An optimization method for environmentally friendly low flow resistance intake muffler

Abstract

The present invention relates to the technical field of vehicle exhaust silencer, and in particular to an environmentally friendly low-resistance intake silencer optimization method, comprising: acquiring silencer data information; determining a target model state according to an abnormal data value proportion and a silencer simulation coefficient, and determining whether to optimize the target model according to the target model state; determining a simulation state according to an abnormal value distribution coefficient and a contour difference, and determining a model optimization method according to the simulation state; when selecting desensitized associated training data for supplementary training, determining a desensitization method of the associated training data according to a sub-data correlation degree, and selecting the desensitized associated training data for training the target model; when extracting complementary training data for supplementary training, determining a complementary data extraction method according to an abnormal factor proportion and an abnormal factor discreteness. The present invention can optimize the model training data according to the model simulation result to improve the accuracy of the model simulation result.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle exhaust silencer, and in particular to an environmentally friendly low flow resistance intake silencer optimization method. Background Art

[0002] When optimizing an environmentally friendly low-resistance intake muffler, a simulation is usually performed based on a transmission loss model to optimize the muffler structure based on the simulation results. However, when training the model, the quality of the muffler training data is inconsistent, resulting in abnormal data values ​​in the model simulation results, which deviate greatly from the actual results. Therefore, how to optimize the model training data based on the model simulation results and improve the accuracy of the model simulation results is a technical problem that needs to be urgently solved by technical personnel in this field.

[0003] Chinese patent publication number CN113626931A discloses a modeling method, device, storage medium and computer equipment for a muffler, including: constructing a linear constraint function; constructing a nonlinear constraint function; calculating the obtained objective function, linear constraint function and nonlinear constraint function through a genetic algorithm function to generate a muffler parameter set; extracting multiple muffler parameters from the muffler parameter set; modeling the muffler based on multiple muffler parameters. It can be seen that the above technical solution has the following problems: the simulation effect of the muffler model is not detected, and the model training data is not adaptively optimized when there is a deviation between the model simulation effect and the actual result, resulting in poor accuracy of the model simulation. Summary of the invention

[0004] To this end, the present invention provides an environmentally friendly low flow resistance intake muffler optimization method to overcome the problem that the prior art does not detect the muffler model effect and does not adaptively optimize the model training data when there is a deviation between the model simulation effect and the actual result, resulting in poor accuracy of the model simulation.

[0005] To achieve the above object, the present invention provides an environmentally friendly low flow resistance intake muffler optimization method, comprising:

[0006] Get muffler data information;

[0007] Determine the state of the target model according to the proportion of abnormal data values ​​and the noise elimination simulation coefficient, and determine whether to optimize the target model according to the state of the target model;

[0008] In the optimization of the target model, the simulation state is determined according to the abnormal value distribution coefficient and the contour difference, and the model optimization method is determined according to the simulation state to select the desensitized associated training data for supplementary training or extract complementary training data for supplementary training;

[0009] When selecting the desensitized associated training data for supplementary training, the desensitization method of the associated training data is determined according to the sub-data correlation degree, which is to desensitize all sub-data or desensitize part of the sub-data, and the desensitized associated training data is selected for training the target model;

[0010] When extracting complementary training data for supplementary training, the complementary data extraction method is determined based on the abnormal factor proportion and the abnormal factor dispersion, that is, complementary data extraction is performed based on the difference influence coefficient or the factor difference threshold;

[0011] The associated training data is training data of a standard model whose similarity ratio with the target model training data is greater than a preset similarity ratio.

[0012] Furthermore, when the state of the target model is that the proportion of abnormal data values ​​is greater than or equal to the preset proportion of abnormal data values ​​or the noise reduction simulation coefficient is greater than or equal to the preset noise reduction simulation coefficient, the target model is optimized.

[0013] Furthermore, the method for confirming the proportion of abnormal data values ​​is to input the muffler parameters into the target model and perform a preset number of simulations to obtain a number of simulated transmission loss images, and record the data values ​​in the simulated transmission loss images whose data value changes are greater than the preset data value changes or whose adjacent data fluctuation mean is greater than the preset adjacent data fluctuation mean as abnormal data values, and record the ratio of the total amount of abnormal data values ​​to all data values ​​as the proportion of abnormal data values;

[0014] The method for confirming the noise reduction simulation coefficient is to record the ratio of the average value of the transmission loss corresponding to each divided point after dividing a single simulated transmission loss image into M equal parts and the muffler characteristic value as a reference ratio, and the noise reduction simulation coefficient is the average value of the reference ratios corresponding to each simulated transmission loss image.

[0015] Furthermore, the simulation state includes:

[0016] A first simulation state in which the abnormal value distribution coefficient is greater than or equal to a preset abnormal value distribution coefficient or the contour difference is greater than or equal to a preset contour difference;

[0017] A second simulation state in which the abnormal value distribution coefficient is less than the preset abnormal value distribution coefficient and the contour difference is less than the preset contour difference.

