A PINN-kalman filter cooperative road slope intelligent identification method
Patent Information
- Application Number
- CN202611024944.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-10
- Publication Date
- 2026-11-03
AI Technical Summary
Existing technologies rely on additional sensors to identify road slopes, which leads to significant errors in actual working conditions and makes it difficult to accurately estimate slope values when data is limited.
The PINN-Kalman filter collaborative method is adopted, which combines a deep learning model and a Kalman filter model to identify the slope using data from the vehicle controller area network and inertial measurement unit. The loss function of physical constraints is constructed and the Kalman filter model is used for secondary estimation.
Under conditions of sparse, low-frequency, and unlabeled data, stable slope estimation was achieved, improving estimation accuracy and stability while reducing reliance on a large number of labeled samples and data collection costs.
Smart Images
Figure CN123028561A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of slope recognition methods, and more specifically, to a PINN-Kalman filter-coordinated intelligent road slope recognition method. Background Technology
[0002] Slope information has a significant impact on the safety, economy, comfort, and intelligence of modern vehicles. In power control, the Engine Control Unit (ECU) optimizes torque output based on slope information, while the Transmission Control Unit (TCU) can intelligently select shift strategies based on the slope. In stability control, the Electronic Stability Program (ESP), by integrating slope information, can more accurately assess vehicle status and improve braking and stability control performance. In energy management, slope is a key parameter for predicting vehicle load; hybrid and electric vehicles need to decide energy allocation based on road slope to maximize energy efficiency and driving range. Furthermore, Adaptive Cruise Control (ACC) systems integrating slope information can control vehicle speed more smoothly, improving ride comfort. For Level 3 and above autonomous vehicles, slope data is a crucial input for path planning. To meet the demand for accurate road information in the development of intelligent and electrified vehicles, it is urgent to utilize existing vehicle sensors and advanced estimation algorithms to identify road slope in real time under various complex driving conditions.
[0003] In existing technologies, Li Shaohua et al. proposed a joint estimation method for vehicle mass and road slope considering the rate of change of slope. This method first establishes a 3-DOF single-track model and a longitudinal dynamics model to clarify the vehicle's forces and motion equations. Then, it uses RLS with a forgetting factor to estimate the rate of change of road slope, using this as input, and combines it with EKF to construct a joint estimation model. Parameter estimation is achieved through linearization, discretization, and iterative computation. This data fusion method can improve the accuracy of slope estimation. However, this method heavily relies on the actual forces obtained from sensors. In real-world operating conditions, some actual forces on the vehicle are difficult to obtain, leading to significant errors in practical applications. Summary of the Invention
[0004] The purpose of this invention is to overcome the shortcomings of existing technologies that rely on additional sensors to obtain vehicle parameters to identify slope, and to provide a PINN-Kalman filter-assisted intelligent road slope identification method that does not rely on additional sensors and improves the stability of estimating physically consistent slope values when data is limited.
[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for intelligent road slope recognition using PINN-Kalman filtering is provided, comprising the following steps: Historical slope data and historical driving data are collected and input into a deep learning model. The first vehicle kinematics equation is rewritten as a loss function and added to the deep learning model for training, resulting in a trained deep learning model. A Kalman filter model is constructed based on the second vehicle kinematics equation; Collect current driving data; The current driving data is input into the trained deep learning model, and the slope value output by the deep learning model is input into the Kalman filter model for filtering to obtain an accurate slope value.
