Anti-overturning method for hilly and mountainous unmanned operation platform based on multi-sensor fusion

Abstract

The application discloses a kind of based on multi-sensor fusion's hilly and mountainous unmanned operation platform anti-overturning method, comprising: (1) establishing operation platform vehicle dynamics model, side inclination dynamics model characteristics and analyzing vehicle state equation, with load transfer rate definition vehicle's stable area, establish vehicle state estimation model;(2) the data of unmanned operation platform multi-sensor system are collected, the data include obtaining vehicle state information and vehicle surrounding environment information;(3) based on neural network prediction model to the multi-sensor data processing collected, and the load transfer rate and state of vehicle are estimated, complete control strategy prediction;(4) vehicle controller according to the load transfer rate value obtained and the estimated value of vehicle state and the vehicle state estimation model established, realize the selection of hilly and mountainous operation platform anti-rollover control strategy.The anti-overturning method of the application makes the vehicle stability, automation, intelligent level greatly improve.

Description

Technical Field

[0001] This invention relates to the field of agricultural machinery and equipment technology, and in particular to a method for preventing overturning of unmanned operating platforms in hilly and mountainous areas based on multi-sensor fusion. Background Technology

[0002] The climate in southern my country is suitable for plant growth, making it a major planting area for various timber and economic forests. Most of these planting areas are located in artificially developed terraced fields. However, the level of mechanization and intelligentization in hilly and mountainous agricultural and forestry farming is low, severely restricting the development of agricultural mechanization and consequently significantly impacting the modernization of agriculture and rural areas. Currently, crop mechanization in hilly and mountainous provinces is low; most traditional agricultural machinery cannot navigate autonomously or avoid obstacles, making it difficult for machinery to enter and operate in the fields, resulting in rapid machine wear.

[0003] The primary task in improving the level of agricultural mechanization in hilly and mountainous areas is to solve the problem of equipment access to mountainous terrain and mobile platforms, which directly affects the future development of forestry mechanization in my country. Production in hilly and mountainous planting areas requires meticulous cultivation and involves a large amount of labor, including planting, tending, fertilization, pruning, plant protection, and harvesting. However, the fragmented and irregular terrain, numerous slopes, ridges, and irregular shapes of hilly and mountainous areas present significant challenges to mobile machinery. Furthermore, due to the delay between driver input and vehicle response, the time delay and limited effective range of the braking system, and the complex relationship between lateral stability control and rollover prevention control, even vehicles equipped with traditional stability control that only manages lateral stability may become unstable, facing a significant risk of rollover.

[0004] Currently, many research institutions in China have conducted research on hilly and mountainous operation machinery to a certain extent. For example, Beijing University of Aeronautics and Astronautics proposed a variable ground clearance and wheel track power platform (VII-HPC) with a balanced rocker arm suspension and H-type transmission. To adapt to the terrain characteristics of hilly areas, it has good obstacle crossing ability, but poor stability. China Agricultural University proposed a pre-detection active leveling method to solve the problem of low leveling accuracy and lag after the agricultural chassis tilts in hilly and mountainous areas. However, the error is large when relying on four height ranging sensors arranged on the chassis to obtain the relative height information of the ground, and it cannot handle terrain that is easy to roll over, and the level of intelligence is low.

[0005] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method for preventing overturning of unmanned operation platforms in hilly and mountainous areas based on multi-sensor fusion, thereby overcoming the shortcomings of operation vehicles being prone to overturning and unstable operation in hilly and mountainous areas.

[0007] To achieve the above objectives, the present invention provides a method for preventing overturning of unmanned operating platforms in hilly and mountainous areas based on multi-sensor fusion, comprising:

[0008] (1) Establish the dynamics model of the work platform vehicle, the characteristics of the roll dynamics model and analyze the vehicle state equation. Use the load transfer rate to define the stable region of the vehicle, and then establish the vehicle state estimation model.

[0009] (2) Collect data from the multi-sensor system of the unmanned vehicle, including information on the vehicle's own status and information on the surrounding environment.

[0010] (3) Based on the neural network prediction model, process the multi-sensor data collected in step (2), estimate the load transfer rate and vehicle status of the vehicle, and complete the prediction of the control strategy.

