. The speed compliance rate directly determines traffic efficiency and is a secondary core indicator. Traffic flow density reasonable rate is the proportion of time that the traffic flow density is in the high-speed saturation density range (0-200 pcu / km), reflecting traffic flow order. 0.2 is the weight of the traffic flow density reasonableness rate. Experiments have verified that reasonable density can reduce the risk of minor collisions. As an auxiliary indicator, its corresponding weight is determined to be 0.2. The duration without abnormal events is the continuous duration without abnormal events such as traffic accidents, spills, or extreme weather. Its weight is 0.1. It is a supplementary indicator reflecting environmental stability, and its weight is determined to be 0.1 through experiments.

[0049] The formula for the road network congestion penalty coefficient is as follows: Road network congestion penalty score = congestion duration × 0.4 + congested road section mileage × 0.3 + vehicle speed non-compliance rate × 0.2 + control response delay time × 0.1; The above formula is a weighted cumulative negative penalty function, used to constrain the algorithm to avoid output strategies that lead to congestion and control failure; the higher the score, the more severe the road network congestion and the worse the control effect, and the algorithm will automatically avoid such unreasonable control actions.

[0050] Among them, congestion duration is the continuous duration during which the average vehicle speed on a road segment is lower than the minimum safe speed (≤40km / h), with a weight of 0.4. It is calculated manually by AHP and fitted experimentally using historical congestion data. Congestion duration is a core indicator of road network failure, with a positive correlation coefficient of 0.88 with traffic loss, hence its highest weight of 0.4. Congested road segment mileage is the total length of road segments where congestion occurs, with a weight of 0.3. It is determined through experimental fitting, and mileage reflects the scope of congestion impact, making it a secondary core indicator. Speed ​​non-compliance rate refers to the proportion of vehicles on a road segment whose actual speed is lower than the safe speed limit, reflecting the traffic flow status, with a weight of 0.2, determined experimentally. Control response delay is the time difference between the occurrence of an abnormal event and the effective implementation of the control strategy, with a weight of 0.1. Control response delay is a supplementary indicator reflecting system response efficiency, with the lowest weight, hence its lowest weight of 0.1.

[0051] In this embodiment, all input parameters of the safe passage reward function and the road network congestion penalty function are first normalized in the 0-1 interval through min-max linear transformation before being weighted and summed to completely eliminate the influence of dimensional differences on the calculation results. The safe passage duration core characterizes the road segment's driving safety status, while the no-abnormal-event duration core characterizes the road environment's stable status. There is no parameter overlap or duplicate calculation problem between the two. The theoretical safe speed limit is the dynamic safe speed limit value of the road segment calculated by combining real-time weather, road surface conditions, and traffic density. Each weight coefficient is determined by constructing a judgment matrix through the analytic hierarchy process, completing consistency checks, and finally verifying the results through a 3-year historical control data fitting experiment and digital twin multi-scenario simulation. The weight determination process can be completely reproduced.

[0052] Step S204: Use historical control samples to conduct offline iterative training and simulation, optimize the speed limit threshold, lane release ratio and induction start-stop parameters cycle by cycle until the model converges to the preset control accuracy, and output the optimized and finalized algorithm parameter set; Specifically, historical traffic control samples from the historical traffic control database for 1-3 years are retrieved and input into a reinforcement learning algorithm to initiate offline iterative training and simulation. In each iteration, the algorithm automatically optimizes speed limit thresholds, lane clearance ratios, induced start-stop times, and detour route planning parameters, and calculates the loss function in real time (loss function = penalty score - reward score). When the loss function fluctuates by ≤0.01 over 100 consecutive iterations and the control accuracy reaches a preset 95% or higher, the model is considered converged, and offline training is stopped. The optimal parameters after training are solidified, generating an optimized and finalized algorithm parameter set, which is stored in the central decision-making layer parameter library for direct use during online runtime.

