A Deep Learning-Based Method for Optimizing Car Sound Pressure Level

Abstract

This invention is a deep learning-based method for optimizing the sound pressure level (SPL) of automotive engine sounds. It includes: 1. Loading data from various operating conditions for a specific vehicle model; 2. Setting input and output data; 3. Setting training and testing sets; 4. Normalizing the training and testing sets; 5. Solving for the optimal number of hidden layers, using mean squared error to determine if the number of hidden layers is optimal, and if so, performing network training and testing; 6. Outputting the final result; 7. Optimizing the SPL of the automotive engine sounds based on the final output result. This invention predicts subsequent torque information in advance based on accelerator pedal position changes and real-time motor speed; it assigns different weights to the pedal gradient under different operating conditions, and achieves engine sound pressure level control by reasonably modifying the SPL mapping relationship. This solves the problem of poor responsiveness caused by torque lag during sudden pedal changes, which reduces the user experience of active sound synthesis.

Description

Technical Field

[0001] This invention belongs to the field of automotive technology, specifically a method for optimizing the sound pressure level of automotive sound waves based on deep learning. Background Technology

[0002] With the current automotive industry's "five major trends"—lightweighting, electrification, intelligence, networking, and sharing—new energy vehicles are developing rapidly, gradually replacing traditional internal combustion engines. To enhance driving pleasure and experience, active sound systems have emerged. These systems use electronically controlled ECUs to detect and synchronize the vehicle's dynamic response in real time, synthesizing sound data corresponding to the vehicle's feedback through algorithms. This sound data is amplified by a power amplifier and then emitted through an external booster unit. Currently, there are many types of synthesized sounds, such as traditional engine sounds and technologically advanced sound types. This diversity of sound options enhances the enjoyment of modification and driving, creating endless possibilities for sound timbre.

[0003] The car's sound system uses CAN to obtain basic information such as vehicle speed and pedal opening to synthesize the sound. The pedal is the most direct information output by the user. If only the pedal is relied upon to synthesize the engine sound, it will not be realistic. If only torque is used, the response will be slow. Under normal circumstances, torque values ​​can be used, but in the case of sudden pedal changes, the torque value will lag and may cause poor responsiveness, thus reducing the user experience of active sound synthesis. Summary of the Invention

[0004] This invention provides a deep learning-based method for optimizing the sound pressure level of automotive engines. It predicts subsequent torque information in advance based on accelerator pedal position change information and real-time motor speed. Different weights are assigned to the pedal gradient under different operating conditions. By reasonably modifying the sound pressure level mapping relationship, the sound pressure level of the engine is controlled. This solves the problem that the torque value lags behind when the pedal changes abruptly, which can lead to poor responsiveness and reduce the user experience of active sound synthesis.

[0005] The technical solution of this invention is described below in conjunction with the accompanying drawings:

[0006] A deep learning-based method for optimizing the sound pressure level of car exhaust sounds includes the following steps:

[0007] Step 1: Load data for a specific vehicle model under various operating conditions;

[0008] Step 2: Set the input and output data;

[0009] Step 3: Set up the training and testing sets;

[0010] Step 4: Normalize the training and test sets;

[0011] Step 5: Find the optimal number of hidden layers. Use the mean squared error to determine if the number of hidden layers is optimal. If it is, perform network training and testing.

[0012] Step 6: Output the final result;

[0013] Step 7: Optimize the sound pressure level of the car engine sound based on the final output.

[0014] Furthermore, in step one, the various operating conditions include constant speed operating conditions, acceleration operating conditions, and deceleration operating conditions.

[0015] Furthermore, in step one, the data under various operating conditions includes accelerator pedal position, real-time motor speed, vehicle speed, torque loss, and torque.

[0016] Furthermore, in step two, the input data includes the accelerator pedal position, real-time motor speed, vehicle speed, and torque loss; the output data is torque.

[0017] Furthermore, in step three, the training set is an M-row, 5-column matrix; the test set dataset is an N-row, 5-column matrix.

