Determination method of side slip angle, model training method and equipment

Abstract

The invention relates to a method for determining a side slip angle, a model training method and equipment, and relates to the technical field of vehicles, and the method comprises the steps: obtaining a first driving state parameter and an initial side slip angle of a vehicle; and inputting the first driving state parameter and the initial side slip angle into a target prediction model, so that the target prediction model predicts a target side slip angle of the vehicle. Therefore, the side slip angle of the vehicle can be accurately calculated.

Description

Technical Field

[0001] This application relates to the technical field of vehicles, in particular to the technical field of vehicle control, and specifically relates to a method for determining the sideslip angle of the center of mass, a model training method, and a device. Background Art

[0002] The sideslip angle of the center of mass is an important parameter for evaluating the driving stability of a vehicle. If the calculated sideslip angle of the center of mass is inaccurate, it may cause the driver or the vehicle control system to be unable to accurately judge the driving state of the vehicle, and further, cause the control strategy to fail or result in misoperations. Especially under high-speed driving or bad road conditions, an inaccurate sideslip angle of the center of mass may cause the vehicle to suddenly get out of control and lead to serious traffic accidents.

[0003] In a related technology, it is proposed to predict and estimate the vehicle dynamics parameters based on an improved unscented particle filter, and use the vehicle state parameters to build a back propagation (BP) neural network to fit the residuals. In another related technology, it is proposed to judge the vehicle state according to the data of the inertial measurement unit and the wheel speed odometer, and calculate an adaptive dynamic switching judgment equation in real time to estimate the vehicle's non-linear variables through the Kalman filter method.

[0004] However, the related technologies mainly estimate the sideslip angle of the center of mass of the vehicle based on the Kalman filter, resulting in poor accuracy of the calculated sideslip angle of the center of mass. Therefore, it is necessary to explore an effective way to accurately calculate the sideslip angle of the center of mass during the vehicle driving process. Summary of the Invention

[0005] This application provides a method for determining the sideslip angle of the center of mass, a model training method, and a device, so as to at least solve the technical problem that it is difficult to accurately calculate the sideslip angle of the center of mass in the related technologies. The technical solutions of this application are as follows: According to the first aspect provided by this application, a method for determining the sideslip angle of the center of mass is provided, including: obtaining the first driving state parameter and the initial sideslip angle of the vehicle; inputting the first driving state parameter and the initial sideslip angle into a target prediction model, so that the target prediction model predicts the target sideslip angle of the vehicle.

[0006] According to the above technical means, this application can use the initial sideslip angle and the first driving state parameter as the input features of the target prediction model, so as to utilize the non-linear fitting ability of the target prediction model to achieve heterogeneous fusion estimation of the sideslip angle of the center of mass, and avoid the limitations in the related technologies when calculating the sideslip angle of the center of mass only through the Kalman filter (such as the Kalman filter is based on linear calculation, while the vehicle dynamics system is essentially non-linear). Therefore, this application can accurately estimate the sideslip angle of the center of mass.

[0007] In a possible implementation manner, the first driving state parameter includes at least one of a steering wheel angle, a longitudinal vehicle speed, a yaw rate, a lateral acceleration, and a road surface adhesion coefficient at which the vehicle is currently driving.

[0008] According to the above technical means, in the present application, due to the mutual correlation or complementarity between different parameters, these parameters can be comprehensively analyzed to accurately predict the centroid sideslip angle of the vehicle. Moreover, these parameters include parameters of the vehicle under different driving conditions (such as straight driving, turning, accelerating, decelerating, etc.), enabling the prediction model to better adapt to the prediction requirements of the centroid sideslip angle under different driving conditions.

[0009] In a possible implementation manner, the initial centroid sideslip angle is obtained through the following steps: obtaining the second driving state parameter of the vehicle; there are parameters in the second driving state parameter that are different from the first driving state parameter; processing the second driving state parameter based on a Kalman filter to obtain the initial centroid sideslip angle.

[0010] According to the above technical means, in the present application, the initial centroid sideslip angle can be determined based on the Kalman filter, facilitating subsequent prediction of the target centroid sideslip angle of the vehicle according to the initial centroid sideslip angle, the first driving state parameter, and the target prediction model, realizing the heterogeneous fusion estimation of the centroid sideslip angle by the Kalman filter and the neural network, and avoiding the limitations in calculating the centroid sideslip angle only through the Kalman filter in the related art. Therefore, the present application can combine the respective advantages of the Kalman filter and the neural network to accurately estimate the centroid sideslip angle.

[0011] In a possible implementation manner, inputting the first driving state parameter and the initial centroid sideslip angle into the target prediction model to enable the target prediction model to predict the target centroid sideslip angle of the vehicle includes: extracting features from the first driving state parameter and the initial centroid sideslip angle through the target prediction model to obtain a feature vector; performing a multi-layer non-linear transformation on the feature vector through the target prediction model to determine the target centroid sideslip angle.

[0012] According to the above technical means, in the present application, the centroid sideslip angle can be estimated through the non-linear fitting ability of the target prediction model to accurately determine the centroid sideslip angle of the vehicle. Additionally, by considering the initial centroid sideslip angle (the estimated value of the centroid sideslip angle), the prediction model can more accurately capture the future sideslip trend of the vehicle.

[0013] In a possible implementation manner, the road surface adhesion coefficient is obtained through the following steps: obtaining the image information of the currently driving road surface; determining the road surface adhesion coefficient based on the image information.

[0014] According to the above technical means, the present application can determine the road surface adhesion coefficient through the image information of the road surface, avoiding the problems of low efficiency and low accuracy of the traditional equipment measurement method, and being able to determine the road surface adhesion coefficient efficiently and accurately.

[0015] In a possible implementation manner, determining the road surface adhesion coefficient based on the image information includes: recognizing the image information based on an image recognition model to obtain the road surface adhesion coefficient; the image recognition model is trained according to an image training set; the image training set includes the image information of the road surface with different road surface attributes and the corresponding road surface adhesion coefficients; the road surface attributes include at least one of material, humidity, and icing state.

[0016] According to the above technical means, the present application can quickly process the captured real-time road surface image through the image recognition model, so as to immediately determine the adhesion coefficient of the current road surface. Moreover, the image training set of the image recognition model covers the image information of different road surface attributes and the corresponding road surface adhesion coefficients, which enables the image recognition model to more comprehensively understand the road surface conditions and accurately determine the road surface adhesion coefficient.

[0017] According to the second aspect provided by the present application, a model training method is provided, including: obtaining a training sample set; the training sample set includes the centroid sideslip angle of the first sample and the input sample corresponding to the centroid sideslip angle of the first sample; the input sample includes the first sample driving state parameter and the centroid sideslip angle of the second sample; training the initial prediction model based on the training sample set to obtain the target prediction model.

[0018] According to the above technical means, the present application can obtain the training sample set and train the initial prediction model through these sample sets, enabling the initial prediction model to learn the non-linear relationship between the driving state parameter and the centroid sideslip angle of the vehicle, so as to obtain the target prediction model, and then predict the centroid sideslip angle of the vehicle through the target prediction model in the follow-up, avoiding the calculation error caused by the traditional Kalman filter relying on a linear or linearized model.

[0019] In a possible implementation manner, the centroid sideslip angle of the second sample is obtained by processing the driving state parameter of the second sample based on a Kalman filter; there are parameters in the driving state parameter of the second sample that are different from those in the driving state parameter of the first sample.

[0020] According to the above technical means, the present application can enrich the content of the training sample set by introducing the centroid sideslip angle of the second sample processed based on the Kalman filter, enabling the initial prediction model to learn the non-linear relationship between the centroid sideslip angle and different data, and realizing the heterogeneous fusion estimation of the centroid sideslip angle.

