Automatic transmission calibration method and system based on driving data characteristics

Abstract

The invention discloses a driving data characteristic-based automatic transmission calibration method and system. The method comprises the following steps of S1, acquiring vehicle driving data of various typical gear shifting processes under each condition; s2, constructing a training data set; s3, constructing a gear shifting parameter optimization module, and generating virtual benign data; obtaining a true and false data set; s4, constructing a gear shifting feature recognition module, and extracting vehicle gear shifting stability related features of the true and false data sets; s5, game training is conducted on the gear shifting parameter optimization module and the gear shifting feature recognition module in an iterative alternating mode, and a final gear shifting parameter optimization module is obtained; s6, the trained gear shifting parameter optimization module is used for optimizing the inferior data of the given vehicle, electric control parameters of the automatic transmission are optimized according to an optimization result, and a final intelligent calibration result is formed. According to the invention, the electric control parameters of the automatic transmission can be rapidly and intelligently calibrated.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning, and in particular relates to an automatic transmission calibration method and system based on driving data characteristics. Background Art

[0002] The automatic transmission is one of the core components in the vehicle power system, which closely affects the smoothness of vehicle shifting. At present, my country's automatic transmission has achieved initial technological breakthroughs and has occupied part of the low-end and mid-end passenger car market both domestically and internationally. However, the electromechanical and hydraulic control technology related to domestic automatic transmissions is still in a rapid development stage, and the scope of application of electronic control technology is not enough to cover the complete life cycle of automatic transmission products. Some vehicles in the market require secondary intervention by calibration engineers, which consumes a lot of manpower, capital and time costs.

[0003] Artificial intelligence technology has been widely used in the research of control algorithms, such as camera calibration, drone adaptation, magnetic suspension workbench leveling and other multi-modal control needs. This technology is based on end-to-end algorithm design and relies on data-driven. It can automatically extract shifting features from existing empirical data to achieve high-precision multi-modal control and has the potential to be applied in the calibration of automatic transmissions.

[0004] At present, there is an urgent need to study a technology for quickly and intelligently calibrating the electronic control parameters of automatic transmissions in order to reduce the secondary intervention costs during after-sales service. Summary of the invention

[0005] The object of the present invention is to provide an automatic transmission calibration method and system based on driving data characteristics, which method is conducive to fast and intelligent calibration of electronic control parameters of the automatic transmission.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] In one aspect, the present invention provides a transmission calibration method based on driving data characteristics, which specifically comprises the following steps:

[0008] Step S1: driving the vehicle under different throttle opening conditions and different road conditions, and performing various typical gear shifting processes under each condition, and collecting vehicle driving data of various typical gear shifting processes under each condition;

[0009] Step S2: According to the parameters related to vehicle gear shifting smoothness and the gear shifting principle of the automatic transmission, the gear shifting data is selected from the driving data obtained in step 1 and divided into good data and bad data, and then data cleaning and data preprocessing are performed to construct a training data set;

[0010] Step S3: construct a shift parameter optimization module, extract characteristic information of the bad data through the shift parameter optimization module, and generate virtual good data; merge the generated virtual good data with the real good data to obtain a true and false data set;

[0011] Step S4: constructing a shift feature recognition module, and extracting vehicle shift smoothness-related features of the true and false data sets through the constructed shift feature recognition module;

[0012] Step S5: constructing physical constraints based on the automatic transmission shifting principle, and performing game training on the shift parameter optimization module and the shift feature recognition module in an iterative alternating manner, so that the shift parameter optimization module can eventually generate accurate good data based on the bad data, and obtain the final shift parameter optimization module;

[0013] Step S6: Use the trained shift parameter optimization module to optimize the inferior data of a given vehicle, and optimize the electronic control parameters of the automatic transmission according to the optimization result to form a final intelligent calibration result.

[0014] Furthermore, in step S1, considering the influence of the throttle opening, the throttle needs to be stabilized at different percentages for testing; considering the influence of the road slope and flatness, it is necessary to test on a flat road, a bumpy road, an uphill road and a downhill road respectively.

[0015] Furthermore, step S2 specifically includes the following steps:

[0016] Step S21: According to the vehicle shifting smoothness related parameters and the automatic transmission shifting principle, the vehicle driving data of various typical shifting processes obtained in step 1 are locally sampled to obtain the shifting data therein, and the inferior data and the good data are distinguished to obtain the initial data set; for each vehicle shifting smoothness related parameter, the data with the worst shifting smoothness is selected as the inferior data of the parameter, and the data with the best shifting smoothness is selected as the good data of the parameter;

[0017] Step S22: cleaning the initial data set to obtain valid data in the gear shifting phase;

[0018] Step S23: perform data preprocessing, specifically: through cluster analysis, select the automatic transmission electronic control parameters that most affect the vehicle driving stability from the valid data of the shifting stage obtained in step 22; group the automatic transmission electronic control parameters that most affect the vehicle driving stability according to the typical shifting process; in the grouped data, merge the good data and the bad data in pairs to obtain a training data set.

