Highway service area vehicle safety management and control method and system based on cloud control platform

Abstract

An expressway service area vehicle safety management and control system based on a cloud control platform comprises a multi-mode sensing device used for collecting vehicle OBD data, driver biological characteristic data and environment parameters in real time; the edge computing node is integrated with a lightweight AI inference engine and can realize vehicle type identification and preliminary risk scoring; the cloud management and control platform is provided with a dynamic risk assessment module which can fuse the improved D-S evidence theory and the LSTM neural network; the intelligent scheduling decision module is used for optimizing resource allocation based on a reinforcement learning algorithm; the digital twinning early warning module is used for constructing a three-dimensional scene and mapping a physical space in real time; and the execution terminal comprises an LED induction screen, an automatic gate and emergency response equipment. According to the invention, through total factor risk identification, dynamic safety decision, intelligent resource scheduling, scene early warning and closed-loop emergency response, the comprehensiveness, dynamic nature and intelligent level of two-passenger one-risk vehicle safety management and control in the expressway service area are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation and the Internet of Things technology, and in particular to a method and system for vehicle safety management and control in highway service areas based on a cloud control platform. Background Art

[0002] With the "14th Five-Year Plan" for the Development of a Modern Integrated Transportation System increasing safety control requirements for "two passenger and one dangerous goods" vehicles (long-distance buses, chartered tourist buses, and dangerous goods transport vehicles), existing technologies present significant deficiencies: manual inspection coverage is less than 80%, the accuracy of identifying dangerous goods vehicle markings is only 75%, multi-source data integration is less than 30%, and risk assessment relies on static thresholds, resulting in a misjudgment rate exceeding 20%. Emergency response times average 42 minutes, and service area resource utilization fluctuates by as much as 40%. In 2023, 35% of service area accidents involving "two passenger and one dangerous goods" vehicles nationwide were caused by spatial conflicts, making traditional two-dimensional monitoring systems unable to provide dynamic early warning of risk areas.

[0003] Through research and analysis, the causes of the above-mentioned significant defects mainly include the following issues: Lack of multi-dimensional data collection system: Traditional systems only obtain basic information such as license plate and vehicle model, and do not integrate vehicle OBD real-time data (such as engine speed, brake pressure), driver biometrics (PERCLOS value, steering wheel grip) and environmental parameters (road slip coefficient, visibility). The data fusion rate is less than 30% (test data from the Ministry of Transport in 2024).

[0004] Key parameters such as the tank temperature and urea level of hazardous materials vehicles were not included in the monitoring scope, resulting in 22% of service area accidents in 2023 being caused by vehicle mechanical failures (Accident Analysis Report of the Ministry of Emergency Management).

[0005] Weak dynamic risk assessment capabilities: It relies on static thresholds to determine safety status (e.g., an alarm will be issued when the vehicle speed is greater than 10km / h), ignoring dynamic factors such as vehicle load and weather conditions, resulting in a misjudgment rate of up to 28%.

[0006] The lack of a time series prediction model makes it impossible to identify high-risk conditions such as fatigue driving and brake system abnormalities 30 minutes in advance, resulting in an average emergency response delay of 42 minutes.

[0007] Blanks in 3D scenario-based management and control: Traditional two-dimensional monitoring systems cannot intuitively display the spatial relationship between vehicle location and risk level. In 2023, 35% of service area accidents involved spatial conflicts between vehicles and dangerous areas such as charging piles and steep slopes (special survey by China Transportation News).

[0008] Due to the lack of centimeter-level positioning technology, the vehicle trajectory tracking error is greater than 2 meters, making it difficult to achieve accurate path guidance and risk warning.

[0009] Smart scheduling and safety management are out of sync: Resource allocation in service areas relies on manual empirical formulas, inspection station utilization fluctuates by 40%, and the mixing of high-risk vehicles and ordinary vehicles increases the probability of accidents by 1.8 times (data from the "Highway Service Area Safety Management Specifications").

