Vehicle rollover early warning system and method based on state prediction and working condition clustering

Abstract

The invention provides a vehicle rollover early warning system and method based on state prediction and working condition clustering, and the system comprises a state prediction module which is used for predicting the motion state of an electric vehicle through a neural network; the state prediction module is used for determining the current working condition of the electric vehicle through clustering; the model updating module is used for updating a vehicle roll dynamics model of the electric vehicle in real time based on the predicted motion state and the determined current working condition; and the rollover early warning module is used for analyzing and early warning the rollover risk of the vehicle based on the vehicle rollover dynamic model updated in real time. Through four steps of state prediction, working condition clustering, model updating and risk early warning, real-time monitoring and early warning of the rollover risk of the electric vehicle are realized. Model parameters can be adjusted according to real-time states and working conditions, and the early warning accuracy is improved. The rollover risk can be predicted and a driver can be reminded in time, so that the driving safety is improved. And an innovative solution is provided for the safety of the intelligent electric vehicle.

Description

Technical Field

[0001] The present invention relates to the technical field of electric vehicle control, and in particular to a vehicle rollover warning system and method based on state prediction and working condition clustering. Background Art

[0002] Vehicle rollover accidents are a major issue in traffic safety, particularly in complex driving scenarios such as high-speed turns and emergency avoidance maneuvers. Therefore, developing an efficient and accurate vehicle rollover warning system is an urgent need to improve driving safety.

[0003] Existing vehicle rollover warning systems are mostly based on traditional dynamic models, incorporating sensors that monitor parameters such as lateral acceleration and roll angle in real time to determine rollover risk. However, these systems have significant limitations. For one thing, fixed model parameters are difficult to adapt to changing operating conditions and vehicle states, resulting in insufficient warning accuracy. Furthermore, in dynamic and complex driving environments, the systems' real-time performance is difficult to guarantee, limiting their effectiveness in practical applications.

[0004] In response to the above problems, the present invention proposes a vehicle rollover warning system and method based on state prediction and working condition clustering. Summary of the Invention

[0005] One of the purposes of the present invention is to provide a vehicle rollover warning system and method based on state prediction and working condition clustering to solve the problems pointed out in the background technology.

[0006] In a first aspect, an embodiment of the present invention provides a vehicle rollover warning system based on state prediction and operating condition clustering, comprising:

[0007] A state prediction module is used to predict the motion state of electric vehicles through a neural network;

[0008] An operating condition clustering module, used to determine the current operating condition of the electric vehicle through clustering;

[0009] A model update module for updating a vehicle roll dynamics model of an electric vehicle in real time based on the predicted motion state and the determined current operating conditions;

[0010] The rollover warning module is used to analyze the vehicle rollover risk and issue a vehicle rollover risk warning based on the real-time updated vehicle roll dynamics model.

[0011] Optionally, the state prediction module predicts the motion state of the electric vehicle through a neural network, including:

[0012] Use a pre-trained recurrent neural network as the neural network;

[0013] Taking vehicle parameters of the electric vehicle, including at least vehicle wheel speed, steering wheel angle, yaw rate, longitudinal acceleration, lateral acceleration, body roll angle, suspension travel sensor data, and motor torque / speed, as input, and inputting them into a neural network;

[0014] The vehicle state of the electric vehicle outputted by the neural network at least includes the predicted vehicle yaw rate, the predicted vehicle lateral acceleration, the predicted vehicle body roll angle and the predicted tire vertical load as the predicted motion state.

[0015] Optionally, the operating condition clustering module determines the current operating condition of the electric vehicle by clustering, including:

[0016] constructing a clustering feature vector set based on vehicle parameters of the electric vehicle including at least a current vehicle speed, a steering wheel angle, a yaw rate, a longitudinal acceleration, a lateral acceleration, a body roll angle, a tire slip rate, and a motor output torque of the electric vehicle;

[0017] Based on the K-means algorithm, the working conditions are clustered according to the clustering feature vector set to obtain the determined current working conditions;

[0018] The categories of current operating conditions include at least: straight-line driving, steady-state steering, transient steering, acceleration, braking, driving conditions under different road adhesion coefficients, and different road types.

[0019] Optionally, the model updating module updates the vehicle roll dynamics model of the electric vehicle in real time based on the predicted motion state and the determined current working condition, including:

[0020] Adjusting corresponding parameters or coefficients in the vehicle roll dynamics model based on the predicted motion state and the determined current operating conditions;

[0021] The parameters or coefficients include at least: roll stiffness, roll damping, roll moment of inertia, and tire cornering stiffness.

