Pavement unevenness identification method and equipment based on physical information neural network, medium and vehicle

Abstract

The invention discloses a road surface unevenness recognition method and device based on a physical information neural network, a medium and a vehicle, and relates to the field of road recognition, and the method comprises the steps: combining a kinetic equation of a suspension system with a neural network, and constructing a neural network model based on physical information; acquiring acceleration data of the suspension system on the to-be-identified road surface; adopting a neural network model based on physical information to obtain a pavement elevation recognition result based on the acceleration data; and representing the road surface unevenness by adopting a road surface elevation recognition result. According to the method, the complexity of an algorithm can be reduced, meanwhile, the physical interpretability is improved, richer road surface information is provided for dynamic control of a suspension system, and then the driving smoothness and riding comfort of a vehicle are improved.

Description

Technical Field

[0001] The present application relates to the field of road recognition, and in particular to a road surface roughness recognition method, equipment, medium and vehicle based on a physical information neural network. Background Art

[0002] Road surface roughness refers to the fluctuations in the road surface's height relative to a reference horizontal plane along the road's course. During vehicle travel, the excitation generated by road surface roughness on the wheels is attenuated by the suspension system and then transmitted to the vehicle body, thereby affecting the vehicle's ride smoothness and comfort. Accurately identifying road surface roughness is a key prerequisite for establishing a dynamic interaction mechanism between vehicle and road surface and for improving and optimizing vehicle suspension control strategies.

[0003] In the prior art, traditional methods for identifying road surface roughness can be roughly divided into two categories. One involves using instruments such as levels and laser levelers to directly measure road surface elevation while the vehicle is moving. The other involves using onboard sensors to collect specific physical quantities and then inferring road surface roughness through vehicle dynamics modeling. For example, Chinese patent application CN117786819A, known to the inventors, uses acceleration sensors and dynamic travel sensors to determine the vertical acceleration of the vehicle body, suspension dynamic travel, and suspension motion speed. Then, based on a preset empirical formula, road surface state parameters representing the road surface grade are derived, thereby identifying the current road surface grade.

[0004] With the rapid development of artificial intelligence technology, research has begun on road surface roughness recognition methods combined with neural networks. Long Short-Term Memory (LSTM) is a special recurrent neural network. The memory units and gating mechanisms it introduces make it more efficient in processing long sequence tasks, and it is particularly suitable for scenarios such as time series. As the inventors are aware of, Chinese patent application CN116842385A proposes a road surface roughness recognition method using LSTM based on the vibration characteristics of tracked vehicles. The collected three-dimensional acceleration signals are sliding sampled and normalized before training the LSTM. This allows the model to establish a complex mapping relationship between the three-dimensional acceleration signals of the tracked vehicle and the road surface roughness, thereby achieving accurate recognition of the road surface grade.

[0005] Based on the above description, the direct measurement method of road roughness is costly and relies on specialized vehicles and equipment, which limits its large-scale and normalized application. It is only suitable for high-precision road surveys and cannot be integrated into passenger vehicles.

[0006] Compared with the direct measurement method, the road surface roughness inverse method based on vehicle dynamic response only requires the installation of common on-board sensors. It is low-cost, easy to assemble and disassemble, does not require large-scale modifications to the vehicle, and will not significantly affect the normal driving performance of the vehicle. Therefore, it is more suitable for large-scale deployment on mass-produced vehicles. However, this method also has limitations such as the back-end algorithm is relatively complex and the actual application requires advance calibration. For example, the Chinese patent application CN117786819A known to the inventor introduces a preset empirical formula to calculate the road surface state parameters used to identify the road surface grade when inversely calculating the road surface grade. The empirical formula is often refined based on a large amount of actual data and engineering practice, which significantly increases the complexity of this method. In addition, since the road surface state parameters are used to identify the road surface grade, these parameters need to be calibrated in advance so that when the vehicle travels to a road surface of unknown grade, the road surface grade can be judged based on the calibrated state parameters.