[0018] Furthermore, in the first simulation state, if the sub-data association degree is greater than or equal to the preset sub-data association degree, the desensitization method is to desensitize all sub-data.

[0019] Furthermore, in the first simulation state, if the sub-data relevance is less than the preset sub-data relevance, the desensitization method is partial sub-data desensitization, wherein:

[0020] Desensitization is performed on each data set in each training data. When desensitization is performed on a single data set, the data set is recorded as the target data set, and the desensitization ratio of the sub-data that needs to be desensitized in the target data set is determined according to the average matching coefficient;

[0021] The desensitization ratio is positively correlated with the average matching coefficient.

[0022] Furthermore, in the second simulation state, complementary training data is extracted for supplementary training, wherein:

[0023] Detect the set of training data of the target model corresponding to the muffler parameters and abnormal values, and record it as the abnormal data set, record the data samples outside the abnormal data set as the normal data set, detect the abnormal factors in the abnormal data set, record the standard model with abnormal factors and model difference value less than the preset model difference value as the model to be extracted, and record the data samples extracted by complementary data extraction method as complementary training data, and extract the complementary training data to perform supplementary training on the target model.

[0024] Furthermore, if the abnormal factor proportion is greater than or equal to the preset abnormal factor proportion and the abnormal factor dispersion is less than the preset abnormal factor dispersion, the complementary data extraction method is to extract complementary data according to the difference influence coefficient, wherein,

[0025] The training data in the model to be extracted whose difference influence coefficient with the normal data set is less than the preset difference influence coefficient is recorded as complementary training data.

[0026] Furthermore, if the abnormal factor proportion is less than the preset abnormal factor proportion or the abnormal factor dispersion is greater than or equal to the preset abnormal factor dispersion, the complementary data extraction method is to extract complementary data according to the factor difference threshold, wherein,

[0027] The training data whose factor difference threshold between the abnormal data set of the target model and the training data in the model to be extracted is less than the preset factor difference threshold is recorded as complementary training data.

[0028] Furthermore, the factor type is determined according to the factor difference and factor completeness, and the factor types include:

[0029] Abnormal factors whose factor difference is greater than or equal to the preset factor difference or whose factor integrity is less than the preset factor integrity;

[0030] A normal factor whose factor difference is less than the preset factor difference and whose factor integrity is greater than or equal to the preset factor integrity.

[0031] Compared with the prior art, the beneficial effect of the present invention lies in that, in the technical scheme of the present invention, the target model state is determined according to the proportion of abnormal data values ​​and the silencer simulation coefficient, and the accuracy of the target model simulation results is effectively reflected by the proportion of abnormal data values ​​and the silencer simulation coefficient, and then it is determined whether to optimize the target model according to the target model state, thereby avoiding the defect of the prior art that the silencer model simulation effect is not detected, thereby reducing the deviation between the target model simulation results and the actual results.

[0032] Furthermore, in the present invention, the simulation state is determined according to the abnormal value distribution coefficient and the contour difference, and the distribution state of the abnormal values ​​and the degree of difference in the simulation results are effectively reflected by the abnormal value distribution coefficient and the contour difference. Then, different model optimization methods are adaptively selected according to the simulation state, so that the selection of the model optimization method is more in line with the actual application scenario, and the problems existing in the model can be judged more accurately, avoiding the problem of poor accuracy of model simulation caused by failure to adaptively optimize the model training data when there is a deviation between the model simulation effect and the actual result in the prior art, and also avoiding the problem of poor model simulation effect caused by defects in the existing training data, thereby improving the overall performance of the model.

[0033] Furthermore, in the first simulation state, the desensitized associated training data is selected for supplementary training. When the distribution of abnormal values ​​is relatively dispersed or the contour difference is relatively large, the standard model training data with a high degree of correlation with the target model training data can be selected, which is helpful to optimize the performance of the model and improve the accuracy of the model. In the second simulation state, complementary training data is extracted for supplementary training. When the distribution of abnormal values ​​is relatively dense and the contour difference is small, the abnormal data set can be effectively identified, and then some training data in the standard model with a high correlation with the abnormal data set can be accurately selected for supplementary training, thereby improving the accuracy of the target model simulation results.

[0034] Furthermore, in the present invention, the degree of correlation of the standard model training data is effectively reflected through the sub-data correlation, and then different desensitization methods are adaptively selected according to the sub-data correlation, so that the selection of the desensitization method is more in line with the actual application scenario, and can reduce the cost of the desensitization processing while protecting the desensitization effect, thereby reducing the risk of data leakage when selecting associated training data for training.

[0035] Furthermore, the present invention effectively reflects the number of abnormal factors in the abnormal data set and the degree of difference of the abnormal factor data through the abnormal factor proportion and the abnormal factor dispersion, and then adaptively selects different complementary data extraction methods according to the abnormal factor proportion and the abnormal factor dispersion, so that the selection of complementary data extraction method is more in line with the actual application scenario, avoiding the problem that when there are abnormalities in the model training data, it is impossible to select complementary training data for supplementary training in a targeted manner, resulting in the model optimization effect being difficult to meet user needs, thereby improving the accuracy of the target model simulation results.