[0011] (4) The vehicle controller selects the anti-rollover control strategy for the hilly and mountainous operation platform based on the vehicle state estimation model established in step (1) and the load transfer rate value and vehicle state estimation value obtained in step (3).

[0012] Preferably, in the above technical solution, the dynamic characteristics of the vehicle in step (1) of establishing the vehicle dynamics model include longitudinal movement, lateral movement and yaw direction movement.

[0013] Preferably, in the above technical solution, the longitudinal motion equation is:

[0014]

[0015] The equation of lateral motion is:

[0016]

[0017] The equation of motion in the yaw direction is:

[0018]

[0019] In the formula, m is the vehicle mass, and V is the mass of the vehicle. x and V y It refers to longitudinal and lateral speeds. It is the yaw rate of the vehicle, F xij and F yij These are the longitudinal and lateral forces on the four wheels, with subscripts ij representing fl, fr, and rl, respectively, and rr representing the left front wheel, right front wheel, left rear wheel, and right rear wheel; ij used in the following equations have the same meaning; δ is the steering angle of the front wheels, Iz It is the moment of inertia about the vertical axis, a and b are the distances from the center of mass to the front and rear axles, respectively, and T is the wheelbase of the vehicle;

[0020] Estimate the state changes of the unmanned vehicle during operation and establish an estimation model; calculate the state equation and measurement equation of the three-degree-of-freedom operating platform from the longitudinal motion equation, the lateral motion equation and the yaw direction motion equation.

[0021] The state equation of the work platform is:

[0022]

[0023] The measurement equation is:

[0024]

[0025] Forward wheel rotation angle δ and longitudinal acceleration a x Let u be the system input vector, i.e., u = [δ, a x ]; with yaw rate The sideslip angle β and the longitudinal vehicle speed Vx are the state vectors, i.e. With lateral acceleration a y The output vector is y = [a y ]; where: k f and k r These are the lateral stiffness of the front and rear tires, respectively.

[0026] Preferably, in the above technical solution, the vehicle roll dynamics model in step (1) is as follows: the roll dynamics characteristics of the vehicle when the wheels are not lifted are solved based on the roll dynamics model, and the roll dynamics characteristic equation is:

[0027]

[0028] Where φ is the roll angle of the vehicle at its center of gravity. It is the vehicle's roll rate, O f and O r These are the roll centers of the front and rear suspensions, I xO It is the roll moment of inertia about the roll axis, ay is the lateral acceleration of the center of gravity, and h is the roll moment of inertia about the roll axis. O The height h is the distance from the center of gravity to the center of roll. g K is the height of the center of gravity, C is the rolling stiffness of the suspension, and C is the rolling damping of the suspension.

[0029] Preferably, in the above technical solution, step (1) defines the stable region of the vehicle using the load transfer rate as follows:

[0030] Based on the vehicle roll dynamics model, a stable region is defined considering the load transfer ratio (LTR). The vehicle attitude is adjusted by controlling the extension and retraction of the leveling hydraulic cylinders. Within the stable region, the load transfer ratio is less than a safe threshold, allowing the vehicle to travel smoothly. The load transfer ratio is defined as:

[0031]

[0032] The threshold for LTR is set to: LTR min =0.6 and LTR max =0.85.

[0033] Preferably, in the above technical solution, the multi-sensor system of the unmanned vehicle in step (2) includes: multi-line hybrid solid-state lidar, laser rangefinder, camera device, global positioning system, microelectromechanical inertial measurement unit, wheel speed sensor, steering wheel angle sensor, lateral acceleration sensor, and yaw rate sensor;

[0034] Data is collected from each sensor according to the required data, then the geometric constraint relationship between each sensor is established, and then the data communication between each module is completed.