[0021] In a possible implementation manner, based on a training sample set, an initial prediction model is trained to obtain a target prediction model, including: updating the model parameters of the initial prediction model multiple times until the updated initial prediction model meets a preset condition to obtain the target prediction model; wherein, in each update process, an input sample is input into the initial prediction model to obtain a predicted centroid sideslip angle; based on the difference between the predicted centroid sideslip angle and the centroid sideslip angle of the first sample, the model parameters of the initial prediction model are updated.

[0022] According to the above technical means, the present application can update the model parameters through multiple iterations, enabling the initial prediction model to gradually learn the complex relationship between the input sample and the centroid sideslip angle of the first sample, so as to more accurately predict the centroid sideslip angle.

[0023] According to the third aspect provided by the present application, a device for determining the centroid sideslip angle is provided, including: an acquisition unit and a determination unit; the acquisition unit is used to acquire the first driving state parameter and the initial centroid sideslip angle of the vehicle; the determination unit is used to input the first driving state parameter and the initial centroid sideslip angle into the target prediction model so that the target prediction model predicts the target centroid sideslip angle of the vehicle.

[0024] In a possible implementation manner, the acquisition unit is further used to acquire the second driving state parameter of the vehicle; there are parameters in the second driving state parameter that are different from the first driving state parameter; the determination unit is further used to process the second driving state parameter based on a Kalman filter to obtain the initial centroid sideslip angle.

[0025] In a possible implementation manner, the determination unit is specifically used for: extracting features from the first driving state parameter and the initial centroid sideslip angle through the target prediction model to obtain a feature vector; performing a multi-layer non-linear transformation on the feature vector through the target prediction model to determine the target centroid sideslip angle.

[0026] In a possible implementation manner, the acquisition unit is further used to acquire the image information of the current driving road surface; the determination unit is further used to determine the road surface adhesion coefficient based on the image information.

[0027] In a possible implementation manner, the determination unit is specifically used for: identifying the image information based on an image recognition model to obtain the road surface adhesion coefficient; the image recognition model is trained according to an image training set; the image training set includes the image information of roads with different road surface attributes and the corresponding road surface adhesion coefficients; the road surface attributes include at least one of material, humidity, and icing state.

[0028] According to a fourth aspect provided by the present application, there is provided a model training device, including: an acquisition unit and a training unit; the acquisition unit is configured to obtain a training sample set; the training sample set includes the centroid sideslip angle of the first sample and an input sample corresponding to the centroid sideslip angle of the first sample; the input sample includes the first sample driving state parameter and the centroid sideslip angle of the second sample; the training unit is configured to train an initial prediction model based on the training sample set to obtain a target prediction model.

[0029] In a possible implementation manner, the training unit is specifically configured to: update the model parameters of the initial prediction model multiple times until the updated initial prediction model meets a preset condition to obtain a target prediction model; wherein, in each update process, the input sample is input into the initial prediction model to obtain a predicted centroid sideslip angle; based on the difference between the predicted centroid sideslip angle and the centroid sideslip angle of the first sample, the model parameters of the initial prediction model are updated.

[0030] According to a fifth aspect provided by the present application, there is provided a processing device, including: a processor; a memory for storing instructions executable by the processor; wherein, the processor is configured to execute the instructions to implement the method according to the first aspect or the second aspect and any possible implementation manner thereof.

[0031] According to a sixth aspect provided by the present application, there is provided a vehicle, including the processing device in the fifth aspect.

[0032] According to a seventh aspect provided by the present application, there is provided a computer-readable storage medium, when the instructions in the computer-readable storage medium are executed by the processor of the processing device, enabling the processing device to execute the method according to the first aspect and any possible implementation manner thereof.

[0033] According to an eighth aspect provided by the present application, there is provided a computer program product, the computer program product includes computer instructions, when the computer instructions run on the processing device, enabling the processing device to execute the method according to the first aspect and any possible implementation manner thereof.

[0034] It should be noted that the technical effects brought by any implementation manner in the second aspect to the eighth aspect can refer to the technical effects brought by the corresponding implementation manner in the first aspect, and will not be elaborated here.

[0035] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application, and do not constitute an improper limitation to the present application.

[0037] Figure 1 It is a schematic diagram of a system for determining the centroid slip angle shown according to an exemplary embodiment; Figure 2 It is a flowchart of a method for determining the centroid slip angle shown according to an exemplary embodiment; Figure 3 It is a schematic diagram of a mapping relationship shown according to an exemplary embodiment; Figure 4 It is a schematic flowchart of a model training method shown according to an exemplary embodiment; Figure 5 It is a schematic diagram of a process for determining the centroid slip angle shown according to an exemplary embodiment; Figure 6 It is a schematic diagram of another process for determining the centroid slip angle shown according to an exemplary embodiment; Figure 7 It is a block diagram of a device for determining the centroid slip angle shown according to an exemplary embodiment; Figure 8 It is a block diagram of a model training device shown according to an exemplary embodiment; Figure 9 It is a block diagram of a processing device shown according to an exemplary embodiment. Detailed implementation manners

[0038] In order to enable those of ordinary skill in the art to better understand the technical solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0039] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that such used data may be interchanged under appropriate circumstances so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0040] In the embodiments of the present application, words such as "exemplary", "for example", or "such as" are used to represent examples, illustrations, or explanations. Any embodiment or design solution described as "exemplary", "for example", or "such as" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary", "for example", or "such as" is intended to present related concepts in a specific manner.

[0041] First, some terms involved in the present application are explained to facilitate the understanding of those skilled in the art.

[0042] The sideslip angle of the vehicle's center of mass: The sideslip angle of the vehicle's center of mass is used to describe the lateral offset degree between the direction of the vehicle's center of mass movement and the actual driving direction of the vehicle body, that is, the angle between the instantaneous velocity direction of the vehicle's center of mass and the longitudinal axis of the vehicle body (i.e., the theoretical driving direction of the vehicle). The larger the sideslip angle of the vehicle's center of mass, the more significant the lateral offset of the vehicle's center of mass, and it may be close to the critical state of loss of control (such as understeer or oversteer).

[0043] The road surface adhesion coefficient: The road surface adhesion coefficient is a key parameter describing the friction characteristics between the tire and the road surface, which directly affects the acceleration, braking, and steering performance of the vehicle. The higher the road surface adhesion coefficient, the greater the friction force between the tire and the road surface, and the higher the vehicle handling limit. The lower the road surface adhesion coefficient, the easier the vehicle is to skid.

[0044] The Kalman filter: The Kalman filter is a recursive state estimation algorithm based on probability, which is used to extract the true state of the system from the observed data with noise. Its core idea is to dynamically correct the state estimation value through the prediction-update loop, combining the system model and sensor measurements, to achieve the optimal linear unbiased estimation.

[0045] The vehicle coordinate system: The vehicle coordinate system is a local reference system fixed on the vehicle body, which is used to describe the motion state of the vehicle, the position of components, and the force relationship. The vehicle coordinate system is the core reference framework for vehicle dynamics analysis, control system design, and autonomous driving algorithms.

[0046] The above is a simple introduction to the terms involved in the embodiments of the present application, and will not be elaborated below.

[0047] As in the background art, to solve the problem of difficult to accurately calculate the sideslip angle of the center of mass, the present application provides a method for determining the sideslip angle of the center of mass, which can obtain the driving state parameters of the vehicle and the road surface adhesion coefficient of the current driving road surface of the vehicle, and determine the target sideslip angle of the vehicle based on the first driving state parameter, the road surface adhesion coefficient, and the target prediction model, where the target prediction model is used to predict the sideslip angle of the vehicle at least according to the first driving state parameter and the road surface adhesion coefficient.

[0048] Based on this, the present application can non-linearly predict the centroid side slip angle of the vehicle through the target prediction model, avoiding the calculation errors caused by the traditional Kalman filter relying on linear or linearized models, so as to accurately calculate the centroid side slip angle of the vehicle.

[0049] Next, the technical solutions in the embodiments of the present application will be described with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.