[0022] Optionally, the rollover warning module analyzes the vehicle rollover risk based on a real-time updated vehicle roll dynamics model, including:

[0023] For each preset vehicle rollover risk index, calculate the index value of the vehicle rollover dynamics model updated in real time under the vehicle rollover risk index, compare it with the corresponding preset index threshold, and determine the vehicle rollover risk based on the comparison result;

[0024] Among them, the vehicle rollover risk indicators include at least: roll angle, load transfer rate and rollover coefficient.

[0025] Optionally, the vehicle rollover risk warning includes:

[0026] Step a: Divide the safety critical time window into at least two consecutive decision nodes, and repeat steps b to f in each consecutive decision node;

[0027] Step b: collecting cognitive load assessment evidence of the electric vehicle driver in real time;

[0028] Step c: Based on the cognitive load assessment basis, determine the driver's current cognitive load and match it with a corresponding solution reception sensitivity level; wherein the solution reception sensitivity level is negatively correlated with the driver's acceptable maximum assistance intervention intensity;

[0029] Step d: Based on the driver's current scenario reception sensitivity level, dynamically select an assistance execution scenario of a corresponding level from a preset rollover prevention assistance strategy library; wherein the rollover prevention assistance strategy library includes multiple levels of assistance execution scenarios, from an active control layer to a passive prompt layer, and the level of each level of assistance execution scenario is positively correlated with the corresponding scenario reception sensitivity level, and the assistance intervention intensity of each level of assistance execution scenario decreases as the level increases;

[0030] Step e: executing the dynamically selected auxiliary execution plan for the driver;

[0031] Step f: when it is detected that the similarity between the active operation behavior vector of the driver operating the electric vehicle and the auxiliary target vector of the auxiliary execution plan executed on the driver exceeds a threshold, stop executing the corresponding auxiliary execution plan for the driver.

[0032] Optionally, the step of obtaining the safety critical time window includes:

[0033] The time window from the time when the vehicle rollover risk is analyzed and determined to the future target time is regarded as the safety critical time window;

[0034] The future target time is the smaller value between the physical rollover prevention limit time and the driver's cognitive buffer time;

[0035] The physical rollover prevention limit time includes: the conservative time of irreversible rollover of the electric vehicle in the future determined based on the real-time updated vehicle roll dynamics model;

[0036] The cognitive buffer time includes the maximum allowable time for the driver to complete the recognition of the vehicle rollover risk and the operational response, which is determined based on the driver's cognitive load when analyzing and determining the vehicle rollover risk.

[0037] Optionally, the cognitive load assessment is based on: the driver's eye gaze trajectory, changes in steering wheel grip pressure, and a history of operational response delays within a previously preset time period.

[0038] In a second aspect, an embodiment of the present invention provides a vehicle rollover warning method based on state prediction and operating condition clustering, including:

[0039] Predicting the motion state of electric vehicles through neural networks;

[0040] Determine the current operating condition of the electric vehicle through clustering;

[0041] Based on the predicted motion state and the determined current operating conditions, the vehicle roll dynamics model of the electric vehicle is updated in real time;

[0042] Based on the real-time updated vehicle roll dynamics model, the vehicle rollover risk is analyzed and a vehicle rollover risk warning is issued.

[0043] Optionally, the vehicle rollover risk warning includes:

[0044] Step a: Divide the safety critical time window into at least two consecutive decision nodes, and repeat steps b to f in each consecutive decision node;

[0045] Step b: collecting cognitive load assessment evidence of the electric vehicle driver in real time;

[0046] Step c: Based on the cognitive load assessment basis, determine the driver's current cognitive load and match it with a corresponding solution reception sensitivity level; wherein the solution reception sensitivity level is negatively correlated with the driver's acceptable maximum assistance intervention intensity;

[0047] Step d: Based on the driver's current scenario reception sensitivity level, dynamically select an assistance execution scenario of a corresponding level from a preset rollover prevention assistance strategy library; wherein the rollover prevention assistance strategy library includes multiple levels of assistance execution scenarios, from an active control layer to a passive prompt layer, and the level of each level of assistance execution scenario is positively correlated with the corresponding scenario reception sensitivity level, and the assistance intervention intensity of each level of assistance execution scenario decreases as the level increases;

[0048] Step e: executing the dynamically selected auxiliary execution plan for the driver;

[0049] Step f: when it is detected that the similarity between the active operation behavior vector of the driver operating the electric vehicle and the auxiliary target vector of the auxiliary execution plan executed on the driver exceeds a threshold, stop executing the corresponding auxiliary execution plan for the driver.