[0007] Combining neural networks with road surface roughness recognition technology to establish a data-driven road surface roughness recognition model can fully utilize the powerful nonlinear fitting capabilities of neural networks, allowing the model to automatically learn the underlying deep relationship between input data and road surface roughness, thereby simplifying the recognition process and improving generalization capabilities. However, training data-driven neural network models relies on a large amount of sample data, and the training samples are generally required to cover the main operating conditions of the target scene as much as possible. For example, in Chinese patent application CN116842385A, known to the inventors, to accurately predict the four road surface grades A, B, C, and D, a large number of three-dimensional acceleration signals of tracked vehicles traveling on these four different road surface grades are collected as sample data to train the designed LSTM model. In addition, because data-driven neural networks are generally black box models, their internal working mechanisms are difficult to fully analyze and explicitly describe, making the model's interpretability far inferior to traditional road surface roughness recognition methods.

[0008] In summary, among the current methods for identifying road roughness, the direct measurement method is costly and difficult to popularize; the road roughness inverse method based on vehicle dynamic response has a complex back-end algorithm and requires calibration before application; and the data-driven road roughness identification model requires a large sample size and has poor model interpretability. Summary of the Invention

[0009] In order to solve the above-mentioned problems existing in the prior art, the present application provides a road roughness identification method, equipment, medium and vehicle based on physical information neural network.

[0010] To achieve the above objectives, this application provides the following solutions:

[0011] In a first aspect, the present application provides a method for identifying road roughness based on a physical information neural network, comprising:

[0012] Combine the dynamic equations of the suspension system with the neural network to build a neural network model based on physical information;

[0013] Acquiring acceleration data of the suspension system on a road surface to be identified;

[0014] The physical information-based neural network model is used to obtain a road surface elevation recognition result based on the acceleration data; and the road surface elevation recognition result is used to characterize road surface roughness.

[0015] Optionally, the dynamic equations of the suspension system are combined with a neural network to construct a neural network model based on physical information, including:

[0016] Combine the dynamic equations of the suspension system with the neural network to build an initial neural network model based on physical information;

[0017] Acquire a training data set; the training data set includes a plurality of training sample pairs; each of the training sample pairs includes acceleration data and a road elevation corresponding to the acceleration data; the acceleration data includes sprung mass acceleration data and unsprung mass acceleration data;

[0018] The training data set is used, the acceleration data is used as input, and the road surface elevation is used as output to train the initial neural network model. During the training process, the loss between the predicted road surface elevation of the initial neural network model and the corresponding road surface elevation in the training sample is determined and back propagation is performed to optimize the parameters of the initial neural network model until the set conditions are met, thereby obtaining a trained initial neural network model. The trained initial neural network model is a neural network model based on physical information.

[0019] Optionally, the dynamic equations of the suspension system are combined with the neural network to construct an initial neural network model based on physical information, including:

[0020] Performing kinematic modeling on the suspension system to obtain a kinematic modeling module; the kinematic modeling module takes the displacement, acceleration, and velocity of the sprung mass at a previous moment, the displacement, acceleration, and velocity of the unsprung mass at a previous moment, the acceleration of the sprung mass at a current moment, and the acceleration of the unsprung mass at a current moment as input, and takes the displacement and velocity of the sprung mass at a current moment, and the displacement and velocity of the unsprung mass at a current moment as output;

[0021] A first neural network, a second neural network, and a third neural network are constructed to obtain the initial neural network model based on physical information; the first neural network and the second neural network are respectively used to obtain elastic force prediction results and damping force prediction results of the suspension system based on the displacement and velocity of the sprung mass at the current moment and the displacement and velocity of the unsprung mass at the current moment output by the kinematic modeling module; according to force balance, the wheel elastic force is obtained based on the elastic force prediction results and the damping force prediction results; the third neural network is used to obtain a road elevation prediction result based on the wheel elastic force and the displacement of the unsprung mass at the current moment.

[0022] Optionally, the first neural network and the second neural network are both neural networks with a single input, a single output, and four hidden layers of neurons; the activation function of the first neural network and the activation function of the second neural network are both LeakyReLU functions.