[0050] The method for determining the centroid side slip angle provided in the embodiments of the present application can be applied to vehicles. Vehicles can also be referred to as transportation means (vehicle), mobile carriers (mobile carrier), electric vehicles (electric vehicle, EV), hybrid electric vehicles (hybrid electric vehicle, HEV), plug-in hybrid electric vehicles (plug-in hybrid electric vehicle, PHEV), fuel cell vehicles (fuel cell vehicle, FCV), autonomous vehicles (autonomous vehicle), intelligent and connected vehicles (intelligent and connected vehicle, ICV), driverless vehicles (driverless vehicle), etc.

[0051] In the embodiments of the present application, the vehicle can be a car, a sport utility vehicle (SUV), a truck, a special vehicle (such as an ambulance, a fire truck, a police car, etc.), a driverless taxi, an intelligent and connected bus, an autonomous logistics vehicle, an electric truck, or other four-wheel vehicles. In addition, this method is also applicable to various special vehicles, such as agricultural vehicles, mining vehicles, forestry vehicles, airport vehicles, port vehicles, and other four-wheel vehicles. The present application does not make specific limitations on this.

[0052] As Figure 1 shown, Figure 1 is a schematic diagram of a system for determining the centroid side slip angle shown according to an exemplary embodiment. The system for determining the centroid side slip angle can include a processing device 101, a model training device 102, and a centroid side slip angle determination device 103. Communication connections can be established between the processing device 101 and the model training device 102. Communication connections can be established between the processing device 101 and the centroid side slip angle determination device 103.

[0053] For example, it can be connected in a wired or wireless manner. Among them, the wireless manner can include Bluetooth, wireless fidelity (WIFI), communication networks, etc. The wired manner can include network cables, system buses, etc.

[0054] The model training device 102 and the centroid sideslip angle determination device 103 can be integrated into one module of the processing device 101.

[0055] The centroid sideslip angle determination device 103 can obtain the first driving state parameter and the initial centroid sideslip angle of the vehicle, and input the first driving state parameter and the initial centroid sideslip angle into the target prediction model, so that the target prediction model predicts the target centroid sideslip angle of the vehicle.

[0056] The model training device 102 can obtain a training sample set, and based on the training sample set, train the initial prediction model to obtain the target prediction model.

[0057] For ease of understanding, the following specifically introduces the centroid sideslip angle determination method and the model training method provided in this application in conjunction with the accompanying drawings.

[0058] Figure 2 is a flowchart of a centroid sideslip angle determination method shown according to an exemplary embodiment. As Figure 2 shown, the centroid sideslip angle determination method includes the following steps: S201 - S202.

[0059] S201. Obtain the first driving state parameter and the initial centroid sideslip angle of the vehicle.

[0060] Among them, the first driving state parameter can include at least one of steering wheel angle, longitudinal vehicle speed, yaw rate, lateral acceleration, and road surface adhesion coefficient.

[0061] In a possible implementation manner, the first driving state parameter is related to the eight - degree - of - freedom model of the vehicle and the expression of the vehicle centroid acceleration in the vehicle coordinate system.

[0062] In an example, the expression of the vehicle centroid acceleration in the vehicle coordinate system satisfies the following first formula and second formula:

[0063]

[0064] Among them, can be used to characterize the longitudinal acceleration of the vehicle. can be used to characterize the lateral acceleration of the vehicle. can be used to characterize the yaw rate of the vehicle. It can be used to characterize the longitudinal vehicle speed. It can be used to characterize the lateral vehicle speed. It can be used to characterize the derivative of It can be used to characterize the derivative of

[0065] It should be noted that the eight-degree-of-freedom model of the vehicle includes longitudinal motion, lateral motion, yaw motion, roll motion, rotational motion of the four wheels, and the definition of the sideslip angle of the center of mass.

[0066] In one example, according to Newton's second law and D'Alembert's principle, the dynamic equations of the motion directions in the eight-degree-of-freedom model of the vehicle satisfy the following third formula to tenth formula: Longitudinal motion:

[0067] Among them, m can be used to characterize the weight of the vehicle. It can be used to characterize the longitudinal acceleration of the vehicle. It can be used to characterize the sprung mass of the vehicle. It can be used to characterize the height from the center of mass of the sprung mass to the roll axis. It can be used to characterize the yaw angular velocity of the vehicle. It can be used to characterize the roll angle the derivative of. C D It can be used to characterize the air resistance coefficient. A D It can be used to characterize the frontal area of the vehicle. It can be used to characterize the air density. It can be used to characterize the wheel angle. It can be used to characterize the longitudinal force of the tire in its respective coordinate system, It can be used to characterize the lateral force of the tire in its respective coordinate system, , representing the left front, right front, left rear, and right rear wheels respectively. It can be used to characterize the longitudinal vehicle speed.

[0068] ‌Lateral motion:

[0069] Among them, m can be used to characterize the weight of the vehicle. It can be used to characterize the lateral acceleration of the vehicle. It can be used to characterize the sprung mass of the vehicle. It can be used to characterize the height from the center of mass of the sprung mass to the roll axis. It can be used to characterize the derivative of

[0070] Yaw motion:

[0071] Among them, can be used to characterize the derivative of the yaw angular velocity . can be used to characterize the moment of inertia of the vehicle about the x-axis of the vehicle coordinate system. can be used to characterize 's derivative. can be used to characterize the lever arm of the tire longitudinal force on the vehicle yaw moment. can be used to characterize the lever arm of the tire lateral force on the vehicle yaw moment. can be used to characterize the longitudinal force of the tire in its respective coordinate system. can be used to characterize the lateral force of the tire in its respective coordinate system. can be used to characterize the wheel angle. , representing the left front, right front, left rear, and right rear wheels respectively.

[0072] Roll motion:

[0073] Among them, is used to characterize the moment of inertia of the vehicle about the x-axis of the vehicle coordinate system. can be used to characterize the roll angle of the vehicle. can be used to characterize the roll angle 's derivative. can be used to characterize 's derivative. can be used to characterize the sprung mass of the vehicle. can be used to characterize the height from the center of mass of the sprung mass to the roll axis. can be used to characterize the yaw angular velocity 's derivative. can be used to characterize the moment of inertia of the vehicle about the z-axis of the vehicle coordinate system. can be used to characterize the roll stiffness of the vehicle front axle. can be used to characterize the roll stiffness of the vehicle rear axle. can be used to characterize the roll damping of the vehicle front axle. can be used to characterize the roll damping of the vehicle rear axle. g can be used to characterize the acceleration due to gravity. can be used to characterize the lateral acceleration of the vehicle.

[0074] Rotational motion of the four wheels:

[0075]

[0076]

[0077] Among them, can be used to characterize the wheel torque. can be used to characterize the derivative of the wheel rotational speed, where i = 1, 2, 3, 4, representing the left front, right front, left rear, and right rear wheels respectively. can be used to characterize the wheel moment of inertia. can be used to characterize the longitudinal tire-road adhesion coefficient. can be used to characterize the lateral tire-road adhesion coefficient. can be used to characterize the relative longitudinal slip ratio of the tire during longitudinal slip and sideslip. can be used to characterize the relative lateral slip ratio of the tire during longitudinal slip and sideslip. s can be used to characterize the relative total slip ratio of the tire during longitudinal slip and sideslip, and . can be used to characterize the longitudinal slip ratio of the tire. can be used to characterize the tire sideslip angle. can be used to characterize the longitudinal slip and sideslip stiffness of the tire. can be used to characterize the lateral slip and sideslip stiffness of the tire. can be used to characterize the longitudinal force of a single tire in the tire model. can be used to characterize the lateral force of a single tire in the tire model. can be used to characterize the dimensionless total tangential force. can be used to characterize the longitudinal load of the tire. can be used to characterize the longitudinal force of the tire in its respective coordinate system. R can be used to characterize the wheel rolling radius.

[0078] Optionally, the tire model can be a unified tire (UniTire) model with high accuracy and good fitting performance, or a brush model. This application does not make specific restrictions on this.