[0050] The present invention has achieved the following beneficial effects:

[0051] The system uses a neural network to predict the electric vehicle's motion state and clustering technology to identify the current operating conditions, thereby updating the vehicle's roll dynamics model in real time. This dynamically adaptive design significantly improves the accuracy and real-time nature of warnings, providing an innovative solution for the safety of smart electric vehicles. In practical applications, such as high-speed cornering scenarios, the system can predict rollover risks in advance and promptly alert the driver, effectively reducing the probability of accidents.

[0052] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0053] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0055] Figure 1 Schematic diagram of a vehicle rollover warning system based on state prediction and working condition clustering in an embodiment of the present invention;

[0056] Figure 2 Schematic diagram of execution steps of each module in a vehicle rollover warning system based on state prediction and working condition clustering in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0058] The research and development approach of this application is to more accurately assess and warn of electric vehicle rollover risks by combining neural network state prediction with operating condition clustering technology to update the vehicle roll dynamics model in real time. This approach can dynamically adapt to different driving conditions and vehicle states, improving the accuracy and timeliness of warnings and providing technical support for intelligent driving safety.

[0059] Figure 1 A schematic diagram of a vehicle rollover warning system based on state prediction and working condition clustering provided in an embodiment of the present application is shown in FIG. Figures 1 to 2 As shown, the system includes the following four main modules:

[0060] The state prediction module 100 is used to execute step 101 and predict the motion state of the electric vehicle through a neural network. Step 101 includes:

[0061] 201. Use a pre-trained recurrent neural network as the neural network.

[0062] In this step, a pre-trained recurrent neural network (RNN) is used as the core prediction model. RNNs are capable of processing time series data and, through their recurrent structure, capture the temporal dependencies of vehicle parameters. They are suitable for predicting future dynamic states based on historical and current vehicle parameters. The RNN is pre-trained using historical data collected by electric vehicle sensors as training data, with the goal of minimizing the error between the predicted and actual states.

[0063] 202. Vehicle parameters of an electric vehicle, including at least vehicle wheel speed, steering wheel angle, yaw rate, longitudinal acceleration, lateral acceleration, body roll angle, suspension travel sensor data, and motor torque / speed, are input into a neural network.

[0064] In this step, the above input data is collected in real time through sensors on the vehicle and input into the RNN in time series.

[0065] 203. The vehicle state of the electric vehicle outputted by the neural network includes at least the predicted vehicle yaw rate, the predicted vehicle lateral acceleration, the predicted vehicle body roll angle, and the predicted tire vertical load as the predicted motion state.

[0066] A specific implementation example: Take an electric SUV traveling at high speed. Sensors collect data such as a current wheel speed of 120 km / h, a steering wheel angle of 15°, and a yaw rate of 0.2 rad / s. After receiving these inputs, the RNN predicts the vehicle's motion state over the next one second: a yaw rate of 0.25 rad / s, a lateral acceleration of 0.4 g, a body roll angle of 5°, and a tire vertical load distribution change rate of 10%. This predicted data provides a real-time status basis for subsequent modules.

[0067] The operating condition clustering module 200 is used to execute step 102 and determine the current operating condition of the electric vehicle through clustering. Step 102 includes:

[0068] 301. Construct a clustering feature vector set based on vehicle parameters of the electric vehicle, including at least the current vehicle speed, steering wheel angle, yaw rate, longitudinal acceleration, lateral acceleration, body roll angle, tire slip rate, and motor output torque of the electric vehicle.

[0069] In this step, the above parameters are normalized and combined into a multi-dimensional feature vector (such as [vehicle speed, steering wheel angle, yaw angular velocity, longitudinal acceleration, lateral acceleration, body roll angle, tire slip rate, motor output torque]) to form a feature vector set for clustering.

[0070] 302. Based on the K-means algorithm, cluster the operating conditions according to the clustering feature vector set to obtain a determined current operating condition. The current operating condition categories include at least: straight-line driving, steady-state steering, transient steering, acceleration, braking, driving conditions under different road adhesion coefficients, and different road types.

[0071] In this step, the K-means algorithm is used to cluster the feature vector set. The algorithm classifies similar feature vectors through iterative optimization and finally outputs the current working condition category.

[0072] Continuing with the previous implementation example, the feature vector is [120 km / h, 15°, 0.2 rad / s, 0.1 g, 0.3 g, 4°, 0.05, 500 Nm]. After K-means analysis, this is classified as a "high-speed steady-state turning" condition. This result reflects the dynamic characteristics of the vehicle during high-speed cornering and provides a basis for model updates.