[0036] Furthermore, in the present invention, when the proportion of abnormal factors is greater than or equal to the preset proportion of abnormal factors and the discreteness of abnormal factors is less than the preset discreteness of abnormal factors, complementary data extraction is performed according to the difference influence coefficient. The difference influence coefficient is used to effectively reflect the degree of influence of the external environment on the muffler. By selecting complementary training data according to the difference influence coefficient in the normal data set, the degree of influence of the external environment on the target model can be reduced, thereby improving the accuracy of the target model simulation.

[0037] Furthermore, in the present invention, when the proportion of abnormal factors is less than the preset abnormal factor proportion or the abnormal factor dispersion is greater than or equal to the preset abnormal factor dispersion, complementary data extraction is performed according to the factor difference threshold. The factor difference threshold is used to effectively reflect the degree of difference between the sub-data corresponding to the factors that are the same as the normal factors in the target model and the target normal factors. Then, complementary training data is selected according to the factor difference threshold, thereby improving the accuracy of the target model simulation results.

[0038] Furthermore, the factor difference and factor completeness in the present invention effectively reflect the degree of difference between the sub-data corresponding to the same factor in the abnormal data set and the normal data set, as well as the degree of completeness of the sub-data corresponding to each factor, and then determine the factor type according to the factor difference and factor completeness, so that the determination of the factor type is more in line with the actual application scenario, and can accurately distinguish different factors, which is conducive to the subsequent extraction of complementary training data for supplementary training. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of an optimization method for an environmentally friendly low flow resistance intake muffler according to the present invention;

[0040] Figure 2 It is a flow chart of determining the simulation state according to the abnormal value distribution coefficient and the contour difference of the present invention;

[0041] Figure 3 A flow chart of the present invention for determining a model optimization method according to a simulation state;

[0042] Figure 4The present invention is a flow chart of determining a complementary data extraction method according to anomaly factor proportion and anomaly factor dispersion. DETAILED DESCRIPTION

[0043] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0044] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0045] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0046] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0047] See also Figures 1 to 4 As shown, the present invention provides an environmentally friendly low flow resistance intake muffler optimization method, comprising:

[0048] Get muffler data information;

[0049] Determine the state of the target model according to the proportion of abnormal data values ​​and the noise elimination simulation coefficient, and determine whether to optimize the target model according to the state of the target model;

[0050] In the optimization of the target model, the simulation state is determined according to the abnormal value distribution coefficient and the contour difference, and the model optimization method is determined according to the simulation state to select the desensitized associated training data for supplementary training or extract complementary training data for supplementary training;

[0051] When selecting the desensitized associated training data for supplementary training, the desensitization method of the associated training data is determined according to the sub-data correlation degree, which is to desensitize all sub-data or desensitize part of the sub-data, and the desensitized associated training data is selected for training the target model;

[0052] When extracting complementary training data for supplementary training, the complementary data extraction method is determined based on the abnormal factor proportion and the abnormal factor dispersion, that is, complementary data extraction is performed based on the difference influence coefficient or the factor difference threshold;

[0053] The associated training data is training data of a standard model whose similarity ratio with the target model training data is greater than a preset similarity ratio.

[0054] The application scenario of the present invention is the optimization process in the process of constructing a transmission loss model of an environmentally friendly low-resistance intake silencer. The silencer data information includes training data information corresponding to each transmission loss model, silencer parameters, and simulated transmission loss images. The training data information corresponding to a single transmission loss model includes a number of training data, and a single training data includes a data set corresponding to a number of data names. A single data set includes sub-data corresponding to a number of factors, and the sub-data includes a number of numerical values; the data names include but are not limited to muffler geometric structure parameters, acoustic performance parameters, working environment parameters, and experimental data, and the factors include but are not limited to size data, shape data, transmission loss data, and boundary conditions. The muffler parameters are parameters input into the target model for simulation, including sound pressure, inlet cross-sectional area, outlet cross-sectional area, expansion ratio, wavelength, wave number, sound wave frequency, sound wave velocity, cavity length, etc.; the simulated transmission loss image is an image output by the target model with frequency as the horizontal coordinate and transmission loss as the vertical coordinate. This is content that is easy for technicians in this field to understand, and will not be elaborated in detail;

[0055] The transmission loss model in the present invention includes an established target model and several standard models. The target model is a model that determines whether to be optimized according to actual conditions, and the standard model is a model whose degree of deviation between the model simulation effect and the actual result can meet the user's needs.

[0056] In the first simulation state, the model optimization method is to select the desensitized associated training data for supplementary training; in the second simulation state, the model optimization method is to extract complementary training data for supplementary training;

[0057] The method for confirming the proportion of similar features is, for the training data of a standard model, the number of factors corresponding to the training data of the standard model is recorded as A, and the number of factors in the training data corresponding to the target model is recorded as B. The proportion of similar features = the number of the same factors in the associated training data and the training data corresponding to the target model / the larger value of A and B.

