Real-time risk early warning and evaluation method in vehicle transportation process based on Internet of Things

Abstract

The invention relates to the technical field of risk analysis, in particular to a real-time risk early warning and evaluation method in the vehicle transportation process based on the Internet of Things. The method comprises the steps that a time-varying Kalman filtering model is adopted, and the driving state deviation degree representing the deviation between the driver manipulation intention and the vehicle dynamic feedback is evaluated and quantified; inputting the vehicle IMU data and the cargo pressure distribution data into a preset vehicle-cargo rigid-flexible coupling dynamic model, and resolving a vehicle-cargo system instability index representing that the vehicle is under the current motion disturbance in real time; and fusing real-time traffic flow and meteorological data serving as dynamic attributes and road geometry and historical accident data serving as static attributes by using a space-time diagram convolutional network, and predicting a risk potential field value of each road section along a transportation line in a future time window. Compared with the prior art, the method has the advantages that a person-vehicle-cargo-road-environment multi-dimensional collaborative risk assessment system is constructed, and the accuracy and foresight of risk early warning are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of risk analysis. More specifically, the present invention relates to a real-time early warning and assessment method for risks in vehicle transportation based on the Internet of Things. Background Art

[0002] With the rapid development of IoT and sensor technologies, safety monitoring and risk management during vehicle transportation have attracted significant attention. Existing technologies primarily use onboard terminals to collect real-time vehicle operating status information, such as GPS location, speed, and acceleration, and also acquire some vehicle operating data via the Controller Area Network (CAN) bus. Some solutions also incorporate video surveillance equipment to identify driver fatigue (such as yawning or closing eyes). These technologies, through threshold judgment or simple analysis of single or limited dimensions of information, enable basic monitoring and alerting of high-risk behaviors such as speeding, abrupt acceleration and deceleration, and fatigued driving, significantly improving road transportation safety.

[0003] However, existing technologies lack significant depth and breadth in risk assessment. First, most approaches analyze risk factors in isolation within the human-vehicle-cargo-road-environment system, lacking comprehensive assessment models that deeply integrate the driver's physiological and psychological state, the dynamic characteristics of vehicle-cargo coupling, and time-varying road environmental risks. This results in limited accuracy and predictive power of early warnings. Second, driver status monitoring often focuses on identifying external physiological signs such as fatigue, failing to quantify subtle discrepancies between driving intent and vehicle feedback caused by cognitive load and emotional fluctuations. Furthermore, in freight transportation scenarios, particularly for liquid, bulk, or easily shifting cargo, existing technologies generally overlook the impact of the dynamic coupling between cargo and vehicle on driving stability. Finally, assessments of external environmental risks often rely on static road attributes or real-time meteorological information, making it difficult to effectively predict dynamic risks such as future traffic flow and severe weather developments along the route. Therefore, achieving deep integration of multi-source heterogeneous information and developing a method for accurately, dynamically, and proactively assessing comprehensive transportation risks remains a pressing technical challenge. Summary of the Invention

[0004] In order to solve the technical problem of how to accurately, dynamically and prospectively assess comprehensive transportation risks, the present invention provides the following solution.

[0005] A real-time early warning and assessment method for risks in vehicle transportation based on the Internet of Things comprises the following steps: S1: real-time acquisition of vehicle IMU data and controller area network (CAN) bus data, heart rate variability (HRV) data, cargo pressure distribution data, and external environment and road network data; S2: based on the heart rate variability (HRV) data and the driver's micro-operation sequence extracted from the CAN bus data, a time-varying Kalman filter model is used to evaluate and quantify the driving state deviation that characterizes the deviation between the driver's manipulation intention and the vehicle's dynamic feedback; S3: inputting the vehicle IMU data and cargo pressure distribution data into a preset vehicle-cargo rigid-flexible coupling dynamics model, and solving in real time the driving state deviation that characterizes the vehicle's current state of motion. The instability index of the vehicle-cargo system under dynamic disturbance; S4: using the spatiotemporal graph convolutional network, the real-time traffic flow and meteorological data as dynamic attributes are integrated with the road geometry and historical accident data as static attributes to predict the risk potential field value of each section along the transportation route in the future time window; wherein the real-time traffic flow, meteorological data, road geometry, and historical accident data are derived from the external environment and road network data; S5: using the predefined hierarchical Bayesian network, the driving state deviation, the instability index of the vehicle-cargo system and the risk potential field value corresponding to the current position of the vehicle are used as evidence input to perform probabilistic reasoning and output the comprehensive transportation risk probability; when the comprehensive transportation risk probability exceeds the preset classification threshold, the corresponding level of risk warning is triggered.