[0018] Furthermore, M > N.

[0019] Furthermore, in step four, the input of the training set is normalized to the interval [0,1], and the output of the training set is normalized to the interval [-1,1].

[0020] Furthermore, in step four, the input of the test set is normalized to the interval [0,1], and the output of the test set is normalized to the interval [-1,1].

[0021] Furthermore, in step five, the mapping relationship between the sound pressure level of the sound wave in sound synthesis and the pedal opening, vehicle speed, motor speed, and torque is shown in the following formula, where V, n, p, and T are... L T represents vehicle speed, motor speed, pedal opening, torque loss, and torque, respectively; ω v ω n ω p1 ω TL1 ω T1 The torque gradient is added to the corresponding weights to solve the problem of torque change delay when the pedal changes suddenly. The torque gradient needs to be predicted based on the BP neural network.

[0022]

[0023] Wherein, ΔT is the difference between the predicted value based on the Bp neural network and the actual torque obtained by CAN;

[0024] The learning and prediction are based on a BP neural network with four inputs: vehicle speed, motor speed, pedal opening and torque loss. The activation functions of the neural network are the purelin function and the tansig function, respectively.

[0025] The number of hidden layers m is determined according to the following formula, and a BP neural network with the optimal number of hidden layers is constructed.

[0026]

[0027] Where, n in n out Let 'a' be the number of input and output layers, and 'a' be a random number.

[0028] Finally, the accuracy of the constructed neural network model is judged based on the correlation coefficient R and the mean squared error.

[0029] Furthermore, the value of 'a' is between 1 and 10.

[0030] The beneficial effects of this invention are as follows:

[0031] This invention predicts subsequent torque information in advance based on accelerator pedal position change information and real-time motor speed. It assigns different weights to the pedal gradient under different operating conditions and optimizes the engine sound pressure level control algorithm by reasonably modifying the sound pressure level mapping relationship. This solves the problem that the torque value lags behind when the pedal changes abruptly, which can lead to poor responsiveness and reduce the user experience of active sound synthesis. Attached Figure Description

[0032] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a schematic diagram of the process of the present invention;

[0034] Figure 2 This is a schematic diagram of the correlation coefficient test matrix between sound pressure level and various parameters.

[0035] Figure 3 This includes measured data on vehicle speed, motor speed, pedal opening, torque loss, and torque for a specific vehicle model.

[0036] Figure 4 This is a schematic diagram of the correlation coefficient test matrix between torque and vehicle speed, motor speed, pedal opening, and torque loss.

[0037] Figure 5 A schematic diagram of the correlation coefficients of the optimal hidden layer;

[0038] Figure 6 A schematic diagram comparing the predicted torque value of pedal opening and Bp with the measured torque value;

[0039] Figure 7 A schematic diagram comparing the predicted torque value of Bp with the measured torque value;

[0040] Figure 8 This is a schematic diagram illustrating the prediction error of a BP neural network on a test set.

[0041] Figure 9 This is a schematic diagram comparing the BP neural network prediction data with vehicle speed, engine speed, pedal opening, and torque loss. Detailed Implementation

[0042] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0043] In this invention, unless otherwise expressly specified and limited, the first feature "above" or "below" the second feature may include direct contact between the first and second features, or contact between the first and second features not in direct contact but through another feature between them.

[0044] Furthermore, "above," "over," and "on top" of the first feature and the second feature include the first feature being directly above or diagonally above the second feature, or simply indicating that the first feature is at a higher horizontal level than the second feature. "Below," "under," and "below" of the first feature and the second feature include the first feature being directly below or diagonally below the second feature, or simply indicating that the first feature is at a lower horizontal level than the second feature.

[0045] In the description of this embodiment, terms such as "upper," "lower," "left," and "right" are based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of description and simplification of operation, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.

[0046] Furthermore, the terms "first" and "second" are merely used to distinguish between descriptions and have no special meaning.