[0006] In one alternative approach, the collection of historical slope data and historical driving data, input into a deep learning model, and the rewriting of the first vehicle kinematics equation as a loss function and incorporating it into the deep learning model for training, to obtain a trained deep learning model, specifically includes the following steps: Historical slope data is collected and noise is processed to obtain noise-processed historical slope data, which is then used as a label; the historical slope data is collected through a vehicle inertial measurement unit. Historical driving data is collected and used as input features; the historical driving data includes historical vehicle speed and historical longitudinal acceleration, the historical vehicle speed is collected through the vehicle controller local area network bus, and the historical longitudinal acceleration is collected through the vehicle inertial measurement unit. The features are merged with the input features into a single dataset; The dataset is divided into a training set and a test set; The deep learning model is trained using the training set, and the first vehicle kinematics equation is rewritten as a loss function and added to the deep learning model for training. The accuracy of the deep learning model is verified using the test set during or after training until a well-trained deep learning model that meets the accuracy requirements is obtained.
[0007] In one alternative approach, the first vehicle kinematics equation specifically uses the following function:
[0008] In the formula, This represents the longitudinal acceleration collected by the vehicle's inertial measurement unit; This represents the vehicle's actual longitudinal acceleration; Indicates the vehicle's pitch angle; Represents gravitational acceleration; Indicates the longitudinal slope angle.
[0009] In one alternative approach, the loss function includes physical loss and data loss.
[0010] In an alternative approach, the physical loss is obtained based on a physical residual function, which specifically uses the following function:
[0011] In the formula, Represents physical residuals; Indicates the vehicle speed at the current moment; Indicates the vehicle speed at the previous moment; Indicates time; Indicates the slope.
[0012] In an alternative approach, the loss function specifically uses the following function: .
[0013] In the formula, Represents the loss function; Indicates the total number of samples; Represents slope data; This represents the average value of the slope data.
[0014] In one alternative approach, constructing the Kalman filter model based on the second vehicle kinematics equations specifically includes the following steps: Establish the second equation of motion for the vehicle; Based on the second vehicle kinematics equation, a continuous-time differential equation is established with vehicle speed and longitudinal slope angle as state variables. The continuous-time differential equation is discretized to construct a Kalman filter model.
[0015] In an alternative approach, the second vehicle kinematics equation specifically uses the following function:
[0016] In the formula, This represents the longitudinal acceleration collected by the vehicle's inertial measurement unit; This represents the vehicle's actual longitudinal acceleration; Indicates the vehicle's pitch angle; Represents gravitational acceleration; Indicates the longitudinal slope angle.
[0017] In an alternative approach, the continuous-time differential equation specifically uses the following functions:
[0018] In the formula, Indicates the rate of change of vehicle speed; This represents the rate of change of the longitudinal slope angle.
[0019] In one alternative approach, the Kalman filter model specifically uses the following function:
[0020] In the formula, express The vehicle speed at any given moment; express The longitudinal slope angle at any given time; Indicates time; express The vehicle speed at any given moment; express The longitudinal slope angle at any given time; express The longitudinal acceleration collected by the vehicle's inertial measurement unit at a given time; express Error in calculating the time step; express The estimated slope at time; express The measurement error of time.
[0021] Compared with the prior art, the beneficial effects of the present invention are: 1. The present invention provides a PINN-Kalman filter-coordinated intelligent road slope recognition method. By acquiring historical vehicle speed, longitudinal acceleration, and slope data from the vehicle controller local area network bus and inertial measurement unit as training samples, and embedding the loss function term into the physical information neural network model based on the first vehicle kinematics equation during the training process, the physical information neural network model not only fits the data distribution during the offline learning stage, but also satisfies the vehicle dynamics constraints. This improves the model's generalization ability and estimation accuracy under conditions of sparse data or noise interference, while reducing the model's dependence on a large number of labeled samples, reducing data acquisition costs, and enabling stable estimation of physically consistent slope values even under data-limited (low-dimensional, low-frequency, unlabeled) conditions.