[0006] The present invention constructs a multi-dimensional coordinated risk assessment system for people, vehicles, cargo, roads and the environment, which can accurately, dynamically and prospectively assess comprehensive transportation risks.

[0007] Furthermore, the step of evaluating and quantifying the driving state deviation includes: using the driver's steering wheel angle as input to construct a vehicle dynamics model, and performing real-time correction on the vehicle dynamics model parameters in combination with the heart rate variability (HRV) data; using the vehicle yaw rate and center of mass sideslip angle obtained from the vehicle IMU data as actual observation values; and using a time-varying Kalman filter model to calculate the Mahalanobis distance between the vehicle state predicted by the time-varying Kalman filter model and the actual observed state as the driving state deviation.

[0008] By fusing heart rate variability (HRV) data with vehicle micro-operation data, the present invention accurately quantifies the degree of deviation between driving intention and vehicle feedback, achieving deep insight into the driver's cognitive and control status.

[0009] Furthermore, the vehicle-cargo rigid-flexible coupling dynamic model includes: establishing a 14-degree-of-freedom rigid body dynamic model of the whole vehicle; equating the cargo system to a multi-flexible body unit based on the absolute node coordinate method, and coupling it with the whole vehicle rigid body dynamic model at the suspension system; using the vehicle's three-axis acceleration and angular velocity as model excitation inputs, converting the cargo pressure distribution data into the normal force of the contact surface between the cargo and the car body, calculating the real-time vertical load of the left and right tires of the vehicle through the whole vehicle rigid body dynamic model, and calculating the vehicle-cargo system instability index based on the lateral load transfer rate formula; the vehicle's three-axis acceleration and angular velocity are derived from the vehicle IMU data.

[0010] By introducing the vehicle-cargo rigid-flexible coupling dynamic model and taking the cargo pressure distribution into consideration, the instability index of the vehicle-cargo system can be accurately calculated.

[0011] Furthermore, the step of predicting the risk potential field value of each section along the transport route in a future time window includes: dividing the transport route into continuous sections according to a predetermined length as nodes, and constructing a graph structure based on the physical topological relationship between the sections; constructing a feature vector for each node containing real-time traffic flow, meteorological data, road geometry parameters and historical accident frequency; using a spatiotemporal graph convolutional network containing a graph convolution layer and a recurrent neural network layer for processing, and outputting the risk potential field value of each section in the future predetermined time window.

[0012] By integrating real-time traffic flow, meteorological data, road geometry parameters, historical accident frequency and other data, the foresight of risk potential field prediction for future time periods is increased.

[0013] Furthermore, the step of using a predefined hierarchical Bayesian network for probabilistic reasoning includes: constructing a Bayesian network with driving state deviation, vehicle-cargo system instability index, and risk potential field value as parent nodes, and comprehensive transportation risk as child nodes; according to the statistical distribution characteristics of historical data, setting a threshold value to discretize the continuous value input of each parent node into multiple state levels; generating a conditional probability table based on historical statistical data, and reasoning on the multiple state levels after the discretization of the input to obtain the comprehensive transportation risk probability.

[0014] Furthermore, the step of triggering a risk warning of a corresponding level includes: setting multiple risk levels corresponding to different probability intervals; when the comprehensive transportation risk probability falls into an interval of a certain level, triggering a warning corresponding to that level, and the mode and intensity of the warning increase with the increase of the risk level, and its mode includes one or more combinations of sound prompts, physical warnings, and sending alarm information to the background monitoring center.

[0015] By setting up risk warning registration, data can be easily classified and warnings can be issued.

[0016] Furthermore, the vehicle IMU data includes data collected by a six-axis IMU sensor installed in the vehicle.

[0017] Furthermore, the CAN bus data includes at least a steering wheel angle, an accelerator pedal opening, a brake pedal pressure, and a vehicle speed.

[0018] Furthermore, the vehicle dynamics model is a benchmark vehicle two-degree-of-freedom dynamics model.

[0019] Furthermore, the spatiotemporal graph convolutional network is a spatiotemporal graph convolutional network STGCN model including a graph convolution layer and a one-dimensional convolution or GRU layer.

