Flatness prediction method and system based on wavelet and physical neural network

Abstract

The invention provides a flatness prediction method and system based on a wavelet and a physical neural network, and relates to the technical field of road engineering, and the method comprises the steps: carrying out the calculation through a scale function of a dbN wavelet, and obtaining a first weight coefficient of a pavement longitudinal contour elevation function; solving the quarter vehicle dynamic equation, determining unsprung mass displacement and sprung mass displacement, and obtaining a second weight coefficient and a third weight coefficient corresponding to the unsprung mass displacement and the sprung mass displacement respectively through wavelet transform; training a preset deep learning network model by using the obtained weight coefficient, and determining a loss function of the preset deep learning network model; performing iterative optimization on the preset deep learning network model by using a Bayesian optimization algorithm and the loss function to obtain a target deep learning network model; and inputting the fourth weight coefficient of the to-be-measured pavement into the target deep learning network model, and determining the to-be-measured unsprung mass displacement, the to-be-measured sprung mass displacement and the international flatness index.

Description

TECHNICAL FIELD

[0001] The present application relates to the road engineering technical field, and particularly relates to a flatness prediction method and system based on wavelet and physical neural network. BACKGROUND

[0002] With the gradual perfection of China's transportation infrastructure, the core of the transportation construction management department has gradually transitioned from construction to maintenance stage. Correct maintenance timing not only reduces the cost of maintenance and repair, but also reduces the safety risk suffered by road users. The selection of maintenance timing depends on road performance indicators, and the international roughness index is a recognized road performance indicator worldwide that can represent the driving comfort of the road. The current international roughness calculation method uses the state transition method used by Mrcrmsr W. Savnns in the paper "On the Calculation of International Roughness Index from Longitudinal Road Profile". This method takes a quarter of the car as the prototype, specifies that the sampling interval is uniform sampling, the connection method between the road longitudinal elevation points is linear connection, and the sampling speed is uniform speed of 80km / h. The displacement and velocity of the sprung mass and the unsprung mass are obtained by solving the linear equation set, and finally the international roughness index value is obtained by integration. However, this method is limited by the assumptions of sampling interval and linear continuity of road elevation points, and it is difficult to effectively handle and calculate the sampling interval changes or sudden road surface abnormalities. SUMMARY

[0003] Therefore, the present application provides a flatness prediction method and system based on wavelet and physical neural network.

[0004] The technical scheme of the present application is implemented as follows: The present application provides a flatness prediction method based on wavelet and physical neural network in the first aspect, comprising:

[0005] The first weight coefficient of the road longitudinal profile elevation function is calculated by using the scale function of the dbN wavelet; the dynamic equation of the quarter car is solved to determine the displacement of the unsprung mass and the displacement of the sprung mass, and the second weight coefficient and the third weight coefficient corresponding to the displacement of the unsprung mass and the displacement of the sprung mass are obtained by wavelet transform;

[0006] The first weight coefficient is used as input data, and the second weight coefficient and the third weight coefficient are used as output data, and a preset deep learning network model is trained to determine the loss function of the preset deep learning network model; the preset deep learning network model is iteratively optimized by using the Bayesian optimization algorithm and the loss function to obtain a target deep learning network model;

[0007] input the fourth weight coefficient of the longitudinal profile elevation function of the to-be-tested road surface into the target deep learning network model to obtain a fifth weight coefficient and a sixth weight coefficient corresponding to the to-be-tested unsprung mass displacement and the to-be-tested sprung mass displacement respectively, and determine the to-be-tested unsprung mass displacement, the to-be-tested sprung mass displacement and the international roughness index based on the fifth weight coefficient and the sixth weight coefficient.

[0008] Preferably, on the basis of the above technical solutions, the first weight coefficient of the longitudinal profile elevation function of the road surface is obtained by calculation using the scale function of the dbN wavelet, and the first weight coefficient comprises:

[0009] The inner product of the longitudinal profile elevation function of the road surface and the scale function is integrated in the continuous sampling time range to obtain the first weight coefficient of the longitudinal profile elevation function of the road surface.