[0073] The model update module 300 is configured to execute step 103 and update the vehicle roll dynamics model of the electric vehicle in real time based on the predicted motion state and the determined current operating conditions. The model update module dynamically adjusts the parameters of the vehicle roll dynamics model based on the predicted motion state and current operating conditions to reflect the real-time vehicle behavior. The roll dynamics model is a mathematical model used to describe the dynamic response of the vehicle under lateral forces and can be pre-built using relevant simulation software. Step 103 includes:

[0074] 401. Based on the predicted motion state and the determined current operating condition, adjust corresponding parameters or coefficients in the vehicle roll dynamics model, wherein the parameters or coefficients include at least: roll stiffness, roll damping, roll moment of inertia, and tire cornering stiffness.

[0075] In this step, a mapping relationship between different predicted motion states or determined current operating conditions and model parameters or coefficients can be pre-established. When making adjustments, the corresponding parameters or coefficients are found based on this mapping relationship and updated and adjusted accordingly. Specific implementation example: Under high-speed steady-state steering conditions, the predicted lateral acceleration is 0.4g. The system increases the roll stiffness (from 5000Nm / rad to 5500Nm / rad) and roll damping (from 1000Ns / m to 1200Ns / m) to simulate the increased rigidity of the vehicle when turning. The updated model more accurately reflects the current state.

[0076] The rollover warning module 400 is used to execute step 104, analyze the vehicle rollover risk based on the real-time updated vehicle roll dynamics model, and issue a vehicle rollover risk warning. Step 104 includes:

[0077] 501. For each preset vehicle rollover risk index, calculate the index value of the vehicle rollover dynamics model, which is updated in real time, under the vehicle rollover risk index, compare it with the corresponding preset index threshold, and determine the vehicle rollover risk based on the comparison result. The vehicle rollover risk index includes at least: roll angle, load transfer rate, and rollover coefficient.

[0078] In this step, you can pre-set different vehicle rollover risk indicators and their corresponding calculation formulas and thresholds. The calculated indicator value is compared with the corresponding threshold. If it exceeds the threshold, the vehicle rollover risk is determined. The roll angle reflects the degree of vehicle tilt, the load transfer rate reflects the tire load distribution during cornering, and the rollover coefficient reflects the comprehensive dynamic stability parameter.

[0079] Specific implementation example: The calculated roll angle is 5° (less than 10°), the load transfer rate is 0.6 (less than 0.8), and the rollover coefficient is 0.7 (less than 1.0). The system determines that there is no rollover risk at the moment. However, if the turn intensifies and the roll angle rises to 12°, an alert is triggered, prompting the driver to slow down.

[0080] This system achieves real-time monitoring and early warning of electric vehicle rollover risks through four steps: state prediction, operating condition clustering, model updating, and risk warning. Its core advantage lies in its dynamic adaptability, which allows it to adjust model parameters based on real-time status and operating conditions, improving warning accuracy. In practical applications, such as high-speed cornering scenarios, the system can predict rollover risks and promptly alert the driver, thereby improving driving safety. This technology provides an innovative solution for the safety of smart electric vehicles.

[0081] In some embodiments, the vehicle rollover risk warning includes:

[0082] Step a: Divide the safety critical time window into at least two consecutive decision nodes, and repeat steps b to f in each consecutive decision node.

[0083] In this step, the safety-critical time window is the critical time range within which the vehicle's current state evolves into an irreversible rollover. By dividing this window into at least two consecutive decision nodes, the system enables dynamic monitoring and phased intervention over time. This division ensures that the system can continuously adjust its assistance strategy based on real-time changes in driver status and vehicle dynamics, avoiding the blindness or lag of one-time interventions.

[0084] In actual operation, the length of the safety-critical time window is first determined through the acquisition steps described later (for example, assuming it is 3 seconds). The system divides it into two decision nodes, each lasting 1.5 seconds. Within each 1.5-second node, the system repeats steps b to f to achieve continuous evaluation and response to the driver and vehicle status. For example, when an electric car is driving on a curve at 80 km / h, the system detects an increase in the roll angle and immediately starts the time window division, which is then looped in units of 1.5 seconds.

[0085] Specific implementation example: Imagine an electric vehicle driving on a mountain road. Sensors detect a rollover risk warning of 5 degrees, triggering the vehicle's rollover angle. The system divides the critical safety time window (for example, 3 seconds) into two nodes: Node 1 (0-1.5 seconds) and Node 2 (1.5-3 seconds). Within each node, the system collects data and dynamically adjusts strategies to ensure that intervention measures are synchronized with the evolving risk.