[0058] The value of the preset similar feature ratio can be determined by the user according to the actual application scenario. The higher the user's demand for improving the correlation between the associated training data and the training data corresponding to the target model, the larger the value of the preset similar feature ratio will be. The average value of the similar feature ratio corresponding to the historical records that can meet the user's needs will be recorded as the preset similar feature ratio.

[0059] In the present invention, several historical records are correspondingly set up, and any historical record records the proportion of abnormal data values, anechoic simulation coefficient, abnormal numerical distribution coefficient, contour difference, sub-data correlation and similar feature proportion in the historical process of at least one model optimization, and each historical record corresponds to a qualified mark, which records whether the sub-data confidentiality effect of the associated training data and the degree of deviation between the model simulation effect and the actual result meet user requirements, and the qualified mark can be recorded manually.

[0060] Specifically, when the state of the target model is that the proportion of abnormal data values ​​is greater than or equal to the preset abnormal data value proportion or the anechoic simulation coefficient is greater than or equal to the preset anechoic simulation coefficient, the target model is optimized.

[0061] Among them, when the target model state is that the abnormal data value ratio is less than the preset abnormal data value ratio and the anechoic simulation coefficient is less than the preset anechoic simulation coefficient, there is no need to optimize the target model;

[0062] The values ​​of the preset abnormal data value ratio and the preset noise reduction simulation coefficient can be determined by the user according to the actual application scenario. The greater the user's demand for the optimization model, the smaller the values ​​of the preset abnormal data value ratio and the preset noise reduction simulation coefficient are. A value of the preset abnormal data value ratio and the preset noise reduction simulation coefficient is provided, and the average value of the abnormal data value ratio corresponding to the historical records that can meet the user's needs is recorded as the abnormal data value ratio, and the average value of the noise reduction simulation coefficient corresponding to the historical records that can meet the user's needs is recorded as the preset noise reduction simulation coefficient.

[0063] Specifically, the method for confirming the proportion of abnormal data values ​​is to input the muffler parameters into the target model and perform a preset number of simulations to obtain a number of simulated transmission loss images, and record the data values ​​in the simulated transmission loss images whose data value changes are greater than the preset data value changes or whose adjacent data fluctuation mean is greater than the preset adjacent data fluctuation mean as abnormal data values, and record the ratio of the total amount of abnormal data values ​​to all data values ​​as the abnormal data value proportion;

[0064] The method for confirming the noise reduction simulation coefficient is to record the ratio of the average value of the transmission loss corresponding to each divided point after dividing a single simulated transmission loss image into M equal parts and the muffler characteristic value as a reference ratio, and the noise reduction simulation coefficient is the average value of the reference ratios corresponding to each simulated transmission loss image.

[0065] Wherein, a data value whose data value variation in the simulated transmission loss image is less than or equal to a preset data value variation and whose adjacent data fluctuation mean is less than or equal to a preset adjacent data fluctuation mean is a normal data value;

[0066] Abnormal data value ratio = total amount of abnormal data values / total data values, where the total amount of abnormal data values ​​is the total number of abnormal data values ​​of the data values ​​of the transmission loss corresponding to each frequency in the simulated transmission loss image, and the total data value is the total number of the data values ​​of the transmission loss corresponding to each frequency in the simulated transmission loss image;

[0067] The value of the preset number of times can be determined by the user according to actual needs. The greater the user's demand for more accurate proportion of abnormal data values ​​and noise reduction simulation coefficient, the larger the value of the preset number of times will be. A value of the preset number of times is provided, and the preset number of times is 20 times.

[0068] The method for confirming the data value change amount and the adjacent data fluctuation mean is as follows: for a single frequency, the frequency is recorded as the target frequency, the data value of the transmission loss corresponding to the target frequency in the simulated transmission loss image is recorded as the target data value, the data value change amount = the maximum value among the target data values ​​in each simulated transmission loss image - the minimum value among the target data values ​​in each simulated transmission loss image; the absolute value of the difference between the data value of the transmission loss corresponding to the frequency adjacent to the target frequency in a single simulated transmission loss image and the target data value is recorded as the first difference and the second difference respectively, the adjacent data fluctuation value = |first difference - second difference| / the larger value of the first difference and the second difference, and the average value of the adjacent data fluctuation values ​​corresponding to each simulated transmission loss image is recorded as the adjacent data fluctuation mean;

[0069] The values ​​of the preset data value change amount and the preset adjacent data fluctuation mean can be determined by the user according to the actual application scenario. It can be understood that the smaller the values ​​of the preset data value change amount and the preset adjacent data fluctuation mean are, the greater the probability of determining that the data value of the transmission loss corresponding to a single frequency is an abnormal data value. A value of the preset data value change amount and the preset adjacent data fluctuation mean is provided, and the average value of the data value change amount corresponding to the historical records that can meet the user's needs is recorded as the preset data value change amount, and the average value of the adjacent data fluctuation mean corresponding to the historical records that can meet the user's needs is recorded as the preset adjacent data fluctuation mean;

[0070] A single simulated transmission loss image is divided into M equal parts, and the horizontal coordinate of the simulated transmission loss image is used as a reference to divide the image into M equal parts, so that the transmission loss corresponding to each divided point can be obtained; the value of M can be determined by the user according to the needs, and one value of M is provided, which is 25.