[0023] Optionally, the third neural network is a dual-input, single-output neural network comprising a four-neuron hidden layer; the activation function of the third neural network is a Tanh function.

[0024] Optionally, the first neural network is used to replace the stiffness of the suspension system; the second neural network is used to replace the damping coefficient of the suspension system; and the third neural network is used to replace the stiffness coefficient of the wheel.

[0025] Optionally, the training sample pairs in the training data set are generated by a suspension system model constructed by SIMULINK.

[0026] In a second aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-mentioned method for identifying road roughness based on a physical information neural network.

[0027] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned road surface roughness identification method based on physical information neural network.

[0028] In a fourth aspect, the present application provides a vehicle, comprising: an on-board processor; the on-board processor is used to implement the above-mentioned road roughness recognition method based on physical information neural network.

[0029] According to the specific embodiments provided in this application, this application has the following technical effects:

[0030] This application provides a road surface roughness identification method, device, medium, and vehicle based on a physical information neural network. By combining the dynamic equations of the suspension system with a neural network, a physical information-based neural network model is constructed. This method can directly identify the elevation information of different road surfaces. It does not require relying on the calculation results of empirical formulas to classify road surface grades, nor does it require providing calibration values ​​for the empirical formulas. This reduces the complexity of the algorithm and improves the physical interpretability of the constructed identification network. Furthermore, the road surface roughness identification result of this application does not obtain the road surface grade, but directly identifies the road surface elevation. This can accurately reflect the time coordinate of the road surface, provide richer road surface information for the dynamic control of the suspension system, and thus improve the vehicle's driving smoothness and ride comfort. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0032] Figure 1 A schematic flow chart of a method for identifying road roughness based on a physical information neural network according to an embodiment of the present application;

[0033] Figure 2 A schematic diagram of an implementation flow of a road roughness identification method based on a physical information neural network provided in one embodiment of the present application;

[0034] Figure 3 A schematic structural diagram of a suspension system provided in one embodiment of the present application;

[0035] Figure 4 A schematic diagram of the structure of an initial neural network model based on physical information provided in one embodiment of the present application;

[0036] Figure 5 The neural network NN provided in one embodiment of the present application ks Schematic diagram of the structure;

[0037] Figure 6 The neural network NN provided in one embodiment of the present application bs Schematic diagram of the structure;

[0038] Figure 7 The neural network NN provided in one embodiment of the present application kt Schematic diagram of the structure;

[0039] Figure 8 A schematic diagram of training sample data provided in one embodiment of the present application;

[0040] Figure 9 A schematic diagram of a complete recognition result of a sample road surface elevation provided in an embodiment of the present application;

[0041] Figure 10 Provided for an embodiment of this application Figure 9 Schematic diagram of the first 10s data;

[0042] Figure 11 A schematic diagram of classification results on a test set provided in one embodiment of the present application;

[0043] Figure 12 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0045] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0046] In an exemplary embodiment, the present application provides a road roughness identification method based on a physical information neural network. The method can be executed by a vehicle-mounted terminal alone or jointly by a vehicle-mounted terminal and a server. In the embodiment of the present application, the method is illustrated by applying it to a vehicle-mounted terminal as an example.

[0047] like Figure 1 As shown, the method includes:

[0048] Step 100: Combine the dynamic equations of the suspension system with the neural network to construct a neural network model based on physical information.

[0049] Step 101: Acquire acceleration data of the suspension system on the road surface to be identified.

[0050] Step 102: Using a physical information-based neural network model, obtain a road elevation recognition result based on the acceleration data, and use the road elevation recognition result to characterize road roughness.

[0051] Among them, if the present application is directly applied to road surface grade recognition, then on the basis of road surface elevation recognition, a higher accuracy rate can be obtained based on a simple random forest algorithm, so as to effectively reduce the design difficulty of the road surface grade recognition algorithm.