[0053] Step S205: In actual operation, input the dynamic synchronous simulation data stream, call the optimized and finalized algorithm parameter set for fast strategy calculation, combine the real-time simulation judgment results of the digital twin of the whole road segment, match the hierarchical control logic according to the road segment partition, and generate the original global strategy set including hierarchical speed limit adjustment, dynamic lane closure, long-distance traffic flow detour guidance and multi-mode guidance strategy switching. Specifically, during actual online operation, the system does not require retraining. It directly inputs the dynamically synchronized simulation data stream into the reinforcement learning model with pre-loaded parameters, performing millisecond-level fast policy inference calculations. Combined with the real-time simulation results of the entire road segment's digital twin, it implements hierarchical control logic based on mountainous / plain / bridge / tunnel / interchange zones. Low-risk road sections: Match standard speed limits + normal guidance strategies; Medium-risk road sections: Implement speed reduction and pay attention to warning strategies; High-risk road sections: Match graded speed limits + lane control + long-distance detour strategies; ultimately generate an original global strategy set that includes graded speed limit adjustment, dynamic lane closure, long-distance traffic flow detour guidance, and multi-mode guidance strategy switching.

[0054] The predicted traffic density and risk point evolution trends in the next 5 minutes from the real-time simulation results of the digital twin serve as input to the state space of the reinforcement learning environment. The original policy set generated by reinforcement learning needs to be simulated and verified in the digital twin model. If the simulation results show that the policy will lead to increased congestion or safety risks, it will revert to a suboptimal policy and re-simulate. The deviation between the digital twin simulation results and the actual execution effect is used to correct the mapping parameters of the digital twin model and the weights of the reinforcement learning reward function. Through the above collaborative mechanism, reinforcement learning is responsible for policy generation and optimization, while digital twin is responsible for policy verification and effect prediction, forming a closed loop of "generation-verification-correction".

[0055] Step S206: Standardize the format and verify the rationality of the original global strategy set to solidify it into standardized global control strategy data.

[0056] Specifically, the original global policy set undergoes format standardization processing: unifying command codes, road segment identifiers, parameter units, execution timing, and terminal adaptation formats to conform to the communication protocol of highway control equipment; then, three layers of rationality checks are performed: Safety verification: Verify whether the speed limit and lane closure range comply with highway safety regulations, and eliminate unreasonable speed limits / lane closures; Feasibility verification: Verify whether the detour routes and guidance strategies match the road network structure and whether there are any route conflicts; Timeliness verification: Verify whether the effective time and duration of the policy match real-time traffic / weather conditions; after verification, solidify the standardized global control policy data and distribute it to the edge computing layer through the network transmission layer.

[0057] In one embodiment of the present invention, the specific steps for the edge computing layer to generate conventional control instructions are as follows: Step S301: Perform data normalization and standardization on the global control strategy data and the localized processing results respectively, unify the timestamp, road segment coordinates and parameter units, and complete the precise alignment and matching of spatiotemporal dimensions; Specifically, the timestamps of both types of data are calibrated to UTC international standard time to eliminate time errors between devices and levels, ensuring complete data synchronization; the data locations are uniformly mapped to standard coordinates of highway kilometer markers + specific lane numbers to achieve precise location matching of road segments and lanes; parameters with different dimensions such as speed limits, vehicle speeds, traffic density, and visibility are converted into standardized values ​​in the 0-1 range through linear transformation to eliminate the interference of dimensional differences on subsequent verification; finally, the two types of data are accurately aligned across all dimensions of time, space, and format.

[0058] Step S302: Extract the core feature parameters of the localized processing results, compare and verify them item by item with the built-in standard thresholds of the global control strategy, and identify abnormal points where the global strategy does not match the local field conditions and where parameters conflict. Specifically, from the localized processing results, seven core operating condition parameters are extracted: visibility, road surface temperature, icing risk level, traffic density, average vehicle speed, warning level, and traffic incident type. The core parameters are then compared item by item and interval matching is verified with the speed limit threshold, traffic density threshold, and warning matching threshold built into the global strategy. If there is a mismatch between the global strategy speed limit and local risk, a conflict between the control level and local warning, or a strategy parameter exceeding the reasonable range of local operating conditions, the road segment / lane is immediately marked as an abnormal point, and the abnormality type, location, and conflicting parameters are recorded.

