JP2010078084A - Data generating device and data generating method of vehicle - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To generate adapting input data for adapting output data to a predetermined scope with high precision by the reduced number of experiments by using a support vector machine in a vehicle controlling change of the output data in accordance with input data. <P>SOLUTION: An ECU controls speed change based on an initial experiment point, measures results of the control (S104), and sets a sampling point indicating the relationship between a combination of input constants and an evaluation index (S106). The ECU averages the evaluation index of the sampling point where combinations of the input constants are the same if the sampling point having the same combinations of the input constants exists (YES at S108) among the sampling points set up to the present (S112) and contracts a blind region ε used to infer a response curved surface indicating the evaluation index for the input constant by using a software margin SVM to a smaller region than an ordinary region (S114). <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a technique for generating a combination of control factors that adapts a control result to a predetermined range in a vehicle in which control in which the control result changes according to the combination of control factors is executed.

  In recent years, a method called a support vector machine (hereinafter referred to as “SVM”) has attracted attention in the field of pattern recognition and regression estimation. A technique for setting a vehicle control target value using such SVM is disclosed in, for example, Japanese Patent Application Laid-Open No. 2008-45475 (Patent Document 1).

Japanese Patent Application Laid-Open No. 2008-45475 discloses a predetermined allowance corresponding to each of a plurality of characteristic values obtained from or generated from the operation of the internal combustion engine. A fitting method that is obtained by selecting from parameter values that can fall within a possible range is disclosed. This adaptation method gives a discrete value to each value of each parameter depending on whether or not the continuous value of the characteristic value when each parameter takes a plurality of values is within an allowable range, Based on discrete values, a support vector machine is used to determine the range of each parameter that can be accommodated in a predetermined allowable range corresponding to each of a plurality of characteristic values, and parameters within the determined range From the values, select the target value for each parameter. By this adaptation method, a more optimal parameter value can be calculated with a smaller calculation load.
JP 2008-45475 A

  However, in the above-mentioned Japanese Patent Application Laid-Open No. 2008-45475, when there are a plurality of characteristic values for the same parameter combination, there is no specific description on how to use these characteristic values for the support vector machine.

  The present invention has been made to solve the above-described problems, and its object is to adapt the output data to a predetermined range in a vehicle in which control is performed in which the output data changes according to the input data. To provide a data generation apparatus and a data generation method capable of generating input data with high accuracy with a small number of experiments using a support vector machine.

  According to a first aspect of the present invention, there is provided a data generation device that generates input data for adapting output data to a predetermined range as conforming input data in a vehicle in which predetermined control in which output data changes according to the input data is executed. To do. The data generation device includes: an estimation unit that estimates a response surface indicating a response of output data to input data based on a plurality of sample data each having input data of predetermined control and corresponding output data; and a response And a generating unit that generates matching input data using a curved surface. When the first plurality of sample data each having the same input data is included in the plurality of sample data, the estimation unit averages the plurality of output data respectively included in the first plurality of sample data Is set as new output data for each of the first plurality of sample data, and a response surface is estimated by performing approximation processing for approximating the averaged new output data using a support vector machine.

  In the data generating apparatus according to the second invention, in addition to the configuration of the first invention, the estimation unit reduces the dead area as the number of the averaged first sample data increases.

  In the data generation device according to the third aspect of the invention, in addition to the configuration of the first or second aspect of the invention, the generation unit uses the response surface to temporarily correspond to the temporary matching input data and the temporary matching input data. The result of calculating additional adaptive output data and repeating the additional process of adding the data having temporary adaptive input data and temporary adaptive output data as additional sample data to the target of the approximation process performed by the estimation unit Based on this, final fit input data is generated. Each time additional sample data is added by the additional processing, the estimation unit averages a plurality of output data each of the first plurality of sample data, sets a dead area, estimates a response surface by approximation processing, and a response surface The sample data acquisition with the optimal input data is repeated.