[0019] Furthermore, step S3 specifically includes the following operations: constructing a shift parameter optimization module with a jump connection structure, including an encoding process and a decoding process; the input of the shift parameter optimization module is the vehicle driving data X, and all the inferior data in the training set are processed by the encoding process and the decoding process to output virtual benign data Y, and the dimension of the virtual benign data is the same as that of the inferior data; merging the virtual benign data with the real benign data to obtain a true and false data set X+Y with a ratio of 1:1.

[0020] Furthermore, the encoding process is composed of five first downsampling modules with the same structure connected in series, which can extract high-dimensional features related to vehicle driving stability from the training data set; the first downsampling module is composed of two convolutional layers and one pooling layer connected in sequence; wherein the convolution kernel size of the convolutional layer is 3×3, the step size is 2, and the activation function is ReLU; the convolution kernel size of the pooling layer is 2×2, the step size is 2, and the activation function is Max; after the vehicle driving data X undergoes the encoding process of five downsamplings, the data is converted into features to obtain the feature map EnCode5;

[0021] The decoding process is composed of five upsampling modules with the same structure connected in series, which can generate virtual benign data from the high-dimensional features related to vehicle driving stability; the upsampling module is composed of two convolutional layers and one deconvolutional layer connected in sequence; wherein the convolutional kernel size of the convolutional layer is 3×3, the step size is 2, the activation function is ReLU, and the convolutional kernel size of the deconvolutional layer is 2×2, and the step size is 0.5. There is a jump connection structure between the first downsampling module and the upsampling module of the same size, so that the shift parameter optimization module can learn the underlying information shared between the inferior data and the benign data; after the feature map EnCode5 undergoes five upsampling decoding processes, the feature-to-data conversion is completed to obtain the virtual benign data Y.

[0022] Furthermore, step S4 specifically includes the following operations: constructing a shift feature recognition module with a Markov random field, including a downsampling layer and a block judgment layer; the input of the shift feature recognition module is a true and false data set X+Y, and the output is a Markov random field, which contains local judgment information on the input data. The shift feature recognition module is used to judge the authenticity of the data from the true and false data sets, and the judgment result is used to optimize the shift parameter optimization module during the training process; the downsampling layer: is composed of 5 second downsampling modules with the same structure connected in series, which can extract high-dimensional features related to shift smoothness from true and false driving data; the second downsampling module is composed of 1 convolution layer, the convolution kernel size is 3×3, the step size is 2, and the activation function is ReLU; the high-dimensional features F5 of the true and false driving data are extracted through 5 second downsampling modules connected in series;

[0023] The block judgment layer includes a Markov random field M.

[0024] Furthermore, step S5 specifically includes the following steps:

[0025] Step S51: constructing physical constraints CONS based on the automatic transmission shifting principle, including acceleration loss function, shifting process loss function and shifting action loss function;

[0026] Step S52: obtaining an initial true and false data set, specifically: initializing the shift parameter optimization module so that it generates virtual good data according to the bad data, and merging the virtual good data with the real good data to obtain the true and false data set;

[0027] Step S53: optimizing the gear shift feature recognition module, specifically: using the true and false data sets to train the gear shift feature recognition module to obtain a trained gear shift feature recognition module;

[0028] Step S54: Optimizing the gear shift parameter optimization module, specifically: using the gear shift feature recognition module trained in the previous step to recognize the optimization result of the gear shift parameter optimization module; if the optimization result is the same as the real benign data, the gear shift parameter optimization module remains unchanged; if the optimization result is different from the real benign data, the gear shift parameter optimization module is trained using the real and fake data sets to obtain a trained gear shift parameter optimization module. The trained gear shift parameter optimization module obtained in this step can generate virtual benign data that the gear shift feature recognition module cannot distinguish;

[0029] Step S55: Perform game training in an iterative alternating manner, specifically: generate new virtual benign data using the gear shift parameter optimization module trained in the previous step, merge the new virtual benign data with the real benign data to obtain a new true and false data set. Use the new true and false data set to train the current gear shift feature recognition module again, and then use the new trained gear shift feature recognition module to optimize the gear shift parameter optimization module again, and repeat this cycle;

[0030] Step S56: until the gear shift parameter optimization module can generate sufficiently realistic virtual benign samples, then the final gear shift parameter optimization module is obtained.

