Clutch pedal-based driving state visual prediction system

Abstract

The invention belongs to the technical field of driving state prediction, and particularly discloses a driving state visual prediction system based on a clutch pedal, which comprises a signal acquisition module, a variational mode decomposition noise reduction module, a driving intention module and a driving state prediction module. According to the scheme, a variational mode decomposition noise reduction method is creatively provided, a variational mode decomposition technology combined with a soft threshold value is adopted, the displacement depth and action strength of a driver on a pedal are accurately measured, and the accuracy of driving state prediction is improved; a driving state prediction method is creatively provided, the driving intention and the pedal motion state of a driver are considered, continuous time series data are processed by adopting a long-short-term memory network combined with graph convolution, the driver is assisted in driving in a visual mode, the vehicle passing efficiency, passing safety and driving state prediction precision are effectively improved, and the driving safety is improved. And the development of intelligent transportation is assisted and promoted.

Description

Technical Field

[0001] The invention relates to the technical field of driving state prediction, and in particular to a driving state visualization prediction system based on a clutch pedal. Background Art

[0002] The clutch pedal is a key component for controlling power transmission in manual transmission vehicles. In addition to driver fatigue and driving intention, the pedal state is also a major factor affecting the driving state. The driving state visualization prediction system based on the clutch pedal refers to a system that uses machine learning technology to predict the driving state.

[0003] Among the existing approximate solutions, for example, CN107742193B is a driving risk prediction method based on a time-varying state transition probability Markov chain. This solution addresses the technical problem that the state transition probability estimation method based on the transition frequency in the application of the traditional Markov chain model has defects. It uses input data such as vehicle-to-vehicle data, driver steering wheel data, and driver attention status, and adopts a Markov chain to establish a multinomial regression model of driving risk state transition under different driving risk states to predict the future driving risk of the car. It realizes the estimation of the n-step time-varying state transition probability of the Markov chain, reflects the characteristics that the driving risk state transition probability changes with the characteristics of the traffic system, and meets the technical effect of the real-time warning requirements. However, there is a technical problem that the driver's behavior is ultimately transmitted to the driving state through mechanical structures such as the shift handle and the clutch pedal. Therefore, it is not accurate enough to rely directly on the driver's behavior characteristics, and lacks the ability to grasp the time dynamics in the driving behavior and capture long-term dependencies;

[0004] In addition, for example, CN115171412B discloses a method, system and device for displaying vehicle driving status. This solution addresses the technical problem that the visual screen cannot display the vehicle roadside information in real time and the vehicle status display result is not accurate. This solution achieves the technical effect of improving the accuracy of the vehicle status display result by fusing the camera and the roadside radar to confirm road obstacles, taking into account traffic lights and abnormal road events, and fusing the driving information, map information and contact information. The fusion result is displayed visually, but there is a technical problem that it does not consider the impact of the driving data of the driver on the vehicle driving status. Summary of the invention

[0005] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a driving state visualization prediction system based on a clutch pedal. In view of the technical problem that the existing schemes that directly rely on the driver's behavior characteristics are not accurate enough and lack the ability to grasp the temporal dynamics and capture long-term dependencies in driving behavior, this scheme creatively proposes a variational mode decomposition denoising method, and adopts the variational mode decomposition technology combined with soft threshold to more accurately measure the displacement depth and action intensity of the driver's pedal, thereby improving the accuracy of driving state prediction; in view of the technical problem that the existing schemes do not consider the impact of the driving data of the driver in the vehicle on the driving state of the vehicle, this scheme creatively proposes a driving state prediction method, considering the driver's driving intention and the pedal movement state, and adopts a long short-term memory network combined with graph convolution to process continuous time series data, and assists the driver in driving in a visual way, effectively improving the vehicle's traffic efficiency, traffic safety and the prediction accuracy of the driving state, and helping to promote the development of intelligent transportation.

[0006] The technical solution adopted by the present invention is as follows: The present invention provides a driving state visualization prediction system based on a clutch pedal, and the driving state visualization prediction system based on a clutch pedal includes a signal acquisition module, a variational mode decomposition noise reduction module, a driving intention module and a driving state prediction module;

[0007] The signal acquisition module includes a pedal signal acquisition device and an on-board sensor; the pedal signal acquisition device collects electrical signals from the clutch pedal, the brake pedal and the accelerator pedal respectively, and the electrical signals are used to characterize the displacement and pedaling intensity of the pedals; the on-board sensor is used to collect on-board sensor data stored in the form of time series;

[0008] The variational mode decomposition denoising module uses a variational mode decomposition denoising method to perform denoising processing on the electrical signal;

[0009] The driving intention module uses a hidden Markov model to identify the driver's driving intention;

[0010] The driving state prediction module adopts a driving state prediction method, receives and processes the noise-reduced electrical signal, vehicle-mounted sensor data and driving intention, outputs driving state prediction data and visualizes it.