[0071] The characteristic value of the muffler = pipe length + expansion ratio, where the pipe length is the length of the pipe inside the muffler, and the expansion ratio = the cross-sectional area of ​​the expansion part of the muffler / the cross-sectional area of ​​the contraction part. The expansion part refers to the area where the cross-sectional area inside the muffler is increased relative to the inlet pipe, and the contraction part refers to the area where the cross-sectional area inside the muffler is reduced relative to the outlet pipe.

[0072] Specifically, the simulation states include:

[0073] A first simulation state in which the abnormal value distribution coefficient is greater than or equal to a preset abnormal value distribution coefficient or the contour difference is greater than or equal to a preset contour difference;

[0074] A second simulation state in which the abnormal value distribution coefficient is less than the preset abnormal value distribution coefficient and the contour difference is less than the preset contour difference.

[0075] Among them, the abnormal value distribution coefficient = adjacent distance reference value + distance segment value,

[0076] The adjacent distance reference value and the distance segment value are confirmed in the following manner: the frequency corresponding to each abnormal value is detected and recorded as the abnormal frequency; for a single abnormal frequency, the abnormal frequency is recorded as the target abnormal frequency; the average value of the distance between the abnormal frequencies adjacent to the target abnormal frequency and the target abnormal frequency is recorded as the adjacent distance mean value; the average value of the adjacent distance means corresponding to each abnormal frequency is recorded as the adjacent distance reference value; the distance between two abnormal frequencies is the absolute value of the difference between the frequency values ​​corresponding to the two abnormal frequencies; in the direction from small to large frequency values ​​on the horizontal axis, the distance segment value = the frequency value corresponding to the last abnormal frequency - the frequency value corresponding to the first abnormal frequency;

[0077] The method for confirming the contour difference degree is to perform contour difference reference value detection on each simulated transmission loss image. When performing contour difference reference value detection on a single simulated transmission loss image, the simulated transmission loss image is recorded as a target image, and simulated transmission loss images other than the target image are recorded as reference images. The contour difference reference value between the target image and the reference image is detected, and the contour difference reference value detection is continued for the simulated transmission loss images that have not been detected by the contour difference reference value until the contour difference reference value detection is performed on each simulated transmission loss image, and a number of contour difference reference values ​​can be obtained. The maximum value among the contour difference reference values ​​is recorded as the contour difference degree. The contour difference reference value is the average value of the sum of squares of the differences of the transmission loss data values ​​corresponding to each frequency of the two simulated transmission loss images.

[0078] The values ​​of the preset abnormal numerical distribution coefficient and the preset contour difference can be determined by the user according to the actual application scenario. It can be understood that the smaller the values ​​of the preset abnormal numerical distribution coefficient and the preset contour difference, the greater the user's demand for extracting complementary training data for supplementary training. A preset abnormal numerical distribution coefficient and a preset contour difference are provided. The detection model optimization method is to extract historical records of complementary training data for supplementary training, and the average value of the abnormal numerical distribution coefficient corresponding to the historical records that can meet the user's needs is recorded as the preset abnormal numerical distribution coefficient, and the average value of the contour difference corresponding to the historical records that can meet the user's needs is recorded as the preset contour difference.

[0079] Specifically, in the first simulation state, if the sub-data association degree is greater than or equal to the preset sub-data association degree, the desensitization method is to desensitize all sub-data.

[0080] Among them, the method for confirming the sub-data association degree is to perform factor coefficient detection on the sub-data corresponding to each factor of a single standard model, detect the matching coefficient between each sub-data corresponding to a single factor, perform matching coefficient detection on each sub-data, and when performing matching coefficient detection on a single sub-data, record the sub-data as the first target sub-data, record the other sub-data other than the first target sub-data as the first reference sub-data, detect the matching coefficient between the first target sub-data and each first reference sub-data, and continue to perform matching coefficient detection on the sub-data that has not been detected for the matching coefficient, until the matching coefficient detection of each sub-data in the single data set is completed, and several matching coefficients can be obtained, and the minimum value of each matching coefficient is recorded as the factor coefficient, and the factor coefficient detection is continued for the sub-data corresponding to the factor that has not been detected for the factor coefficient until the factor coefficient corresponding to each factor is obtained, and the average value of the factor coefficient corresponding to each factor is recorded as the sub-data association degree.

[0081] Matching coefficient The calculation formula is:

[0082]

[0083] Where m is the number of values ​​in a single sub-data; and are the values ​​of the kth values ​​in the two sub-data, respectively. for The average value of each value in the corresponding sub-data, for The average value of each value in the corresponding sub-data, k = 1, 2, 3, ..., m.

[0084] The value of the preset sub-data association degree can be determined by the user according to the actual application scenario. The higher the user's demand for improving data confidentiality, the smaller the value of the preset sub-data association degree will be. The desensitization method is to detect historical records with partial sub-data desensitization, and the average value of the sub-data association degrees corresponding to the historical records that can meet the user's needs is recorded as the preset sub-data association degree.