[0052] By implementing steps 100 through 102 above, the suspension system's dynamic equations are combined with a neural network to construct a physics-based neural network model. This reduces algorithm complexity and improves the physical interpretability of the constructed recognition network. Furthermore, by directly identifying road elevation, the road's temporal coordinates can be accurately reflected, providing richer road surface information for the suspension system's dynamic control, thereby improving the vehicle's ride smoothness and passenger comfort.

[0053] In another exemplary embodiment of the present application, Figure 3 Taking the two-degree-of-freedom linear suspension system shown as an example, the specific implementation process of step 100 provided above in this application is described. Among them, the two-degree-of-freedom linear suspension system in this embodiment is only used for illustration and is not a specific limitation of this application. Figure 3 In the equation, the sprung mass is denoted as M s , whose displacement, velocity and acceleration are denoted as Unsprung mass is denoted as M u , whose displacement, velocity and acceleration are denoted as The stiffness coefficient, damping coefficient and wheel stiffness coefficient of the two-degree-of-freedom linear suspension system are denoted as k s 、b s 、k t . The road surface elevation is recorded as u.

[0054] By analyzing the forces acting on the sprung mass and the unsprung mass, we can obtain:

[0055] F as +F ks +F bs =0.

[0056] F au =F ks +F bs -F kt .

[0057] Where, F as 、F au are the inertial forces of the sprung mass and unsprung mass respectively. ks 、F bs 、F kt They are the elastic force, damping force and elastic force generated by the two-degree-of-freedom linear suspension system and the wheel. These five physical quantities can be expressed as:

[0058] F as =M s a s .

[0059] F au =M u a u .

[0060] F ks =k s (x s -x u ).

[0061] F bs =b s (v s -v u ).

[0062] F kt =k t (x u -u).

[0063] As can be seen, if the displacement of the sprung and unsprung masses and the suspension stiffness coefficient are known, the suspension spring force can be calculated. If the velocities of the sprung and unsprung masses and the suspension damping coefficient are known, the suspension damping force can be calculated. If the acceleration of the unsprung mass is known, the unsprung mass inertia force can be calculated. Combining the suspension spring force and damping force with the unsprung mass inertia force can be used to calculate the wheel spring force. If the wheel stiffness coefficient is known, the wheel spring force and the unsprung mass displacement can be combined to calculate the road surface elevation.

[0064] For a two-degree-of-freedom linear suspension system, the acceleration of the sprung and unsprung masses can be measured using accelerometers mounted on the vehicle body and wheels. However, directly measuring their velocity and displacement using sensors is more difficult. Therefore, as an alternative to direct measurement, the velocity and displacement of the sprung and unsprung masses can be calculated using the following formulas:

[0065] v(t)=v(t-Δt)+a(t)Δt.

[0066]

[0067] When Δt approaches 0, based on the above formula, the velocity and displacement at time t-Δt and the acceleration at time t can be used to approximate the actual velocity and displacement at time t.

[0068] In addition, considering that the stiffness and damping coefficient of the suspension and the stiffness coefficient of the wheel are difficult to measure, this embodiment uses a neural network to replace these three parameters respectively, namely:

[0069] F ks =k s (x s -x u )→F ks =NN ks (xs -x u ).

[0070] F bs =b s (v s -v u )→F bs =NN bs (v s -v u ).

[0071] F kt =k t (x u -u)→u=NN kt (F kt ,x u ).

[0072] Where, NN ks ,NN bs ,NN kt They represent the neural networks used to replace the suspension stiffness coefficient, suspension damping coefficient and wheel stiffness coefficient respectively. kt The input and output of is not relative displacement and elastic force, but elastic force and unsprung mass displacement as input and road elevation as output, which can be regarded as fitting relationship.

[0073] Based on the above description, the implementation process of step 100 can be replaced by the following steps 200 to 202.

[0074] Step 200: Combine the dynamic equations of the suspension system with the neural network to construct an initial neural network model based on physical information.