[0079] The definition of the center of mass sideslip angle is obtained by simplifying according to the small angle theorem:

[0080] Among them, can be used to characterize the center of mass sideslip angle of the vehicle. can be used to characterize the longitudinal vehicle speed. can be used to characterize the lateral speed of the vehicle.

[0081] In a possible implementation manner, the processing device can determine the mapping relationship between the center of mass sideslip angle of the vehicle and multiple vehicle driving parameters according to the first formula to the tenth formula.

[0082] Exemplarily, such asFigure 3 As shown Figure 3 is a schematic diagram of a mapping relationship shown according to an exemplary embodiment. Figure 3 The mapping relationship in The processing device can determine the wheel torque according to the seventh formula and the longitudinal force of a single tire in the tire model The mapping relationship A between them:

[0083] The processing device can determine the tire longitudinal slip ratio s, the tire sideslip angle , the tire longitudinal load , the tire-road adhesion coefficient , the lateral tire-road adhesion coefficient and the longitudinal force of a single tire in the tire model and the lateral force of a single tire in the tire model The mapping relationship B between them:

[0084] The processing device can determine the longitudinal force of a single tire in the tire model according to the third formula to the sixth formula , the lateral force of a single tire in the tire model , the wheel angle and the mapping relationship C between the longitudinal acceleration, the lateral acceleration, the yaw rate :

[0085] The processing device can determine the mapping relationship D between the longitudinal acceleration, the lateral acceleration, the yaw rate and the longitudinal speed, the lateral speed according to the first formula and the second formula:

[0086] The processing device can determine the mapping relationship E between the longitudinal speed, the lateral speed and the centroid sideslip angle according to the tenth formula:

[0087] It is easy to understand that Figure 3 There is a mapping relationship between the parameters involved in Figure 3 and the centroid sideslip angle. Among them, the wheel angle in , the lateral acceleration , the longitudinal acceleration , the lateral vehicle speed , Longitudinal load of the tire , Longitudinal force of a single tire , Lateral force of a single tire , Longitudinal slip ratio s of the tire, tire sideslip angle are parameters that are easy to obtain. Longitudinal tire-road adhesion coefficient , Lateral tire-road adhesion coefficient , Wheel torque are parameters that are difficult to obtain.

[0088] Among them, the processing device can obtain the longitudinal vehicle speed , Yaw rate , Lateral acceleration and steering wheel angle through in-vehicle sensors and high-precision positioning devices. The processing device can determine the wheel angle based on the correlation between the steering wheel angle and the wheel angle .

[0089] In a possible implementation manner, in order to determine the road adhesion coefficient in the first driving state parameter, the processing device can obtain the image information of the current driving road surface. The processing device can determine the road adhesion coefficient based on the image information. For the specific implementation manner of the processing device to determine the road adhesion coefficient, reference can be made to the following S301 - S302. Details are not described here.

[0090] In a possible implementation manner, Figure 3 the longitudinal tire-road adhesion coefficient and the lateral tire-road adhesion coefficient are parameters that are not easy to obtain, but the longitudinal tire-road adhesion coefficient , the lateral tire-road adhesion coefficient and the road adhesion coefficient have the following correlation relationship as shown in the following Eleventh formula:

[0091] Based on this, the present application can use at least one of the steering wheel angle, longitudinal vehicle speed, yaw rate, lateral acceleration, and road adhesion coefficient as the first driving state parameter for easy data acquisition and subsequent calculations.

[0092] In a possible implementation manner, the initial sideslip angle of the center of mass can be determined based on a Kalman filter, or based on a sliding mode observer, or determined according to other methods. The present application does not make specific limitations on this.

[0093] In one example, the processing device may obtain the second driving state parameter of the vehicle. The processing device may process the second driving state parameter based on a Kalman filter to obtain an initial centroid side slip angle.

[0094] Among them, there are parameters in the second driving state parameter that are different from the first driving state parameter. For example, the second driving state parameter may include at least one of longitudinal vehicle speed, lateral vehicle speed, yaw rate, longitudinal acceleration, and lateral acceleration.

[0095] Optionally, the Kalman filter may be an Extended Kalman Filter (EKF), or may also be an Unscented Kalman Filter (UKF). This application does not make specific restrictions on this.

[0096] In one possible implementation, taking the extended Kalman filter as an example, the processing device may establish an extended Kalman filter based on a three-degree-of-freedom vehicle model to calculate the initial centroid side slip angle through the extended Kalman filter.

[0097] Specifically, the processing device may build a state space equation based on the three-degree-of-freedom vehicle model, simplifying the vehicle into translational motions in the x-axis and y-axis directions and yaw motion around the z-axis.

[0098] Furthermore, the processing device may establish an extended Kalman filter based on the three-degree-of-freedom vehicle model, including a state space equation and a measurement equation.

[0099] State space equation:

[0100] Measurement equation:

[0101] The processing device may perform Taylor expansion on the non-linear state space equation to convert it into a linear model to obtain the Jacobian matrix Ft and the Jacobian matrix Ht. The Jacobian matrix Ft satisfies the following fourteenth formula, and the Jacobian matrix Ht satisfies the following fifteenth formula:

[0102]

[0103] Among them, in the twelfth formula to the fifteenth formula, m may be used to represent the weight of the vehicle. May be used to represent the moment of inertia of the vehicle about the z-axis of the vehicle coordinate system. May be used to represent the driving force of the front axle of the vehicle. May be used to represent the driving force of the rear axle of the vehicle. It can be used to characterize the longitudinal acceleration of the vehicle. It can be used to characterize the yaw rate of the vehicle. It can be used to characterize the longitudinal vehicle speed. It can be used to characterize the lateral vehicle speed. It can be used to characterize the equivalent cornering stiffness of the vehicle's front axle. It can be used to characterize the equivalent cornering stiffness of the vehicle's rear axle. It can be used to characterize the distance from the vehicle's center of mass to the front axle. It can be used to characterize the distance from the vehicle's center of mass to the rear axle. It can be used to characterize the steering angle of the vehicle's front wheels. It can be used to characterize the steering angle of the vehicle's rear wheels.

[0104] Based on the twelfth formula to the fifteenth formula, the longitudinal vehicle speed, lateral vehicle speed, and yaw rate of the vehicle can be determined as the system state vector of the extended Kalman filter, the longitudinal acceleration and lateral acceleration can be determined as the system measurement vector of the extended Kalman filter, and the wheel steering angle can be determined as the system control vector of the extended Kalman filter to realize the establishment of the extended Kalman filter.

[0105] Based on this, the processing device can determine the initial sideslip angle of the center of mass based on the Kalman filter, so as to subsequently predict the target sideslip angle of the center of mass of the vehicle according to the initial sideslip angle of the center of mass, the first driving state parameter, and the target prediction model, realizing the heterogeneous fusion estimation of the sideslip angle of the center of mass by the Kalman filter and the neural network, and avoiding the limitations when only calculating the sideslip angle of the center of mass by the Kalman filter in the related art. Therefore, the present application can combine the respective advantages of the Kalman filter and the neural network to accurately estimate the sideslip angle of the center of mass.

[0106] S202. Input the first driving state parameter and the initial sideslip angle of the center of mass into the target prediction model so that the target prediction model predicts the target sideslip angle of the vehicle.

[0107] Among them, the target prediction model is used to predict the sideslip angle of the center of mass of the vehicle at least according to the first driving state parameter.

[0108] Optionally, the target prediction model can be a neural network model or a dynamic equation model. The present application does not make specific limitations on this.

[0109] In a possible implementation manner, before deploying the target prediction model on the processing device, the processing device can train the initial prediction model to obtain the target prediction model. The specific implementation manner for the processing device to train and obtain the target prediction model can refer to S401 - S402 below.

[0110] In one embodiment, the processing device may input the first driving state parameter into the target prediction model. The target prediction model may determine the current driving condition characteristics of the vehicle based on the first driving state parameter. The target prediction model may predict the sideslip angle of the vehicle's center of mass based on the driving condition characteristics.