[0035] A multi-line hybrid solid-state lidar, laser rangefinder, and camera device are used to perceive road information and surrounding tree information for the unmanned vehicle. By observing the characteristics of laser scan lines on concave and convex slopes and obstacles, road features are acquired to facilitate platform state estimation and control decisions. Based on the lidar parameters, the distribution of the lidar multi-line scan across obstacles is as follows:

[0036]

[0037] Where M is the number of scan points in one revolution of the scan line, and μ i The angle between the scan lines is represented by D, the distance between the lidar and the obstacle is represented by H, and the installation height of the lidar relative to the ground is represented by d. i This indicates the horizontal distance between two points on the obstacle being scanned.

[0038] The camera device collects images of trees and vegetation ahead to obtain tree location information, which is helpful for vehicle obstacle avoidance and path planning; the microelectromechanical inertial measurement unit, wheel speed sensor, steering wheel angle sensor, lateral acceleration sensor and yaw rate sensor are used to collect parameters required for vehicle attitude estimation such as speed, acceleration and steering angle.

[0039] Preferably, in the above technical solution, the control strategy prediction in step (3) includes:

[0040] 1) The neural network adopts a fully connected neural network structure. The load transfer rate value and vehicle status predicted by the neural network are used as reference values ​​for the controller control mode.

[0041] 2) The road information collected by the lidar is preprocessed to remove point cloud data with insignificant changes in the surrounding area. The features of ground undulations and obstacles after clustering are extracted. These features, along with the vehicle's lateral acceleration, yaw rate, roll angle, roll rate, steering wheel angle, and longitudinal acceleration currently measured by the microelectromechanical inertial measurement unit (MEMS) sensor, are used as input to the neural network. The estimated values ​​of roll angle and roll rate at the next moment are output to predict the load transfer rate and vehicle state at the next moment. At the same time, the load transfer rate is compared with the preset threshold to realize the selection of control strategy.

[0042] Preferably, in the above technical solution, the method for obtaining the load transfer rate value and vehicle status through neural network prediction in step 1) includes:

[0043] The input layer contains information on the distance and size of obstacles, as well as the lateral acceleration, yaw rate, roll angle, roll rate, steering wheel angle, and longitudinal acceleration of the work platform currently measured by sensors such as the microelectromechanical inertial measurement unit. The output layer contains the roll angle and roll rate. There are 7 hidden layers, and the number of neurons in a single network layer is set to 100. The number of network training iterations is 1000.

[0044] Except for the sigmoid function used for output layer neurons, the Leaky ReLU function with faster convergence is used for both input and hidden layers. L2 regularization is employed to avoid overfitting in the neural network model. The Adam optimization algorithm is used instead of the classic stochastic gradient descent method to update network weights more effectively. The loss function is defined as:

[0045]

[0046] In the formula, m is the number of samples involved in the calculation. The output of the network model, y is the output label, λ is the L2 regularization hyperparameter, L is the number of layers in the neural network, and ||W j ||2 is the norm square, which is defined as the sum of the squares of all elements in the matrix.

[0047] Preferably, in the above technical solution, the method for selecting the control strategy in step (4) includes:

[0048] Based on the vehicle model and body sensors, an extended Kalman filter is used for information fusion to estimate the vehicle's state and provide the yaw rate to the vehicle controller. The control strategy is selected based on the center of gravity sideslip angle β and the longitudinal vehicle speed Vx state vector, as well as the load transfer rate estimated based on the side roll angle and side roll velocity predicted by the trained neural network model.

[0049] a) When the load transfer rate is less than the minimum threshold load transfer rate, it means that there will be no rollover. Considering only the yaw stability of the vehicle, the work platform can pass through the section normally.

[0050] b) When the load transfer rate value is between the minimum and maximum load transfer rate, it means that the load transfer between the left and right wheels is large. If it is not controlled, it may roll over. Therefore, yaw stability and rollover prevention should be considered at the same time. Anti-rollover technologies such as active differential braking, active suspension or semi-active suspension adjustment, and active steering angle correction should be activated in advance.

[0051] c) When the load transfer rate is greater than the maximum load transfer rate, the work platform is in great danger of overturning. The road is blocked and impassable. It is necessary to control the front wheels to steer around the road and replan the route according to the work requirements to carry out forest protection and other work.