[0031] Furthermore, in step 51, the expression of the physical constraint CONS is:

[0032] CONS=LOSS acc +LOSS pro +LOSS act

[0033] LOSS acc =acc(eng)+acc(input)+acc(output)

[0034] LOSSpro =acc(shiftprogression)

[0035] LOSS act =match(offgoing)+match(oncoming)

[0036] Among them, LOSS acc is the acceleration loss function, LOSS pro is the loss function of the gear shift process, LOSS act is the shift action loss function, eng is the engine speed, input is the input shaft speed, output is the output shaft speed, acc is the acceleration function, shiftprogression is the shift process curve, acc is the acceleration function, offgoing is the shift stage of disengaging the clutch, oncoming is the shift stage of engaging the clutch, and match is the matching function.

[0037] Further, step S6 specifically includes the following steps:

[0038] Step S61: collecting driving data of the vehicle to be calibrated, and using the trained shift parameter optimization module to generate good data from the bad data of the vehicle, that is, obtaining an optimization result;

[0039] Step S62: Communicate the optimization result with the vehicle and write the optimization result into the controller of the automatic transmission, thereby optimizing the electronic control parameters of the automatic transmission.

[0040] On the other hand, the present invention provides a transmission calibration system based on driving data characteristics, comprising the following modules:

[0041] Vehicle driving data acquisition module: used to obtain vehicle driving data of various typical shifting processes under different throttle opening conditions and different road conditions;

[0042] Training data set construction module: used to select shifting data from the driving data obtained by the vehicle driving data acquisition module and distinguish between good data and bad data according to the vehicle shifting smoothness related parameters and the automatic transmission shifting principle, and then perform data cleaning and data preprocessing to construct the training data set;

[0043] True and false data set determination module: used to construct a shift parameter optimization module, extract characteristic information of bad data through the shift parameter optimization module, generate virtual benign data; merge the generated virtual benign data with the real benign data to obtain the true and false data set;

[0044] Gear shifting stability related feature extraction module: used to construct a gear shifting feature recognition module, and extract vehicle gear shifting stability related features of the true and false data sets through the constructed gear shifting feature recognition module;

[0045] Training module: used to build physical constraints based on the automatic transmission shifting principle, and to perform game training on the shift parameter optimization module and the shift feature recognition module in an iterative alternating manner, so that the shift parameter optimization module can eventually generate accurate good data based on bad data, and obtain the final shift parameter optimization module;

[0046] Calibration module: used to optimize the inferior data of a given vehicle using the trained shift parameter optimization module, and optimize the electronic control parameters of the automatic transmission according to the optimization results to form the final intelligent calibration result.

[0047] Compared with the prior art, the method of the present invention has the following beneficial effects:

[0048] (1) The shifting characteristics in the driving data are used to calibrate the automatic transmission, which solves the limitation of the previous method that requires a lot of manpower, financial and time costs.

[0049] (2) A gear shift parameter optimization module and a gear shift feature recognition module were constructed. The characteristic information of the poor quality data was extracted through the constructed gear shift parameter optimization module to generate virtual good quality data, and the generated virtual good quality data was merged with the real good quality data to obtain the true and false data sets. The gear shift feature recognition module was used to extract the vehicle gear shift smoothness related features of the true and false data sets.

[0050] (3) It can be used for subsequent research to achieve remote operation and one-click intelligent calibration, saving the time and cost of calibration engineers.

[0051] In summary, the method of the present invention can automatically identify shifting features in driving data and intelligently calibrate the automatic transmission without the need for manual calibration by calibration engineers. It also saves the time of calibration engineers for repeatedly collecting and analyzing data on site, effectively improving the efficiency and accuracy of automatic transmission calibration work. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a module structure diagram in an embodiment of the present invention;

[0053] Figure 2 is the inferior data in the embodiment of the present invention;

[0054] Figure 3 is benign data in an embodiment of the present invention;

[0055] Figure 4 This is the intelligent calibration result in the embodiment of the present invention.

[0056] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. DETAILED DESCRIPTION

[0057] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0058] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0059] Embodiment 1:

[0060] This embodiment provides a transmission calibration method based on driving data characteristics, comprising the following steps:

[0061] Step S1: driving the vehicle under different throttle opening conditions and different road conditions, and performing various typical gear shifting processes under each condition, and collecting vehicle driving data of various typical gear shifting processes under each condition.

[0062] Step S2: According to the parameters related to vehicle gear shifting smoothness and the gear shifting principle of the automatic transmission, the gear shifting data is selected from the driving data obtained in step 1 and divided into good data and bad data, and then data cleaning and data preprocessing are performed to construct a training data set;

[0063] Step S3: construct a shift parameter optimization module, extract characteristic information of the bad data through the shift parameter optimization module, and generate virtual good data; merge the generated virtual good data with the real good data to obtain a true and false data set;

[0064] Step S4: constructing a shift feature recognition module, and extracting vehicle shift smoothness-related features of the true and false data sets through the constructed shift feature recognition module;

[0065] Step S5: constructing physical constraints based on the automatic transmission shifting principle, and performing game training on the shift parameter optimization module and the shift feature recognition module in an iterative alternating manner, so that the shift parameter optimization module can eventually generate accurate good data based on the bad data, and obtain the final shift parameter optimization module;

[0066] Step S6: Use the trained shift parameter optimization module to optimize the inferior data of a given vehicle, and optimize the electronic control parameters of the automatic transmission according to the optimization result to form a final intelligent calibration result.