[0086] Step b: collecting cognitive load assessment data of the electric vehicle driver in real time. The cognitive load assessment data includes the driver's eye gaze trajectory, steering wheel grip pressure changes, and operation response delay history within a preset time period.

[0087] In this step, the driver's cognitive load reflects their current ability to process information and respond to risks, and is a key factor in determining assistance strategies. This step comprehensively assesses the driver's psychological and behavioral state through multi-dimensional data collection, including eye gaze trajectory, steering wheel grip pressure changes, and operational response delay history, providing data support for subsequent decision-making.

[0088] The eye gaze trajectory uses the on-board camera and eye tracking algorithm to record the position and movement path of the driver's eye focus (the acquisition of the driver's eye gaze trajectory is an existing technology. For example, BMW has already applied the function of automatic vehicle steering and lane changing based on eye gaze trajectory, which will not be repeated here).

[0089] Steering wheel grip pressure variation uses a pressure sensor mounted on the steering wheel to detect fluctuations in grip force. A sudden increase in pressure may indicate tension or a stress response, while a decrease in pressure may indicate relaxation or fatigue.

[0090] Operation response delay history analyzes the driver's response time to braking, accelerator, or steering commands over a preset period of time (e.g., the previous 10 seconds). A relatively increased average delay may indicate excessive cognitive load.

[0091] A specific implementation example: When an electric car is traveling at 60 km / h, the system detects that the driver's gaze shifts from the road to the central control screen three times within one second, the steering wheel grip pressure increases from 20N to 30N, and the brake response delay increases from 0.3 seconds to 0.5 seconds within the first 10 seconds. This data is recorded in real time and used as the basis for cognitive load assessment.

[0092] Step c: Based on the cognitive load assessment criteria, determine the driver's current cognitive load and match it with a corresponding solution acceptance sensitivity level. The solution acceptance sensitivity level is negatively correlated with the driver's maximum acceptable assistance intervention intensity.

[0093] In this step, cognitive load directly impacts the driver's acceptance of external assistance. Using quantitative assessment data, the system determines the driver's current cognitive load and maps it to a preset sensitivity level for receiving assistance. This sensitivity level is negatively correlated with the driver's maximum acceptable level of assistance intervention. The higher the sensitivity level, the more likely the driver is to accept less-intensive assistance.

[0094] The system inputs the collected data into the cognitive load assessment model (for example, based on a weighted average algorithm, the number of times the eye gaze trajectory deviates from the road, the pressure change, and the average response delay time increase are weighted to obtain a load index) to calculate the load index. For example, the number of eye gaze deviations accounts for 40%, the grip pressure change is 30%, and the response delay increase is 30%, resulting in a load index of 0.75 (full score 1). According to the load index, the sensitivity level is matched: if the load index is greater than 0.7, it is highly sensitive (accepting weak intervention), 0.3≤load index≤0.7 is medium sensitive (accepting medium intervention), and the load index is less than 0.3, it is low sensitive (accepting strong intervention). In this example, the load index of 0.75 corresponds to a high sensitivity level.

[0095] Continuing with the above specific implementation example: The driver's load index is 0.75, indicating that their attention is highly distracted. The system determines that their solution reception sensitivity is high, which means that subsequent assistance strategies should avoid overly complex interventions to avoid increasing the driver's burden. Step d: Based on the driver's current solution reception sensitivity level, dynamically select the corresponding level of assistance execution plan from the preset rollover assistance strategy library. The rollover assistance strategy library contains multiple levels of assistance execution plans, from the active control layer to the passive prompt layer. The level of each level of assistance execution plan is positively correlated with the corresponding level of solution reception sensitivity. The assistance intervention intensity of each level of assistance execution plan decreases as the level increases.

[0096] In this step, the rollover mitigation assistance strategy library contains multiple levels of preset solutions, ranging from active control (such as automatic deceleration) to passive prompts (such as audible warnings). Each level is positively correlated with the sensitivity level, and the intervention intensity decreases as the level increases. Dynamic selection ensures that the assistance measures are matched to the driver's state, effectively controlling risks while avoiding excessive interference.