[0010] Preferably, on the basis of the above technical solutions, the second weight coefficient and the third weight coefficient corresponding to the unsprung mass displacement and the sprung mass displacement respectively are determined by solving the quarter-car dynamic equation, and the solving comprises:

[0011] The quarter-car dynamic equation is created based on the sprung mass, the unsprung mass, the suspension stiffness coefficient, the suspension damping coefficient, the tire stiffness coefficient, the sprung speed, the unsprung speed, the sprung acceleration, the unsprung acceleration, the sprung mass displacement and the unsprung mass displacement.

[0012] The quarter-car dynamic equation is solved by using the Runge-Kutta method, and the second weight coefficient and the third weight coefficient corresponding to the unsprung mass displacement and the sprung mass displacement respectively are determined by wavelet transform on the solving result.

[0013] Preferably, on the basis of the above technical solutions, before the first weight coefficient is taken as input data, the second weight coefficient and the third weight coefficient are taken as output data, and the preset deep learning network model is trained, the method comprises:

[0014] A plurality of groups of road surface samples are used to obtain a training data set; the training data set comprises a plurality of groups of corresponding first weight coefficients, second weight coefficients and third weight coefficients.

[0015] Preferably, on the basis of the above technical solutions, the preset deep learning network model comprises an encoder, a decoder and a residual module; and the first weight coefficient is taken as input data, the second weight coefficient and the third weight coefficient are taken as output data, and the preset deep learning network model is trained, which comprises:

[0016] The time-scale features of the first weight coefficient, the second weight coefficient and the third weight coefficient are processed by using the encoder and the decoder, and the loss function determined by the residual module is combined to adjust the parameters of the preset deep learning network model.

[0017] On the basis of the above technical solutions, preferably, the loss function comprises a first loss function, a second loss function and a third loss function; the preset deep learning network model is trained by taking the first weight coefficient as input data and taking the second weight coefficient and the third weight coefficient as output data, and the loss function of the preset deep learning network model is determined, comprising:

[0018] The square of the difference between the predicted value and the true value of the second weight coefficient is summed with the square of the difference between the predicted value and the true value corresponding to the third weight coefficient, and the mean value is determined as the first loss function;

[0019] A vehicle dynamic response loss term is created based on the sprung mass, the suspension stiffness coefficient, the suspension damping coefficient, the connection coefficient of each order, the predicted value of the second weight coefficient and the predicted value of the third weight coefficient, and the sprung mass is used for constraint to determine the second loss function;

[0020] A vehicle dynamic response loss term is created based on the unsprung mass, the tire stiffness coefficient, the suspension stiffness coefficient, the suspension damping coefficient, the connection coefficient of each order, the predicted value of the first weight coefficient, the predicted value of the second weight coefficient and the predicted value of the third weight coefficient, and the unsprung mass is used for constraint to determine the third loss function.

[0021] On the basis of the above technical solutions, preferably, the preset deep learning network model is iteratively optimized by using the Bayesian optimization algorithm and the loss function to obtain a target deep learning network model, comprising:

[0022] In the case that the iteration round is less than a first threshold, the loss function of the preset deep learning network model is determined based on the first loss function;

[0023] In the case that the iteration round is not less than the first threshold, the loss function of the preset deep learning network model is determined based on the weighted sum of the first loss function, the second loss function and the third loss function.

[0024] Further preferably, the second aspect of the present application provides a flatness prediction system based on wavelet and physical neural network, comprising: a data acquisition module, a training and optimization module and a parameter determination module; wherein,

[0025] The data acquisition module is configured to calculate a first weight coefficient of a longitudinal profile elevation function of a road surface by using a scaling function of a dbN wavelet; solve a quarter-car dynamic equation to determine a sprung mass displacement and an unsprung mass displacement, and obtain a second weight coefficient and a third weight coefficient corresponding to the sprung mass displacement and the unsprung mass displacement respectively by wavelet transform;

[0026] The training optimization module is configured to take the first weight coefficient as input data, take the second weight coefficient and the third weight coefficient as output data, train a preset deep learning network model, determine a loss function of the preset deep learning network model, and iteratively optimize the preset deep learning network model by using a Bayesian optimization algorithm and the loss function to obtain a target deep learning network model.