[0067] Specifically, in step S1, in this embodiment, the vehicle data acquisition software is Array Technology Industry (ATI VISION 3.8.4), bus parameters are set and messages are read and written through Kvaser, and the OBD interface is used to connect the bus for communication to collect driving data in real time.

[0068] Specifically, considering the influence of the throttle opening, the throttle needs to be stabilized at different percentages for testing, such as: 0%, 25%, 50%, 75%, 100%; considering the influence of the road slope and flatness, it is necessary to test on a flat road, a bumpy road, an uphill road, and a downhill road. In the above-mentioned cases, the vehicle is driven to reproduce various typical shifting processes (typical shifting processes include: shifting up on the spot, shifting down on the spot, coasting upshift, coasting downshift, power upshift, power downshift, etc.), and the vehicle driving data of each typical shifting process in each case is obtained.

[0069] In this embodiment, step S2 specifically includes the following steps:

[0070] Step S21: Based on the vehicle shifting smoothness related parameters and the automatic transmission shifting principle, the vehicle driving data of various typical shifting processes obtained in step 1 are locally sampled to obtain the shifting data therein, and the poor data and the good data are distinguished to obtain the initial data set.

[0071] Specifically, the parameters related to vehicle shifting stability include automatic transmission parameters and vehicle parameters. Among them, automatic transmission parameters include: input shaft speed, output shaft speed, clutch torque, clutch pressure, oil temperature, clutch pressure, control current value, coefficients of related formulas, etc.; vehicle parameters include: throttle opening, engine speed, vehicle speed, brake pedal opening, etc. Shifting data refers to the data corresponding to the parameters related to vehicle shifting stability.

[0072] For each vehicle shift smoothness-related parameter, the worst shift smoothness data is selected as the bad data of the parameter, and the best shift smoothness data is selected as the good data of the parameter. Figure 2 and Figure 3 Two important parameters of the automatic transmission are shown: input shaft speed and output shaft speed, whose fluctuation affects the smoothness of vehicle driving. Figure 2 For inferior data, Figure 3 It is benign data.

[0073] Step S22: Clean the initial data set to obtain valid data in the gear shifting phase.

[0074] Among them, data cleaning specifically involves removing data samples with damaged parameters and abnormal working conditions in the initial data set;

[0075] Step S23: perform data preprocessing. Specifically, through cluster analysis, select the automatic transmission electronic control parameters (including clutch pressure, control current value, coefficients of related formulas, etc.) that most affect the vehicle driving stability from the effective data of the shifting stage obtained in step 22; group the automatic transmission electronic control parameters that most affect the vehicle driving stability according to the typical shifting process; merge the good data and the bad data in pairs in the grouped data to obtain the training data set. The training data set is randomly divided into a training set, a validation set, and a test set in a ratio of 6:2:2.

[0076] In this embodiment, step S3 specifically includes the following operations:

[0077] A shift parameter optimization module with a skip connection structure is constructed, including an encoding process and a decoding process. The input of the shift parameter optimization module is the vehicle driving data X. The encoding process and the decoding process are used to process all the inferior data in the training set, and the virtual benign data Y is output. The dimension of the virtual benign data is the same as that of the inferior data. The virtual benign data is merged with the real benign data to obtain a true and false data set X+Y with a ratio of 1:1.

[0078] Preferably, the encoding process is composed of five first downsampling modules with the same structure connected in series, which can extract high-dimensional features related to vehicle driving stability from the training data set; the first downsampling module is composed of two convolutional layers and one pooling layer connected in sequence; wherein the convolution kernel size of the convolutional layer is 3×3, the step size is 2, and the activation function is ReLU; the convolution kernel size of the pooling layer is 2×2, the step size is 2, and the activation function is Max; after the vehicle driving data X undergoes the encoding process of five downsamplings, the conversion from data to features is completed to obtain the feature map EnCode5, and the specific expression is:

[0079] EnCode1=MAX(Conv(Conv(X)))

[0080] EnCode2=MAX(Conv(Conv(EnCode1)))

[0081] …

[0082] EnCode5=MAX(Conv(Conv(EnCode4)))

[0083] Among them, X is the vehicle driving data, Conv is the convolution layer, and MAX is the pooling layer.