[0059] Step S303: Perform adaptive correction and compensation for abnormal points, calculate the working condition adaptation similarity score, and retain only the matching parameter combinations with adaptation scores higher than the preset threshold, and adapt the corrected fusion parameter package. Specifically, based on the principle of prioritizing local operating conditions, the global strategy parameters for abnormal locations are dynamically adjusted: in the case of local fog (visibility <200m), the global speed limit of 80km / h is adjusted to 20km / h; in the case of local traffic overload, the full lane opening is adjusted to the closed overtaking lane; a weighted matching algorithm is used to calculate the adaptation score (0-100 points) of the corrected parameters, and the scoring dimensions include parameter rationality, risk matching degree, and traffic efficiency adaptation degree; a preset adaptation threshold of 80 points is set, and only the parameter combination with a score ≥80 points with high adaptability is retained; the optimal parameters are integrated and packaged to form a fused parameter package after adaptation and correction.

[0060] In this embodiment, the working condition adaptation similarity score is calculated using a weighted matching algorithm. The calculation formula is: similarity score = speed limit matching degree × 0.4 + risk matching degree × 0.3 + traffic efficiency matching degree × 0.2 + equipment adaptation degree × 0.1, where each matching degree takes a value range of 0-1. The preset adaptation threshold is 80 points. At the same time, the adaptive correction compensation sets a safety boundary. The speed limit correction range does not exceed ±30% of the original global strategy speed limit value, and the number of lanes closed in one direction does not exceed 1 / 2 of the total number of lanes. Corrections exceeding the safety boundary need to be reported to the central decision-making level for confirmation before taking effect.

[0061] Step S304: Based on the modified and adapted fusion parameter combination, generate standardized routine control instructions that include dynamic roadside lighting guidance, variable information hierarchical text warnings, and human-vehicle collaborative interactive broadcasts.

[0062] Specifically, the brightness, flashing frequency, and color of the guidance lights are set (yellow flashing for deceleration warnings and green constant light for lane guidance); corresponding text is generated according to the warning level (regular road condition reminders, caution reminders to reduce speed, and warning reminders to avoid obstacles); vehicle-mounted OBU push information and roadside voice broadcast content are generated to achieve vehicle-road cooperative interaction; all commands adopt the standard communication format of highway control equipment, and the terminal can directly parse and execute them.

[0063] It should be noted that the terminal interaction layer includes a roadside terminal unit, a vehicle-mounted terminal unit, and a management platform terminal unit; The roadside terminal unit is used to perform intelligent guidance light dimming, variable message sign graphic and text display, and roadside directional voice warning operations. Specifically, after receiving routine control instructions or emergency control instructions from the edge computing layer, the roadside terminal unit will synchronously schedule the intelligent guidance lights, variable message signs, and roadside directional voice warning devices to work together according to the instructions. Under normal conditions, green lights indicate normal traffic lanes and display real-time traffic conditions and regular speed limits on variable message signs. Under emergency conditions, red flashing lights warn of danger and closed lanes, emergency speed limits and detour instructions are pushed on variable message signs, and warning information is broadcast to vehicles through directional voice devices, thus completing on-site visual and voice-based guidance and warning.

[0064] The vehicle-mounted terminal unit is used to perform operations such as vehicle-road cooperative information push, AR real-scene navigation guidance, and vehicle assisted driving strategy reminders; Specifically, the vehicle terminal unit takes the vehicle-mounted OBU device as its core, receives control commands issued by the edge computing layer through the vehicle-road cooperative communication link, and pushes information such as road speed limits, road condition warnings, and lane control to the driver in the form of screen display or voice broadcast. At the same time, it combines the vehicle's Beidou positioning with the real-scene image of the road section and overlays AR real-scene navigation guidance signs to issue driving assistance reminders such as deceleration, lane changing, and avoidance, providing the driver with precise and intelligent driving guidance services.

[0065] The management platform terminal unit is used to perform operations such as visual display of the overall control status, sound and light alarms for on-site abnormal events, and receiving and issuing manual intervention instructions.

[0066] Specifically, the management platform terminal unit provides a full-domain visual display of the traffic operation status, equipment working status, and execution progress of control commands across the entire road section. When emergency conditions such as traffic accidents, extreme weather, or equipment failures are detected, it immediately triggers an audible and visual alarm. At the same time, it receives manual intervention commands issued by highway operation and management personnel and transmits the commands to the edge computing layer to complete the supplementary issuance of control commands and full-process supervision, thereby achieving a closed-loop connection between back-end control and front-end execution.

