Control device and control method for vehicle drive unit

Abstract

The present invention relates to a control device and a control method for a vehicle drive unit. The control device for the vehicle drive unit is configured to control the vehicle drive unit having one or more power sources based on the operating state of the vehicle. The control device includes a processor and a storage device. The storage device is configured to store a vehicle fore-and-aft acceleration prediction model as a machine learning model, which receives a command torque as an input and outputs a predicted acceleration. The processor is configured to: perform a predicted acceleration calculation process using the vehicle fore-and-aft acceleration prediction model; and perform a command torque calculation process to calculate a command torque that minimizes an evaluation function. The evaluation function minimizes the deviation of the predicted acceleration relative to a target vehicle fore-and-aft acceleration according to a target torque while reducing the deviation of the command torque relative to the target torque, the target torque being based on the operating state.

Description

Technical Field

[0001] The present disclosure relates to a control device and a control method for a vehicle drive unit. Background Art

[0002] JP 2005-119359 A discloses a vehicle control device applied to a vehicle equipped with a drive system having an internal combustion engine and a transmission. The vehicle control device includes a control type selection device and a control device.

[0003] The control pattern selection device selects one output pattern from a plurality of types of drive torque output patterns prepared in advance based on the detection results of the detection device for detecting various state quantities related to the vehicle. The control device controls the internal combustion engine and the drive system based on the output pattern selected by the control pattern selection device. Summary of the invention

[0004] During vehicle acceleration or deceleration, vehicle fore-and-aft vibrations (vibrations from vehicle fore-and-aft acceleration) may be generated by torsion of the vehicle drive shaft, or may be generated by the elimination of backlash in gears located in the power transmission path from one or more of the vehicle's power sources to the drive shaft.

[0005] In order to achieve a good balance between the reduction of the front and rear vibration of the vehicle and the vehicle's sporting performance (i.e., responsiveness), the vehicle control device disclosed in JP2005-119359 A has multiple types of drive torque output patterns as described above. In addition, these drive torque output patterns are set differently according to the detection results of various state quantities related to the vehicle. In this torque control method disclosed in JP2005-119359 A, many adaptation values ​​may be required depending on the running state of the vehicle.

[0006] The present disclosure has been made in view of the above-mentioned problems, and an object of the present disclosure is to provide a control device and a control method for a vehicle drive unit, which can achieve both reduction of vehicle front and rear vibration and vehicle acceleration and deceleration performance while reducing the number of adaptations of parameters used for vehicle drive torque control.

[0007] A control device for a vehicle drive unit according to one aspect of the present disclosure is configured to control a vehicle drive unit having one or more power sources based on an operating state of the vehicle. The control device includes a processor and a storage device. The storage device is configured to store a vehicle fore-and-aft acceleration prediction model as a machine learning model, which receives a command torque as an input and outputs a predicted acceleration as a predicted value of the vehicle fore-and-aft acceleration, the command torque being a vehicle drive torque commanded to the vehicle drive unit. The processor is configured to: perform a predicted acceleration calculation process to calculate a predicted acceleration using the vehicle fore-and-aft acceleration prediction model; and perform a command torque calculation process to calculate a command torque that minimizes an evaluation function. The evaluation function minimizes a deviation of the predicted acceleration relative to a target vehicle fore-and-aft acceleration according to a target torque while reducing a deviation of the command torque relative to a target torque, the target torque being a target value of the vehicle drive torque based on the operating state.

[0008] The processor can be configured to further perform acceleration correction processing to correct the predicted acceleration output from the vehicle fore-and-aft acceleration prediction model based on the difference between the predicted acceleration and the actual vehicle fore-and-aft acceleration generated when the vehicle drive torque according to the command torque is output from the vehicle drive unit.

[0009] The processor can be configured to perform predicted acceleration calculation processing and command torque calculation processing for each time step for data of target torque, command torque, target vehicle front and rear acceleration and predicted acceleration for multiple time steps included in a prediction period, wherein the prediction period is a specified period starting from the current time step.

[0010] The prediction period can be a minimum period for predicting at least one of vibrations of actual vehicle fore-and-aft acceleration caused by torsion of the vehicle drive shaft and vibrations of actual vehicle fore-and-aft acceleration caused by elimination of backlash of gears located in a power transmission path from the one or more power sources to the drive shaft.

[0011] The processor may be configured to cause the target torque at each time step of the plurality of time steps included in the prediction period to be constant at the value of the target torque at the current time step.