[0022] 2. This invention provides a PINN-Kalman filter-coordinated intelligent road slope identification method. It employs a physical information neural network (PIN) model to perform preliminary slope estimation based on current vehicle speed and longitudinal acceleration. The PIN model's estimation result is then used as the measurement input to a Kalman filter model. A secondary optimal estimation is performed using a Kalman filter model constructed from the second vehicle kinematics equation. This achieves a complementary advantage between the physically consistent output of the PIN model and the real-time dynamic smoothing capability of the Kalman filter model. This coordinated mechanism can both utilize the PIN model to capture nonlinear characteristics under complex working conditions and utilize the Kalman filter model to filter out high-frequency fluctuations and measurement noise in the PIN model's output. This results in a final slope identification result with higher accuracy and stability than a single model, enabling stable estimation of physically consistent slope values even under data constraints (low-dimensional, low-frequency, unlabeled).
[0023] The above description is merely an overview of the technical solutions of the embodiments of the present invention. In order to better understand the technical means of the embodiments of the present invention and to implement them in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the embodiments of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description
[0024] Figure 1 This is a flowchart illustrating the intelligent road slope recognition method of the present invention with PINN-Kalman filter collaboration in Embodiment 1; Figure 2 This is a flowchart illustrating the intelligent road slope recognition method of the present invention with PINN-Kalman filter collaboration in Embodiment 4; Figure 3 This is the vehicle speed change curve in Example 4; Figure 4 This is the longitudinal acceleration variation curve in Example 4; Figure 5 This refers to the historical slope data directly collected by the IMU in Example 4; Figure 6 This is the filtered historical slope data from Example 4; Figure 7 This is the predicted result of the slope estimate in Example 4; Figure 8 This is the predicted result of identifying the slope value in Example 4. Detailed Implementation
[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The present invention will be further described below with reference to specific embodiments.
[0026] Furthermore, if the embodiments of the present invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the meaning of "and / or" throughout the text is to include three parallel solutions; for example, "A and / or B" includes solution A, solution B, or a solution that simultaneously satisfies A and B.
[0027] Example 1 This embodiment is the first embodiment of a PINN-Kalman filter-coordinated intelligent road slope recognition method, such as... Figure 1 As shown, it includes the following steps: Historical slope data and historical driving data are collected and input into a deep learning model. The first vehicle kinematics equation is rewritten as a loss function and added to the deep learning model for training, resulting in a trained deep learning model. The specific steps include: Historical slope data is collected, and high-frequency noise in the historical slope data is filtered out using a low-pass filter method to obtain noise-processed historical slope data, which is then used as a label. The historical slope data is collected by a vehicle inertial measurement unit (IMU). In some embodiments, a high-precision IMU is preferably used to collect the historical slope data.
[0028] Historical driving data is collected and used as input features. This historical driving data includes historical vehicle speed and historical longitudinal acceleration. The historical vehicle speed is collected via the vehicle Controller Area Network (CAN) bus, and the historical longitudinal acceleration is collected via the vehicle inertial measurement unit. In this embodiment, only basic vehicle CAN bus data (vehicle speed and acceleration) is relied upon, without additional sensors or parameter dependencies, thus constituting low-dimensional feature learning.
[0029] The features are merged with the input features into a single dataset; The dataset is divided into a training set and a test set; The deep learning model is trained using the training set, and the first vehicle kinematics equation is rewritten as a loss function and added to the deep learning model for training; it is clearly designed under low sampling rate conditions, and the physical loss function provides key constraints when the data is sparse to ensure that the model does not "go astray".
[0030] The accuracy of the deep learning model is verified using the test set during or after training until a well-trained deep learning model that meets the accuracy requirements is obtained.
[0031] In some embodiments, the deep learning model may use a Physics-Informed Neural Network (PINN) model.
[0032] A Kalman filter model is constructed based on the second vehicle kinematics equation.
[0033] Collect current driving data. Specifically, obtain the current vehicle speed and current longitudinal slope angle based on the vehicle controller area network bus; obtain the current longitudinal acceleration based on the vehicle inertial measurement unit.
[0034] The current driving data is input into the trained deep learning model, and the deep learning model outputs preliminary slope and speed values. The initial slope and initial velocity values are input into the Kalman filter model for filtering to obtain the accurate slope value.