[0085] Specifically, in the first simulation state, if the sub-data relevance is less than the preset sub-data relevance, the desensitization method is partial sub-data desensitization, where:

[0086] Desensitization is performed on each data set in each training data. When desensitization is performed on a single data set, the data set is recorded as the target data set, and the desensitization ratio of the sub-data that needs to be desensitized in the target data set is determined according to the average matching coefficient;

[0087] The desensitization ratio is positively correlated with the average matching coefficient.

[0088] The method for confirming the average matching coefficient is as follows: for each sub-data in a single data set, the matching coefficient between each sub-data is detected, and the matching coefficient detection is performed on each sub-data. When performing the matching coefficient detection on a single sub-data, the sub-data is recorded as the second target sub-data, and other sub-data other than the second target sub-data are recorded as the second reference sub-data. The matching coefficient between the second target sub-data and each second reference sub-data is detected, and the matching coefficient detection is continued for the sub-data that has not been detected for the matching coefficient until the matching coefficient detection of each sub-data in the single data set is completed, and a number of matching coefficients can be obtained, and the average value of the matching coefficients is recorded as the average matching coefficient;

[0089] Desensitization ratio = the number of sub-data that need to be desensitized in a single data set / the total amount of sub-data in a single data set;

[0090] The methods that can be used for desensitizing sub-data include but are not limited to replacement, masking, encryption, and data disturbance. Users can choose according to their actual needs, as long as they can meet the user's confidentiality requirements;

[0091] Selecting the desensitized associated training data for supplementary training includes: using the desensitized associated training data and the training data corresponding to the target model as training samples to train the target model.

[0092] Specifically, in the second simulation state, complementary training data is extracted for supplementary training, wherein:

[0093] Detect the set of training data of the target model corresponding to the muffler parameters and abnormal values, and record it as the abnormal data set, record the data samples outside the abnormal data set as the normal data set, detect the abnormal factors in the abnormal data set, record the standard model with abnormal factors and model difference value less than the preset model difference value as the model to be extracted, and record the data samples extracted by complementary data extraction method as complementary training data, and extract the complementary training data to perform supplementary training on the target model.

[0094] The extracting complementary training data for supplementary training includes: using the complementary training data and the training data corresponding to the target model as training samples to train the target model.

[0095] Detecting a set of training data of a target model corresponding to the muffler parameters and the abnormal values ​​includes: finding training data containing both the muffler parameters and the abnormal values ​​according to the muffler parameters and the abnormal values, that is, the training data corresponding to the abnormal values.

[0096] The method for confirming the model difference value is to record the frequency corresponding to each normal data value corresponding to the target model as the normal frequency, record the average value of each normal data value of a single simulated transmission loss image corresponding to the target model as the normal mean, record the average value of the normal means of each simulated transmission loss image corresponding to the target model as the first reference value, input the muffler parameters into the standard model and perform a preset number of simulations to obtain simulated transmission loss images corresponding to several standard models, record the average value of the transmission loss data values ​​corresponding to each normal frequency of a single simulated transmission loss image corresponding to the model to be matched as the standard mean, record the average value of the standard means of each simulated transmission loss image corresponding to the model to be matched as the second reference value, and the model difference value is the absolute value of the difference between the first reference value and the second reference value.

[0097] The value of the preset model difference value can be determined by the user according to the actual application scenario. The greater the user's demand for improving the simulation accuracy of the target model, the smaller the value of the preset model difference value will be. A value of the preset model difference value is provided, and the average value of the model difference values ​​corresponding to the historical records that can meet the user's needs is recorded as the preset model difference value.

[0098] Specifically, if the abnormal factor proportion is greater than or equal to the preset abnormal factor proportion and the abnormal factor dispersion is less than the preset abnormal factor dispersion, the complementary data extraction method is to extract complementary data according to the difference influence coefficient, where:

[0099] The training data in the model to be extracted whose difference influence coefficient with the normal data set is less than the preset difference influence coefficient is recorded as complementary training data.

[0100] Among them, the proportion of abnormal factors = the number of abnormal factors in the abnormal data set / the total number of factors in the abnormal data set;

[0101] Abnormal factor dispersion = the maximum value of the abnormal fluctuation values ​​corresponding to each abnormal factor - the minimum value of the abnormal fluctuation values ​​corresponding to each abnormal factor; the abnormal fluctuation value corresponding to a single abnormal factor The calculation formula is: , g is the number of training data in the abnormal data set, is the abnormal mean corresponding to the i-th training data in the abnormal data set. The abnormal mean is the average value of the values ​​corresponding to a single abnormal factor in a single training data in the abnormal data set. is the average of the abnormal means corresponding to each training data in the abnormal data set;

[0102] The values ​​of the preset abnormal factor proportion and the preset abnormal factor discreteness can be determined by the user according to the actual application scenario. The greater the user's demand for improving the simulation accuracy of the target model, the greater the value of the preset abnormal factor proportion and the smaller the value of the preset abnormal factor discreteness. The values ​​of the preset abnormal factor proportion and the preset abnormal factor discreteness are provided, and the complementary data extraction method is to detect the historical records of complementary data extraction based on the difference influence coefficient. The average value of the abnormal factor proportion corresponding to the historical records that can meet the user's needs is recorded as the preset abnormal factor proportion, and the average value of the abnormal factor discreteness corresponding to the historical records that can meet the user's needs is recorded as the preset abnormal factor discreteness.