[0012] A control method for a vehicle drive unit according to another aspect of the present disclosure controls a vehicle drive unit having one or more power sources based on the operating state of the vehicle. The control method includes: a predicted acceleration calculation process for calculating a predicted acceleration as a predicted value of the vehicle fore-and-aft acceleration using a vehicle fore-and-aft acceleration prediction model as a machine learning model, the machine learning model receiving a command torque as an input and outputting a predicted acceleration, the command torque being a vehicle drive torque commanded to the vehicle drive unit; and a command torque calculation process for calculating a command torque that minimizes an evaluation function. The evaluation function minimizes the deviation of the predicted acceleration relative to a target vehicle fore-and-aft acceleration according to a target torque while reducing the deviation of the command torque relative to a target torque, the target torque being a target value of the vehicle drive torque based on the operating state.

[0013] According to the control device or control method for a vehicle drive unit disclosed in the present invention, a command torque that minimizes an evaluation function is calculated. The evaluation function minimizes the deviation of the predicted acceleration relative to the target vehicle front and rear acceleration. Therefore, according to the command torque that minimizes the evaluation function, the front and rear vibration of the vehicle can be satisfactorily reduced. Then, a predicted acceleration for calculating the command torque is calculated using a vehicle front and rear acceleration prediction model as a machine learning model. As a result, the number of adaptation values ​​for vehicle drive torque control can be reduced. In addition, according to the evaluation function, a command torque that not only minimizes the deviation of the predicted acceleration relative to the target vehicle front and rear acceleration but also reduces the deviation of the command torque relative to the target torque is calculated. Therefore, the command torque can be determined so as to advantageously meet the target torque while reducing the front and rear vibration of the vehicle. Therefore, according to the control device or control method disclosed in the present invention, the front and rear vibration reduction of the vehicle and the vehicle acceleration and deceleration performance can be achieved while reducing the number of adaptations of the parameters for vehicle drive torque control. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a diagram showing an example of a system configuration of a vehicle on which a vehicle drive unit according to a first embodiment is mounted;

[0015] Figure 2 is a block diagram showing an outline of a control structure of a vehicle driving torque according to a first embodiment;

[0016] Figure 3 is a flowchart showing an example of a process flow related to vehicle driving torque control according to the first embodiment;

[0017] Figure 4 is a time chart for describing a method of predicting a target torque Tpfrq during a prediction period tp;

[0018] Figure 5 It is shown by Figure 2A time chart showing an example of the target vehicle longitudinal acceleration Greq for the prediction period tp calculated by the target value conversion unit shown;

[0019] Figure 6 is a block diagram showing an outline of a control structure of a vehicle driving torque according to a second embodiment;

[0020] Figure 7 is shown with Figure 6 A flowchart showing an example of a processing flow associated with a model correction unit and a model relearning unit;

[0021] Figure 8 Is used to describe Figure 7 A timing diagram of an example of a determination method of step S200 in FIG.

[0022] Fig. 9A Is used to describe Figure 6 A diagram showing an example of a method of correcting the time delay and amplitude of the predicted acceleration Gpre by the model correction unit shown;

[0023] Fig. 9B Is used to describe Figure 6 A diagram showing an example of a method in which the model correction unit corrects the time delay and amplitude of the predicted acceleration Gpre; and

[0024] Fig.10 is shown in Figure 6 The illustrated timing chart is an example of waveforms of the predicted acceleration Gpre and the actual acceleration Gact in a case where the model relearning unit 50a performs relearning of the predicted acceleration Gpre. DETAILED DESCRIPTION

[0025] It should be understood that even if the number, quantity, amount, range or other numerical attributes of elements are mentioned in the description of the following embodiments, the present disclosure is not limited to the mentioned numerical attributes unless otherwise explicitly described or unless the present disclosure is theoretically explicitly specified by the numerical attributes.

[0026] 1. First Embodiment

[0027] 1-1. Vehicle system configuration example

[0028] Figure 11 is a diagram showing an example of a system configuration of a vehicle 100 on which a vehicle drive unit 10 according to a first embodiment is mounted. The vehicle 100 includes the vehicle drive unit 10. The vehicle drive unit 10 is configured to drive the vehicle 100 (wheels 102). The vehicle drive unit 10 includes a first motor generator (MG) 1, a second motor generator (MG) 2, and an internal combustion engine 3 as power sources thereof. MG 1 is mainly used as a generator, and MG 2 is mainly used as an electric motor that drives the wheels 102. In addition, as an example, the vehicle drive unit 10 includes a power distribution device 12.

[0029] The power distribution device 12 includes a first planetary gear unit 14, a second planetary gear unit 16, a low speed clutch 18, a high speed clutch 20 and a speed reduction mechanism 22. The first planetary gear unit 14 includes a first sun gear 14a, a first planet carrier 14b and a first ring gear 14c. The second planetary gear unit 16 includes a second sun gear 16a, a second planet carrier 16b and a second ring gear 16c. The first sun gear 14a is connected to the MG 1, and the first planet carrier 14b is connected to the internal combustion engine 3. The first ring gear 14c is coupled to the second sun gear 16a.