[0011] The variational mode decomposition denoising module adopts a variational mode decomposition denoising method. The variational mode decomposition denoising method has the following specific operations:

[0012] Step S1: signal decomposition, specifically, filtering the electrical signal, and decomposing the filtered electrical signal into multiple sub-signals using variational mode decomposition technology;

[0013] Step S2: Optimizing the sub-signal, specifically, converting the sub-signal into a frequency domain analytical signal through Hilbert transform, derivatizing the spectrum of the frequency domain analytical signal to obtain the signal gradient, minimizing the Euclidean norm of the signal gradients of all sub-signals with the constraint that the sum of all sub-signals must be equal to the electrical signal, recording the minimization task as a constrained optimization problem, converting the constrained optimization problem into an unconstrained optimization problem using exponential Lagrange multipliers, solving the unconstrained optimization problem using an alternating direction multiplier method, and obtaining the frequency domain representation and center frequency of the optimized sub-signal;

[0014] Step S3: noise reduction processing, specifically, noise reduction is performed on the sub-signal, and the formula used is as follows: ; ;

[0015] In the formula, represents the distinction threshold, represents the noise level of the sub-signal, Indicates the length of the signal, represents the sub-signal after noise reduction, represents the symbolic function, Represents a sub-signal;

[0016] Step S4: Signal reconstruction, specifically, recombining the sub-signals according to the frequency domain representation and the center frequency to obtain a noise-reduced electrical signal.

[0017] The driving state prediction module adopts a driving state prediction method, and the driving state prediction method specifically operates as follows:

[0018] Step M1: Curve drawing, specifically, calculating the acceleration threshold, and establishing a two-dimensional coordinate plane with velocity and acceleration as the axis according to the electrical signal after noise reduction. The acceleration threshold calculation formula used is as follows: ;

[0019] In the formula, represents the acceleration threshold, represents the acceleration scaling factor, Indicates the instantaneous driving speed, , and represents the empirical fitting parameters;

[0020] Step M2: safe driving zoning, specifically, drawing an acceleration threshold curve according to the acceleration threshold, dividing the two-dimensional coordinate plane into a safe zone and an unsafe zone;

[0021] Step M3: construct a sensor relationship graph. Specifically, according to the vehicle sensor data, among all the vehicle sensors, quantify the linear correlation between every two vehicle sensors and construct a sensor weighted adjacency matrix. The calculation formula of the sensor weighted adjacency matrix is ​​as follows: ;

[0022] In the formula, Indicates that the coordinates in the sensor weighted adjacency matrix are The edge weights of represents the linear correlation number, Indicates Vehicle sensor data, Indicates Vehicle sensor data, represents the covariance, and Respectively and The standard deviation of

[0023] Step M4: Parameterization is used to reduce the computational complexity and improve the efficiency of graph convolution operations. Specifically, a graph convolution filter is used to parameterize the graph convolution filter through Chebyshev polynomials. The formula used is as follows: ;

[0024] In the formula, represents the eigenvalue diagonal matrix, represents a graph convolution filter that operates directly in the frequency domain of the graph on the diagonal matrix of the eigenvalues ​​of the Laplacian matrix, represents the normalized eigenvalue diagonal matrix, represents the kth order Chebyshev polynomial, represents the learnable parameters of Chebyshev polynomials, represents the total order of the polynomial, and k represents the order subscript index;

[0025] Step M5: Graph convolution, which is used to extract spatial features using graph convolution operations. Specifically, the vehicle-mounted sensors are used as nodes, and the relationships between the vehicle-mounted sensors are used as edges. A weighted graph is constructed by combining the nodes, edges, and sensor weighted adjacency matrices. The weighted graph and the two-dimensional coordinate plane are used as input graph signals. The parameterized graph convolution filter bypasses feature decomposition, and a graph convolution operation is performed directly in the node domain to obtain spatial features. The formula used is as follows: ;

[0026] In the formula, represents the result of graph convolution, i.e., spatial features, represents the input graph signal, It is a graph convolution filter acting on the Laplacian matrix that operates directly in the node domain of the graph. represents the Laplacian matrix;

[0027] Step M6: using a long short-term memory network, inputting the spatial features and the driving intention into the long short-term memory network for time series prediction, and outputting driving state prediction data;

[0028] Step M7: Visually present the driving status prediction data on a screen.