[0052] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a method for preventing overturning of unmanned operation platforms in hilly and mountainous areas based on multi-sensor fusion. It mainly involves detecting the status of unmanned operation vehicles in hilly and mountainous areas and road information in the direction of travel by fusing and complementing data from multiple sensors, and estimating the vehicle load transfer rate (LTR) through a predictive model, which is then provided to the control system to finally determine the control strategy and realize the anti-overturning of unmanned operation platforms in hilly and mountainous areas. First, the characteristics of the vehicle dynamics model and roll dynamics model of the work platform are established, and the vehicle state equation is analyzed. Then, LiDAR, camera devices, GPS, MEMS inertial measurement units, wheel speed sensors, steering wheel angle sensors, lateral acceleration sensors, and yaw rate sensors are used to acquire the vehicle's own state information and the information of the surrounding environment. Then, the load transfer rate of the work vehicle is estimated based on a neural network prediction model using the vehicle attitude data and ground information. Finally, the vehicle controller selects the vehicle control strategy based on the input vehicle attitude data and LTR value, thereby avoiding the rollover of the work vehicle and obstacle avoidance path planning for unmanned driving. This allows the unmanned work vehicle in hilly and mountainous areas to take preventive measures before the rollover actually occurs. The controller can take action earlier, and the vehicle stability, automation, and intelligence level are greatly improved. Attached Figure Description

[0053] Figure 1 This is a flowchart of the anti-tipping method for unmanned operation platforms in hilly and mountainous areas based on multi-sensor fusion according to the present invention;

[0054] Figure 2This is a three-degree-of-freedom vehicle dynamics model according to the present invention;

[0055] Figure 3 It is a side-tilt dynamics model of unmanned vehicles operating in hilly and mountainous terrain;

[0056] Figure 4 It is the structure of a neural network prediction model. Detailed Implementation

[0057] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings, but it should be understood that the scope of protection of the present invention is not limited to the specific embodiments.

[0058] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0059] This invention discloses a method for preventing overturning of unmanned operating platforms in hilly and mountainous areas based on multi-sensor fusion, comprising the following steps:

[0060] Step 1: Establish the vehicle dynamics model and roll dynamics model of the work platform, and determine their characteristics and state equations.

[0061] Specifically, the dynamics model of the work platform vehicle and the characteristics of the roll dynamics model are established and the vehicle state equation is analyzed. The stable region of the vehicle is defined by the load transfer rate, and then the vehicle state estimation model is established.

[0062] Including the following methods:

[0063] (1) Dynamics model of the work platform vehicle and its state equations and measurement equations

[0064] The dynamics model of a three-degree-of-freedom work platform vehicle is as follows: Figure 2 As shown, the vehicle's dynamic characteristics include longitudinal motion, lateral motion, and yaw motion.

[0065] The longitudinal motion equation is:

[0066]

[0067] The equation of lateral motion is:

[0068]

[0069] The equation of motion in the yaw direction is:

[0070]

[0071] In the formula, m is the vehicle mass, and V is the mass of the vehicle. x and Vy It refers to longitudinal and lateral speeds. It is the yaw rate of the vehicle, F xij and F yij These are the longitudinal and lateral forces on the four wheels, with subscripts ij representing fl, fr, and rl, respectively, and rr representing the left front wheel, right front wheel, left rear wheel, and right rear wheel; ij used in the following equations have the same meaning; δ is the steering angle of the front wheels, I z It is the moment of inertia about the vertical axis, a and b are the distances from the center of mass to the front and rear axles, respectively, and T is the wheelbase of the vehicle.

[0072] To accurately estimate the state changes of the unmanned operation platform during operation, the established estimation model uses the front wheel rotation angle δ and longitudinal acceleration a. x Let u be the system input vector, i.e., u = [δ, a x ]; with yaw rate The sideslip angle β and the longitudinal vehicle speed Vx are the state vectors, i.e. With lateral acceleration a y The output vector is y = [a y ].

[0073] The state equation and measurement equation of the three-degree-of-freedom work platform can be calculated from the above formulas as follows:

[0074] The state equation of the work platform is:

[0075]

[0076] The measurement equation is:

[0077]

[0078] Forward wheel rotation angle δ and longitudinal acceleration a x Let u be the system input vector, i.e., u = [δ, a x ]; with yaw rate The sideslip angle β and the longitudinal vehicle speed Vx are the state vectors, i.e. With lateral acceleration a y The output vector is y = [a y ]; where: k f and k r These are the lateral stiffness of the front and rear tires, respectively.