[0075] Among them, the structure of the initial neural network model constructed is as follows Figure 4 The initial neural network model takes the acceleration of the sprung and unsprung masses at each moment as input and the road elevation at the corresponding moment as output. The model can be divided into two parts: one is the kinematic modeling module, which calculates the displacement x of the sprung and unsprung masses at the previous moment. s (t-Δt), x u (t-Δt) and velocity v s (t-Δt), v u (t-Δt) and the acceleration a of the sprung and unsprung masses input at the current moment s (t), a u (t) Calculate the displacement x of the sprung and unsprung masses at the current moment s (t), x u (t) and velocity v s (t), v u(t); The second is the dynamic modeling and neural network module, which inputs the current sprung and unsprung mass displacement and velocity calculated by kinematic modeling into the neural network NN ks , neural network NN bs , predict the elastic force and damping force of the two-degree-of-freedom linear suspension system, obtain the wheel elastic force according to the force balance, and finally use the neural network NN kt Predicting road elevation. Based on this, the implementation process of step 200 includes:

[0076] Step 200-1: Perform kinematic modeling of the suspension system to obtain a kinematic modeling module. The kinematic modeling module takes as input the displacement, acceleration, and velocity of the sprung mass at the previous moment, the displacement, acceleration, and velocity of the unsprung mass at the previous moment, the acceleration of the sprung mass at the current moment, and the acceleration of the unsprung mass at the current moment, and outputs the displacement and velocity of the sprung mass at the current moment, as well as the displacement and velocity of the unsprung mass at the current moment.

[0077] Step 200-2: Construct the first neural network (ie neural network NN ks ), the second neural network (i.e. neural network NN bs ) and the third neural network (i.e. neural network NN kt ) to obtain an initial neural network model based on physical information. The first and second neural networks are used to respectively derive predictions for the suspension system's elastic force and damping force based on the current displacement and velocity of the sprung mass and the current displacement and velocity of the unsprung mass output by the kinematic modeling module. Based on the force balance, the wheel elastic force is derived from the elastic force and damping force predictions. The third neural network is used to derive a road elevation prediction based on the wheel elastic force and the current displacement of the unsprung mass.

[0078] like Figures 5 and 6 As shown, the neural network NN ks and neural network NN bs The structures of are exactly the same, both are single-input, single-output neural networks containing a hidden layer with four neurons, and the activation function used is the LeakyReLU function. Figure 7 As shown, the neural network NN kt It is a neural network with two inputs and one output, containing one hidden layer with four neurons, and the activation function used is the Tanh function.

[0079] Step 201: Obtain a training dataset. The training dataset includes multiple training sample pairs. Each training sample pair includes acceleration data and the road elevation u corresponding to the acceleration data. Each training sample pair also needs to include time t. Corresponding to this time, the acceleration data includes the sprung mass acceleration a. s, unsprung mass acceleration a u .

[0080] Step 202: Using the training data set, acceleration data as input, and road elevation as output, an initial neural network model is trained. During the training process, the loss between the predicted road elevation of the initial neural network model and the corresponding road elevation in the training sample pair is determined and backpropagation is performed to optimize the parameters of the initial neural network model. When a set condition is met (e.g., a maximum number of training iterations is reached or an evaluation index reaches a set value), a trained initial neural network model is obtained. The trained initial neural network model is the neural network model based on physical information.

[0081] Furthermore, a dataset for road roughness recognition was created. This dataset contained the same data as the training dataset, but did not need to include the road elevation u. During the recognition process, the physics-based neural network model made predictions based on the input sprung and unsprung mass accelerations, thereby obtaining the elevation recognition results for the unknown road surface.

[0082] Based on the above description, Figure 8 The first training sample shown in part (a) is Figure 8 Taking the training of the initial neural network model using the second training sample shown in part (b) as an example, the implementation process of the above step 202 is described. Figure 8 The two training sample data shown are generated by a two-degree-of-freedom linear suspension system model built in SIMULINK, which includes a B-level road elevation input with a time step Δt of 1 ms and sprung and unsprung mass acceleration response outputs.