[0030] The low-speed clutch 18 is configured so that the second sun gear 16a and the second planetary carrier 16b can be connected to each other. The high-speed clutch 20 is configured so that the second planetary carrier 16b and the second ring gear 16c can be connected to each other. The speed reduction mechanism 22 includes an intermediate shaft (hereinafter also referred to as a "transmission shaft") 24 and a differential gear 26. The transmission shaft 24 is parallel to the rotation axis of the second ring gear 16c and the rotation axis 2a of the MG 2. The transmission shaft 24 is connected to each of the second ring gear 16c and the MG 2 through a gear. That is, the transmission shaft 24 is arranged so that the torque of the MG 2 can be added to the torque from the second ring gear 16c (mainly the torque from the internal combustion engine 3). In addition, the transmission shaft 24 is connected to the differential gear 26 through a gear. Therefore, the torque from the second ring gear 16c and the torque of the MG2 are transmitted to the wheels 102 via the speed reduction mechanism 22 and the drive shaft 104. It should be noted that the differential gear 26 may be built-in or not built-in in the vehicle drive unit 10.

[0031] According to the above-mentioned vehicle drive unit 10, the speed can be changed by controlling the engagement / disengagement of the clutches 18 and 20. Specifically, the driving mode of the vehicle drive unit 10 includes a high speed mode suitable for the high speed of the vehicle 100 and a low speed mode suitable for the low speed. According to the power distribution device 12, the high speed mode can be selected by engaging the high speed clutch 20 while releasing the low speed clutch 18, and the low speed mode can be selected by engaging the low speed clutch 18 while releasing the high speed clutch 20.

[0032] As described above, as an example, the vehicle 100 is a hybrid electric vehicle including a power distribution type vehicle drive unit 10. However, the "vehicle" according to the present disclosure may be a hybrid electric vehicle including a vehicle drive unit of a type other than the power distribution type. In addition, the vehicle may be, for example, a vehicle including a vehicle drive unit having only one or more electric motors as its power source (e.g., a battery electric vehicle or a fuel cell electric vehicle), or a vehicle including a vehicle drive unit having only an internal combustion engine as its power source.

[0033] In addition, the vehicle 100 includes an electronic control unit (ECU) 30. The ECU 30 is a computer configured to perform various processes related to the vehicle 100, and corresponds to a "control device" configured to control the vehicle drive unit 10. Specifically, the processes performed by the ECU 30 include processes related to the control of each of the MG 1, MG 2, and the internal combustion engine 3, and processes related to the control of the clutches 18 and 20 (speed control). The ECU 30 includes a processor 30a and a storage device 30b. The processor 30a reads and executes a program stored in the storage device 30b. As a result, the above-mentioned various processes of the processor 30a are implemented. A plurality of ECUs may be provided as the ECU 30. For example, the ECU 30 may be configured to include: an ECU for overall control of the vehicle drive unit 10; an ECU for controlling the respective MG 1 and MG 2; an ECU for controlling the internal combustion engine 3; and an ECU for controlling the clutches 18 and 20.

[0034] The ECU 30 receives sensor signals from the sensors 32 for the above-mentioned various processes. The sensors 32 mentioned here include, for example, a vehicle acceleration sensor, a vehicle speed sensor, an accelerator position sensor, a rotation speed sensor for MG 1, a rotation speed sensor for MG 2, a crank angle sensor, and a gear position (vehicle speed mode) sensor. In addition, the storage device 30 b stores a vehicle longitudinal acceleration prediction model 38, which will be referred to below. Figure 2 Give a description.

[0035] 1-2. Vehicle drive torque control

[0036] Figure 2 ECU 30 executes “vehicle driving torque control” to drive vehicle 100 . Figure 21 shows a functional configuration of the ECU 30 associated with the vehicle driving torque control. The ECU 30 includes a target torque calculation unit 34, a target value conversion unit 36, a vehicle front and rear acceleration prediction model 38, and a driving force optimizer 40. When the processor 30a executes a program stored in the storage device 30b, these target torque calculation unit 34, target value conversion unit 36, vehicle front and rear acceleration prediction model 38, and driving force optimizer 40 are realized by software.

[0037] The vehicle driving torque that is the object of the vehicle driving torque control according to the first embodiment is, for example, the torque (transmission shaft torque) Tp of the transmission shaft 24. Hereinafter, the target value (i.e., the required value) of the transmission shaft torque Tp is referred to as the "target torque Tpfrq". In addition, the "command torque" that is the vehicle driving torque commanded to the vehicle drive unit 10 is, for example, the "command torque Tpcm" that is the command value of the transmission shaft torque Tp.