  In the data generating device according to the fourth invention, in addition to the configuration of the third invention, the generating unit uses the temporary matching input data when a predetermined convergence condition is satisfied as the final matching input data. Generate. The predetermined convergence condition includes a condition that the temporary fitting input data has converged to a predetermined value, a condition that the temporary fitting output data has converged to a predetermined value, and a condition that the approximation error of the response surface has converged to a predetermined area. At least one of the conditions.

  In the data generation device according to the fifth invention, in addition to the configurations of the first to fourth inventions, the vehicle includes an internal combustion engine and a transmission having a hydraulic friction engagement element. The predetermined control is a shift control for changing a transmission gear ratio of the transmission. The input data is a combination of a torque command value for the internal combustion engine in the shift control, a hydraulic pressure command value for the transmission in the shift control, and a hydraulic pressure command time for the transmission in the shift control. The output data is a value indicating a shift time by the shift control and a shock by the shift control.

  According to a sixth aspect of the present invention, there is provided a data generation method for generating input data for adapting output data within a predetermined range as conforming input data in a vehicle in which predetermined control in which output data changes according to the input data is executed. This is a data generation method for a vehicle. The data generation method includes a step of estimating a response surface indicating a response of output data to input data based on a plurality of sample data each having input data of predetermined control and corresponding output data, and a response surface Generating matching input data using. In the step of estimating the response surface, when the plurality of sample data includes the first plurality of sample data each having the same input data, the plurality of output data respectively included in the first plurality of sample data The averaged value is set for each new output data of the first plurality of sample data, the dead area is set according to the number of the averaged first plurality of sample data, and the averaged new data Response surface is estimated by approximating the output data using a support vector machine.

  According to the present invention, adaptive input data can be generated with high accuracy with a small number of experiments.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the following description, the same parts are denoted by the same reference numerals. Their names and functions are also the same. Therefore, detailed description thereof will not be repeated.

  A vehicle to which a data generation device according to an embodiment of the present invention is applied will be described with reference to FIG.

  This vehicle is an FR (Front engine Rear drive) vehicle including an engine 1000 and an automatic transmission 2000. The vehicle to which the data generation device according to the present invention is applicable is not limited to the vehicle shown in FIG. For example, an FF (Front engine Front drive) vehicle may be used. Further, a vehicle using a motor as a power source instead of or in addition to engine 1000 may be used.

  The vehicle includes an engine 1000, an automatic transmission 2000, a torque converter 2100, a planetary gear unit 3000 that forms part of the automatic transmission 2000, a hydraulic circuit 4000 that forms part of the automatic transmission 2000, a propeller shaft 5000, A differential gear 6000, a drive shaft 6100, a rear wheel 7000, and an ECU (Electronic Control Unit) 8000 are included.

  Engine 1000 is an internal combustion engine that burns a mixture of fuel and air injected from an injector (not shown) in a combustion chamber of a cylinder. The piston in the cylinder is pushed down by the combustion, and the crankshaft is rotated.

  The output of the engine 1000 is adjusted not only by the fuel injection amount, but also by the air-fuel mixture ignition timing controlled by the igniter 1010 and the intake air amount (air amount sucked into the combustion chamber) controlled by the electronic throttle valve 8016. The igniter 1010 controls the ignition timing according to a control signal from the ECU 8000. Electronic throttle valve 8016 controls the amount of intake air in accordance with a control signal from ECU 8000.

  Automatic transmission 2000 is connected to engine 1000 via torque converter 2100. Automatic transmission 2000 changes the rotational speed of the crankshaft to a desired rotational speed by forming a desired gear stage.

  The driving force output from automatic transmission 2000 is transmitted to left and right rear wheels 7000 via propeller shaft 5000, differential gear 6000, and drive shaft 6100.

  Planetary gear unit 3000 includes a plurality of gear mechanisms and a plurality of friction engagement elements (clutch and brake).

  Combinations of operating states of the clutches and brakes are predetermined for each transmission gear stage. For example, when forming a second gear, the C1 clutch and the B1 brake (both not shown) are engaged, and the other clutches and brakes are released. When a third gear is formed, the C1 clutch and the C3 clutch (not shown) are engaged, and the other clutches and brakes are released. When performing upshift transmission control from the second speed to the third speed, while maintaining the C1 clutch in an engaged state, the engagement hydraulic pressure of the B1 brake is decreased to release the B1 brake and the engagement hydraulic pressure of the C3 clutch. Is increased to engage the C3 clutch. The engagement hydraulic pressure of each clutch and brake is controlled by a hydraulic circuit 4000.