[0035] This embodiment presents a PINN-Kalman filter collaborative intelligent road slope recognition method. First, historical speed and acceleration data from the vehicle's CAN bus are used as input features for the model, along with historical slope data collected by the IMU as training labels. The vehicle's kinematics equations are rewritten as a loss function and added to the training loss to perform offline training of the machine learning model. This allows for model selection and optimization to obtain a suitable offline model. Next, speed and acceleration data from the target vehicle are read, and the data is filtered to eliminate noise. The PINN combined with Kalman filtering method is then used to obtain the road slope value during normal vehicle movement.
[0036] The PINN model of this invention directly outputs the slope, and its physical loss term directly constrains the physical consistency of the slope estimate. Compared with the indirect constraint method of estimating intermediate variables such as acceleration, the loss function has a more direct gradient on the network weights, the training process is more stable, and it can learn the accurate slope mapping relationship more effectively when the sensor data quality is not high or the sample is small.
[0037] The PINN model leverages its nonlinear fitting capabilities to learn preliminary slopes conforming to physical laws under complex operating conditions from historical data. Meanwhile, the back-end Kalman filter uses a kinematic model to smooth the PINN output and filter out its high-frequency fluctuations. PINN's 'direct physical constraint' training method ensures that even with noise, its output's trend closely matches physical laws, perfectly matching the model assumptions of the Kalman filter. The combination of these two approaches produces an accuracy improvement effect greater than the sum of its parts (1+1>2).
[0038] By employing a physical information neural network (PIN) model to perform preliminary slope estimation based on the current vehicle speed and longitudinal acceleration, and using the PIN model's estimation results as the measurement input to a Kalman filter (KFL) model, a secondary optimal estimation is performed using a KFL model constructed based on the second vehicle kinematics equation. This achieves a complementary advantage between the physically consistent output of the PIN model and the real-time dynamic smoothing capability of the KFL model. This collaborative mechanism can both utilize the PIN model to capture nonlinear characteristics under complex operating conditions and use the KFL model to filter out high-frequency fluctuations and measurement noise in the PIN model's output, thereby obtaining a final slope identification result with higher accuracy and stability than a single model. This enables stable estimation of physically consistent slope values even under data constraints (low-dimensional, low-frequency, unlabeled).
[0039] Example 2 This embodiment is the second embodiment of the PINN-Kalman filter-coordinated intelligent road slope recognition method. This embodiment is similar to the first embodiment, except that the first vehicle kinematics equation specifically uses the following function:
[0040] In the formula, This represents the longitudinal acceleration collected by the vehicle's inertial measurement unit; This represents the vehicle's actual longitudinal acceleration; Indicates the vehicle's pitch angle; Represents gravitational acceleration; Indicates the longitudinal slope angle.
[0041] The loss function includes physical loss and data loss.
[0042] The physical loss is obtained based on the physical residual function, which specifically uses the following function:
[0043] In the formula, Represents physical residuals; Indicates the vehicle speed at the current moment; Indicates the vehicle speed at the previous moment; Indicates time; Indicates the slope.
[0044] The loss function specifically uses the following function: .
[0045] In the formula, Represents the loss function; Indicates the total number of samples; Represents slope data; This represents the average value of the slope data.
[0046] Example 3 This embodiment is the third embodiment of the PINN-Kalman filter collaborative intelligent road slope recognition method. This embodiment is similar to the first embodiment, except that the construction of the Kalman filter model based on the second vehicle kinematics equation specifically includes the following steps: Establish the second equation of motion for the vehicle; Based on the second vehicle kinematics equation, a continuous-time differential equation is established with vehicle speed and longitudinal slope angle as state variables; The continuous-time differential equation is discretized to construct a Kalman filter model.