[0029] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0030] (1) In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a driving state visualization prediction system based on the clutch pedal. In view of the technical problem that the existing solutions that directly rely on the driver's behavior characteristics are not accurate enough and lack the ability to grasp the temporal dynamics of driving behavior and capture long-term dependencies, this solution creatively proposes a variational mode decomposition denoising method, which uses the variational mode decomposition technology combined with the soft threshold to more accurately measure the displacement depth and action intensity of the driver's pedal, thereby improving the accuracy of driving state prediction;

[0031] (2) In response to the technical problem that the existing solutions do not consider the impact of the driving data of the driver on the vehicle's driving state, this solution creatively proposes a driving state prediction method that takes into account the driver's driving intention and pedal movement state, and uses a long short-term memory network combined with graph convolution to process continuous time series data. It assists the driver in driving in a visual way, effectively improving the vehicle's traffic efficiency, traffic safety and the prediction accuracy of the driving state, and helping to promote the development of intelligent transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A module connection diagram of a clutch pedal-based driving state visualization prediction system provided by the present invention;

[0033] Figure 2 It is a flow chart of the variational mode decomposition denoising method;

[0034] Figure 3 Schematic diagram of the flow of the driving state prediction method.

[0035] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0037] Example 1: See Figures 1 to 3 , this embodiment provides a driving state visualization prediction system based on a clutch pedal, the driving state visualization prediction system based on a clutch pedal comprises a signal acquisition module, a variational mode decomposition noise reduction module, a driving intention module and a driving state prediction module;

[0038] The signal acquisition module includes a pedal signal acquisition device and an on-board sensor; the pedal signal acquisition device collects electrical signals from the clutch pedal, the brake pedal and the accelerator pedal respectively, and the electrical signals are used to characterize the displacement and pedaling intensity of the pedals; the on-board sensor is used to collect on-board sensor data stored in the form of time series;

[0039] The variational mode decomposition denoising module uses a variational mode decomposition denoising method to perform denoising processing on the electrical signal;

[0040] The driving intention module uses a hidden Markov model to identify the driver's driving intention;

[0041] The driving state prediction module adopts a driving state prediction method, receives and processes the noise-reduced electrical signal, vehicle-mounted sensor data and driving intention, outputs driving state prediction data and visualizes it.

[0042] Example 2: See Figure 1 to Figure 2 This embodiment is based on the above embodiment. The variational mode decomposition denoising module adopts a variational mode decomposition denoising method. The variational mode decomposition denoising method has the following specific operations:

[0043] Step S1: signal decomposition, specifically, filtering the electrical signal, and decomposing the filtered electrical signal into multiple sub-signals using variational mode decomposition technology;

[0044] Step S2: Optimizing the sub-signal, specifically, converting the sub-signal into a frequency domain analytical signal through Hilbert transform, derivatizing the spectrum of the frequency domain analytical signal to obtain the signal gradient, minimizing the Euclidean norm of the signal gradients of all sub-signals with the constraint that the sum of all sub-signals must be equal to the electrical signal, recording the minimization task as a constrained optimization problem, converting the constrained optimization problem into an unconstrained optimization problem using exponential Lagrange multipliers, solving the unconstrained optimization problem using an alternating direction multiplier method, and obtaining the frequency domain representation and center frequency of the optimized sub-signal;

[0045] Step S3: noise reduction processing, specifically, noise reduction is performed on the sub-signal, and the formula used is as follows: ; ;

[0046] In the formula, represents the distinction threshold, represents the noise level of the sub-signal, Indicates the length of the signal, represents the sub-signal after noise reduction, represents the symbolic function, Represents a sub-signal;

[0047] Step S4: Signal reconstruction, specifically, recombining the sub-signals according to the frequency domain representation and the center frequency to obtain a noise-reduced electrical signal.

[0048] Example 3: See Figures 1 to 3 This embodiment is based on the above embodiment. The driving state prediction module adopts a driving state prediction method. The driving state prediction method has the following specific operations:

[0049] Step M1: Curve drawing, specifically, calculating the acceleration threshold, and establishing a two-dimensional coordinate plane with velocity and acceleration as the axis according to the electrical signal after noise reduction. The acceleration threshold calculation formula used is as follows: ;

[0050] In the formula, represents the acceleration threshold, represents the acceleration scaling factor, Indicates the instantaneous driving speed, , and represents the empirical fitting parameters;

[0051] Step M2: safe driving zoning, specifically, drawing an acceleration threshold curve according to the acceleration threshold, dividing the two-dimensional coordinate plane into a safe zone and an unsafe zone;