  The hydraulic circuit 4000 includes a plurality of linear solenoid valves. These linear solenoid valves are controlled by a control signal from ECU 8000. When a control signal from ECU 8000 is output to hydraulic circuit 4000, hydraulic pressure corresponding to the control signal from ECU 8000 is supplied from hydraulic circuit 4000 to each clutch and brake.

  The ECU 8000 includes a vehicle speed sensor 8002, a position switch 8006 of a shift lever 8004, an accelerator opening sensor 8010 of an accelerator pedal 8008, a pedaling force sensor 8014 of a brake pedal 8012, a throttle opening sensor 8018 of an electronic throttle valve 8016, An engine speed sensor 8020, an input shaft speed sensor 8022, and an output shaft speed sensor 8024 are connected via a harness or the like.

  Vehicle speed sensor 8002 detects vehicle speed V from the rotational speed of drive shaft 6100. The position switch 8006 detects the position (shift position) SP of the shift lever 8004. The accelerator opening sensor 8010 detects the opening (accelerator opening) ACC of the accelerator pedal 8008. The pedaling force sensor 8014 detects the pedaling force of the brake pedal 8012 (force that the driver steps on the brake pedal 8012). The throttle opening sensor 8018 detects the opening (throttle opening) TH of the electronic throttle valve 8016. The engine speed sensor 8020 detects a crankshaft speed (engine speed) NE of the engine 1000. Input shaft rotational speed sensor 8022 detects input shaft rotational speed (turbine rotational speed of torque converter 2100) NT of automatic transmission 2000. Output shaft rotational speed sensor 8024 detects the rotational speed (output shaft rotational speed) NOUT of the output shaft of automatic transmission 2000. Each of these sensors transmits a signal representing the detection result to ECU 8000.

  Further, a personal computer 9000 and a torque sensor 9002 are connected to the ECU 8000.

  An experimenter inputs a driving pattern for experimentally driving the vehicle, a combination of initial input constants for performing an initial experiment (initial experimental point), an adaptation map, and the like to the personal computer 9000. The personal computer 9000 outputs information input by the experimenter to the ECU 8000.

  Torque sensor 9002 detects the torque of the output shaft of automatic transmission 2000 and outputs the detection result to ECU 8000.

  The personal computer 9000 and the torque sensor 9002 are connected when the vehicle is experimentally operated to generate data to be described later, and may be removed during normal vehicle travel.

  ECU 8000 controls each device so that the vehicle is in a desired running state based on signals sent from each sensor, a map and a program stored in ROM, and the like.

  When the shift lever 8004 is in the D (drive) position and the D (drive) range is selected as the shift range of the automatic transmission 2000, the ECU 8000 forms one of the plurality of forward gears. In this manner, the hydraulic circuit 4000 is controlled.

  For example, when performing so-called clutch-to-clutch shift control, such as the above-described upshift from the second speed to the third speed, the ECU 8000 includes a release-side hydraulic pressure command value for releasing the release-side frictional engagement element, An engagement side hydraulic pressure command value for engaging the engagement side frictional engagement element is output to the hydraulic circuit 4000. Further, in order to enable quick shifting, a retard required torque command value for delaying the ignition timing of engine 1000 and temporarily reducing engine torque during shift control is output to igniter 1010.

  FIG. 2 shows changes in the release side hydraulic pressure command value, the engagement side hydraulic pressure command value, the retard request torque command value output from the ECU 8000 during the shift control, and the turbine rotational speed NT and the output shaft torque during the shift control.

  The release side hydraulic pressure (release side hydraulic pressure command value at a predetermined time point) shown in FIG. 2 is set in advance for each shift pattern. The release side hydraulic pressure command value is a command value determined by the value of the release side hydraulic pressure.