[0010] A differentiated management and control strategy based on risk levels has not been established. The average waiting time for inspection for high-risk vehicles at levels 4-5 is as long as 18 minutes, missing the best time for handling. Summary of the Invention

[0011] The purpose of the present invention is to propose a method and system for vehicle safety management and control in highway service areas based on a cloud control platform to solve the technical defects pointed out in the background technology.

[0012] In order to solve the above technical problems, the technical solution adopted by the present invention is: A highway service area vehicle safety management and control system based on a cloud control platform, including: Multimodal sensing device for real-time collection of vehicle OBD data, driver biometric data, and environmental parameters; Edge computing nodes with integrated lightweight AI inference engines that can identify vehicle types and perform preliminary risk scoring; The cloud-based management and control platform includes: Dynamic risk assessment module, which can integrate improved DS evidence theory and LSTM neural network; Intelligent scheduling decision module, which optimizes resource allocation based on reinforcement learning algorithms; A digital twin early warning module, which constructs three-dimensional scenes and maps physical spaces in real time; Execution terminal, including LED induction screen, automatic gate and emergency response equipment.

[0013] Furthermore, the multimodal sensing device is composed of the following parts: The OBD data acquisition module is configured to parse the ISO 15031 protocol and can obtain 12 parameters, including engine speed, brake pressure, coolant temperature, fuel pressure, intake air temperature, throttle position, oxygen sensor voltage, vehicle speed, engine load, ignition advance angle, battery voltage, and urea level; Driver monitoring module, integrating infrared camera and steering wheel grip sensor, can detect PERCLOS value and fatigue status; The environmental monitoring module, including millimeter-wave radar and weather station, can obtain road slip coefficient and visibility data in real time.

[0014] Furthermore, the dynamic risk assessment module specifically includes: The data preprocessing unit performs spatiotemporal alignment of multi-source data with a time error of no more than 50ms and performs standardization. The conflict evidence correction unit uses the Yager correction method to process high-conflict data in DS evidence theory, and the conflict degree threshold is set to 0.7; The time series prediction unit uses a bidirectional LSTM network to predict the risk level in the next 30 minutes, with a prediction accuracy of no more than 0.35 in terms of root mean square error.

[0015] Furthermore, the intelligent scheduling decision module adopts a reinforcement learning algorithm, and its state space covers: Service area resource status, including the number of charging stations, parking spaces, and inspection stations; Real-time traffic flow data, with a prediction error R² of no less than 0.85; The detection accuracy of the current vehicle queue length reaches 99.2%.

[0016] Furthermore, the reward function of the reinforcement learning algorithm is: R = 0.4*(T0-T) / T0 + 0.3*U + 0.3*(5-L) / 5 Among them, T0 represents the historical average stay time, T is the current stay time, U is the resource utilization rate, and L is the risk level.

[0017] Furthermore, the digital twin early warning module includes: 3D scene construction unit, which generates a 1:100 scale model based on BIM+oblique photography technology; Real-time positioning unit, integrating 5G+UWB technology to achieve centimeter-level positioning with an accuracy of ±15cm; The dynamic warning unit sets the vehicle speed exceeding warning rule, and the threshold is set to 15km / h; the safety distance insufficient warning rule is set, and the dynamic formula adopts 0.5v+3m.

[0018] A method for vehicle safety management and control in highway service areas based on a cloud control platform, with the following implementation steps: Step S1. Multimodal data acquisition: using a sensor array to acquire vehicle status, driver behavior, and environmental parameters; Step S2. Edge layer preprocessing: denoising, feature extraction, and lightweight model inference on the raw data; Step S3. Cloud risk assessment: Use improved DS evidence theory to fuse multi-source data; Use LSTM neural network to predict future risk levels, which are divided into levels 1-5; Step S4. Intelligent scheduling decision: Generate dynamic scheduling instructions based on risk level and resource status; Optimize inspection station priority and resource allocation through Q-learning algorithm; Step S5. Digital twin warning: Mark vehicle location and risk level in real time in a 3D scene; Trigger AR alerts, such as reminders of entering dangerous areas; Step S6. Execute hierarchical control: Execute measures such as automatic release, guided inspection, or forced parking based on the risk level.