[0067] It should be noted that the vehicle-mounted terminal unit is an OBU device that integrates a Beidou-3 positioning module, a 5G dual-mode communication module, and an inertial navigation unit.

[0068] This integrated vehicle-mounted OBU device serves as the physical carrier of the vehicle terminal unit, integrating the BeiDou-3 positioning module, 5G dual-mode communication unit, and inertial navigation unit into a single hardware unit for coordinated operation. The BeiDou-3 positioning module collects high-precision latitude and longitude, speed, and direction of travel satellite positioning information in real time. The 5G dual-mode communication unit establishes a stable, low-latency communication link with roadside equipment and the edge computing layer through SA / NSA dual-mode, quickly receiving routine control commands and emergency control commands and simultaneously transmitting vehicle driving status data. In scenarios where BeiDou satellite signals are weak or interrupted, such as in tunnels, mountainous areas, or obstructed road sections, the inertial navigation unit autonomously calculates the vehicle's position, attitude, and motion parameters, seamlessly connecting with BeiDou-3 positioning data to achieve uninterrupted continuous navigation. Relying on the precise positioning and stable communication capabilities of the multi-module collaboration, the device pushes information such as road speed limits, road condition warnings, lane guidance, and AR real-scene navigation to the driver in real time, completing precise interaction and driving guidance services in vehicle-road collaboration.

[0069] It should be noted that the central decision-making layer is also used to receive on-site execution data of instructions from the terminal interaction layer, and to iteratively optimize the parameters of the digital twin 3D model and the configuration of the reinforcement learning algorithm based on the feedback execution data.

[0070] The central decision-making layer receives real-time execution data from the terminal interaction layer via the network transmission layer. This data covers a full range of measured information, including the response status of roadside terminal equipment, the success rate of information push from vehicle terminals, changes in actual vehicle driving parameters, improvements in road section traffic efficiency, the effectiveness of safety warnings, and the results of handling abnormal conditions. The central decision-making layer first performs preprocessing operations on the execution data, such as data cleaning, normalization, and spatiotemporal alignment calibration, to remove invalid, abnormal, and duplicate data. Then, it uses a pre-set control effect evaluation model to quantitatively score the implementation effect of the issued global control strategy. Based on the scoring results, it specifically adjusts the road network simulation parameters, environment mapping parameters, traffic flow projection parameters, and equipment linkage parameters of the digital twin 3D model to ensure that the virtual situation of the twin model is highly consistent with the actual operating state of the highway. At the same time, based on the execution data and quantitative evaluation results, it dynamically adjusts the core configurations of the reinforcement learning algorithm, such as the reward and penalty function weights, learning rate, number of iterations, state space threshold, and control action levels. It then conducts offline iterative training and digital twin simulation verification again to optimize and upgrade the algorithm configuration, ensuring that the subsequently generated global control strategy data is more in line with the actual operating needs of the highway.

[0071] In this embodiment, the central decision-making layer conducts concept drift detection on a weekly basis. The KS test is used to statistically analyze the distribution difference between the on-site execution data of the current week's instructions and the historical training samples. When the test p-value is <0.05, concept drift is determined to have occurred. After determining that concept drift has occurred, the effective control data of the most recent 3 months is retrieved as incremental samples. Combined with the original historical samples, the reinforcement learning algorithm is incrementally iteratively trained and the algorithm parameter set is updated. The updated parameter set must be verified by digital twin simulation and the control accuracy must be ≥95% before it can be officially released. At the same time, the effective parameter set of the previous version is retained as a backup. If the control effect of the new version parameter set is lower than the preset threshold for 24 consecutive hours after online operation, it will automatically roll back to the effective parameter set of the previous version and start incremental training again.