  The engagement annealing time and the engagement side hydraulic pressure (the engagement side hydraulic pressure command value when the engagement annealing time has elapsed) shown in FIG. 2 are set in advance for each shift pattern. The engagement side hydraulic pressure command value is a command value determined by the engagement annealing time and the engagement side hydraulic pressure.

  The retard torque reduction amount shown in FIG. 2 is preset for each shift pattern. The retard required torque command value is a value determined by the retard torque reduction amount.

  The waveforms of the turbine rotational speed NT and the output shaft torque during the shift control vary depending on the combination of the release side hydraulic pressure command value, the engagement side hydraulic pressure command value, and the retard request torque command value. Therefore, the shift time determined from the waveform of the turbine speed NT and the shift shock quantitative value determined from the waveform of the output shaft torque are the engagement smoothing time, the engagement side hydraulic pressure, the release side hydraulic pressure, and the retarded torque down amount shown in FIG. Varies depending on the combination.

  The data generation device according to the present embodiment sets a control factor as an input constant, a control result as an evaluation index, and a response curved surface indicating a response of the evaluation index to the input constant (or a combination thereof when there are a plurality of input constants) Is approximated from data obtained through experiments using a support vector machine (SVM), and on this response surface, a combination of input constants (adaptation constants) for adapting the evaluation index to the characteristics of the request is generated. It is.

  In the following description, in the shift control, a total of four values of the engagement smoothing time, the engagement side hydraulic pressure, the release side hydraulic pressure, and the retard torque reduction amount shown in FIG. When a total of two values is set as an evaluation index, and a combination of input constants for adapting the balance of the evaluation index (ratio between shift time and shift shock quantitative value) to the desired balance is generated as an adaptation constant explain.

  FIG. 3 shows a functional block diagram of the data generation apparatus according to the present embodiment. In the following description, a case where the data generation device according to the present embodiment is realized by ECU 8000 will be described. Note that the data generation device according to the present embodiment may be realized by another control device different from ECU 8000.

  ECU 8000 includes an input interface 8100, a calculation processing unit 8200, a storage unit 8300, and an output interface 8400.

  The input interface 8100 receives the experimental pattern and initial experimental point from the personal computer 9000, the turbine rotational speed NT from the input shaft rotational speed sensor 8022, the output shaft torque from the torque sensor 9002, and the like, and transmits them to the arithmetic processing unit 8200. To do.

  The experimental pattern is a condition necessary for conducting an experiment for obtaining a response curved surface. The experimental pattern includes, for example, information such as a shift pattern (which gear stage to shift to which gear stage) and what value the accelerator opening at the time of shifting is set.

The initial experimental point is a combination of input constants of the minimum limit number necessary for obtaining a response surface. In the present embodiment, as shown in FIG. 4, combinations of the upper limit value and the lower limit value of each of the four input constants of the engagement side hydraulic pressure, the engagement smoothing time, the release side hydraulic pressure, and the retarding torque down amount are set. This is the initial experimental point. That is, the initial experimental score shown in FIG. 4 is 16 points (= 2 ⁴ ).

  Returning to FIG. 3, various types of information, programs, threshold values, maps, and the like are stored in the storage unit 8300, and data is read out by the arithmetic processing unit 8200 as necessary, or the arithmetic result of the arithmetic processing unit 8200 is calculated. Is stored.

  The arithmetic processing unit 8200 includes an initial experiment unit 8210, an evaluation index detection unit 8220, a response surface estimation unit 8230, a matching constant generation unit 8240, and an additional experiment unit 8250.

  The initial experiment unit 8210 sequentially performs shift control according to the experiment pattern for each initial experiment point.

  Evaluation index detection unit 8220 detects an evaluation index (shift time, shift shock quantitative value) for each experimental point from the waveform of turbine rotational speed NT and output shaft torque during shift control obtained by each experiment.

  The response surface estimation unit 8230 estimates, by approximation, a response surface that continuously indicates responses of the two evaluation indexes with respect to the combination of the four input constants based on the evaluation indexes for the respective experimental points. By estimating the response surface, it is possible to estimate an evaluation index for an untested point other than the experimental point.