[0079] 2. Roll dynamics model

[0080] Roll dynamics model, such as Figure 3 As shown, the roll dynamics characteristics of the vehicle when the wheels are not lifted are solved based on the roll dynamics model. The roll dynamics characteristic equation is:

[0081]

[0082] In the formula, φ is the roll angle of the vehicle at its center of gravity. It is the vehicle's roll rate, O f and O r These are the roll centers of the front and rear suspensions, I xO It is the roll moment of inertia about the roll axis, ay is the lateral acceleration of the center of gravity, and h is the roll moment of inertia about the roll axis. O The height h is the distance from the center of gravity to the center of roll. g K is the height of the center of gravity, C is the rolling stiffness of the suspension, and C is the rolling damping of the suspension.

[0083] To ensure the stability of the work platform in hilly and mountainous terrain and prevent rollovers due to the common uneven terrain, a rollover prevention control strategy considering the load transfer rate (LTR) is proposed based on the vehicle roll dynamics model. Simultaneously, this strategy controls... Figure 3 The extension / retraction of the leveling hydraulic cylinder L in the system is used to adjust the vehicle's posture, preventing excessive roll angles that could increase the risk of rollover. Load transfer ratio (LTR) is typically used to define the vehicle's stability region. Within the stability region, the LTR is less than a safe threshold, indicating vehicle stability. The load transfer ratio is defined as:

[0084]

[0085] The threshold for LTR is set to: LTR min =0.6 and LTR max =0.85.

[0086] Step 2: Data Acquisition by the Multi-Sensor System of the Unmanned Operation Platform

[0087] Data is collected from the multi-sensor system of the unmanned operation vehicle, including information on the vehicle's own status and the surrounding environment.

[0088] Specifically, the unmanned vehicle's perception system includes a multi-line hybrid solid-state lidar, a laser rangefinder, a camera device, a global positioning system, a microelectromechanical inertial measurement unit, four wheel speed sensors, a steering wheel angle sensor, a lateral acceleration sensor, and a yaw rate sensor.

[0089] First, data is collected from each sensor according to the required data. Then, geometric constraints between the sensors are established to facilitate data communication between the modules.

[0090] LiDAR and cameras are used to perceive road information and surrounding tree information for the unmanned vehicle. By observing the characteristics of the laser scanning lines on concave and convex slopes and obstacles, road features are acquired to facilitate platform state estimation and control decisions. Based on the LiDAR parameters, the distribution of LiDAR multi-line scanning across obstacles is as follows:

[0091]

[0092] Where M is the number of scan points in one revolution of the scan line, and μ i The angle between the scan lines is represented by D, the distance between the lidar and the obstacle is represented by H, and the installation height of the lidar relative to the ground is represented by d. i This indicates the horizontal distance between two points on the obstacle being scanned.

[0093] The camera device collects images of trees and vegetation ahead to obtain tree location information, which helps the vehicle avoid obstacles and plan its path; the other sensors are used to collect parameters needed for vehicle attitude estimation, such as speed, acceleration, and turning angle.

[0094] Step 3: Process multi-sensor data and predict control strategies based on neural network prediction models.

[0095] The multi-sensor data collected in step (2) is processed based on the neural network prediction model, and the load transfer rate and vehicle status of the vehicle are estimated to complete the prediction of the control strategy.

[0096] The method is as follows:

[0097] (1) The neural network adopts a fully connected neural network structure, as shown in Figure 4. The LTR value and the status of the work platform predicted by the neural network are used as reference quantities for the controller control mode. The input layer contains information on the distance and size of concave and convex obstacles and the lateral acceleration, yaw rate, roll angle, roll rate, steering wheel angle and longitudinal acceleration of the work platform currently measured by sensors such as the microelectromechanical inertial measurement unit. The output layer is the roll angle and roll rate. There are 7 hidden layers. The number of neurons in a single network layer is set to 100, and the number of network training times is 1000.