[0072] This invention employs a five-layer collaborative architecture comprising a perception layer, a network transmission layer, an edge computing layer, a central decision-making layer, and a terminal interaction layer. It leverages multi-source perception collected from both roadside and vehicle-mounted systems to gather comprehensive road condition and environmental data, overcoming the shortcomings of traditional highway management systems, such as limited perception dimensions and blind spots. The edge computing layer performs data fusion calibration, event identification, and risk classification. It combines three-dimensional judgment logic to identify emergency situations and transmits emergency commands via a dedicated low-latency 5G slicing channel, overcoming the problems of traditional centralized cloud architectures that rely on communication links, experience control failures during network outages, and suffer from delayed emergency responses. The central decision-making layer integrates real-time digital twin simulation with offline reinforcement learning-based qualitative reasoning to generate globally optimal strategies, avoiding the drawbacks of high latency in online computation and poor adaptability of static strategies. The edge layer performs spatiotemporal alignment and adaptive correction of the global strategy, addressing the mismatch between the global solution and local actual conditions. Relying on an integrated BeiDou positioning, 5G communication, and inertial navigation vehicle-mounted OBU, coupled with a three-in-one terminal interaction system, it achieves precise guidance and early warning in multiple scenarios, improving upon the shortcomings of incomplete information push coverage in traditional systems. Meanwhile, based on the closed-loop iterative optimization of the model and algorithm parameters using on-site execution feedback data, the system overcomes the shortcomings of being fixed and unable to dynamically adapt to environmental changes, effectively ensuring the stability of control under severe weather and emergency conditions, improving traffic efficiency, and reducing the incidence of traffic accidents.

[0073] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An induction and control system for ensuring all-weather operation of intelligent highways, characterized in that, It includes a perception layer, a network transmission layer, an edge computing layer, a central decision-making layer, and a terminal interaction layer that are connected in sequence. The sensing layer is used to collect environmental sensing data and road condition sensing data; The network transport layer is used to receive and transmit data; The edge computing layer is used to perform multiple localization processes based on environmental sensing data and road condition sensing data, obtain localization processing results, and upload them to the central decision-making layer. The central decision-making layer is used to generate global management and control strategy data based on localized processing results, through digital twin modeling and reinforcement learning algorithm iterative optimization, and then distribute it to the edge computing layer. The edge computing layer combines localized processing results and global management and control strategy data to generate regular management and control instructions and output them to the terminal interaction layer. When an emergency occurs, it generates emergency control instructions. The terminal interaction layer is used to execute routine control commands and emergency control commands.

2. The guidance and control system for ensuring all-weather passage on intelligent highways according to claim 1, characterized in that, The perception layer includes roadside sensing components and vehicle-mounted sensing components. The roadside sensing components include millimeter-wave radar, a meteorological sensor array, a video analysis component, and a road surface condition sensor. The vehicle-mounted sensing components are vehicle-mounted OBU terminals. The roadside sensing components are used to collect environmental sensing data and some road condition sensing data, while the vehicle-mounted components are used to collect vehicle driving status data. The environmental sensing data includes visibility, precipitation intensity, road surface temperature, and icing risk parameters. The road condition sensing data includes vehicle location, driving speed, vehicle driving status, traffic accidents, and road debris parameters.

3. The induction and control system for ensuring all-weather operation of intelligent expressways according to claim 1, characterized in that, The specific steps for the edge computing layer to perform multiple localization processes are as follows: Based on environmental and road condition sensor data, multi-source data fusion calibration is performed to remove error and redundant data and generate calibrated sensor data. Traffic incident detection is performed based on calibrated sensor data to identify abnormal events and generate incident detection results; Based on the calibrated sensor data and event detection results, a warning level determination is performed, the warning level is matched, and a warning determination result is generated. The warning levels include Level 1 regular warning, Level 2 attention warning, and Level 3 emergency warning. Based on the warning determination result, the calibrated sensor data is classified and graded. The calibrated sensor data corresponding to Level 3 emergency warning is marked as emergency event data, and the remaining calibrated sensor data is marked as regular data, generating a data classification and grading result. The calibrated sensor data, event detection results, early warning judgment results, and data classification results are integrated to form localized processing results, which are then uploaded to the central decision-making level.

4. The induction and control system for ensuring all-weather operation of intelligent expressways according to claim 3, characterized in that, An emergency condition is determined when any one of the following criteria is met: The traffic incident types corresponding to the emergency incident data have a correspondence with the items in the preset emergency incident type list; The network transport layer communication link was abnormally interrupted; The global control strategy data was not received for more than a preset time period.