[0019] Furthermore, the implementation method of the improved DS evidence theory in step S3 is: Introducing a weight coefficient matrix determined based on historical data training; Yager correction is performed on the conflicting evidence, and the conflict degree of the synthesis result is reduced by 40%.

[0020] Furthermore, the specific content of the intelligent scheduling decision in step S4 is: Dynamically adjust the inspection station priority, with priorities ranging from 1 to 5, and the response time does not exceed 30 seconds; Publish guidance information to the LED screen with an update frequency of 5Hz to guide vehicles to choose the optimal path.

[0021] Furthermore, the specific content of the digital twin warning in step S5 is: Map 5G+UWB positioning data to a 3D model in real time with an update frequency of 1Hz; Conduct simulation predictions for dangerous scenarios, such as calculating braking distance in heavy rain, with a prediction error of no more than 2%.

[0022] Compared with the prior art, the present invention has the following beneficial effects: Full-factor risk identification: Integrates vehicle OBD data, driver biometrics, and environmental parameters to break through the limitations of traditional single-dimensional detection and build a three-dimensional risk assessment system covering mechanical status, human behavior, and environmental conditions, significantly improving the comprehensiveness of risk identification.

[0023] Dynamic security decision-making: Improve the DS evidence theory and LSTM neural network fusion model, analyze the spatiotemporal correlation characteristics of multi-source data in real time, dynamically adjust the risk judgment threshold, avoid misjudgment problems caused by static rules, and achieve a management and control upgrade from passive response to active prevention.

[0024] Intelligent resource scheduling: Automatically optimizes inspection station allocation strategies based on reinforcement learning algorithms, dynamically adjusts priorities based on real-time traffic flow, vehicle risk level, and resource status, significantly improving service area resource utilization efficiency and reducing the risk of high-risk vehicles and ordinary vehicles mixing.

[0025] Scenario-based early warning and disposal: The digital twin system uses centimeter-level positioning and real-time three-dimensional scene mapping to intuitively display the spatial relationship between vehicle position and risk level. Combined with AR warning technology, it can achieve instant early warning of scenarios such as intrusion into dangerous areas and insufficient safety distance, thereby enhancing situational awareness capabilities.

[0026] Closed-loop emergency response: The hierarchical management and control strategy is deeply linked with the execution terminal to achieve full process automation from risk identification, instruction generation to measure execution, significantly shortening the emergency response time, ensuring the timely handling of high-risk incidents, and reducing the probability of accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0028] Figure 1 This is a logic block diagram of the implementation steps of the highway service area vehicle safety management and control method based on the cloud control platform of the present invention. DETAILED DESCRIPTION

[0029] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. 1. System deployment and hardware configuration (refer to Figure 1 )

[0030] 1.1 Multimodal Perception Device OBD data acquisition module: This module uses an on-board diagnostic interface that complies with the ISO 15031 standard and acquires 12 parameters (such as engine speed and brake pressure, as detailed in Claim 2) in real time via the CAN bus. The data acquisition frequency is 10 Hz, and the transmission protocol uses MQTT (QoS = 1).

[0031] Driver monitoring module: An infrared camera (resolution 1920×1080) is installed above the instrument panel, with a PERCLOS value calculation cycle of 5 seconds; the steering wheel grip force sensor has a range of 0-500N, an accuracy of ±5%, and a sampling frequency of 20Hz.

[0032] Environmental monitoring module: Millimeter-wave radar (operating frequency band 77GHz) is deployed at the service area entrance with a detection range of 0-50m. The road slip coefficient is calculated based on Doppler shift. The weather station updates visibility (accuracy ±5m) and temperature (±0.5℃) data every 30 seconds.

[0033] 1.2 Edge Computing Nodes Hardware platform: NVIDIA Jetson AGX Orin (275 TOPS computing power), integrated 5G communication module (Sub-6GHz frequency band, peak rate 10Gbps).

[0034] Software framework: Ubuntu 20.04 operating system, TensorRT 8.5 optimized YOLOv8n model (vehicle detection latency <20ms), lightweight risk scoring model (based on XGBoost, 95% accuracy for level 1-3 risk classification).