[0052] Step M3: construct a sensor relationship graph. Specifically, according to the vehicle sensor data, among all the vehicle sensors, quantify the linear correlation between every two vehicle sensors and construct a sensor weighted adjacency matrix. The calculation formula of the sensor weighted adjacency matrix is ​​as follows: ;

[0053] In the formula, Indicates that the coordinates in the sensor weighted adjacency matrix are The edge weights of represents the linear correlation number, Indicates Vehicle sensor data, Indicates Vehicle sensor data, represents the covariance, and Respectively and The standard deviation of

[0054] Step M4: Parameterization is used to reduce the computational complexity and improve the efficiency of graph convolution operations. Specifically, a graph convolution filter is used to parameterize the graph convolution filter through Chebyshev polynomials. The formula used is as follows: ;

[0055] In the formula, represents the eigenvalue diagonal matrix, represents a graph convolution filter that operates directly in the frequency domain of the graph on the diagonal matrix of the eigenvalues ​​of the Laplacian matrix, represents the normalized eigenvalue diagonal matrix, represents the kth order Chebyshev polynomial, represents the learnable parameters of Chebyshev polynomials, represents the total order of the polynomial, and k represents the order subscript index;

[0056] Step M5: Graph convolution, which is used to extract spatial features using graph convolution operations. Specifically, the vehicle-mounted sensors are used as nodes, and the relationships between the vehicle-mounted sensors are used as edges. A weighted graph is constructed by combining the nodes, edges, and sensor weighted adjacency matrices. The weighted graph and the two-dimensional coordinate plane are used as input graph signals. The parameterized graph convolution filter bypasses feature decomposition, and a graph convolution operation is performed directly in the node domain to obtain spatial features. The formula used is as follows: ;

[0057] In the formula, represents the result of graph convolution, i.e., spatial features, represents the input graph signal, It is a graph convolution filter acting on the Laplacian matrix that operates directly in the node domain of the graph. represents the Laplacian matrix;

[0058] Step M6: using a long short-term memory network, inputting the spatial features and the driving intention into the long short-term memory network for time series prediction, and outputting driving state prediction data;

[0059] Step M7: Visually present the driving status prediction data on a screen.

[0060] Example 4: See Figures 1 to 3 This embodiment is based on the above embodiment. In the driving intention module, the driving intention is divided into lateral intention and longitudinal intention. The lateral intention mainly includes changing lanes, turning, and going straight; the longitudinal intention includes acceleration, deceleration, cruising, acceleration, rapid acceleration, cruising, deceleration, and rapid deceleration.

[0061] Example 5: See Figures 1 to 3 This embodiment is based on the above embodiment. In step M1, =0.198, −0.592, 0.569.

[0062] Example 6: See Figures 1 to 3 This embodiment is based on the above embodiment. In step M3, the vehicle-mounted sensor data includes engine load, fuel consumption, intake manifold pressure, engine speed, steering wheel angle and angular acceleration, and brake pressure.

[0063] Example 7: See Figures 1 to 3 This embodiment is based on the above embodiment. The vehicle-mounted sensor is used to collect vehicle-mounted sensor data stored in the form of time series. The vehicle-mounted sensor includes an engine speed sensor, an intake manifold pressure sensor, a fuel consumption sensor, an engine load sensor, a gearbox position sensor, a brake pressure sensor, a steering angle sensor, a steering angular velocity sensor, a vehicle speed sensor, a body acceleration sensor and a wheel speed sensor. The specific method of storing in the form of time series is to read the vehicle-mounted sensor data through the CAN bus, and pack the data into data frames according to a period of 10ms, arrange the data in chronological order in JSON format, and send and store them in the memory of the vehicle-mounted controller. In this embodiment, the collected vehicle-mounted sensor data is as follows:

[0064] Engine speed: 3000RPM, intake manifold pressure: 100kPa, fuel consumption: 5 L / 100km, engine load: 70%, gearbox gear: 3rd gear, brake pressure: 0 bar, steering angle: 5°, steering angle speed: 1° / s, vehicle speed: 60 km / h, vehicle acceleration: 2 m / s², wheel speed: 50 km / h;

[0065] In this embodiment, the vehicle sensor data corresponding to the JSON format is as follows: { "timestamp": 1634567890, "sensors": { "engine_rpm": 3000, "map": 100, "fuel_consumption": 5, "engine_load": 70, "gear_position": 3, "brake_pressure": 0, "steering_angle": 5, "steering_angle_speed": 1, "vehicle_speed": 60, "body_acceleration": 2, "wheel_speed": 50 } }.