[0083] use Figure 8 The physical information neural network trained with the training samples shown in the figure is used to identify the road elevations of different levels (B, C, D and E). The recognition results are shown in Table 1 and Table 2. Figure 9 and Figure 10 As shown in Table 1, the correlation coefficient R, mean absolute error MAE and root mean square error RMSE are used as evaluation indicators to measure the recognition accuracy. The road elevation recognition is performed on 5 groups of 100s training samples of the same road surface grade, and the average value of the three evaluation indicators of each group of training samples is calculated. The results are summarized in Table 1. A group of samples is selected from each of the three road surface grades C, D and E, and the road elevation recognition curve is drawn. Figure 9 and Figure 10 As shown, Figure 9 For the complete recognition results of the training sample road surface elevation, the overall trend is highly consistent with the real road surface elevation, which verifies the effectiveness of the method provided in this application. Figure 10 Then intercepted Figure 9 From the data of the first 10 seconds, we can see that Figure 10 There is only a slight difference between the recognition result shown and the actual road elevation. Figure 9 Part (a) shows the C-level road elevation recognition result. Figure 9 Part (b) shows the D-level road elevation recognition result. Figure 9 Part (c) shows the E-level road elevation recognition result. Figure 10 Part (a) shows the recognition results of the first 10 seconds of the C-level road elevation. Figure 10 Part (b) shows the recognition results of the first 10 seconds of the D-level road elevation. Figure 10 Part (c) shows the recognition results of the first 10 seconds of the E-level road elevation.

[0084] The five groups of elevation recognition data for each of the three road surface grades were divided into 50 groups of samples with a length of 10 seconds. After converting the time domain data to the frequency domain using fast Fourier transform, the training set and test set were randomly divided into two groups with a ratio of 4:1. Subsequently, a road surface grade classification model was constructed based on random forest. The classification results of the model on the test set are shown in the following figure. Figure 11 As shown in the figure, the recognition accuracy of road surface grade can reach 95%.

[0085] Table 1 Results of three evaluation indicators for training samples

[0086]

[0087] In summary, the overall implementation process of the road roughness recognition method based on physical information neural network provided by this application is as follows: Figure 2 Compared with the prior art, this application has the following advantages:

[0088] 1. The neural network model based on physical information designed in this application enables the intermediate variables involved in training to automatically meet the constraints of physical laws, thereby improving the physical interpretability of the model.

[0089] 2. This application uses a neural network to replace the stiffness and damping coefficients of the suspension system, as well as the stiffness coefficient of the wheel, and can be applied to scenarios where the suspension system parameters are unknown and difficult to measure.

[0090] 3. Compared to inverse methods for determining road roughness based on vehicle dynamics, this application eliminates the need for complex back-end processing algorithms. Training with sample data containing known road elevations and the corresponding sprung and unsprung mass acceleration responses of the suspension system yields a highly efficient road roughness recognition model. Furthermore, this application can directly identify elevation information for different road grades, without relying on empirical formulas to classify road grades or requiring calibration values ​​for these formulas, thus reducing algorithmic complexity.

[0091] 4. Compared with the data-driven road surface roughness recognition model, this application combines the dynamic equations of the suspension system with the neural network, which improves the physical interpretability of the constructed recognition network. At the same time, this application requires a small amount of data for training samples and does not require calibration. Combined with the above description, the network training can be completed using two samples with a duration of 1s and an interval of 1ms. In addition, this application has a strong generalization ability. Although the two training samples are the acceleration responses of the sprung and unsprung masses of the suspension system when it is excited by a Class B road surface, it can accurately identify the elevation information of other grades of roads when used for recognition, and the time to ensure recognition accuracy can reach 100s.

[0092] 5. The result of the road surface roughness identification in this application is not to obtain the road surface grade, but to directly identify the road surface elevation, which can accurately reflect the time coordinates of the road surface, provide richer road surface information for the dynamic control of the suspension system, and improve the vehicle's driving smoothness and ride comfort.

[0093] In an exemplary embodiment, a vehicle is provided, which includes an on-board processor; the on-board processor is used to implement the road surface roughness recognition method based on physical information neural network provided in the above embodiment.