[0035] 1.3 Cloud Management and Control Platform Server cluster: Use Alibaba Cloud ACK container service and deploy Spring Cloud microservice architecture, including: Vehicle health management service: Based on HBase to store OBD historical data, the remaining range prediction algorithm has an error of less than 5%; Driver Status Analysis Service: Integrates an emotion recognition model trained using FER-2013 (92% accuracy), with a fatigue threshold set at a blink frequency >15 times / minute. Environmental risk assessment service: Linked with meteorological satellite data, heavy rain warnings are triggered 30 minutes in advance, and road icing warnings are based on infrared temperature measurement (threshold 0°C).

[0036] 1.4 Execution Terminal LED guidance screen: P4 model, display resolution 1920×1080, update frequency 5Hz, support dynamic path planning guidance; Automatic gate: response time <500ms, supports license plate recognition and RFID dual verification; Emergency response equipment: Deploy 5G smart terminals (Huawei 5G CPE Pro 3) to support video calls and command push (latency < 200ms). 2. Algorithm Implementation and Data Flow

[0037] 2.1 Dynamic Risk Assessment Model Data preprocessing: Multi-source data were aligned by timestamp (accuracy ±1ms), normalized by Z-score, and outlier detection of OBD data was based on the IQR method (threshold range 1.5 times the interquartile range).

[0038] Improved DS evidence theory: The weight coefficient matrix is ​​determined through historical data training (such as OBD data weight 0.4, driver status 0.3, environmental factors 0.3), and conflict evidence processing adopts the Yager correction method (activated when the conflict degree is greater than 0.7).

[0039] LSTM prediction model: The input layer is 12-dimensional features (including 3 time-step historical data), the bidirectional LSTM layer is set with 128 neurons, the loss function uses cross entropy, the training data contains more than 100,000 accident samples (2020-2024), and the prediction accuracy RMSE=0.32.

[0040] 2.2 Intelligent Scheduling Decision Algorithm Reinforcement learning training: A digital twin simulation environment was built using the Unity engine, with 100 training cycles and 5,000 iterations per cycle. The state space included resource status (3 dimensions), traffic prediction (1 dimension), and queue length (1 dimension). The action space consisted of inspection station priority adjustment (levels 1-5).

[0041] Reward function optimization: The Q network was trained using TensorFlow, with a learning rate of 0.001 and a discount factor of γ = 0.95. After training, the reward value stabilized above 0.85, reducing waiting time at inspection stations by 32%.

[0042] 2.3 Digital Twin Early Warning System 3D modeling: The BIM model is constructed using Revit 2024, with an oblique photography data resolution of 0.05m, dynamic traffic flow data update frequency of 1Hz, and 5 types of risk points (buffer zones, sharp bends, etc.) marked in dangerous areas.

[0043] 5G+UWB positioning: UWB base stations are deployed 200 meters apart and use a time-of-flight (TOF) positioning algorithm with a positioning accuracy of ±15cm. Real-time location data is transmitted to the digital twin scene via the UDP protocol (latency <100ms).

[0044] Warning rule execution: A red warning is triggered when the vehicle speed exceeds the limit (15km / h), and a yellow warning is activated when the safety distance is insufficient (0.5v+3m). The AR warning is presented through the Hololens 2 device. 3. System workflow

[0045] 3.1 Vehicle entry stage The ground sensor coil triggers multi-sensor acquisition, with a license plate recognition accuracy of 99.7% (OCR algorithm based on CRNN+CTC); The edge node completes vehicle type identification (YOLOv8n) and preliminary risk scoring (levels 1-3). High-risk vehicles (levels 4-5) trigger in-depth analysis in the cloud.

[0046] 3.2 Real-time processing stage The cloud-based dynamic risk assessment module integrates multi-source data to generate risk levels from 1 to 5 (response time < 800ms); The intelligent scheduling decision module generates dynamic scheduling instructions (such as guiding to the quick inspection channel) based on the risk level and resource status.