[0027] The parameter determination module is configured to input a fourth weight coefficient of a longitudinal profile elevation function of a to-be-tested road surface into the target deep learning network model to obtain a fifth weight coefficient and a sixth weight coefficient corresponding to a to-be-tested sprung mass displacement and a to-be-tested unsprung mass displacement respectively, and determine the to-be-tested sprung mass displacement, the to-be-tested unsprung mass displacement and an international roughness index based on the fifth weight coefficient and the sixth weight coefficient.

[0028] Further preferably, the third aspect of the present application provides an electronic device comprising a processor and a memory; the memory has a computer program stored therein, wherein the computer program, when executed by the processor, implements the roughness prediction method based on wavelets and physical neural networks of the first aspect.

[0029] Further preferably, the fourth aspect of the present application provides a non-transitory computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the roughness prediction method based on wavelets and physical neural networks of the first aspect.

[0030] The roughness prediction method and system based on wavelets and physical neural networks of the present application have the following beneficial effects over the prior art:

[0031] 1. The mapping relationship between the wavelet parameters of the road elevation and the wavelet parameters of the vehicle and suspension is established by using a neural network. Through wavelet decomposition, the elevation function is mapped to the coefficient space of different scales and positions. Even if the sampling interval is uneven, the local characteristics can still be accurately captured through the compact support of the wavelet basis function. After projecting the longitudinal profile elevation of the road, the mass displacement above and below the spring to the wavelet domain, the physical constraints are forced to be satisfied through the loss function, and the localization characteristics of the wavelet basis function make the abnormal points appear as isolated peaks in the coefficient space. Through the explicit suppression of false responses by the loss function, the robust processing of the change of sampling interval and sudden road abnormalities is realized, and the accuracy and calculation efficiency of the international smoothness evaluation are improved.

[0032] 2. In the model training, a phased training strategy and different loss functions are introduced. In the early stage, only the first loss function is used to train the model, and the gradient direction focuses on data fitting, so that the model quickly learns the coarse-grained mapping relationship between the road elevation and the vehicle displacement. In the later stage, the second loss function and the third loss function are introduced, the model parameters are refined through gradient superposition, and the output is ensured to meet the physical law. Through the phased switching of the loss function combination, the dynamic adjustment of the loss weight is indirectly realized, so that the model has the ability of multi-scale feature learning, and is suitable for the scene of change of sampling interval or sudden abnormality. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0034] Figure 1 A flowchart of a smoothness prediction method based on wavelet and physical neural network provided by an embodiment of the present application;

[0035] Figure 2 A structural schematic diagram of a quarter car model provided by an embodiment of the present application;

[0036] Figure 3 A structural schematic diagram of a smoothness prediction system based on wavelet and physical neural network provided by an embodiment of the present application;

[0037] Figure 4 A structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0039] In some embodiments, as shown in Figure 1 Figure 1 A flowchart of a flatness prediction method based on a wavelet and a physical neural network provided by an embodiment of the present application; the flatness prediction method based on a wavelet and a physical neural network provided by the present application comprises:

[0040] S110, calculating a first weight coefficient of a longitudinal profile elevation function of a road surface by using a scale function of a dbN wavelet; solving a quarter-car dynamic equation to determine a sprung mass displacement and an unsprung mass displacement, and obtaining a second weight coefficient and a third weight coefficient corresponding to the sprung mass displacement and the unsprung mass displacement respectively by wavelet transform.

[0041] The dbN wavelet is a kind of orthogonal wavelet basis function, which projects a signal to different frequency subbands while maintaining orthogonality through multi-scale decomposition. The scale function and the wavelet function of the dbN wavelet have N-1 order continuous derivatives, for example, the db4 wavelet has 3 order continuous derivatives, and the scale function and the wavelet function have continuous and smooth characteristics, and the support length is 2N, where N is the wavelet order. Taking N=8 as an example, the scale function can be expressed as:

[0042] ;

[0043] wherein, denotes the scale function decomposition layer number, denotes the scale function displacement coefficient, and Z represents an integer.