[0111] Exemplarily, the first driving state parameter may include: longitudinal vehicle speed of 30 meters per second (m / s), yaw rate of 5 degrees per second, lateral acceleration of 4.5 m / s 2 , steering wheel angle of 90 degrees, and road surface adhesion coefficient of the current road surface of 0.85. The target prediction model may determine that the sideslip angle of the vehicle's center of mass is 1.32 degrees based on the first driving state parameter.

[0112] In another embodiment, the processing device may input the first driving state parameter and the initial sideslip angle of the vehicle's center of mass into the target prediction model, so that the target prediction model predicts the target sideslip angle of the vehicle.

[0113] In a possible implementation manner, the processing device may perform feature extraction on the first driving state parameter and the initial sideslip angle of the vehicle's center of mass through the target prediction model, map the first driving state parameter and the initial sideslip angle of the vehicle's center of mass to a high-dimensional feature space, and obtain a feature vector containing key information. Furthermore, the processing device may perform a multi-layer non-linear transformation on the feature vector through the target prediction model to capture the complex mapping relationship between the feature vector and the target sideslip angle of the vehicle's center of mass, so as to determine the target sideslip angle of the vehicle's center of mass.

[0114] Based on the above technical solutions, the present application may use the initial sideslip angle of the vehicle's center of mass and the first driving state parameter as input features of the target prediction model, so as to utilize the non-linear fitting ability of the target prediction model to achieve heterogeneous fusion estimation of the sideslip angle of the vehicle's center of mass, and avoid the limitations in the related art when only calculating the sideslip angle of the vehicle's center of mass through a Kalman filter (such as the Kalman filter is based on linear calculation, while the vehicle dynamics system is essentially non-linear). Therefore, the present application can accurately estimate the sideslip angle of the vehicle's center of mass.

[0115] In some embodiments, in order to obtain the road surface adhesion coefficient, the method for determining the sideslip angle of the vehicle's center of mass provided in the embodiments of the present application further includes the following steps: S301 - S302.

[0116] S301. Obtain the image information of the current driving road surface.

[0117] In a possible implementation manner, an image acquisition device may be deployed on the vehicle. The image acquisition device may acquire the image information of the current driving road surface. The device for determining the sideslip angle of the vehicle's center of mass may receive the image information sent by the image acquisition device.

[0118] Optionally, the image information may be video information of the current driving road surface, or it may be photo information of the current driving road surface. This application does not make specific restrictions on this.

[0119] Optionally, the image acquisition device may be an in-vehicle camera or an in-vehicle radar. This application does not make specific restrictions on this.

[0120] S302. Determine the road surface adhesion coefficient based on the image information.

[0121] In one embodiment, the processing device may identify the image information based on an image recognition model to obtain the road surface adhesion coefficient.

[0122] Among them, the image recognition model may be obtained by training according to an image training set. The image training set may include image information of road surfaces with different road surface attributes and corresponding road surface adhesion coefficients. The road surface attributes may include at least one of material, humidity, and icing state.

[0123] Exemplarily, the material may include at least one of asphalt, concrete, soil, bricks, stones, and metal. The humidity may include at least one of dry, wet, waterlogged, and partially waterlogged. The icing state may include at least one of snow, ice-water mixture, frost, and freezing rain.

[0124] Exemplarily, as shown in Table 1, Table 1 shows the road surface adhesion coefficients corresponding to different road surface attributes.

[0125] Table 1

[0126] In a possible implementation manner, the processing device may determine the image recognition model based on the following steps.

[0127] S3001. Obtain an image training set and preprocess the image training set.

[0128] Exemplarily, the image acquisition device may collect real vehicle driving image information in multiple seasons such as summer and winter, multiple weather conditions such as sunny, cloudy, rainy, and snowy days, and multiple time periods such as day, evening, and night.

[0129] In a possible implementation manner, when the image information is video information, the preprocessing may include: splitting a video file with a long time into short-time videos at fixed intervals, and classifying the video information according to the road surface attributes. For example, the road surface information may be divided into dry asphalt road surface, wet asphalt road surface, and ice and snow road surface, so as to further classify and store the preprocessed video information based on the road surface attributes.

[0130] S3002. Augment the original image information based on the multi-modal data augmentation strategy to obtain more training samples.

[0131] Among them, the multi-modal data augmentation strategy may include translation, rotation, scaling, mirroring, and adding noise.

[0132] S3003. Use the semantic segmentation network to perform pixel-level segmentation on the image, and remove redundant information such as pedestrians and vehicles in the image information.

[0133] In a possible implementation manner, after removing the redundant information in the image information, accurate road surface feature information can be obtained, thereby improving the subsequent network's recognition ability for the image.

[0134] S3004. Construct an initial recognition model to realize the recognition and classification of the image training set.

[0135] Optionally, the initial recognition model may be constructed based on the ResNet18 convolutional neural network or based on the Transformer architecture. This application does not make specific limitations on this.

[0136] S3005. Divide the image training set into a training set and a test set according to a preset ratio, and design the model parameters of the initial recognition model.

[0137] Optionally, the preset ratio may be 8:2 or 9:1. This application does not make specific limitations on this.

[0138] Exemplarily, the model parameters of the initial recognition model may include: the total number of training rounds is 30, the learning rate adjustment step size is 6; the learning rate decay coefficient is 0.4, that is, every 6 training rounds, the learning rate is reduced to 0.4 of the original. Use the above model parameters to carry out the training work, and use the test set to verify the training effect to obtain the image recognition model.

[0139] Based on this, this application can determine the road surface adhesion coefficient through the image information of the road surface, avoiding the problems of low efficiency and low accuracy of the traditional device measurement method, and being able to determine the road surface adhesion coefficient efficiently and accurately.

[0140] In some embodiments, the processing device may also determine the road surface adhesion coefficient through the dynamic data collected in real time during the vehicle driving process.

[0141] In a possible implementation manner, the processing device can collect dynamic data during the vehicle driving in real time through multiple sensors deployed on the vehicle, for example, wheel speed, braking pressure, etc. The processing device can perform filtering processing on the collected original data to eliminate the influence of noise and outliers. The processing device can calculate the slip ratio of the wheel according to the collected dynamic data, and thus determine the road adhesion coefficient of the current road surface based on the slip ratio of the wheel and a preset mapping relationship.

[0142] Wherein, the preset mapping relationship can include the measured road adhesion coefficients corresponding to multiple measured slip ratios respectively. The preset mapping relationship can be a mapping relationship obtained by performing braking tests on the vehicle under different road surfaces (dry asphalt, wet asphalt, snow, etc.).

[0143] In some embodiments, Figure 4 is a schematic flowchart of a model training method shown according to an exemplary embodiment, as Figure 4 shown, the model training method includes the following steps: S401 - S402.

[0144] S401. Obtain a training sample set.

[0145] Wherein, the training sample set can include the centroid sideslip angle of the first sample, and the input sample corresponding to the centroid sideslip angle of the first sample. The input sample can include the first sample driving state parameters and the centroid sideslip angle of the second sample. The centroid sideslip angle of the second sample can be obtained by processing the second sample driving state parameters based on a Kalman filter. There can be parameters in the second sample driving state parameters that are different from the first sample driving state parameters.

[0146] It can be understood that the centroid sideslip angle of the first sample can be used to represent the true centroid sideslip angle corresponding to the input sample.

[0147] In a possible implementation manner, the processing device can collect the true centroid sideslip angle (the centroid sideslip angle of the first sample) of the vehicle under various working conditions (such as slalom and emergency lane change) through a high-precision inertial measurement unit, and collect the first sample driving state parameters and the second sample driving state parameters corresponding to the true centroid sideslip angle. The processing device can process the second sample driving state parameters based on a Kalman filter to obtain the centroid sideslip angle of the second sample. The processing device can determine the true centroid sideslip angle, and the first sample driving state parameters and the centroid sideslip angle of the second sample corresponding to the true centroid sideslip angle as the training sample set.