5. The induction and control system for ensuring all-weather operation of intelligent expressways according to claim 4, characterized in that, The network transport layer uses 5G slicing technology to build a dedicated communication network, pre-dividing low-latency emergency data transmission channels and regular bandwidth data transmission channels; When an emergency occurs, the edge computing layer generates an emergency control command based on the localized processing results and sends a channel scheduling command to the network transport layer to start a low-latency emergency data transmission channel, pushing the emergency control command to the terminal interaction layer for immediate execution in real time.

6. The guidance and control system for ensuring all-weather passage on intelligent highways according to claim 1, characterized in that, The specific steps for the central decision-making layer to generate global management and control strategy data are as follows: Accept the results of localization processing and simultaneously retrieve the static road network ledger of the expressway, the historical traffic control database and the shared big data of the whole region's meteorology as the basic dataset; A high-precision real-scene digital twin model is built based on the basic dataset. The localized processing results are then mounted and bound to the corresponding road segment units of the digital twin model in real time, generating a dynamic synchronous simulation data stream and real-time simulation and judgment results of the digital twin of the entire road segment. Construct a reinforcement learning algorithm operating architecture, define a safe passage reward function and a road network congestion penalty coefficient, and delineate the environmental state space and the control action space; Using historical control samples, offline iterative training and simulation are carried out to optimize the speed limit threshold, lane release ratio and induced start-stop parameters cycle by cycle until the model converges to the preset control accuracy and outputs the optimized and finalized algorithm parameter set. In actual operation, the dynamic synchronous simulation data stream is input, the optimized and finalized algorithm parameter set is called to perform fast strategy calculation, and combined with the real-time simulation and judgment results of the digital twin of the whole road segment, the hierarchical control logic is matched according to the road segment partition to generate the original global strategy set including hierarchical speed limit adjustment, dynamic lane closure, long-distance traffic flow detour guidance and multi-mode guidance strategy switching. The original set of global strategies is standardized in format and validated for rationality, and then solidified into standardized global control strategy data.

7. The induction and control system for ensuring all-weather operation of intelligent expressways according to claim 1, characterized in that, The specific steps for the edge computing layer to generate routine control commands are as follows: The global control strategy data and the localized processing results are normalized and standardized respectively to unify the timestamps, road segment coordinates and parameter units, and achieve accurate alignment and matching in the spatiotemporal dimensions. Extract the core feature parameters of the localized processing results and compare them item by item with the built-in standard thresholds of the global control strategy to identify abnormal points where the global strategy does not match the local field conditions and where parameters conflict. Adaptive correction and compensation are performed for abnormal locations, and the working condition adaptation similarity score is calculated. Only matching parameter combinations with adaptation scores higher than a preset threshold are retained, and the fusion parameter package after adaptation and correction is generated. Based on the modified and adapted fusion parameter combination, standardized routine control instructions are generated, including dynamic roadside lighting guidance, variable information hierarchical text warnings, and human-vehicle collaborative interactive broadcasts.

8. The induction and control system for ensuring all-weather operation of intelligent expressways according to claim 1, characterized in that, The terminal interaction layer includes a roadside terminal unit, a vehicle-mounted terminal unit, and a management platform terminal unit. The roadside terminal unit is used to perform intelligent guidance light dimming, variable message sign graphic and text display, and roadside directional voice warning operations. The vehicle-mounted terminal unit is used to perform operations such as vehicle-road cooperative information push, AR real-scene navigation guidance, and vehicle assisted driving strategy reminders; The management platform terminal unit is used to perform operations such as visual display of the overall control status, sound and light alarms for on-site abnormal events, and receiving and issuing manual intervention instructions.

9. The induction and control system for ensuring all-weather operation of intelligent expressways according to claim 8, characterized in that, The vehicle-mounted terminal unit is an OBU device that integrates a Beidou-3 positioning module, a 5G dual-mode communication module, and an inertial navigation unit.

10. The induction and control system for ensuring all-weather operation of intelligent expressways according to claim 6, characterized in that, The central decision-making layer is also used to receive on-site execution data of instructions from the terminal interaction layer, and to iteratively optimize the parameters of the digital twin 3D model and the configuration of the reinforcement learning algorithm based on the feedback execution data.

Images