[0038] In addition, the "vehicle driving torque" used in the vehicle driving torque control is not limited to the propeller shaft torque Tp (i.e., the torque on the input side of the differential gear 26), as long as it is uniformly used in the vehicle driving torque control. That is, for example, the torque after the final deceleration by the differential gear 26 (i.e., the torque of the drive shaft 104) can be used as the vehicle driving torque.

[0039] The target torque calculation unit 34 calculates the target torque Tpfrq. The target torque Tpfrq can be calculated, for example, as follows. That is, the target torque calculation unit 34 calculates the transmission shaft speed based on the accelerator position (accelerator depression amount) and the vehicle speed of the vehicle 100. Then, the target torque calculation unit 34 calculates the target torque Tpfrq from a mapping (not shown) that defines the relationship between the target torque Tpfrq and the accelerator position, the transmission shaft speed, and the gear according to the accelerator position, the transmission shaft speed, and the gear. The accelerator position, the vehicle speed, and the gear used to calculate the target torque Tpfrq are obtained using, for example, the above-mentioned sensor 32.

[0040] The target torque Tpfrq calculated by the target torque calculation unit 34 is input to the target value conversion unit 36 ​​and the driving force optimizer 40. The target value conversion unit 36 ​​calculates the "target vehicle longitudinal acceleration (or simply "target acceleration") Greq", which is the target value (required value) of the vehicle longitudinal acceleration G. Figure 3 A specific calculation method of the target acceleration Greq is described. The target acceleration Greq calculated by the target value conversion unit 36 ​​is input to the driving force optimizer 40 .

[0041] The vehicle front-rear acceleration prediction model (or simply referred to as "acceleration prediction model") 38 is configured to predict the vehicle front-rear acceleration G generated in the vehicle 100 when the command torque Tpcm is commanded to the vehicle drive unit 10. That is, the acceleration prediction model 38 is a machine learning model constructed using the command torque Tpcm as an input and using the "predicted acceleration Gpre" ​​as an output. The predicted acceleration Gpre is a predicted value of the vehicle front-rear acceleration G. Specifically, the command torque Tpcm output from the driving force optimizer 40 is input to the acceleration prediction model 38. In addition, one or more variables x indicating the operating state of the vehicle 100 are input to the acceleration prediction model 38 together with the command torque Tpcm. The one or more variables x are, for example, the torque of MG 2 and the vehicle speed.

[0042] The acceleration prediction model 38 is constructed using, for example, a deep neural network. The learning of the acceleration prediction model 38 is performed in advance using learning data, which is a combination of "time series data of command torque Tpcm and one or more variables x (which are explanatory variables (i.e., inputs)" and "time series data of vehicle longitudinal acceleration G (more specifically, actual vehicle longitudinal acceleration Gact) (which is the target variable (i.e., output)"). The learning data is acquired during the specified data acquisition period during acceleration and deceleration of the vehicle 100. In addition, the acceleration prediction model 38 is constructed to be updateable on the vehicle 100.

[0043] In addition, the learning data is acquired by actually using the vehicle 100 including the vehicle drive unit 10. Therefore, the acceleration prediction model 38 is constructed as a machine learning model that is well adapted to the characteristics of the vehicle 100. Therefore, according to the “vehicle longitudinal acceleration prediction model” according to the present disclosure, the prediction of the vehicle longitudinal acceleration G can be performed according to the input command torque while satisfactorily reflecting the characteristics of the vehicle used to acquire the learning data.

[0044] The predicted acceleration Gpre predicted by the acceleration prediction model 38 is input to the driving force optimizer 40 together with the target torque Tp and the target acceleration Greq. Upon receiving these inputs, the driving force optimizer 40 calculates the evaluation function J (see below) Figure 3 ) minimizes the command torque Tpcm.

[0045] When the command torque Tpcm calculated by the driving force optimizer 40 is input to the acceleration prediction model 38, the acceleration prediction model 38 calculates the latest predicted acceleration Gpre based on the input latest command torque Tpcm, and outputs the latest predicted acceleration Gpre to the driving force optimizer 40. If the value of the evaluation function J does not meet the specified optimality criterion, the driving force optimizer 40 repeatedly calculates (i.e., corrects) the command torque Tpcm while receiving the latest predicted acceleration Gpre from the acceleration prediction model 38. On the other hand, if the value of the evaluation function J meets the optimality criterion, the driving force optimizer 40 outputs (i.e., commands) the latest command torque (i.e., corrected torque) Tpcm to the vehicle drive unit 10 as the command value of the final propeller shaft torque Tp (final command torque Tpcmf).