[0020] In summary, the present invention offers the following beneficial effects: Compared to existing technologies, it constructs a multi-dimensional, coordinated risk assessment system encompassing person, vehicle, cargo, road, and environment, significantly improving the accuracy and foresight of risk warnings. Regarding driver state assessment, this invention transcends the limitations of traditional fatigue detection. By integrating heart rate variability (HRV) data with vehicle micro-operation data, it accurately quantifies the deviation between driving intent and vehicle feedback, providing deep insight into the driver's cognitive and operational state. Regarding vehicle stability, it introduces a vehicle-cargo rigid-flexible coupling dynamic model, incorporating cargo state (such as cargo pressure distribution) to accurately calculate the vehicle-cargo system instability index, effectively filling the gap in existing technologies that ignore the impact of cargo on driving safety. Regarding external environmental risks, by integrating real-time traffic flow, meteorological data, road geometry, and historical accident frequencies, it proactively predicts the risk potential field along the transport route for future periods. Finally, through a hierarchical Bayesian network, these multi-source, heterogeneous risk factors are probabilistically inferred to output a comprehensive risk probability, providing a more comprehensive and reliable basis for early warning decisions and effectively ensuring safety throughout the entire transport process. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a conceptual diagram schematically illustrating a method for real-time early warning and assessment of vehicle transportation process risks based on the Internet of Things according to an embodiment of the present invention; Figure 2 Schematic diagram of reasoning of driving state deviation according to an embodiment of the present invention; Figure 3 The flowchart schematically illustrates a method for real-time early warning and assessment of vehicle transportation process risks based on the Internet of Things according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] like Figure 3As shown, a real-time early warning and assessment method for risks in vehicle transportation based on the Internet of Things includes the following steps: S1: Real-time acquisition of vehicle IMU data representing vehicle operating status, Controller Area Network (CAN) bus data, heart rate variability (HRV) data, cargo pressure distribution data, and external environment and road network data. The HRV data represents driver physiological status data, while the cargo pressure distribution data represents cargo status sensor data.

[0024] Specifically, the vehicle's three-axis acceleration and three-axis angular velocity are collected through a six-axis IMU sensor installed in the vehicle. IMU stands for On-Board Inertial Measurement Unit. A CAN acquisition card is connected to the vehicle's OBD interface to read and analyze CAN bus data such as steering wheel angle, accelerator pedal opening, brake pedal pressure, and vehicle speed at a frequency of, for example, 100 Hz, i.e., vehicle operation data. The driver wears a smart bracelet with an integrated photoplethysmography sensor to monitor and calculate the beat-to-beat heart rate interval in real time, and can further calculate heart rate variability (HRV) data, which refers to the tiny differences in the time intervals between consecutive heartbeats. By laying a flexible thin film pressure sensor array on the bottom of the cargo compartment, the pressure distribution map generated by the cargo during vehicle movement can be obtained in real time. By calling, for example, the AutoNavi Map API and the China Weather Network API, the real-time road conditions, traffic flow density, future weather forecasts, road curvature, slope, and historical accident point data at the vehicle's current location, i.e., external environment and road network data, can be obtained.

[0025] like Figure 1 As shown, based on the technical problem addressed by the present invention - a comprehensive assessment model that deeply integrates the driver's physiological and psychological state, the dynamic characteristics of vehicle-cargo coupling, and the time-varying road environment risks - the main challenge faced by the present invention is how to integrate information such as "the driver's physiological and psychological state, the dynamic characteristics of vehicle-cargo coupling, and the time-varying road environment" into the assessment model. The concept of the present invention is: The "human characteristics" that represent the relationship between people and cars are converted into driving state deviations to represent the deviation between people's driving intentions and vehicle feedback. The so-called "human characteristics" not only reflect the characteristics of people, but also include the interaction with the car, thus reflecting the "characteristics of people's driving state", which can better mine and utilize information for subsequent reasoning.

[0026] The "vehicle characteristics" representing the relationship between the vehicle and the cargo are converted into a vehicle system instability index, which is used to maximize the coupling relationship between the vehicle and the cargo to facilitate subsequent reasoning.

[0027] The "environmental characteristics" representing the road environment are converted into risk potential field values. The risk potential field value can integrate various types of multimodal information, such as the correlation between various roads and weather, and the correlation with historical information, so as to better mine and utilize information for subsequent reasoning.

[0028] Based on this, the present invention focuses on the specific implementation of the above three features, which are described below through steps S2 to S4. It should be noted that S2, S3 and S4 are parallel steps, so the order of S2 to S4 does not mean that they must be performed in the same order.

[0029] S2: Based on the heart rate variability (HRV) data and the driver's micro-operation sequence extracted from the CAN bus data, a time-varying Kalman filter model is used to evaluate and quantify the driving state deviation that represents the deviation between the driver's manipulation intention and the vehicle dynamic feedback.