[0097] Here is a simple example of an anti-rollover assistance strategy library:

[0098] Level 1 (low solution reception sensitivity, high auxiliary intervention intensity): active control, such as automatic adjustment of vehicle speed or steering angle, with high intervention intensity;

[0099] Level 2 (medium solution reception sensitivity, medium assistance intervention intensity): semi-active assistance, such as light braking accompanied by instrument panel warnings, with medium intervention intensity;

[0100] Level 3 (high solution reception sensitivity, low auxiliary intervention intensity): Passive prompts, such as buzzer alarms or voice reminders, with low intervention intensity.

[0101] Specific implementation example: In the case of mountain curves, the driver has a high cognitive load and high sensitivity. The system selects the level 3 solution from the strategy library and plays "Please slow down, be aware of the risk of rollover" through the car audio at the first node to avoid direct intervention in the steering wheel.

[0102] Step e: executing the dynamically selected auxiliary execution plan for the driver.

[0103] In this step, the selected assistance strategy is executed to reduce the risk of rollover through appropriate intervention, while ensuring that the driver understands and cooperates with the system's behavior. This step converts the strategy into specific actions, which are then applied to the vehicle or driver in real time.

[0104] Specific implementation example: For a Level 3 solution, the system activates the vehicle's audio system, plays a pre-recorded voice prompt, and displays a red warning icon on the instrument panel. For a Level 1 solution, the electronic control unit (ECU) may reduce the vehicle speed to 50 km / h and fine-tune the steering angle. Step f: When the similarity between the driver's active operation behavior vector and the auxiliary target vector of the auxiliary execution plan for the driver exceeds a threshold, the corresponding auxiliary execution plan for the driver is terminated.

[0105] In this step, the system compares the driver's actual actions with the expected behavior of the assistance target to determine whether the driver has effectively responded to the risk. When the similarity between the two exceeds a preset threshold, indicating that the driver is in control of the situation, the system can stop intervening to avoid redundancy.

[0106] Specific implementation example: An active action vector is constructed using parameters such as vehicle speed change and steering angle adjustment. For example, if the driver decelerates by 5 km / h and adjusts the steering angle by 2 degrees, the voice prompt's goal is to slow down and maintain stability. An auxiliary target vector is constructed by reducing the vehicle speed by 5-10 km / h and adjusting the steering angle by 1-3 degrees. Cosine similarity is used to calculate the similarity between the active action vector and the auxiliary target vector. If the result exceeds a preset threshold of 0.9, assistance is terminated.

[0107] In summary, steps a through f ensure accurate and efficient rollover risk warnings by defining critical safety time windows, assessing cognitive load in real time, dynamically selecting and executing assistance strategies, and incorporating driver behavioral feedback into a closed-loop optimization mechanism. This approach significantly improves safety and driver experience in complex driving scenarios, further providing an innovative solution for the safety of smart electric vehicles.

[0108] In some embodiments, the step of obtaining the safety critical time window includes:

[0109] The time window from the time when the vehicle rollover risk is analyzed and determined to the future target time is regarded as the safety critical time window;

[0110] The future target time is the smaller value between the physical rollover prevention limit time and the driver's cognitive buffer time;

[0111] The physical rollover prevention limit time includes: the conservative time of irreversible rollover of the electric vehicle in the future determined based on the real-time updated vehicle roll dynamics model;

[0112] The cognitive buffer time includes the maximum allowable time for the driver to complete the recognition of the vehicle rollover risk and the operational response, which is determined based on the driver's cognitive load when analyzing and determining the vehicle rollover risk.

[0113] The safety critical time window is the basis of the early warning system. It needs to be determined by combining the physical limits of the vehicle and the cognitive ability of the driver. The smaller value of the physical anti-rollover limit time and the cognitive buffer time is taken to ensure that the system takes action before the risk is irreversible.

[0114] Using a vehicle roll dynamics model (taking into account speed, roll angle, center of gravity height, etc.), a conservative time for irreversible rollover is predicted. Technicians can then set physical rollover prevention limits based on different vehicle rollover risks through experiments or on demand. For example, at a speed of 80 km / h and a roll angle of 5 degrees, the model calculates a limit time of 4 seconds.

[0115] Based on the driver's current cognitive load and historical response data, the maximum allowable time for them to recognize the risk and take action is estimated. Technicians can also conduct experiments to determine the cognitive buffer time for different driver cognitive loads and different vehicle rollover risks. For example, a load of 0.75 corresponds to a buffer time of 3 seconds.

[0116] The smaller value between the physical anti-rollover limit time and the driver's cognitive buffer time is taken as the future target time, and the time window from the time when the vehicle rollover risk is analyzed and determined to the future target time is taken as the safety critical time window.