[0046] In the vehicle 100, when the target torque Tp is given to the vehicle drive unit 10 during acceleration or deceleration, vehicle front-rear vibrations caused by the torsion of the drive shaft 104 may be generated, and vehicle front-rear vibrations caused by the elimination of the backlash of the gears of each part of the vehicle drive unit 10 may also be generated. In order to evaluate the waveform of such vehicle front-rear vibrations (more specifically, the vibration of the vehicle front-rear acceleration G), it is necessary to obtain the waveform (more specifically, time series data) of the predicted acceleration Gpre. Hereinafter, the period for obtaining the waveform (time series data) of the predicted acceleration Gpre is referred to as the "prediction period tp". The prediction period tp (for example, see the prediction period tp described below) Figure 4 ) is the specified period starting from the current time step.

[0047] In order to ensure the prediction accuracy of the predicted acceleration Gpre and reduce the calculation load required for the prediction, it is advantageous that the prediction period tp is short. On the other hand, in order to properly evaluate the waveform of the front-rear vibration of the vehicle, it is necessary to obtain the time series data of the predicted acceleration Gpre of at least half a cycle of the front-rear vibration of the vehicle. The reason is that the amplitude of the front-rear vibration of the vehicle can be obtained. Then, in order to obtain the time series data of the predicted acceleration Gpre of half a cycle of the front-rear vibration of the vehicle, a prediction period tp of 0.3 seconds or longer is required. Therefore, in the first embodiment, 0.3 seconds is used as an example of the prediction period tp. It should be noted that a prediction period tp longer than 0.3 seconds can be used.

[0048] Due to the above reasons, in the first embodiment, obtaining Figure 2 The target acceleration Greq, command torque Tpcm, and predicted acceleration Gpre in the vehicle driving torque control shown are time series data of the prediction period tp, as shown below with reference to Figure 3 Described in detail.

[0049] Next, Figure 31 is a flowchart showing an example of a process flow associated with vehicle drive torque control according to the first embodiment. The process of the flowchart is repeatedly executed at a specified control cycle (in other words, at each time step) during operation of the vehicle 100. Therefore, for each time step, the prediction period tp advances by one time step.

[0050] exist Figure 3 In the above process, first, in step S100 , the ECU 30 (processor 30 a ) calculates a target vehicle longitudinal acceleration (target acceleration) Greq for the prediction period tp based on the target torque Tpfrq. This step S100 corresponds to the processing of the target value conversion unit 36 ​​.

[0051] Figure 4 is a time chart for describing a method of predicting the target torque Tpfrq during the prediction period tp. It should be noted that Figure 4 The waveform of the target torque Tpfrq during acceleration is shown in FIG. The waveform of the target torque Tpfrq during deceleration is similar to that of the target torque Tpfrq in the positive and negative directions. Figure 4 The waveforms shown are opposite.

[0052] like Figure 4 As shown, the target torque Tpfrq, which is the basis for calculating the target acceleration Greq of the prediction period tp in step S100, is predicted to be constant at the current time value (at the current time step value). As described above, in the calculation of the target torque Tpfrq in the target torque calculation unit 34, the accelerator position, vehicle speed, transmission shaft speed and gear are used, and the accelerator position, vehicle speed, transmission shaft speed and gear are parameters indicating the operating state of the vehicle 100. By assuming that the target torque Tpfrq in the prediction period tp is constant and continuous, the target torque Tpfrq in the prediction period tp can be obtained without predicting parameters such as the accelerator position in the prediction period tp. That is, the technology does not need to predict the driver's accelerator pedal operation during the prediction period tp.

[0053] In step S100, based on the motion equation of the vehicle 100 as the object of the vehicle driving torque control, the target acceleration Greq may be calculated using the following equation (1).

[0054]

[0055] In the calculation of the target acceleration Greq using equation (1), the target torque Tpfrq is not used as it is, but the target torque Tpfrq_d after conversion as follows is used. The target value conversion unit 36 ​​is configured to include, for example, the following torque prediction model. The torque prediction model is configured to predict the response characteristics of the drive shaft torque Tp by simulating communication delay, dead time and first-order lag. The target torque Tpfrq_d is calculated by inputting the target torque Tpfrq at the current time into this torque prediction model. In addition, in equation (1), the final reduction ratio DR of the differential gear 26, the tire radius r of the wheel 102 and the vehicle weight VM are known. The driving resistance RL is calculated or detected separately. The value "9.81" in equation (1) is the acceleration of gravity.