[0030] Specifically, micro-operation features of the driver, such as the steering wheel angle change rate, throttle and brake pedal switching frequency, within a 2-second time window, are extracted from the CAN bus data to form an observation vector; the driver's ideal control input and the vehicle's expected response state, such as yaw angular velocity, are used as the system state vector; a time-varying Kalman filter model is constructed, in which the state transfer matrix is ​​established based on the vehicle's two-degree-of-freedom dynamics model, and the observation matrix associates the system state with the driver's micro-operation; the real-time HRV index is used as a regulating factor for the model noise covariance. For example, when the HRV index decreases, indicating an increase in the driver's cognitive load, the process noise covariance is increased; the model continuously predicts the micro-operation observation value at the next moment, and the Mahalanobis distance between the predicted value and the actual observation value is used as the driving state deviation. The larger the distance, the greater the deviation between the driver's actual operation and his intention and the vehicle's expected feedback.

[0031] In an optional embodiment, the step of evaluating and quantifying the driving state deviation includes: A vehicle dynamics model is constructed using the driver's steering wheel angle as input, and the model parameters are corrected in real time in combination with the heart rate variability (HRV) data. The vehicle yaw rate and center of mass sideslip angle obtained from the vehicle IMU data are used as actual observation values. A time-varying Kalman filter model is used, and the Mahalanobis distance between the vehicle state predicted by the time-varying Kalman filter model and the actual observed state is calculated as the driving state deviation.

[0032] Specifically, a baseline two-degree-of-freedom vehicle dynamics model (also known as a bicycle model or slip model) can be established. This vehicle dynamics model receives the steering wheel angle applied by the driver while awake as input and predicts the theoretical vehicle yaw rate and slip angle based on this input. This predicted value represents the dynamic response of the vehicle under ideal driving conditions. Simultaneously, the driver's real-time heart rate variability (HRV) data collected by a wearable device is used to dynamically adjust the parameters of the vehicle dynamics model. For example, if the HRV value is detected to drop from the normal 50 milliseconds to 25 milliseconds, indicating that the driver may be fatigued or nervous, the driver's reaction delay time parameter in the model will be increased by 0.1 second, making the model prediction more closely aligned with the driver's current physiological state.

[0033] like Figure 2 As shown, the real-time Kalman prediction model, whose core is the aforementioned vehicle dynamics model, can convert input driver micro-operations (such as steering wheel angle) into outputs of the vehicle's yaw rate and center-of-mass slip angle. The real-time Kalman prediction model can further fine-tune the parameters of the vehicle dynamics model based on real-time heart rate variability (HRV) data. These parameters include the front axle equivalent tire cornering stiffness and the rear axle equivalent tire cornering stiffness. This is based on the fact that real-time heart rate variability (HRV) data reflects the driver's driving state. If the driver's driving state is poor, resulting in slow reactions or inadequate operation, this can be equivalently translated into changes in the front axle equivalent tire cornering stiffness and the rear axle equivalent tire cornering stiffness. Ultimately, the real-time Kalman prediction model outputs the vehicle's yaw rate and center-of-mass slip angle, reflecting the driver's current physiological state.

[0034] Specifically, to quantify the degree of deviation, the predicted yaw rate and slip angle output by the HRV-corrected vehicle dynamics model are used as the predicted state. The yaw rate and estimated slip angle measured in real time by the onboard inertial measurement unit (IMU) are used as the actual observed state. Ultimately, the driving state deviation is calculated by calculating the Mahalanobis distance between these two state vectors. The Mahalanobis distance comprehensively considers the variance and covariance of each state variable and is a dimensionless statistical distance (a measure of the similarity between two data points. It considers the correlation and scale differences between features in a dataset and is more statistically reasonable than the Euclidean distance). For example, a Mahalanobis distance value greater than a preset threshold, such as 5, clearly indicates that the driver's control behavior has significantly deviated from the normal range.

[0035] S3: Inputting the vehicle IMU data and cargo pressure distribution data into a preset vehicle-cargo rigid-flexible coupling dynamic model, and calculating in real time the vehicle-cargo system instability index that characterizes the vehicle under the current motion disturbance.

[0036] Specifically, a multi-body dynamic model of the entire trailer or truck is established in ADAMS software in advance, including rigid components such as the frame, suspension, and tires; the finite element method is used to establish its flexible body dynamic model according to the type of cargo, such as liquid or bulk cargo; the real-time collected cargo pressure distribution data is used as the load boundary condition and applied to the flexible body dynamic model to simulate the real-time changes in the cargo shape and center of gravity; the real-time acceleration and angular velocity of the vehicle collected by the IMU are used as the motion excitation of the entire coupling model; the dynamic equations of the rigid-flexible coupling system are solved through joint simulation, and the vehicle's lateral load transfer rate LTR is calculated in real time. This value is the instability index of the vehicle-cargo system. The closer the value is to 1, the higher the risk of vehicle rollover.