[0084] Preferably, the decoding process is composed of five upsampling modules with the same structure connected in series, which can generate virtual benign data from the high-dimensional features related to vehicle driving stability; the upsampling module is composed of two convolutional layers and one deconvolutional layer connected in sequence; wherein, the convolution kernel size of the convolutional layer is 3×3, the step size is 2, the activation function is ReLU, and the convolution kernel size of the deconvolution layer is 2×2, and the step size is 0.5. There is a jump connection structure between the first downsampling module and the upsampling module of the same size, so that the shift parameter optimization module can learn the underlying information shared between the inferior data and the benign data; after the feature map EnCode5 undergoes five upsampling decoding processes, the feature-to-data conversion is completed to obtain the virtual benign data Y, the specific expression is:

[0085] DeCode1=DeConv(Conv(Conv(EnCode5)))+EnCode5

[0086] DeCode2=DeConv(Conv(Conv(DnCode1)))+EnCode4

[0087] …

[0088] Y=DeConv(Conv(Conv(DnCode4)))+EnCode1

[0089] Among them, Conv is the convolution layer and DeConv is the deconvolution layer.

[0090] In this embodiment, step S4 specifically includes the following operations: constructing a shift feature recognition module with a Markov random field, including a downsampling layer and a block judgment layer; the input of the shift feature recognition module is a true and false data set X+Y, and the output is a Markov random field, which contains local judgment information on the input data. The shift feature recognition module is used to judge the authenticity of the data from the true and false data set, and the judgment result is used to optimize the shift parameter optimization module during the training process.

[0091] Preferably, the downsampling layer is composed of five second downsampling modules with the same structure connected in series, which can extract high-dimensional features related to gear shifting smoothness from the true and false driving data; the second downsampling module is composed of one convolution layer, the size of the convolution kernel is 3×3, the step size is 2, and the activation function is ReLU; the high-dimensional features F5 of the true and false driving data are extracted by five second downsampling modules connected in series, and the specific expression is:

[0092] F1=Conv(X+Y)

[0093] F2=Conv(F1)

[0094] …

[0095] F5=Conv(F4)

[0096] Among them, X is the vehicle driving data, Y is the virtual benign data, and Conv is the convolutional layer.

[0097] Preferably, the block judgment layer includes a Markov random field M, which can judge the authenticity of each data block output by the downsampling layer. The random field M is an n×n matrix, each element of the matrix represents a segment in the driving data (i.e., local data somewhere in the vehicle driving data), and the value of the element is 1 or 0, representing that the judgment result of the data is true or false.

[0098] Furthermore, step S5 specifically includes the following steps:

[0099] Step S51: construct physical constraints CONS based on the automatic transmission shifting principle, including acceleration loss function, shifting process loss function and shifting action loss function. The specific expression is:

[0100] CONS=LOSS acc +LOSS pro +LOSS act

[0101] Among them, LOSS acc is the acceleration loss function, LOSS pro is the loss function of the gear shift process, LOSS act is the gear shift action loss function, and its specific meanings are as follows:

[0102] According to the engine speed, input shaft speed and output shaft speed, the engine acceleration, input shaft acceleration and output shaft acceleration are calculated to construct the acceleration loss function LOSS acc , the specific expression is:

[0103] LOSS acc =acc(eng)+acc(input)+acc(output)

[0104] Among them, eng is the engine speed, input is the input shaft speed, output is the output shaft speed, and acc is the acceleration function.

[0105] According to the gear shift process curve, the gear shift process acceleration is calculated to construct the gear shift process loss function LOSS pro , the specific expression is:

[0106] LOSS pro =acc(shiftprogression)

[0107] Among them, shiftprogression is the shift process curve, and acc is the acceleration function.

[0108] According to the shifting phases of the clutch separation and clutch engagement, the gear shifting action matching degree is calculated to construct the gear shifting action loss function LOSS act , the specific expression is:

[0109] LOSS act =match(offgoing)+match(oncoming)

[0110] Among them, offgoing is the shifting stage of disengaging the clutch, oncoming is the shifting stage of engaging the clutch, and match is the matching function.

[0111] The physical constraint CONS is used as an additional loss function of PatchGAN to constrain the learning tendency of the shift parameter optimization module. The improved PatchGAN structure is as follows: Figure 2 shown.