[0056] Figure 5 It is shown by Figure 2 2. The target value conversion unit 36 ​​calculates the target vehicle longitudinal acceleration Greq for the prediction period tp as shown in FIG. 2. In step S100, by substituting the target torque Tpfrq_d converted as described above into equation (1), as shown in FIG. Figure 5 The waveform (i.e., time series data) of the target acceleration Greq for the calculation prediction period tp is shown. It should be noted that Figure 5 The waveform of the target acceleration Greq during acceleration is shown, and the waveform of the target acceleration Greq during deceleration is different from that of the target acceleration Greq in the positive and negative directions. Figure 5 The waveforms shown are opposite.

[0057] The processing of step S102 after step S100 corresponds to the processing of the driving force optimizer 40. In step S102, the ECU 30 randomly determines the search starting point of the command torque Tpcm in the prediction period tp using a random number. If the control cycle is, for example, 20 milliseconds, then the prediction period tp of 0.3 seconds will include 15 time steps after the current time step. Therefore, in this step S102, fifteen values ​​of the command torque Tpcm are determined as the search starting point of the command torque Tpcm in the prediction period tp. These values ​​of the command torque Tpcm are determined within the specified upper and lower limits. In addition, the upper limit is, for example, the value of the target torque Tpfrq that is regarded as a constant value in step S100.

[0058] The process of step S104 following step S102 corresponds to the process of the driving force optimizer 40 and the process of the acceleration prediction model 38. The latter process corresponds to an example of the "predicted acceleration calculation process" according to the present disclosure.

[0059] Specifically, in step S104, the ECU 30 (driving force optimizer 40) inputs the search starting point determined in step S102 or the search point corrected in the following step S110 to the acceleration prediction model 38. The ECU 30 (acceleration prediction model 38) generates (calculates) the time series data of the predicted acceleration Gpre in the prediction period tp according to the input search starting point or the corrected search point (i.e., according to the time series data of the command torque (corrected torque) Tpcm in the prediction period tp). The generated time series data of the predicted acceleration Gpre is input to the driving force optimizer 40.

[0060] The process of step S106 after step S104 corresponds to the process of driving force optimizer 40. In step S106, ECU 30 calculates the value of evaluation function J using target torque Tpfrq (constant value), target acceleration Greq (time series data) and predicted acceleration Gpre (time series data) in prediction period tp.

[0061] The evaluation function J is an evaluation function used to minimize the deviation of the predicted acceleration Gpre relative to the target acceleration Greq while reducing the deviation of the command torque Tpcm (search point) relative to the target torque Tpfrq (constant value) based on the operating state of the vehicle 100, and the evaluation function J is formulated as expressed by the following equation (2).

[0062]

[0063] In equation (2), “t” is time, and “w t ” and “w g " is the weight and is freely set. The right side of equation (2) is the square of the deviation of the command torque Tpcm(t) relative to the target torque Tpfrq (constant value) and w t The product of the predicted acceleration Gpre(t) and the target acceleration Greq(t) is equal to w g The definite integral of the sum of the products of / 2. Specify the integration interval of equation (2) from time point t0 to time point t h The period corresponds to the above prediction period tp. In addition, according to equation (2), by adjusting the weight w t and w g The size of can freely set the degree of influence of each of the above two deviations on the minimization of the evaluation function J.

[0064] The processing of steps S108 and S110 after step S106 corresponds to the processing of the driving force optimizer 40. The processing of these steps S108 and S110 is performed according to the algorithm of the particle swarm optimization method, which is one of the specific methods for optimization (minimization) of the driving force (i.e., command torque Tpcm) performed by the driving force optimizer 40. It should be noted that other optimization methods such as the shooting method can be used for the optimization.

[0065] In step S108, ECU 30 determines whether the convergence condition of evaluation function J to the optimal solution is satisfied. More specifically, ECU 30 determines whether the value of evaluation function J calculated in step S106 satisfies the reference value of optimality derived by the particle swarm optimization method (e.g., the absolute value of evaluation function J).

[0066] When the convergence condition is not satisfied in step S108, the process proceeds to step S110. In step S110, the ECU 30 corrects the search point (i.e., the time series data of the command torque Tpcm in the prediction period tp) according to the value of the evaluation function J calculated in step S106. More specifically, in the example using the particle swarm optimization method, the coordinates of the corrected search point are calculated according to the algorithm of the particle swarm optimization method.

[0067] After the search point is corrected in step S110, the process returns to step S104. When the convergence condition of step S108 is not satisfied, the ECU 30 (driving force optimizer 40) repeatedly executes the processes of steps S104, S106 and S110. As a result, the value of the evaluation function J gradually decreases.