[0037] In an optional embodiment, the vehicle-cargo rigid-flexible coupling dynamic model includes: Establish a 14-degree-of-freedom rigid body dynamic model of the vehicle (the 14 degrees of freedom usually include: 6 degrees of freedom of the vehicle chassis: translation along the X, Y, and Z axes (longitudinal, lateral, and vertical) and rotation around the X, Y, and Z axes (roll, pitch, and yaw). 1 vertical degree of freedom for each of the four wheels: describes the vertical jump of each wheel relative to the vehicle body. 1 rotational degree of freedom for each of the four wheels: describes the rotation of each wheel).

[0038] The cargo system is equated to a multi-flexible body element (ANCF) based on the absolute nodal coordinate method and coupled to the full vehicle rigid body model at the suspension system. This is a key innovation, as it takes cargo deformation into account and simulates the interaction between the cargo and road excitations through the vehicle's suspension system during real-world driving. Equating the cargo system to a multi-flexible body element means that the cargo is no longer treated as a simple rigid load, but rather as a deformable entity. The absolute nodal coordinate method is an advanced flexible body dynamics modeling method, particularly suitable for handling large deformations and complex contact problems. The absolute coordinates of the nodes of the ANCF element during motion effectively avoid the limitations of traditional finite element methods in handling large rotations and simplify the solution of the dynamic equations. This means that the cargo system can be decomposed into multiple ANCF elements to simulate its complex flexible behavior, rather than a single flexible body. Coupling to the full vehicle rigid body model at the suspension system means that cargo motion and deformation are transmitted to the vehicle through the suspension system, and vice versa.

[0039] The vehicle's three-axis acceleration and angular velocity measured by the IMU are used as the excitation input of the vehicle-cargo rigid-flexible coupling dynamic model. The cargo pressure distribution data is converted into the normal force on the contact surface between the cargo and the car body. The real-time vertical loads on the left and right tires of the vehicle are calculated through the model solution, and the instability index of the vehicle-cargo system is calculated based on the lateral load transfer rate formula.

[0040] Specifically, the vehicle-cargo rigid-flexible coupling dynamic model abstracts the entire vehicle into a 14-degree-of-freedom rigid body system that includes the longitudinal, lateral, and vertical motions of the vehicle body, as well as roll, pitch, and sway motions, and independent rotation and vertical jump of the four wheels. This can accurately describe the macroscopic posture changes of the vehicle during driving. For the cargo in the compartment, especially non-solid materials such as liquids or bulk particles, this method does not regard it as a rigid body, but instead discretizes it into multiple interconnected flexible units using the absolute node coordinate method. This method can accurately simulate the swaying, accumulation, and deformation of the cargo when the vehicle turns or accelerates or decelerates, and transmits the additional forces and torques generated by the swaying of the cargo to the rigid body model of the entire vehicle through the suspension system, realizing the dynamic mechanical coupling between the vehicle and the cargo.

[0041] Specifically, the operation of the vehicle-cargo rigid-flexible coupling dynamics model requires real-time data input. The vehicle's IMU data, including triaxial acceleration and angular velocity data, such as a lateral acceleration of 4 meters per second squared, serves as the dynamic excitation for the entire vehicle-cargo rigid-flexible coupling dynamics model. Simultaneously, pressure sensor arrays mounted on the vehicle floor and sidewalls collect cargo pressure distribution data, such as significantly higher pressure on the left side than on the right side. This data is converted into normal forces acting on the contact surface and serves as boundary conditions for the cargo-flexible system. The vehicle-cargo rigid-flexible coupling dynamics model is numerically solved at each time step, outputting key vehicle state parameters, particularly the real-time vertical loads on the left and right tires. For example, during an emergency evasive maneuver, the vehicle-cargo rigid-flexible coupling dynamics model calculated vertical loads of 8,000 Newtons on the left tire and 18,000 Newtons on the right. Subsequently, the instability index, calculated using the lateral load transfer rate formula, reached 0.8, indicating that the vehicle was nearing a critical rollover state.