[0044] In some embodiments, the first weight coefficient of the longitudinal profile elevation function of the road surface is calculated by using the scale function of the dbN wavelet, comprising:

[0045] In a continuous sampling time range, the inner product of the longitudinal profile elevation function of the road surface and the scale function is integrated to obtain the first weight coefficient of the longitudinal profile elevation function of the road surface.

[0046] In this embodiment, the first weight coefficient of the longitudinal profile elevation function can be calculated by the following formula:

[0047] ;

[0048] wherein, ​Represents the total sampling time. This represents the sampling time.

[0049] In some embodiments, the dynamic equations of a quarter-car are solved to determine the unsprung mass displacement and the sprung mass displacement. Wavelet transform is then used to obtain the second and third weighting coefficients corresponding to the unsprung mass displacement and the sprung mass displacement, respectively, including:

[0050] The dynamic equations for a quarter-car are created based on sprung mass, unsprung mass, suspension stiffness coefficient, suspension damping coefficient, tire stiffness coefficient, sprung velocity, unsprung velocity, sprung acceleration, unsprung acceleration, sprung mass displacement, and unsprung mass displacement.

[0051] The dynamic equations of a quarter-car were solved using the Runge-Kutta method, and wavelet transform was performed on the solution to determine the second and third weighting coefficients corresponding to the unsprung mass displacement and the sprung mass displacement, respectively.

[0052] In this embodiment, please refer to Figure 2 , Figure 2 This is a schematic diagram of a quarter-car model provided in an embodiment of the present invention; the quarter-car model includes unsprung mass, sprung mass, suspension springs, shock absorbers, and tire stiffness. Unsprung mass Represents the mass of the wheels, tires, and some suspension components; sprung mass Representing the mass of the vehicle body and its load, the suspension spring is an elastic element connecting the unsprung mass and the sprung mass, simulating the stiffness of the suspension, which can be measured by the suspension stiffness coefficient. This indicates that the shock absorber is connected in parallel with the suspension spring to dissipate vibration energy and reduce vehicle body bumps. Its performance parameters can be obtained from the suspension damping coefficient. This indicates tire stiffness. Simulates tire elasticity, connecting unsprung mass and road surface. V represents the speed of the quarter-car model.

[0053] The dynamic equation for the quarter-car is shown in Equation 5:

[0054] ;

[0055] in, Indicates the displacement of the spring mass. This indicates the displacement of the unsprung mass. , They are respectively The first and second derivatives, , They are respectively The first and second derivatives.

[0056] The answer can be obtained using Runge-Kutta or other methods for solving differential equations. , The value of is calculated using equation (2). , The corresponding second weight coefficients and the third weighting coefficient As shown in equations (6) and (7):

[0057] .

[0058] In some embodiments, before training a preset deep learning network model using a first weight coefficient as input data and a second and third weight coefficient as output data, the following steps are included:

[0059] The training dataset is obtained by calculating multiple sets of road surface samples; the training dataset includes multiple sets of corresponding first weight coefficients, second weight coefficients and third weight coefficients.

[0060] In this embodiment, high-precision road surface measurement equipment, such as laser road surface rangefinders and inertial measurement units, is used to collect multiple sets of road surface elevation data, ensuring that the samples cover different types of road surfaces, such as flat roads, slopes, potholes, cracks, etc., and data under different driving speeds. For each set of road surface samples, multi-scale decomposition is performed using the dbN wavelet scaling function to calculate the first weight coefficient. Using a quarter-car model, dynamic simulation is performed on each set of road surface samples to calculate the unsprung mass displacement and sprung mass displacement, and then the corresponding second and third weight coefficients are obtained through wavelet transform. The first, second, and third weight coefficients corresponding to each set of road surface samples are integrated into a single data sample.