  In the present embodiment, the response surface is estimated by performing regression using a soft margin SVM (support vector machine). Here, the soft margin SVM will be briefly described. The SVM uses an identification plane that maximizes the distance (margin) from each sample point among the identification planes that separate points (hereinafter also referred to as “sample points”) indicating the relationship between the input data and output data of each experiment. Obtained using a predetermined function. In the soft margin SVM, in consideration of the fact that each sample point includes an error (variation), the distance from the identification plane is equal to or less than a predetermined allowable error in order to suppress the degree of freedom of the sample point used for calculation of the identification plane. The insensitive area ε is set, and an identification plane that maximizes the distance (margin) from the remaining sample points other than the sample points included in the insensitive area ε is obtained using a predetermined function.

  In the present embodiment, the identification plane obtained using the soft margin SVM is used as the response curved surface. As a result, it is possible to estimate the response of untested points with a small number of experiments, and to effectively suppress not only the experimental point approximation error but also the untested point estimation error.

FIG. 5 shows an image of a response surface when the input constant combination is taken as the input X on the horizontal axis and the evaluation index combination is taken as the output Y on the vertical axis. As shown in FIG. 5, the response surface (identification plane) is a plane that minimizes the objective function τ (W) = (1/2) × ‖W‖ 2 + C × Σiεi. Here, εi indicates a distance (error) between each sample point included in the insensitive area ε and the upper limit plane (or lower limit plane) of the insensitive area ε, and Σiεi indicates an error in each sample point included in the insensitive area ε. Indicates the total value. W is a parameter corresponding to the synaptic load, and W and the insensitive area ε are mutually dependent. C is an adjustment parameter that determines which of the margin (= 1 / ‖W‖ ² ) or the insensitive area ε is to be emphasized. The above objective function τ (W) is merely an example, and other objective functions may be set.

  Returning to FIG. 3, the adaptation constant generation unit 8240 adapts the balance of the evaluation index (ratio between the shift time and the shift shock quantitative value) to the desired balance based on the response surface obtained by the response surface estimation unit 8230. A combination of input constants (adaptation constants) is generated. Specifically, as shown in FIG. 6, the adaptation constant generation unit 8240 does not have a Pareto solution of the evaluation index on the response surface (other points where both the shift time and the shift shock quantitative value are smaller than themselves). A set of points) is obtained, the intersection of the Pareto solution and the craving line is taken as a sought point, and a combination of input variables corresponding to this sought point is obtained as a temporary fitness constant. Here, the craving line indicates a ratio (adapted balance) between the shift time and the shift shock quantitative value preset by the experimenter.

  When a predetermined convergence condition (for example, a condition that the temporary adaptation constant has converged to a predetermined combination) is established, the adaptation constant generation unit 8240 performs an experiment using the temporary adaptation constant at that time as a final adaptation constant. Terminate. The adaptation constant generation unit 8240 outputs an additional experiment command to the additional experiment unit 8250 when a predetermined convergence condition is not satisfied.

  When the additional experiment command from the adaptation constant generation unit 8240 is input, the additional experiment unit 8250 receives a combination of input constants corresponding to the desired points (temporary adaptation constants) and points on the Pareto solution other than the desired points. An additional experiment is further performed using a total of two combinations of corresponding input constants as additional experimental points.

  Each process of the evaluation index detection unit 8220, the response surface estimation unit 8230, the adaptive constant generation unit 8240, and the additional experiment unit 8250 is repeatedly executed until the above-described convergence condition is satisfied.

  As a result, as shown in FIG. 7, additional experimental points are intensively added in the vicinity of the sought point and in the vicinity of the Pareto solution other than the sought point. Therefore, in particular, the estimation accuracy of the response surface near the sought point is improved, and the adaptation constant converges to a predetermined combination.