[0042] S4: Utilizing a spatiotemporal graph convolutional network, this method fuses dynamic attributes (real-time traffic flow and meteorological data) with static attributes (road geometry and historical accident data) to predict the risk potential field value for each road segment along the transport route within a future time window. The real-time traffic flow, meteorological data, road geometry, and historical accident data are derived from the external environment and road network data. Real-time traffic flow refers to the actual operating conditions of vehicles on the road, including volume, speed, density, and occupancy, within the current or recent period. Risk potential field value is a concept used in the fields of intelligent driving, environmental perception, and risk assessment to quantify and spatially represent potential hazards and uncertainties in the environment.

[0043] Specifically, the planned transport route of the vehicle is first divided into continuous sections of 500 meters in length, and these sections are constructed into a topological graph, with sections as nodes and the connection relationships between sections as edges; the feature vector of each node contains static attributes such as road curvature, slope, number of lanes, historical accident frequency, and dynamic attributes such as real-time traffic congestion index, average speed, rainfall, and visibility; the node feature sequence every five minutes in the past hour is used as input and sent to a spatiotemporal graph convolutional network STGCN model containing a graph convolution layer and a one-dimensional convolution or GRU layer; the spatiotemporal graph convolutional network STGCN model aggregates the spatial risk information of adjacent sections through graph convolution, and learns the law of risk evolution over time through the temporal convolution layer; the spatiotemporal graph convolutional network STGCN model finally outputs the risk prediction value of each section node in the next 15 minutes, and the prediction value sequence of all sections together constitutes the risk potential field along the transportation route.

[0044] In an optional embodiment, the step of predicting the risk potential field value of each section along the transport route in a future time window includes: The transport route is divided into continuous sections of predetermined length as nodes in the graph network, and a graph structure is constructed based on the physical topological relationship between the sections. A feature vector containing real-time traffic flow, meteorological data, road geometry parameters and historical accident frequencies is constructed for each node. The constructed graph data (including graph structure and node feature vectors) is input into a spatiotemporal graph convolutional network containing a graph convolution layer and a recurrent neural network layer, and the risk potential field value of each section within a predetermined future time window is output.

[0045] Specifically, a transport route of hundreds of kilometers is first divided into sections of 500 meters each, and each section becomes a node in the graph network. Subsequently, connections between nodes are established based on the geographical connections between these sections, forming a chain or more complex graph structure. For each section node, the system aggregates multiple data sources to form a multi-dimensional feature vector. For example, for a section of the Shanghai-Nanjing Expressway, its feature vector at 10 a.m. may include real-time traffic flow of 1,500 vehicles per hour, light rain, a road curvature radius of 800 meters, and an average monthly accident number of 0.5 times in the past three years.

[0046] Specifically, the constructed spatiotemporal graph data is input into a deep learning model that can process spatial and temporal features. For example, in existing technologies, the core of a deep learning model is a spatiotemporal graph convolutional network, which contains two key parts: the graph convolution layer is responsible for capturing spatial interactions. For example, traffic congestion on upstream sections of a road will increase the risk of downstream sections. The graph convolution operation can learn and quantify this spatial dependency. The recurrent neural network layer, such as the long short-term memory network, is responsible for processing the changing patterns in the time series. For example, it can learn the periodic fluctuations in risk values ​​during peak hours in the morning and evening. After being trained with a large amount of historical data, the deep learning model can predict the risk potential field value of each road node in a preset time window in the future, such as the next 30 minutes, and output a value between 0 and 10. The higher the value, the greater the comprehensive risk of the road section in the future time period.

[0047] In the above embodiments, the calculation of the risk potential field is based on a combination of multiple factors; in other embodiments, it can also be implemented in a simpler way, such as removing some factors.

[0048] S5: Using a predefined hierarchical Bayesian network, the driving state deviation, the vehicle-cargo system instability index, and the risk potential field value corresponding to the vehicle's current position are used as evidence input to perform probabilistic reasoning and output the comprehensive transportation risk probability; when the comprehensive transportation risk probability exceeds the preset classification threshold, a risk warning of the corresponding level is triggered.

[0049] Specifically, a three-layer Bayesian network is constructed, in which the top-level root node is the comprehensive transportation risk, and its status is divided into three levels: low, medium, and high; the second layer is the parent node, including driver risk, vehicle-cargo system risk, and road environment risk; the third layer is the leaf node, which corresponds to three continuous input evidences: driving state deviation, vehicle-cargo system instability index, and risk potential field value; based on historical data and expert knowledge, the conditional probability table between each node is pre-set, such as P(driver risk|driving state deviation); while the vehicle is driving, the three indicator values ​​calculated by S2, S3, and S4 are input as evidence to the corresponding leaf nodes; the belief propagation algorithm is used for probabilistic reasoning to calculate the posterior probability that the comprehensive transportation risk of the root node is at a high level; the first-level warning threshold is set to 0.6, and the second-level warning threshold is set to 0.8. When the calculated comprehensive transportation risk probability is 0.7, it exceeds the first-level threshold, and the system issues a second-level risk warning to the driver through a pop-up window on the vehicle screen and voice broadcast.