[0148] Optionally, the processing device can perform preprocessing on the training data set. For example, perform moving average filtering on parameters such as vehicle speed and yaw rate, and perform normalization processing on the numerical features of the input sample. This application does not make specific limitations on this.

[0149] S402. Train the initial prediction model based on the training sample set to obtain the target prediction model.

[0150] Optionally, the architecture of the initial prediction model can be set according to actual needs. For example, the initial prediction model can be a neural network architecture or a decision tree model architecture. This application does not make specific limitations in this regard.

[0151] Exemplarily, the architecture of the initial prediction model can be as follows: The first layer: input layer, with 6 input feature quantities, used to receive serialized input data.

[0152] Among them, the input features include: road surface adhesion coefficient, steering wheel angle, longitudinal vehicle speed, yaw rate, lateral acceleration, and initial centroid side slip angle.

[0153] In a possible implementation manner, the first layer can map the input features into a tensor form that the model can process, providing initial data input for subsequent calculations and ensuring that the data format and dimensions meet the requirements of subsequent processing.

[0154] The second layer: Long Short-Term Memory (LSTM) layer 1. LSTM layer 1 can be composed of 20 neurons, used to process time series data and extract the time dependencies in the features, and output the processed features to the fully connected layer 1, providing a basis for subsequent non-linear transformation and feature learning.

[0155] In an example, when processing time series data, LSTM layer 1 can consider the change of the initial centroid side slip angle over time and the time dependencies between the initial centroid side slip angle and other features. For example, the combination of the initial centroid side slip angle and the longitudinal vehicle speed at different times may contain important information about the change of the vehicle motion state. LSTM layer 1 extracts these time-dependent features through its internal memory units and gating mechanisms, providing a basis for the non-linear transformation of the subsequent fully connected layer.

[0156] The third layer: fully connected layer 1. Fully connected layer 1 can be composed of 20 neurons.

[0157] In a possible implementation, the fully connected layer 1 can be specifically used to perform a non-linear transformation on the output of the LSTM layer 1, so as to output a feature matrix based on the feature combination between different features, thereby mapping time-dependent features to a high-dimensional space and enhancing the feature expression ability (for example, combining the non-linear relationship between the road surface adhesion coefficient and the longitudinal vehicle speed, and the non-linear relationship between the road surface adhesion coefficient and the longitudinal vehicle speed can be used to characterize the non-linear influence of the road surface adhesion coefficient on the longitudinal vehicle speed). The fully connected layer 1 can output the feature matrix to the LSTM layer 2 to provide a high-dimensional feature representation for capturing long-term time dependencies.

[0158] In a possible implementation, the fully connected layer 1 can determine the complex relationships between input features through dense connections between neurons. Each neuron can receive all features from the previous layer and determine the importance of different feature combinations through weight parameters.

[0159] In one example, when it is detected that a sharp steering wheel turn (such as 90 degrees) and a low road surface adhesion coefficient occur simultaneously, the neuron can activate "skidding warning", enhance the weight of the product term of the two features of the steering wheel angle and the road surface adhesion coefficient, and anticipate the possible centroid offset in advance.

[0160] In another example, for the dynamic delay caused by vehicle speed changes, the fully connected layer 1 can determine the response lag at different speeds. For example, at 80 km / h, the yaw rate signal can be processed in advance to simulate the physical delay of the actual vehicle steering.

[0161] In another example, the fully connected layer 1 can determine the complex relationships between the initial centroid side slip angle and other input features through dense connections between neurons, and determine the importance of different feature combinations through weight parameters, so as to provide a high-dimensional feature representation for capturing long-term time dependencies.

[0162] The fourth layer: LSTM layer 2. The LSTM layer 2 can be composed of 30 neurons. The LSTM layer 2 can be used to process time series data of the feature matrix output by the fully connected layer 1 to capture the long-term time dependencies between different features (such as the cumulative influence of longitudinal vehicle speed changes on the centroid side slip angle). The LSTM layer 2 can output the processed features to the fully connected layer 2 to provide a basis for extracting key features.

[0163] In a possible implementation, the LSTM layer 2 can control the flow of information through the input gate, forget gate, and output gate, so as to capture the long-term trends and cumulative effects in the time series.

[0164] In one example, when the longitudinal vehicle speed changes, the input gate can decide whether to add the vehicle speed information to the memory unit according to the current vehicle speed and the hidden state at the previous moment. Additionally, as time goes by, if the longitudinal vehicle speed continues to change, the forget gate can gradually adjust the historical information in the memory unit according to the vehicle speed change and other feature information. The output gate can output the cumulative effect of the longitudinal vehicle speed change on the centroid side slip angle.

[0165] In another example, the LSTM layer 2 can control the information flow through the input gate, the forget gate, and the output gate. During the processing, it can adjust the historical information related to the initial centroid side slip angle according to other input features, so as to capture the features that have a long-term impact on the vehicle motion state, such as the cumulative effect of the longitudinal vehicle speed change on the initial centroid side slip angle.

[0166] The fifth layer: the fully connected layer 2. The fully connected layer 2 can be composed of 6 neurons and has the same function as the fully connected layer 1. The fully connected layer 2 can reduce the dimension of the output of the LSTM layer 2, extract key features to compress the feature dimension, and retain the core information strongly related to the centroid side slip angle (such as the coupling relationship between the yaw rate and the lateral acceleration, and the coupling relationship between the yaw rate and the lateral acceleration can be used to characterize the influence of the combination of the yaw rate and the lateral acceleration on the centroid side slip angle). The fully connected layer 2 can output the core information strongly related to the centroid side slip angle to the activation layer to provide input for subsequent stable training.

[0167] In a possible implementation manner, the fully connected layer 2 can combine different features into a new feature representation through weighted summation and non-linear activation.

[0168] In one example, when the yaw rate is relatively large and the lateral acceleration is also relatively large, after weighted summation and non-linear transformation, the output of the neurons in the fully connected layer 2 increases, indicating the influence of this coupling situation on the centroid side slip angle.

[0169] In another example, the fully connected layer 2 can further refine the information related to the initial centroid side slip angle and retain the core information strongly related to the centroid side slip angle. For example, the influence of the coupling relationship between the yaw rate and the lateral acceleration on the centroid side slip angle may also have a certain correlation with the initial centroid side slip angle. The fully connected layer 2 presents this correlation in the form of a new feature representation through weighted summation and non-linear activation, providing key input for the subsequent activation layer and the final prediction.

[0170] Sixth layer: Activation layer. The hyperbolic tangent function is used as the activation function to compress the output of the fully connected layer 2 to (-1, 1), enhancing the stability of the model output. That is, the small-value fluctuations of the input features are restricted within a stable range to reduce the impact of fluctuations on the model output (such as dealing with the small-value fluctuations of the road surface adhesion coefficient). The activation layer can output the processed features to the fully connected layer 3 to provide stable input for the final prediction.

[0171] In one example, the eigenvalue related to the road surface adhesion coefficient output by the fully connected layer 2 is x. After passing through the activation layer, the output is y = tanh(x). When there are small-value fluctuations in the road surface adhesion coefficient, for example, fluctuating from 0.2 to 0.21, due to the non-linear characteristics of the tanh function, such fluctuations can be compressed after passing through the tanh function. For example, tanh(0.2) ≈ 0.1974, tanh(0.21) ≈ 0.2057, and the fluctuation amplitude is relatively reduced.

[0172] In another example, when the eigenvalue related to the initial centroid side slip angle passes through the activation layer, its small-value fluctuations are restricted within the range of (-1, 1), reducing the impact of fluctuations on the model output, making the influence of the information related to the initial centroid side slip angle on the final prediction result (centroid side slip angle) more stable, and improving the overall stability and accuracy of the model.