[0117] In summary, the above technical solution provides a scientific decision-making basis for the system by obtaining the safety critical time window through dual physical and cognitive constraints.

[0118] The present application provides a vehicle rollover warning method based on state prediction and operating condition clustering, including:

[0119] Predicting the motion state of electric vehicles through neural networks;

[0120] Determine the current operating condition of the electric vehicle through clustering;

[0121] Based on the predicted motion state and the determined current operating conditions, the vehicle roll dynamics model of the electric vehicle is updated in real time;

[0122] Based on the real-time updated vehicle roll dynamics model, the vehicle rollover risk is analyzed and a vehicle rollover risk warning is issued.

[0123] In some embodiments, the vehicle rollover risk warning includes:

[0124] Step a: Divide the safety critical time window into at least two consecutive decision nodes, and repeat steps b to f in each consecutive decision node;

[0125] Step b: collecting cognitive load assessment evidence of the electric vehicle driver in real time;

[0126] Step c: Based on the cognitive load assessment basis, determine the driver's current cognitive load and match it with a corresponding solution reception sensitivity level; wherein the solution reception sensitivity level is negatively correlated with the driver's acceptable maximum assistance intervention intensity;

[0127] Step d: Based on the driver's current scenario reception sensitivity level, dynamically select an assistance execution scenario of a corresponding level from a preset rollover prevention assistance strategy library; wherein the rollover prevention assistance strategy library includes multiple levels of assistance execution scenarios, from an active control layer to a passive prompt layer, and the level of each level of assistance execution scenario is positively correlated with the corresponding scenario reception sensitivity level, and the assistance intervention intensity of each level of assistance execution scenario decreases as the level increases;

[0128] Step e: executing the dynamically selected auxiliary execution plan for the driver;

[0129] Step f: when it is detected that the similarity between the active operation behavior vector of the driver operating the electric vehicle and the auxiliary target vector of the auxiliary execution plan executed on the driver exceeds a threshold, stop executing the corresponding auxiliary execution plan for the driver.

[0130] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A vehicle rollover warning system based on state prediction and working condition clustering, characterized by: include: A state prediction module is used to predict the motion state of electric vehicles through a neural network; An operating condition clustering module, used to determine the current operating condition of the electric vehicle through clustering; A model update module for updating a vehicle roll dynamics model of an electric vehicle in real time based on the predicted motion state and the determined current operating conditions; The rollover warning module is used to analyze the vehicle rollover risk and issue a vehicle rollover risk warning based on the real-time updated vehicle roll dynamics model.

2. The vehicle rollover warning system based on state prediction and working condition clustering according to claim 1, characterized in that: The state prediction module predicts the motion state of the electric vehicle through a neural network, including: Use a pre-trained recurrent neural network as the neural network; Taking vehicle parameters of the electric vehicle, including at least vehicle wheel speed, steering wheel angle, yaw rate, longitudinal acceleration, lateral acceleration, body roll angle, suspension travel sensor data, and motor torque / speed, as input, and inputting them into a neural network; The vehicle state of the electric vehicle outputted by the neural network at least includes the predicted vehicle yaw rate, the predicted vehicle lateral acceleration, the predicted vehicle body roll angle and the predicted tire vertical load as the predicted motion state.

3. The vehicle rollover warning system based on state prediction and working condition clustering according to claim 1, characterized in that: The operating condition clustering module determines the current operating condition of the electric vehicle through clustering, including: constructing a clustering feature vector set based on vehicle parameters of the electric vehicle including at least a current vehicle speed, a steering wheel angle, a yaw rate, a longitudinal acceleration, a lateral acceleration, a body roll angle, a tire slip rate, and a motor output torque of the electric vehicle; Based on the K-means algorithm, the working conditions are clustered according to the clustering feature vector set to obtain the determined current working conditions; The categories of current operating conditions include at least: straight-line driving, steady-state steering, transient steering, acceleration, braking, driving conditions under different road adhesion coefficients, and different road types.

4. The vehicle rollover warning system based on state prediction and working condition clustering according to claim 1, characterized in that: The model updating module updates the vehicle roll dynamics model of the electric vehicle in real time based on the predicted motion state and the determined current working condition, including: Adjusting corresponding parameters or coefficients in the vehicle roll dynamics model based on the predicted motion state and the determined current operating conditions; The parameters or coefficients include at least: roll stiffness, roll damping, roll moment of inertia, and tire cornering stiffness.