[0050] In an optional embodiment, the step of performing probabilistic reasoning using a predefined hierarchical Bayesian network includes: A Bayesian network is constructed with the driving state deviation, vehicle-cargo system instability index, and risk potential field value as parent nodes, and the comprehensive transportation risk as child nodes. According to the statistical distribution characteristics of historical data, a threshold is set to discretize the continuous value input of each parent node into multiple state levels. A conditional probability table is generated based on historical statistical data, and the discretized evidence is input for reasoning to obtain the comprehensive transportation risk probability.

[0051] Specifically, this method establishes an intuitive probabilistic graphical model. The network structure clearly defines three independent risk sources as parent nodes, namely the driving state deviation, the vehicle-cargo system instability index, and the risk potential field value. These three parent nodes collectively point to a child node, namely the comprehensive transportation risk. Before reasoning, the continuous numerical inputs obtained from the previous steps need to be discretized. For example, the Mahalanobis distance value of the driving state deviation can be divided into three levels: normal, slight deviation, and severe deviation; the vehicle-cargo system instability index can be divided into three levels: stable, critical, and unstable according to its numerical value; and the risk potential field value can also be divided into three levels: low, medium, and high.

[0052] Specifically, the core of the hierarchical Bayesian network lies in its conditional probability table, derived through statistical analysis of massive amounts of historical transportation data. This table stores the probabilities of child nodes taking different states under all combinations of parent node states. For example, the conditional probability table might contain a record stating that when the driving state is severely deviated, the vehicle-cargo system is critical, and the environmental risk is high, the probability of the overall transportation risk being dangerous is 95%. When the system receives real-time evidence, such as a severely deviated driving state, a stable vehicle-cargo system, and a moderate environmental risk, the Bayesian network applies a probabilistic inference algorithm, combining the entire conditional probability table, to calculate the posterior probability distribution of the overall transportation risk at various levels, such as safe, warning, and dangerous. The final output might be a probability vector, such as a 70% probability of danger.

[0053] In an optional embodiment, the step of triggering a risk warning of a corresponding level includes: Set multiple risk levels corresponding to different probability intervals; when the comprehensive transportation risk probability falls into a certain level, the warning corresponding to the level is triggered. The mode and intensity of the warning increase with the increase of the risk level. Its mode includes one or more combinations of sound prompts, physical warnings, and sending alarm information to the background monitoring center.

[0054] Specifically, the system predefines a hierarchical response mechanism. For example, the probability of a dangerous state for comprehensive transport risk is divided into three intervals corresponding to different risk levels. When the calculated probability of danger is less than 10%, it is defined as a Level 1 risk or safe state, and no active warning is triggered. When the probability of danger is between 10% and 50%, it is defined as a Level 2 risk or warning state. When the probability of danger exceeds 50%, it is defined as a Level 3 risk or dangerous state. These probability thresholds can be flexibly configured based on the importance of the transport mission, the value of the cargo, or regulatory requirements.

[0055] Specifically, once the comprehensive transport risk probability falls into a preset range, the system will automatically trigger an early warning of the corresponding level. For level 2 risks, the system may issue a milder prompt, such as a yellow warning icon lighting up on the dashboard, accompanied by a short beep, to alert the driver. For the highest level 3 risk, early warning measures will be significantly enhanced to ensure that the driver can respond immediately. At this time, the system may issue a continuous and sharp alarm sound, and at the same time activate the vibration function of the driver's seat or steering wheel to produce a strong physical warning. While issuing an alert to the driver, the system will also immediately automatically send the alarm information containing the vehicle's location, speed, risk level and specific cause to the company's background monitoring center via the wireless network, so that management personnel can conduct remote monitoring and necessary intervention support.