[0047] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0049] Example 1

[0050] The key part of the active sound wave synthesis algorithm lies in processing the mapping relationship between sound pressure level and pedal opening, vehicle speed, motor speed, and torque. Its correlation coefficient verification matrix is ​​as follows: Figure 2 As shown, speed represents vehicle speed, n represents motor speed, pedal represents pedal opening, Tloss represents torque loss, and soundSPL represents sound pressure level. Analysis of the Pearson correlation coefficient matrix shows that the sound pressure level is strongly correlated with vehicle speed, motor speed, torque, and pedal opening. That is, these parameters are the dependent variables of the mapping relationship of the sound synthesis algorithm, and sound pressure level is the independent variable.

[0051] Actual test data for a certain car model is as follows Figure 3 As shown, when the pedal opening suddenly changes at around 20 seconds, the torque is delayed by 100ms before the value begins to decrease. Without considering the delay of the sound synthesis algorithm, the driving experience will be greatly reduced.

[0052] To address the torque delay issue, we first analyze the correlation between torque and vehicle speed, motor speed, pedal opening, and torque loss. Based on... Figure 3 It is known that torque is strongly correlated with pedal opening, but in order to ensure the accuracy of torque prediction, the information on the change in accelerator pedal position is the primary factor, supplemented by the real-time motor speed, vehicle speed, and torque loss, to predict subsequent torque information.

[0053] See Figure 1 This invention provides a deep learning-based method for optimizing the sound pressure level of automobile sound waves, comprising the following steps:

[0054] Step 1: Load data for a specific vehicle model under various operating conditions;

[0055] Multiple operating conditions include constant speed, acceleration, and deceleration.

[0056] Data under various operating conditions includes accelerator pedal position, real-time motor speed, vehicle speed, torque loss, and torque.

[0057] Step 2: Set the input and output data;

[0058] The input data includes accelerator pedal position, real-time motor speed, vehicle speed, and torque loss; the output data is torque.

[0059] Step 3: Set up the training and testing sets;

[0060] In this embodiment, the training set data is a 1,463,572 x 5 matrix. The accelerator pedal position, real-time motor speed, vehicle speed, and torque loss are set as the training set inputs (first four columns), and torque is set as the training set output (last column). The test set data is a 17,245 x 5 matrix containing acceleration and deceleration processes. The test set inputs and outputs are set as the training set...

[0061] Step 4: Normalize the training and test sets;

[0062] For ease of calculation, the data is normalized. The training set input is normalized to the range [0,1], and the training set output is normalized to the default range [-1,1]. The test set input uses the same normalization method as the training set input.

[0063] Step 5: Find the optimal number of hidden layers. Use the mean squared error to determine if the number of hidden layers is optimal. If it is, perform network training and testing.

[0064] The mapping relationship between sound pressure level and pedal opening, vehicle speed, motor speed, and torque in the sound synthesis algorithm is shown in the following formula, where V, n, p, and T are... L T represents vehicle speed, motor speed, pedal opening, torque loss, and torque, respectively. ω v ω n ω p1 ω TL1 ω T1 Each of these is assigned a weight, and a torque gradient is added to address the issue of torque delay when the pedal changes abruptly. The torque gradient needs to be predicted using a BP neural network.

[0065]

[0066] Wherein, ΔT is the difference between the predicted value based on the Bp neural network and the actual torque obtained by CAN.

[0067] Due to the large amount of test data and the lack of obvious functional relationship between torque and accelerator pedal position, real-time motor speed, vehicle speed, and torque loss, a backpropagation neural network was used for learning and prediction. The inputs were four: vehicle speed, motor speed, pedal opening, and torque loss. The activation functions of the neural network were the purelin function and the tansig function, respectively.

[0068] The number of hidden layers m is determined according to the following formula to construct a BP neural network with the optimal number of hidden layers and reduce errors.

[0069]

[0070] Where, n in n out The number of input and output layers is denoted by 'a', where 'a' is a random number ranging from 1 to 10.

[0071] The accuracy of the constructed neural network model is judged based on the correlation coefficient R and the mean squared error.