[0173] It can be understood that by restricting the small-value fluctuations of the input features within a stable range, the impact of the small-value fluctuations of the features on the final prediction result (centroid side slip angle) in subsequent calculations can be reduced, thereby improving the stability of the model output.

[0174] Seventh layer: Fully connected layer 3. It synthesizes the learned features generated from the first layer to the sixth layer and outputs the prediction result to map the high-dimensional features to the target space and provide input for the final prediction.

[0175] Eighth layer: Output layer. It is used to output the final prediction result of the model, that is, the centroid side slip angle of the vehicle, and calculate the loss function for subsequent model performance evaluation.

[0176] In one possible implementation, the model can set the upper and lower boundaries of the model output, that is, the output range, according to the value of the initial centroid side slip angle to prevent the model from overfitting the noise and abnormal patterns in the training data and improve the generalization ability of the model.

[0177] In one possible implementation, the processing device can update the model parameters of the initial prediction model multiple times until the updated initial prediction model meets the preset conditions to obtain the target prediction model.

[0178] Among them, in the process of each update, the processing device can input the input sample into the initial prediction model to obtain the predicted centroid sideslip angle, and update the model parameters of the initial prediction model based on the difference between the predicted centroid sideslip angle and the centroid sideslip angle of the first sample.

[0179] Optionally, the preset condition can be set according to actual needs. The preset condition can be that the loss function converges, or the number of update iterations reaches the maximum number of iterations. This application does not make specific limitations on this.

[0180] Optionally, the maximum number of iterations can be set according to actual needs.

[0181] It can be understood that the convergence of the loss function includes that the difference between the predicted centroid sideslip angle and the centroid sideslip angle of the first sample is reduced below the difference threshold, or the loss change in consecutive multiple iterations is less than the change threshold.

[0182] Based on this, this application can obtain a training sample set and train the initial prediction model through these sample sets, so that the initial prediction model can learn the non-linear relationship between the driving state parameters and the centroid sideslip angle of the vehicle, in order to obtain the target prediction model, thereby predicting the centroid sideslip angle of the vehicle through the target prediction model in the subsequent process, and avoiding the calculation error caused by the traditional Kalman filter relying on a linear or linearized model.

[0183] In some embodiments, as Figure 5 shown, Figure 5 is a schematic diagram of a determination process of the centroid sideslip angle shown according to an exemplary embodiment.

[0184] S1. Estimate the road surface adhesion coefficient based on a convolutional neural network.

[0185] S11. Data collection and preprocessing.

[0186] S12. Data processing.

[0187] S13. Select and construct an initial recognition model.

[0188] S14. Train the initial recognition model to obtain an image recognition model.

[0189] S15. Based on the image recognition model, determine the road surface adhesion coefficient of the driving road surface.

[0190] S2. Establish an extended Kalman filter based on a three-degree-of-freedom vehicle model to estimate the centroid sideslip angle of the vehicle through the extended Kalman filter.

[0191] S21. Establish a three-degree-of-freedom vehicle model.

[0192] S22. Establish an extended Kalman filter based on the three-degree-of-freedom vehicle model.

[0193] S23. Estimate the sideslip angle of the vehicle's center of mass based on the extended Kalman filter to obtain the initial sideslip angle of the center of mass.

[0194] S3. Calculate the sideslip angle of the vehicle's center of mass based on the heterogeneous fusion estimator of the target prediction model.

[0195] S31. Determine the input features based on the eight-degree-of-freedom vehicle model.

[0196] Combine Figure 5 , such as Figure 6 shown, Figure 6 is a schematic diagram showing another determination process of the sideslip angle of the center of mass according to an exemplary embodiment.

[0197] In a possible implementation, the input features may include: road surface adhesion coefficient, initial sideslip angle of the center of mass, steering wheel angle, longitudinal vehicle speed, yaw rate, and lateral acceleration.

[0198] Among them, the road surface adhesion coefficient may be determined based on an image recognition model. The initial sideslip angle of the center of mass may be determined based on the extended Kalman filter. The steering wheel angle, longitudinal vehicle speed, yaw rate, and lateral acceleration may be determined based on in-vehicle sensors and high-precision positioning devices.

[0199] S32. Construct a heterogeneous fusion estimator of the target prediction model based on a neural network.

[0200] In a possible implementation, the structure of the heterogeneous fusion estimator of the target prediction model may include: Input layer: The number of input features is 6.

[0201] LSTM layer 1: The number of neurons is 20.

[0202] Fully connected layer 1: The number of neurons is 20.

[0203] LSTM layer 2: The number of neurons is 30.

[0204] Fully connected layer 2: The number of neurons is 6.

[0205] Activation layer: Hyperbolic tangent function.

[0206] Fully connected layer 3: The number of neurons is 1.

[0207] Output layer: The number of neurons is 1.

[0208] In a possible implementation, the functions of different layers in the heterogeneous fusion estimator of the target prediction model can refer to the examples in S402 and will not be elaborated here.

[0209] S33. Output the target centroid sideslip angle.

[0210] Combine Figure 5 , in a possible implementation, input the steering wheel angle, longitudinal vehicle speed, yaw rate, lateral acceleration, initial centroid sideslip angle, and vehicle driving state parameters into the target prediction model, so that the target prediction model predicts the target centroid sideslip angle.

[0211] The above mainly introduces the solution provided by the embodiments of the present application from the perspective of methods. To implement the above functions, the centroid sideslip angle determination device, model training device, or processing device includes the corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should easily realize that, combined with the units and algorithm steps of each example described in the embodiments disclosed in this article, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a certain function is executed in the way of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0212] The embodiments of the present application can, according to the above method, exemplarily divide the function modules of the centroid sideslip angle determination device, model training device, or processing device. For example, the centroid sideslip angle determination device, model training device, or processing device may include each function module corresponding to each function division, or two or more functions may be integrated into one processing module. The above integrated modules can be implemented in the form of hardware or in the form of software function modules. It should be noted that the division of modules in the embodiments of the present application is illustrative, only a logical function division, and there may be other division methods in actual implementation.

[0213] Figure 7 is a block diagram of a centroid sideslip angle determination device shown according to an exemplary embodiment. Refer to Figure 7 , the centroid sideslip angle determination device includes: an acquisition unit 501 and a determination unit 502.

[0214] In a possible implementation, the acquisition unit 501 is used to acquire the first driving state parameter and the initial centroid sideslip angle of the vehicle.

[0215] In a possible implementation, the determination unit 502 is used to input the first driving state parameter and the initial centroid sideslip angle into the target prediction model, so that the target prediction model predicts the target centroid sideslip angle of the vehicle.

[0216] In a possible implementation, the acquisition unit 501 is further used to acquire the second driving state parameter of the vehicle.

[0217] In a possible implementation, the determining unit 502 is further configured to process the second driving state parameter based on a Kalman filter to obtain an initial sideslip angle of the centroid.

[0218] In a possible implementation, the determining unit 502 is specifically configured to: extract features from the first driving state parameter and the initial sideslip angle of the centroid through a target prediction model to obtain a feature vector. Perform a multi-layer non-linear transformation on the feature vector through the target prediction model to determine the target sideslip angle of the centroid.

[0219] In a possible implementation, the obtaining unit 501 is further configured to obtain image information of the current driving road surface.

[0220] In a possible implementation, the determining unit 502 is further configured to determine a road surface adhesion coefficient based on the image information.

[0221] In a possible implementation, the determining unit 502 is specifically configured to: identify the image information through an image recognition model to obtain a road surface adhesion coefficient.

[0222] Figure 8 is a block diagram of a model training device shown according to an exemplary embodiment. Refer to Figure 8 , the model training device includes: a collection unit 601 and a training unit 602.

[0223] In a possible implementation, the collection unit 601 is configured to obtain a training sample set.

[0224] In a possible implementation, the training unit 602 is configured to train an initial prediction model based on the training sample set to obtain a target prediction model.