[0098] Except for the sigmoid function used for output layer neurons, the Leaky ReLU function with faster convergence is used for both input and hidden layers. L2 regularization is employed to avoid overfitting in the neural network model, and the Adam optimization algorithm is used instead of the classic stochastic gradient descent method to update network weights more effectively. The loss function is defined as:

[0099]

[0100] In the formula, m is the number of samples involved in the calculation. λ is the output of the network model, y is the output label, λ is the L2 regularization hyperparameter, L is the number of layers in the neural network, and ||W j ||2 is the norm square, which is defined as the sum of the squares of all elements in the matrix.

[0101] (2) The road point cloud information collected by the lidar is preprocessed, the point cloud data with no obvious changes in the surrounding area is removed, and the features of ground unevenness and slope and obstacles after clustering are extracted. These features are used as the input of the neural network along with the lateral acceleration, yaw rate, roll angle, roll rate, steering wheel angle and longitudinal acceleration of the working platform currently measured by sensors such as the microelectromechanical inertial measurement unit. The estimated values ​​of roll angle and roll rate at the next moment are output to predict the LTR value and working platform status at the next moment. At the same time, the control strategy is selected by comparing the LTR preset threshold.

[0102] Step 4: Select the anti-rollover control strategy for hilly and mountainous work platforms based on LTR values ​​and vehicle status.

[0103] The vehicle controller selects the anti-rollover control strategy for the hilly and mountainous work platform based on the load transfer rate value and the estimated vehicle state obtained in step three and the vehicle state estimation model established in step one.

[0104] Based on the vehicle model and body sensors, an extended Kalman filter is used for information fusion to estimate the vehicle's state and provide the yaw rate to the vehicle controller. The control strategy is selected based on the center of gravity sideslip angle β and the longitudinal vehicle speed Vx state vector, as well as the LTR value estimated by predicting the sideslip angle and sideslip angular velocity based on the trained neural network model.

[0105] a) When the LTR value is less than the threshold LTR min This means that there will be no rollover, and considering only the vehicle's yaw stability, the work platform can pass through the section of road normally;

[0106] b) When the LTR value is between LTR min and LTR max When the yaw rate is between the left and right wheels, this means that the load transfer between the left and right wheels is large. If it is not controlled, it may roll over. Therefore, yaw stability and rollover prevention should be considered at the same time. Anti-rollover technologies such as active differential braking, active suspension or semi-active suspension adjustment, and active steering angle correction should be activated in advance.

[0107] c) When the LTR value is greater than LTR max The work platform is at great risk of overturning. The road section is blocked by obstacles and cannot be passed. It is necessary to control the front wheels to steer around the road and replan the route according to the work requirements to carry out forest protection and other work.

[0108] This invention provides a method for preventing rollover of unmanned work platforms in hilly and mountainous areas based on multi-sensor fusion. It mainly involves using complementary data fusion from multiple sensors to detect the state of the unmanned work platform and road information in its direction of travel. A predictive model is then used to estimate the vehicle load transfer rate (LTR), providing this information to the control system to ultimately determine the control strategy and prevent rollover. First, the characteristics of the work platform's vehicle dynamics model and roll dynamics model are established, and the vehicle's state equation is analyzed. Then, LiDAR, cameras, GPS, MEMS, wheel speed sensors, steering wheel angle sensors, lateral acceleration sensors, and yaw rate sensors are used to acquire the vehicle's own state information and information about the surrounding environment. Next, using the vehicle's attitude data and ground information, a neural network predictive model is used to estimate the load transfer rate of the work platform. Finally, the vehicle controller selects a vehicle control strategy based on the input vehicle attitude data and LTR value, thereby avoiding rollover and obstacle avoidance path planning for unmanned driving. This allows the unmanned work platform to take preventative measures before rollover actually occurs, enabling the controller to act earlier and significantly improving vehicle stability, automation, and intelligence.