[0103] Each training data of the present invention has corresponding driving data, and the driving data includes monitoring parameters corresponding to each time point monitored in real time during the historical use of the muffler with the use time, and the monitoring parameters include driving speed, temperature, and airflow speed, etc., wherein the time point is set by the user, and a method for setting the time point is provided, wherein a time point is set every 5 minutes, that is, the monitoring parameters are recorded once every 5 minutes.

[0104] Difference influence coefficient = the absolute value of the difference between the airflow fluctuation value corresponding to a single training data of the standard model and the mean of the airflow fluctuation / the larger value of the airflow fluctuation value corresponding to a single training data of the standard model and the mean of the airflow fluctuation, the airflow fluctuation value The calculation formula is: , H is the number of time points in the driving data corresponding to the training data, is the airflow velocity corresponding to the jth time point in the driving data corresponding to the training data, It is the average value of the airflow velocity corresponding to each time point in the driving data corresponding to the training data; the airflow velocity is the gas flow velocity monitored at the muffler inlet; the airflow fluctuation mean is the average value of the airflow fluctuation values ​​corresponding to each training data in the abnormal data set.

[0105] The value of the preset difference impact coefficient can be determined by the user according to the actual application scenario. The greater the user's demand for improving the accuracy of complementary data extraction, the smaller the value of the preset difference impact coefficient. A value of the preset difference impact coefficient is provided, and the preset difference impact coefficient is 10%.

[0106] Specifically, if the abnormal factor proportion is less than the preset abnormal factor proportion or the abnormal factor dispersion is greater than or equal to the preset abnormal factor dispersion, the complementary data extraction method is to extract complementary data according to the factor difference threshold, where:

[0107] The training data whose factor difference threshold between the abnormal data set of the target model and the training data in the model to be extracted is less than the preset factor difference threshold is recorded as complementary training data.

[0108] Among them, the factor difference threshold = the maximum value of the matching difference corresponding to each normal factor in the abnormal data set - the minimum value of the matching difference corresponding to each normal factor in the abnormal data set,

[0109] The matching difference is confirmed in the following way: for a single normal factor, the normal factor is recorded as the target normal factor. The matching difference = the average value of the values ​​of the sub-data corresponding to the factor that is the same as the target normal factor in the single training data of the model to be extracted - the mean value of the normal factor corresponding to the target normal factor. The mean value of the normal factor is the average value of each value corresponding to the single normal factor in the abnormal data set.

[0110] The value of the preset factor difference threshold can be determined by the user according to the actual application scenario. The greater the user's demand for improving the accuracy of complementary data extraction, the smaller the value of the preset factor difference threshold is. A value of the preset factor difference threshold is provided, and the historical records of complementary data extraction according to the factor difference threshold are detected. The average value of the factor difference threshold corresponding to the historical records that can meet the user's needs is recorded as the preset factor difference threshold.

[0111] Specifically, the factor type is determined according to the factor difference and factor completeness, and the factor types include:

[0112] Abnormal factors whose factor difference is greater than or equal to the preset factor difference or whose factor integrity is less than the preset factor integrity;

[0113] A normal factor whose factor difference is less than the preset factor difference and whose factor integrity is greater than or equal to the preset factor integrity.

[0114] Among them, factor difference = average difference + extreme difference. The confirmation method of average difference and extreme difference is that for a single factor, the factor is recorded as the first target factor, and the difference between the average value of the numerical value corresponding to the first target factor in the abnormal data set and the average value of the numerical value corresponding to the first target factor in the normal data sample set is recorded as the average difference. Extreme difference = |first reference difference-second reference difference|, the first reference difference is the difference between the maximum and minimum values ​​of the numerical value corresponding to the first target factor in the abnormal data set, and the second reference difference is the difference between the maximum and minimum values ​​of the numerical value corresponding to the first target factor in the normal data set;

[0115] The factor completeness is confirmed in the following way: for a single factor, the factor is recorded as the second target factor, and the factor completeness = the average value of the number of values ​​corresponding to each sub-data in the abnormal data set corresponding to the second target factor / the average value of the number of values ​​corresponding to each sub-data in the normal data set corresponding to the second target factor;

[0116] The values ​​of the preset factor difference and the preset factor completeness can be determined by the user according to the actual application scenario. It can be understood that the larger the value of the preset factor difference and the smaller the value of the preset factor completeness, the greater the probability of judging the factor as an abnormal factor. The values ​​of the preset factor difference and the preset factor completeness are provided, and the average value of the factor difference corresponding to the historical records that can meet the user's needs is recorded as the preset factor difference, and the preset factor completeness is 90%.