[0112] Step S52: Obtaining an initial true and false data set. Initialize the shift parameter optimization module (assign random numbers) so that it generates virtual good data based on the bad data, and merge the virtual good data with the real good data to obtain the true and false data set;

[0113] Step S53: Optimizing the shift feature recognition module. Specifically, the shift feature recognition module is trained using the true and false data sets to obtain a trained shift feature recognition module. This step can accurately determine whether the parameter optimization result is real data or virtual data;

[0114] Step S54: Optimize the gear shift parameter optimization module. Specifically, the gear shift feature recognition module trained in the previous step is used to identify the optimization result of the gear shift parameter optimization module; if the optimization result is the same as the real benign data, the gear shift parameter optimization module remains unchanged; if the optimization result is different from the real benign data, the gear shift parameter optimization module is trained using the real and fake data sets to obtain a trained gear shift parameter optimization module. The trained gear shift parameter optimization module obtained in this step can generate virtual benign data that the gear shift feature recognition module cannot distinguish;

[0115] Step S55: Perform game training in an iterative alternating manner. Generate new virtual benign data using the gear shift parameter optimization module trained in the previous step, merge the new virtual benign data with the real benign data to obtain a new true and false data set. Use the new true and false data set to train the current gear shift feature recognition module again, and then use the new trained gear shift feature recognition module to optimize the gear shift parameter optimization module again, and repeat this cycle;

[0116] In each cycle of step 55, the shift parameter optimization module can be connected in series with the shift feature recognition module, and then training is started until the judgment results in the random field M are unified to 1 or unified to 0, and the cycle ends. At this time, the mean of the Markov random field M is taken as the judgment result R of true and false data. The specific expression is: R = Avg (M), where Avg is the mean function, and the value is 1 or 0.

[0117] Step S56: until the gear shift parameter optimization module can generate sufficiently realistic virtual benign samples, specifically in this embodiment, until the accuracy of the gear shift feature recognition module is lower than 50%, at which time it is considered that the gear shift feature recognition module is completely ineffective, and the gear shift parameter optimization module can generate sufficiently realistic virtual benign samples, at which time the final gear shift parameter optimization module is obtained.

[0118] Furthermore, step S6 specifically includes the following steps:

[0119] Step S61: collecting driving data of the vehicle to be calibrated, and using the trained shift parameter optimization module to generate good data from the bad data of the vehicle, that is, obtaining an optimization result;

[0120] Step S62: embed the optimization results into Array Technology Industry (ATI VISION 3.8.4) or other professional data reading and writing software, communicate with the vehicle through Kvaser and OBD interface, and write the optimization results into the controller of the automatic transmission, so as to optimize the electronic control parameters of the automatic transmission and realize intelligent calibration. Figure 4 As shown, the speed of the input shaft and output shaft of the automatic transmission is stable, which is consistent with Figure 3 The effect of real benign data is similar, achieving the final intelligent calibration result.

[0121] Compared with the prior art, the present invention has the following beneficial effects: the method takes into account the physical constraints related to gear shifting, and uses the gear shifting characteristics in the driving data to calibrate the automatic transmission, which solves the limitation that the previous methods require a lot of manpower costs, financial costs and time costs, and realizes automatic identification of gear shifting characteristics in driving data and intelligent calibration of the automatic transmission, effectively improving the efficiency and accuracy of the automatic transmission calibration work.

[0122] Embodiment 2:

[0123] This embodiment provides a transmission calibration system based on driving data characteristics, including the following modules:

[0124] Vehicle driving data acquisition module: used to obtain vehicle driving data of various typical shifting processes under different throttle opening conditions and different road conditions;

[0125] Training data set construction module: used to select shifting data from the driving data obtained by the vehicle driving data acquisition module and distinguish between good data and bad data according to the vehicle shifting smoothness related parameters and the automatic transmission shifting principle, and then perform data cleaning and data preprocessing to construct the training data set;

[0126] True and false data set determination module: used to construct a shift parameter optimization module, extract characteristic information of bad data through the shift parameter optimization module, generate virtual benign data; merge the generated virtual benign data with the real benign data to obtain the true and false data set;

[0127] Gear shifting stability related feature extraction module: used to construct a gear shifting feature recognition module, and extract vehicle gear shifting stability related features of the true and false data sets through the constructed gear shifting feature recognition module;

[0128] Training module: used to build physical constraints based on the automatic transmission shifting principle, and to perform game training on the shift parameter optimization module and the shift feature recognition module in an iterative alternating manner, so that the shift parameter optimization module can eventually generate accurate good data based on bad data, and obtain the final shift parameter optimization module;

[0129] Calibration module: used to optimize the inferior data of a given vehicle using the trained shift parameter optimization module, and optimize the electronic control parameters of the automatic transmission according to the optimization results to form the final intelligent calibration result.

[0130] The above program design scheme provided in this embodiment can be stored in a computer-readable storage medium in a coded form and implemented in the form of a computer program, and the basic parameter information required for calculation is input through computer hardware, and the calculation result is output.

[0131] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0132] The present invention is described with reference to flowcharts of methods, devices (apparatus), and computer program products according to embodiments of the present invention. It should be understood that each process in the flowchart, as well as the combination of processes in the flowchart, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A device that specifies functions in a process or multiple processes.

[0133] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple functions specified in a flowchart.

[0134] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 The steps of a specified function in a process or multiple processes.