  FIG. 8 shows an example of convergence of the release side hydraulic pressure command value, the engagement side hydraulic pressure command value, the turbine speed NT, and the output shaft torque that are converged by this experiment. The waveforms of the release side hydraulic pressure command value, the engagement side hydraulic pressure command value, the turbine rotational speed NT, and the output shaft torque converge to a predetermined waveform by repeating this experiment, as shown on the right side of FIG. These converged waveforms are waveforms that keep the ratio between the shift time and the shift shock quantitative value in the experimenter's desire balance. In other words, by repeating this experiment, it is possible to generate the release side hydraulic pressure command value and the engagement side hydraulic pressure command value in which both the shift time and the shift shock are good.

  FIG. 9 shows an example of combinations of input constants that converge in this experiment. In FIG. 9, it can be seen that the combination of the four input constants converges to two by adding the experiment. That is, by performing the shift control with a combination of any of these two input constants, the balance between the shift time and the shift shock can be matched to the desired balance.

  As described above, since the data generation apparatus according to the present embodiment adds experimental points near the sought points as further experimental points, the number of experiments in which the combinations of the input constants are inevitably increased. FIG. 10 shows an example. In the example of FIG. 10, among the 96 experiments, there are 32 experiments that have the same combination of input constants.

  In this way, when there are experiments with the same combination of input constants, by averaging the experimental results, variations in the experimental results for the combinations of input constants can be reduced. For example, assuming that the variation in experimental results is a normal distribution, the averaging of N = 2 can reduce the variation in experimental results to 1 / (square root of 2) times as shown in FIG. In general, when N pieces of experimental result data are averaged, the variation in the experimental result data can be reduced to about 1 / (square root of N) times, and the more experimental data used for averaging, the more the variation in the experimental results. Can be reduced.

  This point is focused on in the present embodiment. That is, the response surface estimator 8230 averages the experimental results with the same combination of input constants, and uses the averaged highly reliable experimental results for the calculation of the response surface. The dead area ε (allowable error) when performing regression estimation using the soft margin SVM is reduced according to the number of experimental points.

  In the following, all of the initial experiment unit 8210, the evaluation index detection unit 8220, the response surface estimation unit 8230, the adaptive constant generation unit 8240, and the additional experiment unit 8250 are stored in the storage unit of the CPU that is the arithmetic processing unit 8200. A description will be given assuming that the program functions as software realized by executing the program stored in the 8300. Note that each of the above-described units may be realized by hardware.

  With reference to FIG. 12, a control structure of a program executed by ECU 8000 which is the data generation device according to the present embodiment will be described. Note that this program is repeatedly executed at a predetermined cycle time when the personal computer 9000 is connected to the ECU 8000.

  In step (hereinafter, step is abbreviated as S) 100, ECU 8000 determines whether or not an experimental pattern from personal computer 9000 has been input. When the experimental pattern is input (YES in S100), the process proceeds to S102. Otherwise (NO in S100), this process ends.

  In S102, ECU 8000 determines whether or not an initial experimental point from personal computer 9000 has been input. When the initial experimental point is input (YES in S102), the process proceeds to S104. Otherwise (NO in S102), this process ends.

  In S104, ECU 8000 executes shift control based on the experimental pattern and the initial experimental point, and measures the control result. Specifically, a control signal including a hydraulic pressure command value and a torque command value corresponding to the experimental pattern and the initial experimental point is output to the hydraulic circuit 4000 and the igniter 1010, and the turbine speed NT during the shift control and the output shaft are controlled. Measure torque.

  In S106, ECU 8000 sets a sample point (a point indicating a relationship between a combination of input constants and an evaluation index). Specifically, ECU 8000 detects an evaluation index (shift time, shift shock quantitative value) from the turbine rotational speed NT and the output shaft torque waveform (see FIG. 2 described above), and the detected evaluation index and corresponding to it. The experimental points are sequentially stored in the storage unit 8300 as sample points.

  In S108, ECU 8000 determines whether there are sample points with the same combination of input constants among the sample points set up to now. If there is a sample point having the same combination of input constants (YES in S108), the process proceeds to S112. Otherwise (NO in S108), the process proceeds to S110.

  In S110, ECU 8000 sets insensitive area ε as a normal area. In S112, ECU 8000 averages each evaluation index (shift time and shift shock quantitative value) possessed by sample points having the same combination of input constants.