[0068] On the other hand, when the convergence condition is satisfied in step S108, that is, when the command torque (correction torque) Tpcm that minimizes the evaluation function J is calculated, the process proceeds to step S112. In step S112, the ECU 30 (driving force optimizer 40) outputs (commands) the value of the command torque Tpcm of the next time step included in the time series data of the latest (that is, optimized) command torque Tpcm to the vehicle drive unit 10 as the final command torque Tpcmf.

[0069] In addition, the processes of steps S102 to S110 (however, excluding the process of the acceleration prediction model 38 included in step S104 ) correspond to an example of “command torque calculation processing” according to the present disclosure.

[0070] 1-3. Effect

[0071] As described above, the control structure of the vehicle driving torque according to the first embodiment is configured by combining the vehicle longitudinal acceleration prediction model 38 and the driving force optimizer 40 (see Figure 2). Specifically, according to the vehicle driving torque control, the command torque Tpcm that minimizes the evaluation function J is calculated. The evaluation function J minimizes the deviation (acceleration deviation) of the predicted acceleration Gpre relative to the target acceleration Greq. More specifically, as Figure 5 As shown, the waveform of the target acceleration Greq does not vibrate. Therefore, when the above-mentioned acceleration deviation included in the evaluation function J is reduced, the front-to-rear vibration of the vehicle (that is, the vibration of the front-to-rear acceleration G of the vehicle) is reduced. Therefore, according to the command torque Tpcm that minimizes the evaluation function J, the front-to-rear vibration of the vehicle can be satisfactorily reduced. Then, the predicted acceleration Gpre used to calculate the command torque Tpcm is calculated using the acceleration prediction model 38 as a machine learning model. Therefore, the number of adaptation values ​​used for vehicle drive torque control can be significantly reduced.

[0072] In addition, the evaluation function J is set to be able to calculate the command torque Tpcm, which can not only reduce the acceleration deviation of the predicted acceleration Gpre relative to the target vehicle front-rear acceleration Greq, but also reduce the deviation (torque deviation) of the command torque Tpcm relative to the target torque Tpfrq. Therefore, the command torque Tpcm (more specifically, the final command torque Tpcmf) can be determined while reducing the front-rear vibration of the vehicle so as to advantageously meet the target torque Tpfrq.

[0073] Therefore, according to the vehicle driving torque control of the first embodiment, it is possible to achieve reduction in vehicle front-rear vibration and vehicle acceleration and deceleration performance while reducing the number of adaptations of parameters related to the vehicle driving torque control. In addition, according to the evaluation function J (see equation (2)), by adjusting the weight w t and w g The size can be adjusted to reduce the front and rear vibration of the vehicle and ensure the balance between the vehicle's acceleration and deceleration performance.

[0074] Furthermore, in the vehicle driving torque control according to the first embodiment, for the data of the target torque Tpfrq, the command torque Tpcm, the target vehicle front-rear acceleration Greq, and the predicted acceleration Gpre for a plurality of time steps included in the prediction period tp which is a specified period from the current time step, the calculation of the predicted acceleration Gpre by the acceleration prediction model 38 (predicted acceleration calculation processing) and the calculation of the command torque Tpcm by the driving force optimizer 40 (command torque calculation processing) are performed for each time step. As a result, the command torque Tpcm can be calculated while evaluating the predicted acceleration Gpre by using the waveform (time series data) of the predicted acceleration Gpre.

[0075] In addition, the prediction period tp is set to the minimum period required to predict the front-to-back vibration of the vehicle caused by the torsion of the drive shaft 104 and the front-to-back vibration of the vehicle caused by the elimination of the backlash of the gears of the vehicle drive unit 10 (more specifically, a period corresponding to a half cycle of the front-to-back vibration of the vehicle (for example, 0.3 seconds)). Therefore, by setting the prediction period tp as short as possible, it is possible to calculate the command torque Tpcm (more specifically, the final command torque Tpcmf) that can achieve both the front-to-back vibration of the vehicle and the acceleration and deceleration performance of the vehicle while ensuring the accuracy of the prediction and reducing the calculation load of the prediction. It should be noted that the "prediction period" according to the present disclosure can be the minimum period required to predict "any one" of the vibration of the actual vehicle front-to-back acceleration caused by the torsion of the drive shaft and the vibration of the actual vehicle front-to-back acceleration caused by the elimination of the backlash of the gears located in the power transmission path from one or more power sources to the drive shaft.

[0076] In addition, in the present embodiment, the target torque Tpfrq of each time step in the plurality of time steps included in the prediction period tp is constant at the value of the target torque Tpfrq of the current time step (i.e., the starting point of the prediction period tp). Instead of such an example, the target torque Tpfrq of the prediction period tp may be calculated in any manner while predicting the parameters used to calculate the target torque Tpfrq of the prediction period tp (such as the driver's accelerator pedal operation in the prediction period tp). On the contrary, by predicting the target torque Tpfrq of the prediction period tp as a constant value as in the first embodiment, the target torque Tpfrq of the prediction period tp may be set without predicting parameters such as the driver's accelerator pedal operation. This also results in a reduction in the adaptation value for vehicle drive torque control.