[0135] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A transmission calibration method based on driving data characteristics, characterized in that: The specific steps include: Step S1: driving the vehicle under different throttle opening conditions and different road conditions, and performing various typical gear shifting processes under each condition, and collecting vehicle driving data of various typical gear shifting processes under each condition; Step S2: According to the parameters related to vehicle gear shifting smoothness and the gear shifting principle of the automatic transmission, the gear shifting data is selected from the driving data obtained in step 1 and divided into good data and bad data, and then data cleaning and data preprocessing are performed to construct a training data set; Step S3: construct a shift parameter optimization module, extract characteristic information of the bad data through the shift parameter optimization module, and generate virtual good data; merge the generated virtual good data with the real good data to obtain a true and false data set; Step S4: constructing a shift feature recognition module, and extracting vehicle shift smoothness-related features of the true and false data sets through the constructed shift feature recognition module; Step S5: constructing physical constraints based on the automatic transmission shifting principle, and performing game training on the shift parameter optimization module and the shift feature recognition module in an iterative alternating manner, so that the shift parameter optimization module can eventually generate accurate good data based on the bad data, and obtain the final shift parameter optimization module; Step S6: Use the trained shift parameter optimization module to optimize the inferior data of a given vehicle, and optimize the electronic control parameters of the automatic transmission according to the optimization result to form a final intelligent calibration result.

2. The transmission calibration method based on driving data characteristics according to claim 1, characterized in that: In step S1, considering the influence of the throttle opening, the throttle needs to be stabilized at different percentages for testing; considering the influence of the road slope and flatness, it is necessary to test on a flat road, a bumpy road, an uphill road, and a downhill road respectively.

3. The transmission calibration method based on driving data characteristics as claimed in claim 1, characterized in that: Step S2 specifically includes the following steps: Step S21: According to the vehicle shifting smoothness related parameters and the automatic transmission shifting principle, the vehicle driving data of various typical shifting processes obtained in step 1 are locally sampled to obtain the shifting data therein, and the inferior data and the good data are distinguished to obtain the initial data set; for each vehicle shifting smoothness related parameter, the data with the worst shifting smoothness is selected as the inferior data of the parameter, and the data with the best shifting smoothness is selected as the good data of the parameter; Step S22: cleaning the initial data set to obtain valid data in the gear shifting phase; Step S23: perform data preprocessing, specifically: through cluster analysis, select the automatic transmission electronic control parameters that most affect the vehicle driving stability from the valid data of the shifting stage obtained in step 22; group the automatic transmission electronic control parameters that most affect the vehicle driving stability according to the typical shifting process; in the grouped data, merge the good data and the bad data in pairs to obtain a training data set.

4. The transmission calibration method based on driving data characteristics as claimed in claim 1, characterized in that: Step S3 specifically includes the following operations: constructing a shift parameter optimization module with a jump connection structure, including an encoding process and a decoding process; the input of the shift parameter optimization module is the vehicle driving data X, and all the inferior data in the training set are processed by the encoding process and the decoding process to output virtual benign data Y, and the dimension of the virtual benign data is the same as that of the inferior data; merging the virtual benign data with the real benign data to obtain a true and false data set X+Y with a ratio of 1:

1.

5. The transmission calibration method based on driving data characteristics as claimed in claim 4, characterized in that: The encoding process is composed of five first downsampling modules with the same structure connected in series, which can extract high-dimensional features related to vehicle driving stability from the training data set; the first downsampling module is composed of two convolutional layers and one pooling layer connected in sequence; wherein the convolution kernel size of the convolutional layer is 3×3, the step size is 2, and the activation function is ReLU; the convolution kernel size of the pooling layer is 2×2, the step size is 2, and the activation function is Max; after the vehicle driving data X undergoes the encoding process of five downsamplings, the data is converted into features to obtain the feature map EnCOde5; The decoding process is composed of five upsampling modules with the same structure connected in series, which can generate virtual benign data from the high-dimensional features related to vehicle driving stability; the upsampling module is composed of two convolutional layers and one deconvolutional layer connected in sequence; wherein the convolutional kernel size of the convolutional layer is 3×3, the step size is 2, the activation function is ReLU, and the convolutional kernel size of the deconvolutional layer is 2×2, and the step size is 0.

5. There is a jump connection structure between the first downsampling module and the upsampling module of the same size, so that the shift parameter optimization module can learn the underlying information shared between the inferior data and the benign data; after the feature map EnCOde5 undergoes five upsampling decoding processes, the feature-to-data conversion is completed to obtain the virtual benign data Y.