[0094] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 12 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store road surface roughness identification data based on a physical information neural network. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a road surface roughness identification method based on a physical information neural network is implemented.

[0095] Those skilled in the art will understand that Figure 12The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0096] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0097] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0098] In an exemplary embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0099] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0100] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (RRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0101] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0102] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0103] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A road roughness recognition method based on physical information neural network, characterized in that: include: Combine the dynamic equations of the suspension system with the neural network to build a neural network model based on physical information; Acquiring acceleration data of the suspension system on a road surface to be identified; The physical information-based neural network model is used to obtain a road surface elevation recognition result based on the acceleration data; and the road surface elevation recognition result is used to characterize road surface roughness.

2. The road roughness identification method based on physical information neural network according to claim 1 is characterized in that: Combine the dynamic equations of the suspension system with the neural network to build a neural network model based on physical information, including: Combine the dynamic equations of the suspension system with the neural network to build an initial neural network model based on physical information; Acquire a training data set; the training data set includes a plurality of training sample pairs; each of the training sample pairs includes acceleration data and a road elevation corresponding to the acceleration data; the acceleration data includes sprung mass acceleration data and unsprung mass acceleration data; The training data set is used, the acceleration data is used as input, and the road surface elevation is used as output to train the initial neural network model. During the training process, the loss between the predicted road surface elevation of the initial neural network model and the corresponding road surface elevation in the training sample is determined and back propagation is performed to optimize the parameters of the initial neural network model until the set conditions are met, thereby obtaining a trained initial neural network model. The trained initial neural network model is a neural network model based on physical information.

3. The road roughness identification method based on physical information neural network according to claim 2 is characterized in that: Combine the dynamic equations of the suspension system with the neural network to build an initial neural network model based on physical information, including: Performing kinematic modeling on the suspension system to obtain a kinematic modeling module; the kinematic modeling module takes the displacement, acceleration, and velocity of the sprung mass at a previous moment, the displacement, acceleration, and velocity of the unsprung mass at a previous moment, the acceleration of the sprung mass at a current moment, and the acceleration of the unsprung mass at a current moment as input, and takes the displacement and velocity of the sprung mass at a current moment, and the displacement and velocity of the unsprung mass at a current moment as output; A first neural network, a second neural network, and a third neural network are constructed to obtain the initial neural network model based on physical information; the first neural network and the second neural network are respectively used to obtain elastic force prediction results and damping force prediction results of the suspension system based on the displacement and velocity of the sprung mass at the current moment and the displacement and velocity of the unsprung mass at the current moment output by the kinematic modeling module; according to force balance, the wheel elastic force is obtained based on the elastic force prediction results and the damping force prediction results; the third neural network is used to obtain a road elevation prediction result based on the wheel elastic force and the displacement of the unsprung mass at the current moment.

4. The road roughness identification method based on physical information neural network according to claim 3 is characterized in that: The first neural network and the second neural network are both neural networks with a single input, a single output, and four hidden layers of neurons; the activation function of the first neural network and the activation function of the second neural network are both LeakyReLU functions.

5. The road surface roughness identification method based on physical information neural network according to claim 3 is characterized in that: The third neural network is a dual-input, single-output neural network comprising a four-neuron hidden layer; the activation function of the third neural network is a Tanh function.

6. The road surface roughness identification method based on physical information neural network according to claim 3 is characterized in that: The first neural network is used to replace the stiffness of the suspension system; the second neural network is used to replace the damping coefficient of the suspension system; and the third neural network is used to replace the stiffness coefficient of the wheel.

7. The road surface roughness identification method based on physical information neural network according to claim 2 is characterized in that: The training sample pairs in the training data set are generated by a suspension system model constructed by SIMULINK.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the road surface roughness identification method based on a physical information neural network according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the road surface roughness identification method based on a physical information neural network according to any one of claims 1 to 7 is implemented.

10. A vehicle, characterized in that: include: Onboard processor; The on-board processor is used to implement the road surface roughness recognition method based on physical information neural network according to any one of claims 1 to 7.

Images