[0109] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A method for preventing overturning of unmanned operating platforms in hilly and mountainous areas based on multi-sensor fusion, characterized in that, include: (1) Establish the dynamics model of the work platform vehicle, the characteristics of the roll dynamics model and analyze the vehicle state equation. Use the load transfer rate to define the stable region of the vehicle, and then establish the vehicle state estimation model. (2) Collect data from the multi-sensor system of the unmanned vehicle, including information on the vehicle's own status and information on the surrounding environment; (3) Based on the neural network prediction model, process the multi-sensor data collected in step (2), estimate the load transfer rate and vehicle status of the vehicle, and complete the prediction of the control strategy. (4) The vehicle controller selects the anti-rollover control strategy for the hilly and mountainous operation platform based on the vehicle state estimation model established in step (1) and the load transfer rate value and vehicle state estimation value obtained in step (3). Step (2) The multi-sensor system of the unmanned vehicle includes: multi-line hybrid solid-state lidar, laser rangefinder, camera device, global positioning system, microelectromechanical inertial measurement unit, wheel speed sensor, steering wheel angle sensor, lateral acceleration sensor, and yaw rate sensor; Data is collected from each sensor according to the required data, then the geometric constraint relationship between each sensor is established, and then the data communication between each module is completed. A multi-line hybrid solid-state lidar, laser rangefinder, and camera device are used to perceive road information and surrounding tree information for the unmanned vehicle. By observing the characteristics of laser scan lines on concave and convex slopes and obstacles, road features are acquired to facilitate platform state estimation and control decisions. Based on the lidar parameters, the distribution of the lidar multi-line scan across obstacles is as follows: ， Where M is the number of scan points in one revolution of the scan line. The angle between the scan lines is represented by D, the distance between the lidar and the obstacle is represented by H, and the installation height of the lidar relative to the ground is represented by H. This indicates the horizontal distance between two points on the obstacle being scanned. The camera device collects images of trees and vegetation ahead to obtain tree location information, which is helpful for vehicle obstacle avoidance and path planning; the microelectromechanical inertial measurement unit, wheel speed sensor, steering wheel angle sensor, lateral acceleration sensor and yaw rate sensor are used to collect parameters required for vehicle attitude estimation, such as speed, acceleration and steering angle. Step (3) of the control strategy prediction includes: 1) The neural network adopts a fully connected neural network structure. The load transfer rate value and vehicle status predicted by the neural network are used as reference values ​​for the controller control mode. 2) The road information collected by the lidar is preprocessed to remove data with insignificant changes in the surrounding environment. The features of ground undulations and obstacles after clustering are extracted. These features, along with the vehicle's lateral acceleration, yaw rate, roll angle, roll rate, steering wheel angle, and longitudinal acceleration currently measured by the microelectromechanical inertial measurement unit (MEMS) sensor, are used as input to the neural network. The estimated values ​​of roll angle and roll rate at the next moment are output to predict the load transfer rate and vehicle state at the next moment. At the same time, the load transfer rate is compared with the preset threshold to realize the selection of control strategy.

2. The method for preventing overturning of unmanned operation platforms in hilly and mountainous areas based on multi-sensor fusion according to claim 1, characterized in that, Step (1) Establish the dynamic characteristics of the vehicle in the working platform vehicle dynamic model, including longitudinal movement, lateral movement and yaw direction movement.

3. The method for preventing overturning of unmanned operation platforms in hilly and mountainous areas based on multi-sensor fusion according to claim 2, characterized in that, The longitudinal motion equation is: ， The equation of lateral motion is: ， The equation of motion in the yaw direction is: ， In the formula, m is the vehicle mass. and It refers to longitudinal and lateral speeds. It is the vehicle's yaw rate. and These are the longitudinal and lateral forces on the four wheels, with subscripts ij as fl, fr, rl, and rr representing the left front wheel, right front wheel, left rear wheel, and right rear wheel, respectively; the ij used in the following equations have the same meaning; It's the steering angle of the front wheels. It is the moment of inertia about the vertical axis, a and b are the distances from the center of mass to the front and rear axles, respectively, and T is the wheelbase of the vehicle; Estimate the state changes of the unmanned vehicle during operation and establish an estimation model; calculate the state equation and measurement equation of the three-degree-of-freedom operating platform from the longitudinal motion equation, the lateral motion equation and the yaw direction motion equation. The state equation of the work platform is: ， The measurement equation is: ， Previous wheel angle and longitudinal acceleration The system input vector, i.e. ; with yaw rate , centroid side slip angle and longitudinal speed It is a state vector, that is ; with lateral acceleration The output vector, i.e. In the formula: and These are the lateral stiffness of the front and rear tires, respectively.