6. The transmission calibration method based on driving data characteristics as claimed in claim 1, characterized in that: Step S4 specifically includes the following operations: constructing a shift feature recognition module with a Markov random field, including a downsampling layer and a block judgment layer; the input of the shift feature recognition module is a true and false data set X+Y, and the output is a Markov random field, which contains local judgment information on the input data; using the shift feature recognition module to judge the authenticity of data from the true and false data set, and the judgment result is used to optimize the shift parameter optimization module during the training process; the downsampling layer: is composed of 5 second downsampling modules with the same structure connected in series, and can extract high-dimensional features related to shift smoothness from true and false driving data; the second downsampling module is composed of 1 convolution layer, the size of the convolution kernel is 3×3, the step size is 2, and the activation function is ReLU; the high-dimensional feature F5 of the true and false driving data is extracted through 5 second downsampling modules connected in series; The block judgment layer includes a Markov random field M.

7. The transmission calibration method based on driving data characteristics as claimed in claim 1, characterized in that: Step S5 specifically includes the following steps: Step S51: constructing physical constraints CONS based on the automatic transmission shifting principle, including acceleration loss function, shifting process loss function and shifting action loss function; Step S52: obtaining an initial true and false data set, specifically: initializing the shift parameter optimization module so that it generates virtual good data according to the bad data, and merging the virtual good data with the real good data to obtain the true and false data set; Step S53: optimizing the gear shift feature recognition module, specifically: using the true and false data sets to train the gear shift feature recognition module to obtain a trained gear shift feature recognition module; Step S54: Optimizing the gear shift parameter optimization module, specifically: using the gear shift feature recognition module trained in the previous step to recognize the optimization result of the gear shift parameter optimization module; if the optimization result is the same as the real benign data, the gear shift parameter optimization module remains unchanged; if the optimization result is different from the real benign data, the gear shift parameter optimization module is trained using the real and fake data sets to obtain a trained gear shift parameter optimization module. The trained gear shift parameter optimization module obtained in this step can generate virtual benign data that the gear shift feature recognition module cannot distinguish; Step S55: Perform game training in an iterative alternating manner, specifically: generate new virtual benign data using the gear shift parameter optimization module trained in the previous step, merge the new virtual benign data with the real benign data to obtain a new true and false data set. Use the new true and false data set to train the current gear shift feature recognition module again, and then use the new trained gear shift feature recognition module to optimize the gear shift parameter optimization module again, and repeat this cycle; Step S56: until the gear shift parameter optimization module can generate sufficiently realistic virtual benign samples, then the final gear shift parameter optimization module is obtained.

8. The transmission calibration method based on driving data characteristics as claimed in claim 7, characterized in that: In step 51, the expression of the physical constraint CONS is: CONS=LOSS acc +LOSS pro +LOSS act LOSS acc =acc(eng)+acc(input)+acc(output) LOSS pro =acc(shiftprogression) LOSS act =match(offgoing)+match(oncoming) Among them, LOSS acc is the acceleration loss function, LOSS pro is the loss function of the gear shift process, LOSS act is the shift action loss function, eng is the engine speed, input is the input shaft speed, output is the output shaft speed, acc is the acceleration function, shiftprogression is the shift process curve, acc is the acceleration function, offgoing is the shift stage of disengaging the clutch, oncoming is the shift stage of engaging the clutch, and match is the matching function.

9. The transmission calibration method based on driving data characteristics as claimed in claim 7, characterized in that: Step S6 specifically includes the following steps: Step S61: collecting driving data of the vehicle to be calibrated, and using the trained shift parameter optimization module to generate good data from the bad data of the vehicle, that is, obtaining an optimization result; Step S62: Communicate the optimization result with the vehicle and write the optimization result into the controller of the automatic transmission, thereby optimizing the electronic control parameters of the automatic transmission.

10. A transmission calibration system based on driving data characteristics, characterized in that: Includes the following modules: Vehicle driving data acquisition module: used to obtain vehicle driving data of various typical shifting processes under different throttle opening conditions and different road conditions; Training data set construction module: used to select shifting data from the driving data obtained by the vehicle driving data acquisition module and distinguish between good data and bad data according to the vehicle shifting smoothness related parameters and the automatic transmission shifting principle, and then perform data cleaning and data preprocessing to construct the training data set; True and false data set determination module: used to construct a shift parameter optimization module, extract characteristic information of bad data through the shift parameter optimization module, generate virtual benign data; merge the generated virtual benign data with the real benign data to obtain the true and false data set; Gear shifting stability related feature extraction module: used to construct a gear shifting feature recognition module, and extract vehicle gear shifting stability related features of the true and false data sets through the constructed gear shifting feature recognition module; Training module: used to build physical constraints based on the automatic transmission shifting principle, and to perform game training on the shift parameter optimization module and the shift feature recognition module in an iterative alternating manner, so that the shift parameter optimization module can eventually generate accurate good data based on bad data, and obtain the final shift parameter optimization module; Calibration module: used to optimize the inferior data of a given vehicle using the trained shift parameter optimization module, and optimize the electronic control parameters of the automatic transmission according to the optimization results to form the final intelligent calibration result.

Images