[0225] In a possible implementation, the training unit 602 is specifically configured to: update the model parameters of the initial prediction model multiple times until the updated initial prediction model meets a preset condition to obtain a target prediction model.

[0226] Regarding the device in the above embodiments, the specific manner in which each module performs operations has been described in detail in the embodiments related to the method, and will not be elaborated here.

[0227] Figure 9 is a block diagram of a processing device shown according to an exemplary embodiment. As Figure 9 shown, the processing device includes but is not limited to: a processor 701 and a memory 702.

[0228] Among them, the above-mentioned memory 702 is used to store the executable instructions of the above-mentioned processor 701. It can be understood that the above-mentioned processor 701 is configured to execute instructions to implement the model training method and the SOC estimation method in the above-mentioned embodiments.

[0229] It should be noted that those skilled in the art can understand that Figure 9 the structure of the processing device shown in does not constitute a limitation on the processing device. The processing device may include more or fewer components than Figure 9 shown, or combine certain components, or have different component arrangements.

[0230] The processor 701 is the control center of the processing device, connecting various parts of the entire processing device through various interfaces and lines. By running or executing the software programs and / or modules stored in the memory 702, and calling the data stored in the memory 702, it executes various functions of the processing device and processes data, thereby monitoring the entire processing device. The processor 701 may include one or more processing units. Optionally, the processor 701 may integrate an application processor and a modem processor. Among them, the application processor mainly processes the operating system, user interface, application programs, etc., and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor may not be integrated into the processor 701 either.

[0231] The memory 702 can be used to store software programs and various data. The memory 702 may mainly include a program storage area and a data storage area. Among them, the program storage area may store the operating system, application programs required by at least one functional module (such as the determination unit, the processing unit, etc.). In addition, the memory 702 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0232] In an exemplary embodiment, there is also provided a computer-readable storage medium including instructions, such as the memory 702 including instructions. The above-mentioned instructions can be executed by the processor 701 of the processing device to implement the method in the above-mentioned embodiments.

[0233] In actual implementation, Figure 7 the functions of the acquisition unit 501, the determination unit 502 in, and Figure 8 the functions of the acquisition unit 601, the training unit 602 in can all be implemented by Figure 9 the processor 701 in calling the computer program stored in the memory 702. The specific execution process can refer to the description of the method part in the above embodiments, and will not be elaborated here.

[0234] Optionally, the computer-readable storage medium may be a non-transitory computer-readable storage medium. For example, the non-transitory computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, magnetic tape, a floppy disk, and an optical data storage device, etc.

[0235] In an exemplary embodiment, the embodiments of the present application further provide a computer program product including one or more instructions, and the one or more instructions may be executed by a processor 701 of a processing device to implement the method in the above embodiments.

[0236] It should be noted that when the instructions in the above computer-readable storage medium or the one or more instructions in the computer program product are executed by the processor of the processing device, each process of the above method embodiments is implemented, and the same technical effects as the above method can be achieved. To avoid repetition, details are not described here again.

[0237] Through the description of the above embodiments, those skilled in the art can clearly understand that for the convenience and brevity of description, only the above division of each functional module is used as an example. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.

[0238] In several embodiments provided by the present application, it should be understood that the disclosed device and method can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other may be through some interfaces. The indirect coupling or communication connection of the device or unit may be in an electrical, mechanical or other form.

[0239] The units described as separate components may or may not be physically separated. The components displayed as units may be a physical unit or multiple physical units, that is, they may be located in one place, or may be distributed to multiple different places. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0240] In addition, in each embodiment of the present application, each functional unit may be integrated into one processing unit, may exist separately physically for each unit, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0241] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a readable storage medium. Based on such an understanding, the technical solution of the embodiments of the present application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, may be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions for causing a device (which may be a single-chip microcomputer, a chip, etc.) or a processor to execute all or part of the steps of the methods of the various embodiments of the present application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, ROM, RAM, magnetic disks, or optical discs that can store program codes.

[0242] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for determining a sideslip angle of a center of mass, characterized in that: The method comprises: Obtaining a first driving state parameter and an initial center-of-mass sideslip angle of the vehicle; The first driving state parameter and the initial center-of-mass sideslip angle are input into a target prediction model so that the target prediction model predicts a target center-of-mass sideslip angle of the vehicle.

2. The method according to claim 1, characterized in that The first driving state parameter includes at least one of a steering wheel angle, a longitudinal vehicle speed, a yaw angular velocity, a lateral acceleration, and a road adhesion coefficient on which the vehicle is currently driving.

3. The method according to claim 2, characterized in that The initial center of mass sideslip angle is obtained by: Acquire a second driving state parameter of the vehicle; wherein the second driving state parameter has a parameter different from the first driving state parameter; The second driving state parameter is processed based on a Kalman filter to obtain an initial center of mass sideslip angle.

4. The method according to claim 1, characterized in that: The step of inputting the first driving state parameter and the initial center-of-mass sideslip angle into the target prediction model so that the target prediction model predicts the target center-of-mass sideslip angle of the vehicle comprises: Extracting features of the first driving state parameter and the initial center of mass sideslip angle by using the target prediction model to obtain a feature vector; The target prediction model is used to perform multi-layer nonlinear transformation on the feature vector to determine the target center of mass sideslip angle.

5. The method according to claim 2, characterized in that: The road adhesion coefficient is obtained by: Obtain image information of the current driving road surface; Based on the image information, the road adhesion coefficient is determined.

6. The method according to claim 5, characterized in that The determining the road adhesion coefficient based on the image information includes: The image information is recognized based on an image recognition model to obtain the road adhesion coefficient; the image recognition model is trained based on an image training set; the image training set includes image information of roads with different road properties and corresponding road adhesion coefficients; the road properties include at least one of material, humidity and icing state.

7. A model training method, characterized in that: The method comprises: Acquire a training sample set; the training sample set includes a first sample center of mass sideslip angle, and an input sample corresponding to the first sample center of mass sideslip angle; the input sample includes a first sample driving state parameter and a second sample center of mass sideslip angle; Based on the training sample set, the initial prediction model is trained to obtain the target prediction model.

8. The method according to claim 7, characterized in that The second sample center of mass sideslip angle is obtained by processing the second sample driving state parameters based on the Kalman filter; the second sample driving state parameters contain parameters different from the first sample driving state parameters.

9. The method according to claim 7 or 8, characterized in that: The step of training the initial prediction model based on the training sample set to obtain the target prediction model includes: The model parameters of the initial prediction model are updated multiple times until the updated initial prediction model meets the preset conditions, thereby obtaining the target prediction model; Wherein, during each updating process, the input sample is input into the initial prediction model to obtain a predicted sideslip angle of the center of mass; and based on the difference between the predicted sideslip angle of the center of mass and the first sample sideslip angle of the center of mass, the model parameters of the initial prediction model are updated.

10. A device for determining a sideslip angle of a center of mass, characterized in that: The device comprises: an acquisition unit and a determination unit; The acquisition unit is used to acquire a first driving state parameter and an initial center of mass sideslip angle of the vehicle; The determination unit is used to input the first driving state parameter and the initial center of mass sideslip angle into a target prediction model so that the target prediction model predicts the target center of mass sideslip angle of the vehicle.

11. A model training device, characterized in that: include: Acquisition unit and training unit; The acquisition unit is used to obtain a training sample set; the training sample set includes a first sample center of mass sideslip angle, and an input sample corresponding to the first sample center of mass sideslip angle; the input sample includes a first sample driving state parameter and a second sample center of mass sideslip angle; The training unit is used to train the initial prediction model based on the training sample set to obtain a target prediction model.

12. A processing device, characterized in that: include: processor; a memory for storing instructions executable by the processor; Wherein, the processor is configured to execute the instructions to implement the method for determining the center of mass sideslip angle as described in any one of claims 1-6, or to implement the model training method as described in any one of claims 7-9.

13. A vehicle, characterized in that: Comprising the processing device as claimed in claim 12.

Images