  In S114, ECU 8000 reduces insensitive area ε smaller than the normal area. At this time, only the insensitive area ε corresponding to the combination of the same input constants is reduced from the normal area, not the entire insensitive area ε in the combination.

  In S116, ECU 8000 estimates a response surface using soft margin SVM. In other words, as described above, ECU 8000 obtains an identification plane that minimizes the above-described objective function τ (W) based on the insensitive area ε set in S110 or S114, and sets this identification plane as a response surface. .

  In S118, ECU 8000 calculates an evaluation index of the desired score. That is, ECU 8000 obtains a Pareto solution of the evaluation index on the response surface as shown in FIG.

  In S120, ECU 8000 calculates the input constant corresponding to the evaluation index of the sought point as the input constant of the sought point.

  In S122, ECU 8000 determines whether or not a predetermined convergence condition is satisfied. The predetermined convergence condition includes a condition that the combination of the evaluation indices of the sought points has converged to the predetermined combination, a condition that the input constant corresponding to the evaluation index of the sought points has converged to the predetermined combination, and an approximation error of the response surface (insensitive This is at least one of the conditions that the region ε) has converged to the predetermined region. If the predetermined convergence condition is satisfied (YES in S122), the process proceeds to S124. Otherwise (NO in S122), the process proceeds to S126.

  In S124, ECU 8000 generates a combination of input constants for the current desired point as a final matching constant, and outputs it to personal computer 9000. This process ends the experiment.

  In S126, ECU 8000 sets an additional experimental point. The additional experimental points are set to two points of input constants corresponding to the evaluation index on the Pareto solution other than the combination of the input constants of the sought points. After this process, the process returns to S104 again, and an additional experiment is performed.

  As described above, in the data generation device according to the present embodiment, when the predetermined convergence condition is not satisfied (NO in S122), the current combination of the input constants of the desired points is set as the additional experimental point (S126). Therefore, the number of experiments that inevitably has the same combination of input constants increases.

  Therefore, ECU 8000 has the same combination of input constants when there is an evaluation index having the same combination of input constants among the evaluation indices detected so far in the course of the additional experiment (YES in S108). The evaluation index is averaged (S112), and the dead area ε used when estimating the response surface using the soft margin SVM is reduced more than the normal area (S114).

  Therefore, the variation in the experimental result (evaluation index) is reduced by averaging, and the sample points having the averaged highly reliable experimental result are easily used for calculating the response surface. Thereby, a highly accurate response surface can be obtained with a smaller number of experiments. Therefore, a combination of input constants for adapting the balance of the evaluation index (ratio between the shift time and the shift shock quantitative value) to the desired balance can be obtained with a smaller number of experiments and with higher accuracy.

  Note that the control applicable to the data generation device according to the present invention is not limited to the shift control. For example, you may make it apply to the output control of the engine 1000. FIG.

  The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

It is a figure which shows the structure of the vehicle by which the data generation apparatus which concerns on embodiment of this invention is mounted. It is a figure which shows the change of the turbine rotational speed and output shaft torque by shift control by the releasing side hydraulic pressure command value, engagement side hydraulic pressure command value, retardation request torque command value which are output from ECU at the time of shift control. It is a functional block diagram of a data generation device. It is a figure which shows an initial experimental point. It is a figure which shows the image of a response curved surface. FIG. 6 is a diagram (part 1) illustrating a Pareto solution and a craving line of an evaluation index on a response curved surface. FIG. 5 is a diagram (part 2) illustrating a Pareto solution and a craving line of an evaluation index on a response curved surface. It is a figure which shows the mode of the convergence of a releasing side hydraulic pressure command value, an engagement side hydraulic pressure command value, turbine rotational speed NT, and output shaft torque. It is a figure which shows the mode of the convergence of the combination of input constants. It is a figure which shows the frequency | count of experiment from which the combination of an input constant becomes the same. It is a figure which shows the change of the probability density by averaging. It is a flowchart which shows the control structure of a data generation apparatus.