[0047] Considering the small pitch angle of the vehicle body and the road slope angle, the longitudinal kinematics formula of the vehicle is transformed into a second vehicle kinematics equation, which specifically uses the following function:
[0048] In the formula, This represents the longitudinal acceleration collected by the vehicle's inertial measurement unit; This represents the vehicle's actual longitudinal acceleration; Indicates the vehicle's pitch angle; Represents gravitational acceleration; Indicates the longitudinal slope angle.
[0049] The continuous-time differential equation specifically uses the following functions:
[0050] In the formula, Indicates the rate of change of vehicle speed; This represents the rate of change of the longitudinal slope angle.
[0051] The Kalman filter model specifically uses the following functions:
[0052] in, , express Prior state estimation at time t; , express The control input vector at each time step; ,express The measured value at time 1, which is the value output by the PINN model; , represents the state transition matrix; , represents the control input matrix; , representing the observation matrix; That is, the Kalman filter model is:
[0053] In the formula, express The vehicle speed at any given moment; express The longitudinal slope angle at any given time; Indicates time; express The vehicle speed at any given moment; express The longitudinal slope angle at any given time; express The longitudinal acceleration collected by the vehicle's inertial measurement unit at a given time; express Error in calculating the time step; express The estimated slope at time; express The measurement error of time.
[0054] Example 4 This embodiment is the fourth embodiment of the PINN-Kalman filter-coordinated intelligent road slope recognition method. In this embodiment, combining the specific functions in Embodiments 2 and 3, the specific flow of a PINN-Kalman filter-coordinated intelligent road slope recognition method is as follows: Figure 2 As shown in the figure. In the figure, Loss represents the loss function. ODE Loss represents physical loss. data This indicates data loss.
[0055] This represents the error covariance prediction equation. In the formula, express The prior error covariance matrix at time t; express The posterior error covariance matrix at time t; This represents the transpose of matrix A; This represents the process noise covariance matrix.
[0056] The equation for calculating Kalman gain is given. In the formula, express Kalman gain at time step; This represents the transpose of the H matrix; This represents the measurement noise covariance matrix.
[0057] Represents the state update equation, In the formula, express The posterior state estimate at time step is the identification of the slope value; express Prior state estimation at time t, that is .
[0058] The error covariance update equation is represented by... In the formula, express The posterior error covariance matrix at time t is used for the next iteration; I represents the identity matrix.
[0059] Based on this specific process, the intelligent road slope recognition method was verified.
[0060] like Figure 3 The image shows a vehicle speed change curve obtained based on the CAN bus. Figure 4 The image shows the longitudinal acceleration variation curve obtained from an IMU. Figure 5 The image shows historical slope data collected using a high-precision IMU, such as... Figure 6 The image shows the filtered historical slope data.
[0061] Based on the physical information neural network model, the slope estimate is obtained, such as... Figure 7 As shown.
[0062] Based on the slope estimate and the Kalman filter model, the identified slope value is obtained, such as... Figure 8 As shown.
[0063] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system, or other device. Furthermore, the embodiments of this invention are not directed to any particular programming language.
[0064] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. Similarly, for the sake of brevity and to aid in understanding one or more aspects of the invention, in the description of exemplary embodiments of the invention above, various features of the embodiments are sometimes grouped together in a single embodiment, figure, or description thereof. The claims, which follow the detailed description, are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of the invention.
[0065] Those skilled in the art will understand that the modules in the device of the embodiment can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiment can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components, except that at least some of such features and / or processes or units are mutually exclusive.
[0066] It should be noted that the above embodiments are illustrative of the invention and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be construed as limiting the order of execution.
Claims
1. A road slope intelligent recognition method with PINN-Kalman filter collaboration, characterized in that, Includes the following steps: Historical slope data and historical driving data are collected and input into a deep learning model. The first vehicle kinematics equation is rewritten as a loss function and added to the deep learning model for training, resulting in a trained deep learning model. A Kalman filter model is constructed based on the second vehicle kinematics equation; Collect current driving data; The current driving data is input into the trained deep learning model, and the slope value output by the deep learning model is input into the Kalman filter model for filtering to obtain an accurate slope value.