[0061] S120: Using the first weight coefficient as input data and the second and third weight coefficients as output data, train the preset deep learning network model to determine the loss function of the preset deep learning network model; use the Bayesian optimization algorithm and the loss function to iteratively optimize the preset deep learning network model to obtain the target deep learning network model.

[0062] In some embodiments, a preset deep learning network model includes an encoder, a decoder, and a residual module; training the preset deep learning network model by using a first weight coefficient as input data and a second and a third weight coefficient as output data includes:

[0063] The encoder and decoder process the temporal-scale features of the first, second, and third weight coefficients, and then, combined with the loss function determined by the residual module, adjust the parameters of the preset deep learning network model.

[0064] In some embodiments, the loss function includes a first loss function, a second loss function, and a third loss function; training a preset deep learning network model using the first weight coefficient as input data and the second and third weight coefficients as output data, and determining the loss function of the preset deep learning network model, includes:

[0065] The sum of the square of the difference between the predicted value and the true value corresponding to the second weight coefficient and the square of the difference between the predicted value and the true value corresponding to the third weight coefficient is averaged, and the mean is determined as the first loss function.

[0066] The vehicle dynamic response loss term is created based on the predicted values ​​of sprung mass, suspension stiffness coefficient, suspension damping coefficient, various order connection coefficients, second weight coefficient, and third weight coefficient, and the second loss function is determined by using sprung mass as a constraint.

[0067] A vehicle dynamic response loss term is created based on the predicted values ​​of unsprung mass, tire stiffness coefficient, suspension stiffness coefficient, suspension damping coefficient, various order correlation coefficients, first weight coefficient, second weight coefficient, and third weight coefficient. The third loss function is determined by using unsprung mass as a constraint.

[0068] In this embodiment, the first loss function can be expressed as:

[0069] ;

[0070] The second loss function can be expressed as:

[0071] ;

[0072] The third loss function can be expressed as:

[0073] ;

[0074] in, , The second weighting coefficients are respectively and the third weighting coefficient The corresponding predicted value, The calculation formula is:

[0075]

[0076]

[0077]

[0078] in, , Wavelet basis functions representing different positions, , Characterization The first and second derivatives with respect to time n.

[0079] In some embodiments, a preset deep learning network model is iteratively optimized using a Bayesian optimization algorithm and a loss function to obtain a target deep learning network model, including:

[0080] If the number of iterations is less than the first threshold, the loss function of the preset deep learning network model is determined based on the first loss function;

[0081] If the number of iterations is not less than the first threshold, the loss function of the preset deep learning network model is determined by the weighted sum of the first loss function, the second loss function, and the third loss function.

[0082] Here, by switching the combination of loss functions in stages, the dynamic adjustment of loss weights is indirectly achieved, enabling the model to have multi-scale feature learning capabilities, which is suitable for scenarios with changing sampling intervals or sudden anomalies.

[0083] The training dataset obtained in the preceding steps is divided into a training set and a test set in an 8:2 ratio, and the total number of training rounds for the model is set to [number missing]. The deep learning network is trained using the training set data, and iterative training is performed using the loss function until the final loss function meets the preset requirements, thus obtaining the target deep learning network model.

[0084] S130, the fourth weight coefficient of the longitudinal profile elevation function of the road surface to be measured is input into the target deep learning network model to obtain the fifth and sixth weight coefficients corresponding to the unsprung mass displacement and the sprung mass displacement to be measured, respectively, and the unsprung mass displacement, the sprung mass displacement and the international roughness index are determined based on the fifth and sixth weight coefficients.

[0085] The fourth weight coefficient of the longitudinal profile elevation function of the road surface to be measured is input into the target deep learning network model to obtain the fifth and sixth weight coefficients corresponding to the unsprung mass displacement and the sprung mass displacement to be measured, respectively. Then, the inverse wavelet transform is used to obtain and :

[0086] ;

[0087] ;

[0088] The International Roughness Index can be expressed as:

[0089]

[0090] In an optional embodiment, please refer to Table 1. Table 1 shows the IRI difference before and after interpolation of multiple road surfaces with different missing proportions for the deep learning network of this application at different training rounds and the traditional method. It can be seen that the IRI difference error of the method proposed in this application is lower than that of the traditional method. Moreover, through the paired t-test of statistical hypotheses, the error of the method proposed in this application is significantly lower than that of the traditional method, as shown in Table 2.