Explanation of symbols

  1000 engine, 1010 igniter, 2000 automatic transmission, 2100 torque converter, 3000 planetary gear unit, 4000 hydraulic circuit, 5000 propeller shaft, 6000 differential gear, 6100 drive shaft, 7000 rear wheels, 8002 vehicle speed sensor, 8004 shift lever, 8006 position switch, 8008 Accelerator pedal, 8010 Accelerator opening sensor, 8012 Brake pedal, 8014 Treading force sensor, 8016 Electronic throttle valve, 8018 Throttle opening sensor, 8020 Engine speed sensor, 8022 Input shaft speed sensor, 8024 Output shaft speed sensor, 8000 ECU, 8100 input interface, 8200 arithmetic processing unit, 8 10 Initial experiments unit, 8220 evaluation index detection section, 8230 the response surface estimation unit, 8240 adapted constant generation unit, 8250 an additional experiment section, 8300 a storage section, 8400 output interface, 9000 personal computers, 9002 torque sensor.

Claims (6)

In a vehicle in which predetermined control in which output data changes according to input data is executed, a vehicle data generation device that generates input data for adapting the output data to a predetermined range as conforming input data,
An estimation unit that estimates a response surface indicating a response of output data to input data based on a plurality of sample data each having input data of the predetermined control and corresponding output data;
A generating unit that generates the adaptive input data using the response surface;
When the first plurality of sample data each having the same input data is included in the plurality of sample data, the estimation unit averages a plurality of output data respectively included in the first plurality of sample data The converted value is set in each new output data of the first plurality of sample data, and the insensitive area is set according to the number of the averaged first sample data, and averaged. A vehicle data generation apparatus for estimating the response surface by performing an approximation process for approximating new output data using a support vector machine.   2. The vehicle data generation device according to claim 1, wherein the estimation unit reduces the dead area as the number of the plurality of averaged first sample data increases. The generating unit calculates temporary adaptation input data corresponding to the temporary adaptation input data and the temporary adaptation input data using the response curved surface, and the temporary adaptation input data and the temporary adaptation data. Based on the result of sequentially repeating additional processing for adding data having adaptive output data as additional sample data to the target of the approximation processing performed by the estimation unit, final adaptive input data is generated,
Each time the additional sample data is added by the addition process, the estimation unit averages the plurality of output data included in each of the first plurality of sample data, sets the insensitive area, and the approximation process performs the approximation process. The vehicle data generation device according to claim 1, wherein estimation of a response surface and acquisition of sample data based on optimum input data on the response surface are repeatedly performed. The generation unit generates the temporary matching input data when a predetermined convergence condition is established as the final matching input data,
The predetermined convergence condition includes a condition that the temporary fitting input data has converged to a predetermined value, a condition that the temporary fitting output data has converged to a predetermined value, and an approximation error of the response surface converges to a predetermined region. The vehicle data generation device according to claim 3, wherein the vehicle data generation device is at least one of the conditions. The vehicle includes an internal combustion engine and a transmission having a hydraulic friction engagement element,
The predetermined control is a shift control for changing a transmission gear ratio of the transmission,
The input data is a combination of a torque command value to the internal combustion engine in the shift control, a hydraulic command value to the transmission in the shift control, and a hydraulic command time to the transmission in the shift control.
5. The vehicle data generation device according to claim 1, wherein the output data is a value indicating a shift time by the shift control and a shock by the shift control. A vehicle data generation method for generating input data for adapting the output data to a predetermined range as conforming input data in a vehicle in which predetermined control in which output data changes according to input data is performed.
Estimating a response surface indicating a response of the output data to the input data based on a plurality of sample data each having the input data of the predetermined control and the corresponding output data;
Generating the matched input data using the response surface;
In the step of estimating the response surface, when the plurality of sample data includes a plurality of first sample data each having the same input data, a plurality of the plurality of first sample data respectively include A value obtained by averaging the output data is set in each new output data of the first plurality of sample data, and a dead area is set according to the number of the first plurality of sample data averaged. A vehicle data generation method for estimating the response surface by performing an approximation process for approximating new averaged output data using a support vector machine.

Images