5. The vehicle rollover warning system based on state prediction and working condition clustering according to claim 1, characterized in that: The rollover warning module analyzes the vehicle rollover risk based on the real-time updated vehicle roll dynamics model, including: For each preset vehicle rollover risk index, calculate the index value of the vehicle rollover dynamics model updated in real time under the vehicle rollover risk index, compare it with the corresponding preset index threshold, and determine the vehicle rollover risk based on the comparison result; Among them, the vehicle rollover risk indicators include at least: roll angle, load transfer rate and rollover coefficient.

6. The vehicle rollover warning system based on state prediction and working condition clustering according to claim 1, characterized in that: The vehicle rollover risk warning includes: Step a: Divide the safety critical time window into at least two consecutive decision nodes, and repeat steps b to f in each consecutive decision node; Step b: collecting cognitive load assessment evidence of the electric vehicle driver in real time; Step c: Based on the cognitive load assessment basis, determine the driver's current cognitive load and match it with a corresponding solution reception sensitivity level; wherein the solution reception sensitivity level is negatively correlated with the driver's acceptable maximum assistance intervention intensity; Step d: Based on the driver's current scenario reception sensitivity level, dynamically select an assistance execution scenario of a corresponding level from a preset rollover prevention assistance strategy library; wherein the rollover prevention assistance strategy library includes multiple levels of assistance execution scenarios, from an active control layer to a passive prompt layer, and the level of each level of assistance execution scenario is positively correlated with the corresponding scenario reception sensitivity level, and the assistance intervention intensity of each level of assistance execution scenario decreases as the level increases; Step e: executing the dynamically selected auxiliary execution plan for the driver; Step f: when it is detected that the similarity between the active operation behavior vector of the driver operating the electric vehicle and the auxiliary target vector of the auxiliary execution plan executed on the driver exceeds a threshold, stop executing the corresponding auxiliary execution plan for the driver.

7. The vehicle rollover warning system based on state prediction and working condition clustering according to claim 6, characterized in that: The step of obtaining the safety critical time window includes: The time window from the time when the vehicle rollover risk is analyzed and determined to the future target time is regarded as the safety critical time window; The future target time is the smaller value between the physical rollover prevention limit time and the driver's cognitive buffer time; The physical rollover prevention limit time includes: the conservative time of irreversible rollover of the electric vehicle in the future determined based on the real-time updated vehicle roll dynamics model; The cognitive buffer time includes the maximum allowable time for the driver to complete the recognition of the vehicle rollover risk and the operational response, which is determined based on the driver's cognitive load when analyzing and determining the vehicle rollover risk.

8. The vehicle rollover warning system based on state prediction and working condition clustering according to claim 6, characterized in that: The cognitive load assessment is based on the driver's eye gaze trajectory, changes in steering wheel grip pressure, and the operation response delay history within a preset time.

9. A vehicle rollover warning method based on state prediction and working condition clustering is characterized by: include: Predicting the motion state of electric vehicles through neural networks; Determine the current operating condition of the electric vehicle through clustering; Based on the predicted motion state and the determined current operating conditions, the vehicle roll dynamics model of the electric vehicle is updated in real time; Based on the real-time updated vehicle roll dynamics model, the vehicle rollover risk is analyzed and a vehicle rollover risk warning is issued.

10. The vehicle rollover warning method based on state prediction and working condition clustering according to claim 9, characterized in that: The vehicle rollover risk warning includes: Step a: Divide the safety critical time window into at least two consecutive decision nodes, and repeat steps b to f in each consecutive decision node; Step b: collecting cognitive load assessment evidence of the electric vehicle driver in real time; Step c: Based on the cognitive load assessment basis, determine the driver's current cognitive load and match it with a corresponding solution reception sensitivity level; wherein the solution reception sensitivity level is negatively correlated with the driver's acceptable maximum assistance intervention intensity; Step d: Based on the driver's current scenario reception sensitivity level, dynamically select an assistance execution scenario of a corresponding level from a preset rollover prevention assistance strategy library; wherein the rollover prevention assistance strategy library includes multiple levels of assistance execution scenarios, from an active control layer to a passive prompt layer, and the level of each level of assistance execution scenario is positively correlated with the corresponding scenario reception sensitivity level, and the assistance intervention intensity of each level of assistance execution scenario decreases as the level increases; Step e: executing the dynamically selected auxiliary execution plan for the driver; Step f: when it is detected that the similarity between the active operation behavior vector of the driver operating the electric vehicle and the auxiliary target vector of the auxiliary execution plan executed on the driver exceeds a threshold, stop executing the corresponding auxiliary execution plan for the driver.

Images