[0047] 3.3 Hierarchical Control Implementation Risk Level Control measures Execute terminal response Level 1-2 Automatic release Gate opening (<500ms) Level 3 Guide to the fast lane LED screen update path (<300ms) Level 4-5 Forced parking + emergency response Gate closed + notify security officer (<15 seconds) 3.4 Exit Phase Generate a safety report (PDF format) containing risk analysis maps and treatment recommendations; Data is archived to a cloud database (InfluxDB) with a storage period of 3 years. 4. Test and Verification

[0048] 4.1 Test Environment Guangdong Province Expressway Service Area (average daily traffic volume of 2,000 vehicles, including complex scenarios such as rainy days and nighttime); The test period is 30 days, covering long-distance buses, tourist chartered buses, hazardous chemicals transport vehicles and other models.

[0049] 4.2 Test Results index Test results Risk identification accuracy 98.7% Emergency response time 12.3 seconds Check workstation utilization 85% Average vehicle dwell time 21 minutes 5G+UWB positioning accuracy ±12cm 5. Implementation Effect

[0050] Through multimodal data fusion, dynamic risk assessment models and reinforcement learning scheduling algorithms, the present invention achieves: the accuracy rate of high-risk vehicle identification is increased to 98.7%, a 16.4% improvement over traditional methods; emergency response time is shortened from 42 minutes to 12.3 seconds, meeting the requirements of the "Regulations on the Safety Management of Road Transport of Dangerous Goods"; service area resource utilization is increased by 30%, and waiting time at inspection stations is reduced by 32%; three-dimensional scenario-based early warning reduces the spatial conflict accident rate by 45%.

[0051] The present invention has the following beneficial effects: Full-factor risk identification capability: By integrating vehicle OBD data, driver biometrics, and environmental parameters through multimodal sensing devices, a full-factor risk assessment system covering mechanical status, human behavior, and environmental conditions is constructed, breaking through the limitations of traditional single-dimensional detection and significantly improving the comprehensiveness and accuracy of risk identification.

[0052] Dynamic security decision-making mechanism: An improved fusion model of DS evidence theory and LSTM neural network can analyze the spatiotemporal correlation characteristics of multi-source data in real time, dynamically adjust the risk level judgment threshold, avoid misjudgment problems caused by static rules, and achieve an upgrade from "passive response" to "active prevention" management and control mode.

[0053] Intelligent resource scheduling system: A dynamic scheduling algorithm based on reinforcement learning can automatically optimize the inspection station allocation strategy based on real-time traffic flow, vehicle risk level, and service area resource status, significantly improving resource utilization efficiency while reducing the risk of high-risk vehicles and ordinary vehicles mixing.

[0054] Scenario-based early warning and disposal capabilities: The digital twin early warning system uses centimeter-level positioning technology and real-time mapping of three-dimensional scenes to intuitively display the spatial relationship between vehicle location and risk level. Combined with AR warning technology, it can achieve instant early warning of high-risk scenarios such as intrusion into dangerous areas and insufficient safety distance, enhancing the situational awareness capabilities of drivers and managers.

[0055] Closed-loop emergency response process: The deep linkage between hierarchical management and control strategies and execution terminals realizes the automation of the entire process from risk identification, instruction generation to measure execution, significantly shortens emergency response time, ensures the timely handling of high-risk incidents, and reduces the probability of accidents.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

[0057] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A highway service area vehicle safety management and control system based on a cloud control platform, characterized by: include: Multimodal sensing device for real-time collection of vehicle OBD data, driver biometric data, and environmental parameters; Edge computing nodes with integrated lightweight AI inference engines that can identify vehicle types and perform preliminary risk scoring; The cloud-based management and control platform includes: Dynamic risk assessment module, which can integrate improved DS evidence theory and LSTM neural network; Intelligent scheduling decision module, which optimizes resource allocation based on reinforcement learning algorithms; A digital twin early warning module, which constructs three-dimensional scenes and maps physical spaces in real time; Execution terminal, including LED induction screen, automatic gate and emergency response equipment.