[0091]

[0092] Table 1

[0093]

[0094] Table 2

[0095] In some embodiments, please refer to Figure 3 , Figure 3 This is a schematic diagram of a smoothness prediction system based on wavelet and physical neural network provided in an embodiment of the present invention. The present invention provides a smoothness prediction system 300 based on wavelet and physical neural network, including: a data acquisition module 310, a training and optimization module 320, and a parameter determination module 330; wherein,

[0096] The data acquisition module 310 is configured to use the scaling function of the dbN wavelet to calculate the first weight coefficient of the longitudinal profile elevation function of the road surface; solve the dynamic equation of the quarter-car to determine the unsprung mass displacement and the sprung mass displacement; and obtain the second weight coefficient and the third weight coefficient corresponding to the unsprung mass displacement and the sprung mass displacement respectively through wavelet transform.

[0097] The training optimization module 320 is configured to take the first weight coefficient as input data and the second and third weight coefficients as output data to train the preset deep learning network model and determine the loss function of the preset deep learning network model; and to iteratively optimize the preset deep learning network model using the Bayesian optimization algorithm and the loss function to obtain the target deep learning network model.

[0098] The parameter determination module 330 is configured to input the fourth weight coefficient of the longitudinal profile elevation function of the road surface to be measured into the target deep learning network model, obtain the fifth weight coefficient and the sixth weight coefficient corresponding to the unsprung mass displacement and the sprung mass displacement to be measured, respectively, and determine the unsprung mass displacement, the sprung mass displacement and the international roughness index based on the fifth weight coefficient and the sixth weight coefficient.

[0099] In some embodiments, the data acquisition module 310 is specifically configured as follows:

[0100] Within the continuous sampling time range, the inner product of the road longitudinal profile elevation function and the scale function is integrated to obtain the first weighting coefficient of the road longitudinal profile elevation function.

[0101] In some embodiments, the data acquisition module 310 is specifically configured as follows:

[0102] The dynamic equations for a quarter-car are created based on sprung mass, unsprung mass, suspension stiffness coefficient, suspension damping coefficient, tire stiffness coefficient, sprung velocity, unsprung velocity, sprung acceleration, unsprung acceleration, sprung mass displacement, and unsprung mass displacement.

[0103] The dynamic equations of a quarter-car were solved using the Runge-Kutta method, and wavelet transform was performed on the solution to determine the second and third weighting coefficients corresponding to the unsprung mass displacement and the sprung mass displacement, respectively.

[0104] In some embodiments, the data acquisition module 310 is specifically configured as follows:

[0105] The training dataset is obtained by calculating multiple sets of road surface samples; the training dataset includes multiple sets of corresponding first weight coefficients, second weight coefficients and third weight coefficients.

[0106] In some embodiments, the preset deep learning network model includes an encoder, a decoder, and a residual module; the training optimization module 320 is specifically configured as follows:

[0107] The encoder and decoder process the temporal-scale features of the first, second, and third weight coefficients, and then, combined with the loss function determined by the residual module, adjust the parameters of the preset deep learning network model.

[0108] In some embodiments, the loss function includes a first loss function, a second loss function, and a third loss function; the training optimization module 320 is specifically configured as follows:

[0109] The sum of the square of the difference between the predicted value and the true value corresponding to the second weight coefficient and the square of the difference between the predicted value and the true value corresponding to the third weight coefficient is averaged, and the mean is determined as the first loss function.

[0110] The vehicle dynamic response loss term is created based on the predicted values ​​of sprung mass, suspension stiffness coefficient, suspension damping coefficient, various order connection coefficients, second weight coefficient, and third weight coefficient, and the second loss function is determined by using sprung mass as a constraint.