2. The intelligent road slope recognition method with PINN-Kalman filter collaboration according to claim 1, characterized in that, The process of collecting historical slope data and historical driving data, inputting them into a deep learning model, and rewriting the first vehicle kinematics equation as a loss function and adding it to the deep learning model for training, to obtain a trained deep learning model, specifically includes the following steps: Historical slope data is collected and noise is processed to obtain noise-processed historical slope data, which is then used as a label; the historical slope data is collected through a vehicle inertial measurement unit. Historical driving data is collected and used as input features; the historical driving data includes historical vehicle speed and historical longitudinal acceleration, the historical vehicle speed is collected through the vehicle controller local area network bus, and the historical longitudinal acceleration is collected through the vehicle inertial measurement unit. The features are merged with the input features into a single dataset; The dataset is divided into a training set and a test set; The deep learning model is trained using the training set, and the first vehicle kinematics equation is rewritten as a loss function and added to the deep learning model for training. The accuracy of the deep learning model is verified using the test set during or after training until a well-trained deep learning model that meets the accuracy requirements is obtained.
3. The intelligent road slope recognition method with PINN-Kalman filter collaboration according to claim 1, characterized in that, The first vehicle kinematic equation specifically uses the following functions: In the formula, This represents the longitudinal acceleration collected by the vehicle's inertial measurement unit; This represents the vehicle's actual longitudinal acceleration; Indicates the vehicle's pitch angle; Represents gravitational acceleration; Indicates the longitudinal slope angle.
4. The intelligent road slope recognition method with PINN-Kalman filter collaboration according to claim 3, characterized in that, The loss function includes physical loss and data loss.
5. The intelligent road slope recognition method with PINN-Kalman filter collaboration according to claim 4, characterized in that, The physical loss is obtained based on the physical residual function, which specifically uses the following function: In the formula, Represents physical residuals; Indicates the vehicle speed at the current moment; Indicates the vehicle speed at the previous moment; Indicates time; Indicates the slope.
6. The intelligent road slope recognition method with PINN-Kalman filter collaboration according to claim 5, characterized in that, The loss function specifically uses the following function: In the formula, Represents the loss function; Indicates the total number of samples; Represents slope data; This represents the average value of the slope data.
7. The intelligent road slope recognition method based on PINN-Kalman filtering according to any one of claims 1 to 6, characterized in that, The construction of the Kalman filter model based on the second vehicle kinematics equations specifically includes the following steps: Establish the second equation of motion for the vehicle; Based on the second vehicle kinematics equation, a continuous-time differential equation is established. The continuous-time differential equation is discretized to construct a Kalman filter model.
8. The intelligent road slope recognition method with PINN-Kalman filter collaboration according to claim 7, characterized in that, The second vehicle kinematic equation specifically uses the following functions: In the formula, This represents the longitudinal acceleration collected by the vehicle's inertial measurement unit; This represents the vehicle's actual longitudinal acceleration; Indicates the vehicle's pitch angle; Represents gravitational acceleration; Indicates the longitudinal slope angle.
9. The intelligent road slope recognition method with PINN-Kalman filter collaboration according to claim 8, characterized in that, The continuous-time differential equation specifically uses the following functions: In the formula, Indicates the rate of change of vehicle speed; This represents the rate of change of the longitudinal slope angle.
10. The intelligent road slope recognition method with PINN-Kalman filter collaboration according to claim 9, characterized in that, The Kalman filter model specifically uses the following functions: In the formula, express The vehicle speed at any given moment; express The longitudinal slope angle at any given time; Indicates time; express The vehicle speed at any given moment; express The longitudinal slope angle at any given time; express The longitudinal acceleration collected by the vehicle's inertial measurement unit at a given time; express Error in calculating the time step; express The estimated slope at time; express The measurement error of time.