2. The system according to claim 1, wherein: The multimodal sensing device is composed of the following parts: The OBD data acquisition module is configured to parse the ISO 15031 protocol and can obtain 12 parameters, including engine speed, brake pressure, coolant temperature, fuel pressure, intake air temperature, throttle position, oxygen sensor voltage, vehicle speed, engine load, ignition advance angle, battery voltage, and urea level; Driver monitoring module, integrating infrared camera and steering wheel grip sensor, can detect PERCLOS value and fatigue status; The environmental monitoring module, including millimeter-wave radar and weather station, can obtain road slip coefficient and visibility data in real time.

3. The system according to claim 1, wherein: The dynamic risk assessment module specifically includes: The data preprocessing unit performs spatiotemporal alignment of multi-source data with a time error of no more than 50ms and performs standardization. The conflict evidence correction unit uses the Yager correction method to process high-conflict data in DS evidence theory, and the conflict degree threshold is set to 0.7; The time series prediction unit uses a bidirectional LSTM network to predict the risk level in the next 30 minutes, with a prediction accuracy of no more than 0.35 in terms of root mean square error.

4. The system according to claim 1, wherein: The intelligent scheduling decision module adopts a reinforcement learning algorithm, and its state space covers: Service area resource status, including the number of charging stations, parking spaces, and inspection stations; Real-time traffic flow data, with a prediction error R² of no less than 0.85; The detection accuracy of the current vehicle queue length reaches 99.2%.

5. The system according to claim 4, wherein: The reward function of the reinforcement learning algorithm is: R = 0.4*(T0-T) / T0 + 0.3*U + 0.3*(5-L) / 5 Among them, T0 represents the historical average stay time, T is the current stay time, U is the resource utilization rate, and L is the risk level.

6. The system according to claim 1, wherein: The digital twin early warning module includes: 3D scene construction unit, which generates a 1:100 scale model based on BIM+oblique photography technology; Real-time positioning unit, integrating 5G+UWB technology to achieve centimeter-level positioning with an accuracy of ±15cm; The dynamic warning unit sets the vehicle speed exceeding warning rule, and the threshold is set to 15km / h; the safety distance insufficient warning rule is set, and the dynamic formula adopts 0.5v+3m.

7. A vehicle safety management and control method for highway service areas based on a cloud control platform, characterized in that: The implementation steps are as follows: Step S1. Multimodal data acquisition: using a sensor array to acquire vehicle status, driver behavior, and environmental parameters; Step S2. Edge layer preprocessing: denoising, feature extraction, and lightweight model inference on the raw data; Step S3. Cloud risk assessment: Use improved DS evidence theory to fuse multi-source data; Use LSTM neural network to predict future risk levels, which are divided into levels 1-5; Step S4. Intelligent scheduling decision: Generate dynamic scheduling instructions based on risk level and resource status; Optimize inspection station priority and resource allocation through Q-learning algorithm; Step S5. Digital twin warning: Mark vehicle location and risk level in real time in a 3D scene; Trigger AR alerts, such as reminders of entering dangerous areas; Step S6. Execute hierarchical control: Execute measures such as automatic release, guided inspection, or forced parking based on the risk level.

8. The method according to claim 7, characterized in that The implementation method of the improved DS evidence theory in step S3 is: Introducing a weight coefficient matrix determined based on historical data training; Yager correction is performed on the conflicting evidence, and the conflict degree of the synthesis result is reduced by 40%.

9. The method according to claim 7, characterized in that The specific content of the intelligent scheduling decision in step S4 is: Dynamically adjust the inspection station priority, with priorities ranging from 1 to 5, and the response time does not exceed 30 seconds; Publish guidance information to the LED screen with an update frequency of 5Hz to guide vehicles to choose the optimal path.

10. The method according to claim 7, characterized in that The specific content of the digital twin warning in step S5 is: Map 5G+UWB positioning data to a 3D model in real time with an update frequency of 1Hz; Conduct simulation predictions for dangerous scenarios, such as calculating braking distance in heavy rain, with a prediction error of no more than 2%.

Images