[0111] A vehicle dynamic response loss term is created based on the predicted values ​​of unsprung mass, tire stiffness coefficient, suspension stiffness coefficient, suspension damping coefficient, various order correlation coefficients, first weight coefficient, second weight coefficient, and third weight coefficient. The third loss function is determined by using unsprung mass as a constraint.

[0112] In some embodiments, the training optimization module 320 is specifically configured as follows:

[0113] If the number of iterations is less than the first threshold, the loss function of the preset deep learning network model is determined based on the first loss function;

[0114] If the number of iterations is not less than the first threshold, the loss function of the preset deep learning network model is determined by the weighted sum of the first loss function, the second loss function, and the third loss function.

[0115] It should be noted that the smoothness prediction system based on wavelet and physical neural network provided in this application embodiment and the smoothness prediction method based on wavelet and physical neural network provided in this application embodiment are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned smoothness prediction method based on wavelet and physical neural network, and the repeated parts will not be described again.

[0116] In some embodiments, please refer to Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device 400 provided in this application includes a processor 410 and a memory 420; the memory 420 stores a computer program, wherein the computer program, when executed by the processor, implements the aforementioned smoothness prediction method based on wavelet and physical neural networks.

[0117] Specifically, processor 410 may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. Processor 410 may also include onboard memory for caching purposes. Processor 410 may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0118] Memory 420 may be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, memory 420 may include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, instruments, or propagation media. Specific examples of memory 420 include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and may also be random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0119] This application also provides a non-transitory computer-readable medium storing a computer program thereon, which, when executed by a processor, implements the aforementioned wavelet and physical neural network-based flatness prediction method. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0120] According to embodiments of this application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.

[0121] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments and / or claims of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments, but should be defined not only by the appended claims, but also by their equivalents. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the protection scope of this invention.

Claims

1. A method for predicting smoothness based on wavelet and physical neural network, characterized in that, include: The first weighting coefficient of the road longitudinal profile elevation function is calculated using the scaling function of the dbN wavelet. Solve the dynamic equations of the quarter-car to determine the unsprung mass displacement and the sprung mass displacement, and obtain the second and third weighting coefficients corresponding to the unsprung mass displacement and the sprung mass displacement respectively through wavelet transform; Using the first weight coefficient as input data and the second and third weight coefficients as output data, a preset deep learning network model is trained to determine the loss function of the preset deep learning network model; the preset deep learning network model is iteratively optimized using the Bayesian optimization algorithm and the loss function to obtain the target deep learning network model. The fourth weight coefficient of the longitudinal profile elevation function of the road surface to be tested is input into the target deep learning network model to obtain the fifth and sixth weight coefficients corresponding to the unsprung mass displacement and the sprung mass displacement to be tested, respectively. Based on the fifth and sixth weight coefficients, the unsprung mass displacement, the sprung mass displacement and the international roughness index are determined.

2. The smoothness prediction method based on wavelet and physical neural network as described in claim 1, characterized in that, The first weighting coefficients for calculating the longitudinal profile elevation function of the road surface using the scaling function of the dbN wavelet include: Within a continuous sampling time range, the inner product of the road surface longitudinal profile elevation function and the scale function is integrated to obtain the first weighting coefficient of the road surface longitudinal profile elevation function.

3. The smoothness prediction method based on wavelet and physical neural network as described in claim 1, characterized in that, Solving the dynamic equations of the quarter-car to determine the second and third weighting coefficients corresponding to the unsprung mass displacement and sprung mass displacement, respectively, includes: The dynamic equations for a quarter-car are created based on sprung mass, unsprung mass, suspension stiffness coefficient, suspension damping coefficient, tire stiffness coefficient, sprung velocity, unsprung velocity, sprung acceleration, unsprung acceleration, sprung mass displacement, and unsprung mass displacement. The dynamic equations of the quarter-car are solved using the Runge-Kutta method, and wavelet transform is performed on the solution to determine the second and third weighting coefficients corresponding to the unsprung mass displacement and the sprung mass displacement, respectively.