[0056] Those skilled in the art will come up with many changes, modifications and substitutions without departing from the spirit and concept of the present invention. It should be understood that in practicing the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A real-time early warning and assessment method for risks in vehicle transportation based on the Internet of Things, characterized in that: The following steps are involved: S1: Real-time acquisition of vehicle IMU data, controller area network (CAN) bus data, heart rate variability (HRV) data, cargo pressure distribution data, and external environment and road network data; S2: Based on the heart rate variability (HRV) data and the driver's micro-operation sequence extracted from the CAN bus data, a time-varying Kalman filter model is used to evaluate and quantify a driving state deviation representing a deviation between the driver's manipulation intention and vehicle dynamic feedback; S3: Inputting the vehicle IMU data and cargo pressure distribution data into a preset vehicle-cargo rigid-flexible coupling dynamics model to calculate in real time the vehicle-cargo system instability index representing the vehicle under the current motion disturbance; S4: Using a spatiotemporal graph convolutional network, real-time traffic flow and weather data as dynamic attributes are integrated with road geometry and historical accident data as static attributes to predict the risk potential field value of each section along the transport route in a future time window; wherein the real-time traffic flow, weather data, road geometry, and historical accident data are derived from the external environment and road network data; S5: Using a predefined hierarchical Bayesian network, the driving state deviation, the vehicle-cargo system instability index, and the risk potential field value corresponding to the vehicle's current position are used as evidence input to perform probabilistic reasoning and output a comprehensive transportation risk probability; When the comprehensive transportation risk probability exceeds a preset classification threshold, a risk warning of the corresponding level is triggered.

2. The method according to claim 1, characterized in that The step of evaluating and quantifying the driving state deviation comprises: A vehicle dynamics model is constructed using the driver's steering wheel angle as input, and the vehicle dynamics model parameters are corrected in real time in combination with the heart rate variability (HRV) data; the vehicle yaw rate and center of mass sideslip angle obtained from the vehicle IMU data are used as actual observation values; and a time-varying Kalman filter model is used to calculate the Mahalanobis distance between the vehicle state predicted by the time-varying Kalman filter model and the actual observed state as the driving state deviation.

3. The method according to claim 1, characterized in that The vehicle-cargo rigid-flexible coupling dynamics model includes: A 14-degree-of-freedom rigid-body dynamics model of the vehicle is established. The cargo system is equivalent to a multi-flexible body unit based on the absolute node coordinate method and coupled to the vehicle rigid-body dynamics model at the suspension system. The vehicle's three-axis acceleration and angular velocity are used as model excitation inputs, and the cargo pressure distribution data is converted into the normal force on the cargo-carriage contact surface. The real-time vertical loads on the vehicle's left and right tires are calculated using the vehicle rigid-body dynamics model, and the vehicle-cargo system instability index is calculated based on the lateral load transfer rate formula. The vehicle's three-axis acceleration and angular velocity are derived from the vehicle's IMU data.

4. The method according to claim 1, wherein The step of predicting the risk potential field value of each section along the transport route in the future time window includes: The transport route is divided into continuous sections of predetermined length as nodes, and a graph structure is constructed based on the physical topological relationship between the sections. A feature vector containing real-time traffic flow, meteorological data, road geometry parameters and historical accident frequencies is constructed for each node. The risk potential field value of each section within a predetermined future time window is output using a spatiotemporal graph convolutional network containing a graph convolution layer and a recurrent neural network layer.

5. The method according to claim 1, wherein The steps of performing probabilistic reasoning using a predefined hierarchical Bayesian network include: A Bayesian network is constructed with the driving state deviation, vehicle-cargo system instability index, and risk potential field value as parent nodes, and comprehensive transportation risk as child nodes. According to the statistical distribution characteristics of historical data, a threshold is set to discretize the continuous value input of each parent node into multiple state levels. A conditional probability table is generated based on historical statistical data, and the multiple state levels after discretization of the input are inferred to obtain the comprehensive transportation risk probability.

6. The method according to claim 1, wherein The steps of triggering a risk warning of a corresponding level include: Set multiple risk levels corresponding to different probability intervals; when the comprehensive transportation risk probability falls into a certain level, the warning corresponding to the level is triggered. The mode and intensity of the warning increase with the increase of the risk level. Its mode includes one or more combinations of sound prompts, physical warnings, and sending alarm information to the background monitoring center.

7. The method according to claim 1, characterized in that Vehicle IMU data includes data collected by the six-axis IMU sensor installed in the vehicle.

8. The method according to claim 1, characterized in that The CAN bus data at least includes steering wheel angle, accelerator pedal opening, brake pedal pressure and vehicle speed.

9. The method according to claim 2, characterized in that The vehicle dynamics model is a benchmark vehicle two-degree-of-freedom dynamics model.

10. The method according to claim 4, characterized in that The spatiotemporal graph convolutional network is a spatiotemporal graph convolutional network STGCN model including a graph convolution layer and a one-dimensional convolution or GRU layer.

Images