[0066] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0067] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0068] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A driving state visualization prediction system based on clutch pedal, characterized in that: It includes signal acquisition module, variational mode decomposition noise reduction module, driving intention module and driving state prediction module; The signal acquisition module includes a pedal signal acquisition device and an on-board sensor; the pedal signal acquisition device collects electrical signals from the clutch pedal, the brake pedal and the accelerator pedal respectively, and the electrical signals are used to characterize the displacement and pedaling intensity of the pedals; the on-board sensor is used to collect on-board sensor data stored in the form of time series; The variational mode decomposition denoising module uses a variational mode decomposition denoising method to perform denoising processing on the electrical signal; The driving intention module uses a hidden Markov model to identify the driver's driving intention; The driving state prediction module adopts a driving state prediction method, receives and processes the noise-reduced electrical signal, vehicle-mounted sensor data and driving intention, outputs driving state prediction data and visualizes it.

2. The driving state visualization prediction system based on clutch pedal according to claim 1 is characterized in that: The variational mode decomposition denoising module adopts a variational mode decomposition denoising method. The variational mode decomposition denoising method has the following specific operations: Step S1: signal decomposition, specifically, filtering the electrical signal, and decomposing the filtered electrical signal into multiple sub-signals using variational mode decomposition technology; Step S2: Optimizing the sub-signal, specifically, converting the sub-signal into a frequency domain analytical signal through Hilbert transform, derivatizing the spectrum of the frequency domain analytical signal to obtain the signal gradient, minimizing the Euclidean norm of the signal gradients of all sub-signals with the constraint that the sum of all sub-signals must be equal to the electrical signal, recording the minimization task as a constrained optimization problem, converting the constrained optimization problem into an unconstrained optimization problem using exponential Lagrange multipliers, solving the unconstrained optimization problem using an alternating direction multiplier method, and obtaining the frequency domain representation and center frequency of the optimized sub-signal; Step S3: noise reduction processing, specifically, noise reduction is performed on the sub-signal, and the formula used is as follows: ； ； In the formula, represents the distinction threshold, represents the noise level of the sub-signal, Indicates the length of the signal, represents the sub-signal after noise reduction, represents the symbolic function, Represents a sub-signal; Step S4: Signal reconstruction, specifically, recombining the sub-signals according to the frequency domain representation and the center frequency to obtain a noise-reduced electrical signal.

3. The driving state visualization prediction system based on clutch pedal according to claim 1, characterized in that: The driving state prediction module adopts a driving state prediction method, and the driving state prediction method specifically operates as follows: Step M1: Curve drawing, specifically, calculating the acceleration threshold, and establishing a two-dimensional coordinate plane with velocity and acceleration as the axis according to the electrical signal after noise reduction. The acceleration threshold calculation formula used is as follows: ； In the formula, represents the acceleration threshold, represents the acceleration scaling factor, Indicates the instantaneous driving speed, , and represents the empirical fitting parameters; Step M2: safe driving zoning, specifically, drawing an acceleration threshold curve according to the acceleration threshold, dividing the two-dimensional coordinate plane into a safe zone and an unsafe zone; Step M3: construct a sensor relationship graph. Specifically, according to the vehicle sensor data, among all the vehicle sensors, quantify the linear correlation between every two vehicle sensors and construct a sensor weighted adjacency matrix. The calculation formula of the sensor weighted adjacency matrix is ​​as follows: ； In the formula, Indicates that the coordinates in the sensor weighted adjacency matrix are The edge weights of represents the linear correlation number, Indicates Vehicle sensor data, Indicates Vehicle sensor data, represents the covariance, and Respectively and The standard deviation of Step M4: parameterization, specifically, using a graph convolution filter and parameterizing the graph convolution filter through Chebyshev polynomials. The formula used is as follows: ； In the formula, represents the eigenvalue diagonal matrix, represents a graph convolution filter that operates directly in the frequency domain of the graph on the diagonal matrix of the eigenvalues ​​of the Laplacian matrix, represents the normalized eigenvalue diagonal matrix, represents the kth order Chebyshev polynomial, represents the learnable parameters of Chebyshev polynomials, represents the total order of the polynomial, and k represents the order subscript index; Step M5: Graph convolution, specifically, using the vehicle-mounted sensors as nodes and the relationships between the vehicle-mounted sensors as edges, constructing a weighted graph with the nodes, edges and sensor weighted adjacency matrix, using the weighted graph and the two-dimensional coordinate plane as input graph signals, bypassing feature decomposition through a parameterized graph convolution filter, and performing graph convolution operations directly in the node domain to obtain spatial features; Step M6: using a long short-term memory network, inputting the spatial features and the driving intention into the long short-term memory network for time series prediction, and outputting driving state prediction data; Step M7: Visually present the driving status prediction data on a screen.

Images