[0117] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An environmentally friendly low flow resistance intake muffler optimization method, characterized in that: include: Get muffler data information; Determine the state of the target model according to the proportion of abnormal data values ​​and the noise elimination simulation coefficient, and determine whether to optimize the target model according to the state of the target model; In the optimization of the target model, the simulation state is determined according to the abnormal value distribution coefficient and the contour difference, and the model optimization method is determined according to the simulation state to select the desensitized associated training data for supplementary training or extract complementary training data for supplementary training; When selecting the desensitized associated training data for supplementary training, the desensitization method of the associated training data is determined according to the sub-data correlation degree, which is to desensitize all sub-data or desensitize part of the sub-data, and the desensitized associated training data is selected for training the target model; When extracting complementary training data for supplementary training, the complementary data extraction method is determined based on the abnormal factor proportion and the abnormal factor dispersion, that is, complementary data extraction is performed based on the difference influence coefficient or the factor difference threshold; The associated training data is training data of a standard model whose similarity ratio with the target model training data is greater than a preset similarity ratio; The simulation states include: A first simulation state in which the abnormal value distribution coefficient is greater than or equal to a preset abnormal value distribution coefficient or the contour difference is greater than or equal to a preset contour difference; A second simulation state in which the abnormal value distribution coefficient is less than a preset abnormal value distribution coefficient and the contour difference is less than a preset contour difference; In the first simulation state, if the sub-data relevance is greater than or equal to the preset sub-data relevance, the desensitization method is to desensitize all sub-data; In the first simulation state, if the sub-data relevance is less than the preset sub-data relevance, the desensitization method is partial sub-data desensitization, where: Desensitization is performed on each data set in each training data. When desensitization is performed on a single data set, the data set is recorded as the target data set, and the desensitization ratio of the sub-data that needs to be desensitized in the target data set is determined according to the average matching coefficient; The desensitization ratio is positively correlated with the average matching coefficient; In the second simulation state, complementary training data is extracted for supplementary training, wherein: Detect the set of training data of the target model corresponding to the muffler parameters and abnormal values, and record it as the abnormal data set, record the data samples outside the abnormal data set as the normal data set, detect the abnormal factors in the abnormal data set, record the standard model with abnormal factors and model difference value less than the preset model difference value as the model to be extracted, and record the data samples extracted by complementary data extraction method as complementary training data, and extract the complementary training data to perform supplementary training on the target model.

2. The environmentally friendly low flow resistance intake muffler optimization method according to claim 1 is characterized in that: When the target model state is that the abnormal data value ratio is greater than or equal to the preset abnormal data value ratio or the anechoic simulation coefficient is greater than or equal to the preset anechoic simulation coefficient, the target model is optimized.

3. The environmentally friendly low flow resistance intake muffler optimization method according to claim 2 is characterized in that: The method for confirming the proportion of abnormal data values ​​is to input the muffler parameters into the target model and perform a preset number of simulations to obtain a number of simulated transmission loss images, record the data values ​​in the simulated transmission loss images whose data value changes are greater than the preset data value changes or whose adjacent data fluctuation mean is greater than the preset adjacent data fluctuation mean as abnormal data values, and record the ratio of the total amount of abnormal data values ​​to all data values ​​as the proportion of abnormal data values; The method for confirming the noise reduction simulation coefficient is to record the ratio of the average value of the transmission loss corresponding to each divided point after dividing a single simulated transmission loss image into M equal parts and the muffler characteristic value as a reference ratio, and the noise reduction simulation coefficient is the average value of the reference ratios corresponding to each simulated transmission loss image.

4. The environmentally friendly low flow resistance intake muffler optimization method according to claim 1 is characterized in that: If the abnormal factor proportion is greater than or equal to the preset abnormal factor proportion and the abnormal factor dispersion is less than the preset abnormal factor dispersion, the complementary data extraction method is to extract complementary data according to the difference influence coefficient, where: The training data in the model to be extracted whose difference influence coefficient with the normal data set is less than the preset difference influence coefficient is recorded as complementary training data.

5. The environmentally friendly low flow resistance intake muffler optimization method according to claim 4 is characterized in that: If the abnormal factor proportion is less than the preset abnormal factor proportion or the abnormal factor dispersion is greater than or equal to the preset abnormal factor dispersion, the complementary data extraction method is to extract complementary data according to the factor difference threshold, where: The training data whose factor difference threshold between the abnormal data set of the target model and the training data in the model to be extracted is less than the preset factor difference threshold is recorded as complementary training data.

6. The environmentally friendly low flow resistance intake muffler optimization method according to claim 5 is characterized in that: The factor type is determined based on the factor difference and factor completeness. The factor types include: Abnormal factors whose factor difference is greater than or equal to the preset factor difference or whose factor integrity is less than the preset factor integrity; A normal factor whose factor difference is less than the preset factor difference and whose factor integrity is greater than or equal to the preset factor integrity.

Images