[0072] The optimal number of hidden layer nodes in this embodiment is 9, with a mean squared error of 0.0069497. The prediction results are as follows: Figure 5-9 As shown; via Figure 5 It can be seen that when the number of hidden layers is 9, the correlation coefficient is around 0.98, indicating a high degree of correlation and good data fit. (This is achieved through...) Figure 6-7 It can be seen that when the pedal opening changes suddenly, the curve predicted by the neural network has a low delay, and the torque changes immediately with the pedal opening, while the measured data shows a delay of 100ms. Figure 8 and Figure 9 It can be seen that although there is an error between the predicted value and the measured value, the error is small, which also proves that the algorithm can solve the torque delay problem and ensure the accuracy of the value. Figure 9 A comparison chart of BP neural network prediction data with vehicle speed, engine speed, pedal opening, and torque loss.

[0073] Step 6: Output the final result;

[0074] Step 7: Optimize the sound pressure level of the car engine sound based on the final output.

[0075] In summary, this invention predicts subsequent torque information in advance based on accelerator pedal position change information and real-time motor speed; it assigns different weights to the pedal gradient under different operating conditions, and achieves engine sound pressure level control by reasonably modifying the sound pressure level mapping relationship. This solves the problem that torque values ​​tend to lag and have poor responsiveness when the pedal changes abruptly, thus reducing the user experience of active sound synthesis.

[0076] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the scope of protection of the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, any person skilled in the art can make equivalent substitutions or changes based on the technical solution and inventive concept of the present invention within the scope of the technology disclosed in the present invention. These simple modifications are all within the scope of protection of the present invention.

[0077] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

[0078] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed by the present invention.

Claims

1. A method for optimizing the sound pressure level of automobile noise based on deep learning, characterized in that, Includes the following steps: Step 1: Load data for a specific vehicle model under various operating conditions; Step 2: Set the input and output data; Step 3: Set up the training and testing sets; Step 4: Normalize the training and test sets; Step 5: Find the optimal number of hidden layers. Use the mean squared error to determine if the number of hidden layers is optimal. If it is, perform network training and testing. Step 6: Output the final result; Step 7: Optimize the sound pressure level of the car engine based on the final output; In step one, the various operating conditions include constant speed operating condition, acceleration operating condition and deceleration operating condition; In step one, the data under various operating conditions include accelerator pedal position, real-time motor speed, vehicle speed, torque loss, and torque. In step two, the input data are accelerator pedal position, real-time motor speed, vehicle speed, and torque loss; the output data is torque. In step three, the training set is an M-row, 5-column matrix; the test set dataset is an N-row, 5-column matrix; M > N; In step four, the input of the training set is normalized to the interval [0,1], and the output of the training set is normalized to the interval [-1,1]. In step four, the input of the test set is normalized to the interval [0,1], and the output of the test set is normalized to the interval [-1,1]. In step five, the mapping relationship between the sound pressure level of the sound wave and the pedal opening, vehicle speed, motor speed, and torque in sound synthesis is shown in the following formula, where V, n, p, and T are... L T represents vehicle speed, motor speed, pedal opening, torque loss, and torque, respectively. , , The torque gradient is added to the corresponding weights to solve the problem of torque change delay when the pedal changes suddenly. The torque gradient needs to be predicted based on the BP neural network. ； Wherein, ΔT is the difference between the predicted value based on the Bp neural network and the actual torque obtained by CAN; The learning and prediction are based on a BP neural network with four inputs: vehicle speed, motor speed, pedal opening and torque loss. The activation functions of the neural network are the purelin function and the tansig function, respectively. The number of hidden layers m is determined according to the following formula, and a BP neural network with the optimal number of hidden layers is constructed. ； in, , The number of input and output layers. It is a random number; Finally, the accuracy of the constructed neural network model is judged based on the correlation coefficient R and the mean squared error.

2. The method for optimizing the sound pressure level of automobile sound waves based on deep learning according to claim 1, characterized in that, The The value ranges from 1 to 10.

Images