4. The method for preventing overturning of unmanned operation platforms in hilly and mountainous areas based on multi-sensor fusion according to claim 1, characterized in that, In step (1), the vehicle roll dynamics model is as follows: Based on the roll dynamics model, the roll dynamics characteristics of the vehicle when the wheels are not lifted are solved, and the roll dynamics characteristic equation is: ， in, It is the roll angle of the vehicle at its center of gravity. It is the vehicle's roll rate. and These are the roll centers of the front and rear suspensions, respectively. It is the roll moment of inertia about the roll axis. It is the lateral acceleration of the center of gravity. It is the height from the center of gravity to the center of roll. K is the height of the center of gravity, C is the rolling stiffness of the suspension, and C is the rolling damping of the suspension.

5. The method for preventing overturning of unmanned operation platforms in hilly and mountainous areas based on multi-sensor fusion according to claim 4, characterized in that, Based on step (1), the vehicle stability region is defined by considering the load transfer rate. The vehicle attitude is adjusted by controlling the extension and retraction of the leveling hydraulic cylinder. The load transfer rate defines the vehicle's stability region. In the stability region, the load transfer rate is less than the safety threshold, and the vehicle can drive smoothly. The load transfer rate is defined as: ， The threshold for LTR is set as follows: and .

6. The method for preventing overturning of unmanned operation platforms in hilly and mountainous areas based on multi-sensor fusion according to claim 1, characterized in that, Step 1) The methods for predicting the load transfer rate and vehicle status using neural networks include: The input layer contains information on the distance and size of obstacles, as well as the lateral acceleration, yaw rate, roll angle, roll rate, steering wheel angle, and longitudinal acceleration of the work platform currently measured by the microelectromechanical inertial measurement unit (MEMS) sensor. The output layer contains the roll angle and roll rate. There are 7 hidden layers, and the number of neurons in a single network layer is set to 100. The network training times are 1000. Except for the sigmoid function used for the output layer neurons, the Leaky ReLU function with faster convergence is used for both the input and hidden layers. L2 regularization is employed to avoid overfitting in the neural network model. The Adam optimization algorithm is used instead of the classic stochastic gradient descent method to update the network weights more effectively. The loss function is defined as: ， In the formula, m is the number of samples involved in the calculation. y is the output of the network model, and y is the output label. is the L2 regularization hyperparameter, and is the number of layers in the neural network. 2 is the norm square, which is defined as the sum of the squares of all elements in the matrix.

7. The method for preventing overturning of unmanned operation platforms in hilly and mountainous areas based on multi-sensor fusion according to claim 1, characterized in that, Step (4) The methods for selecting control strategies include: Based on the vehicle model and body sensors, an extended Kalman filter is used for information fusion to estimate the vehicle's state and provide the yaw rate to the vehicle controller. The sideslip angle β and longitudinal speed of the vehicle The state vector is used, and the load transfer rate is estimated based on the roll angle and roll rate predicted by the trained neural network model to select the control strategy. a) When the load transfer rate is less than the minimum threshold load transfer rate, it means that there will be no rollover. Considering only the yaw stability of the vehicle, the work platform can pass through the section normally. b) When the load transfer rate value is between the minimum and maximum load transfer rate, it means that the load transfer between the left and right wheels is large. If it is not controlled, it may roll over. Therefore, yaw stability and rollover prevention should be considered at the same time. Active differential braking, active suspension or semi-active suspension adjustment, and active steering angle correction anti-rollover technology should be activated in advance. c) When the load transfer rate is greater than the maximum load transfer rate, the work platform is in great danger of overturning. The road is blocked and impassable. It is necessary to control the front wheels to steer around the road and replan the route according to the work requirements to carry out forest protection work.

Images