4. The smoothness prediction method based on wavelet and physical neural network as described in claim 1, characterized in that, Before training a preset deep learning network model using the first weight coefficient as input data and the second and third weight coefficients as output data, the method includes: A training dataset is obtained by calculating multiple sets of road surface samples; the training dataset includes multiple sets of corresponding first weight coefficients, second weight coefficients and third weight coefficients.

5. The smoothness prediction method based on wavelet and physical neural network as described in claim 1, characterized in that, The preset deep learning network model includes an encoder, a decoder, and a residual module; the step of training the preset deep learning network model by using the first weight coefficient as input data and the second and third weight coefficients as output data includes: The encoder and decoder process the temporal-scale features of the first weight coefficient, the second weight coefficient, and the third weight coefficient, and then, in conjunction with the loss function determined by the residual module, adjust the parameters of the preset deep learning network model.

6. The smoothness prediction method based on wavelet and physical neural network as described in claim 1, characterized in that, The loss function includes a first loss function, a second loss function, and a third loss function; the step of training a preset deep learning network model using the first weight coefficient as input data and the second and third weight coefficients as output data to determine the loss function of the preset deep learning network model includes: The sum of the square of the difference between the predicted value and the true value of the second weight coefficient and the square of the difference between the predicted value and the true value of the third weight coefficient is averaged, and the average is determined as the first loss function. A vehicle dynamic response loss term is created based on the predicted values ​​of sprung mass, suspension stiffness coefficient, suspension damping coefficient, various order connection coefficients, second weight coefficient, and third weight coefficient, and the second loss function is determined by using the sprung mass as a constraint. A vehicle dynamic response loss term is created based on the predicted values ​​of unsprung mass, tire stiffness coefficient, suspension stiffness coefficient, suspension damping coefficient, various order correlation coefficients, first weight coefficient, second weight coefficient, and third weight coefficient. The third loss function is determined by using the unsprung mass as a constraint.

7. The smoothness prediction method based on wavelet and physical neural network as described in claim 6, characterized in that, The step of iteratively optimizing the preset deep learning network model using the Bayesian optimization algorithm and the loss function to obtain the target deep learning network model includes: If the number of iterations is less than the first threshold, the loss function of the preset deep learning network model is determined based on the first loss function; If the number of iterations is not less than a first threshold, the loss function of the preset deep learning network model is determined based on the weighted sum of the first loss function, the second loss function, and the third loss function.

8. A smoothness prediction system based on wavelet and physical neural network, characterized in that, include: The module consists of a data acquisition module, a training and optimization module, and a parameter determination module; among which, The data acquisition module is configured to calculate the first weighting coefficient of the longitudinal profile elevation function of the road surface using the scaling function of the dbN wavelet; solve the dynamic equation of the quarter-car to determine the unsprung mass displacement and the sprung mass displacement; and obtain the second weighting coefficient and the third weighting coefficient corresponding to the unsprung mass displacement and the sprung mass displacement respectively through wavelet transform. The training optimization module is configured to use the first weight coefficient as input data, the second weight coefficient and the third weight coefficient as output data to train a preset deep learning network model, determine the loss function of the preset deep learning network model, and iteratively optimize the preset deep learning network model using a Bayesian optimization algorithm and the loss function to obtain a target deep learning network model. The parameter determination module is configured to input the fourth weight coefficient of the longitudinal profile elevation function of the road surface to be measured into the target deep learning network model to obtain the fifth weight coefficient and the sixth weight coefficient corresponding to the unsprung mass displacement and the sprung mass displacement to be measured, respectively, and determine the unsprung mass displacement, the sprung mass displacement and the international roughness index based on the fifth weight coefficient and the sixth weight coefficient.

9. An electronic device comprising a processor and a memory; said memory having a storage for a computer program, wherein, When the computer program is executed by the processor, it implements the smoothness prediction method based on wavelet and physical neural network as described in any one of claims 1 to 7.

10. A non-transitory computer storage medium, characterized in that, It stores a computer program, wherein the computer program, when executed by a processor, implements the smoothness prediction method based on wavelet and physical neural network as described in any one of claims 1 to 7.

Images