[0077] 2. Second Embodiment

[0078] The second embodiment differs from the first embodiment in the control structure of the vehicle driving torque described below.

[0079] Figure 6 is a block diagram showing an outline of a control structure of a vehicle driving torque according to a second embodiment. Figure 6 The control structure shown is similar to that according to the first embodiment (see Figure 2 ) differs from the control structure in that an acceleration prediction model 50 is included instead of the acceleration prediction model 38, and a model correction unit 52 is also included.

[0080] Specifically, the acceleration prediction model 50 is the same as the acceleration prediction model 38 except that it includes a model relearning unit 50 a .

[0081] The model correction unit 52 (processor 30a) corrects the predicted acceleration Gpre output from the acceleration prediction model 50 based on the difference of the predicted acceleration Gpre relative to the actual vehicle front-rear acceleration (hereinafter referred to as "actual acceleration") Gact generated when the propeller shaft torque Tp according to the command torque Tpcm (more specifically, the final command torque Tpcmf) is output from the vehicle drive unit 10. After performing this correction of the predicted acceleration Gpre, the model correction unit 52 outputs the corrected predicted acceleration Gpre' to the driving force optimizer 40 instead of the predicted acceleration Gpre as the output of the acceleration prediction model 50. This correction corresponds to an example of "acceleration correction processing" according to the present disclosure. The actual acceleration Gact is detected by a vehicle acceleration sensor (G sensor) included in the sensor 32.

[0082] Figure 7 is shown with Figure 6 The process of the flowchart is repeatedly executed during the execution of the vehicle driving torque control (i.e., with Figure 3 It should be noted that the following description is given by taking acceleration as an example, but this is also applicable to the processing performed during deceleration.

[0083] exist Figure 7 In the embodiment of the present invention, first, in step S200, the ECU 30 (processor 30a) determines whether the difference (i.e., prediction error) of the predicted acceleration Gpre relative to the actual acceleration Gact is equal to or greater than a specified threshold value. More specifically, for example, the determination is performed using the waveforms (time series data) of the actual acceleration Gact and the predicted acceleration Gpre in the specified determination period td.

[0084] Figure 8 Is used to describe Figure 7 2 is a timing diagram of an example of a determination method of step S200 in FIG. Figure 8 1 shows a case where a difference occurs between the actual acceleration Gact (solid line) and the predicted acceleration Gpre (dashed line) when the vehicle 100 is accelerating. Figure 8, "Aa" and "Ap" are the amplitudes of the actual acceleration Gact and the predicted acceleration Gpre, respectively. In addition, "Fa" and "Fp" are the half-periods of the vibrations of the actual acceleration Gact and the predicted acceleration Gpre, respectively. As an example, the determination period td is the same as the above-mentioned prediction period tp used in the driving force optimizer 40. However, the determination period td can be a specified period that is longer than the prediction period tp. When the determination period td is set to be longer than the prediction period tp, in this step S200, the acceleration prediction model 50 can be used to calculate the time series data of the predicted acceleration Gpre of the determination period td to determine the prediction error.

[0085] In step S200, Figure 8 As shown, the ECU 30 calculates the area (integral value) of the region enclosed between the waveform of the actual acceleration Gact and the waveform of the predicted acceleration Gpre in the specified determination period td as the above-mentioned prediction error. Then, the ECU 30 determines whether the calculated prediction error is equal to or greater than a threshold value. The threshold value is preset to, for example, a value equivalent to the minimum value of acceleration that a person in the vehicle 100 can perceive.

[0086] When the predicted error calculated in step S200 is smaller than the threshold value (ie, when the predicted error is smaller than the acceleration that humans can perceive), the ECU 30 ends the current processing cycle. On the other hand, when the calculated predicted error is equal to or larger than the threshold value, the process proceeds to step S202.

[0087] In step S202, the ECU 30 determines whether vibration occurs in the waveform of the predicted acceleration Gpre. For example, the amplitude Ap of the waveform of the predicted acceleration Gpre after the start of acceleration (see, for example, Figure 8 ) is equal to or greater than a specified threshold to make this determination.

[0088] When the determination result of step S202 is positive (i.e., when Figure 8 When vibration is generated in the example of the waveform of the predicted acceleration Gpre shown in FIG. 2 ), the process proceeds to step S204. In step S204, the ECU 30 obtains the half cycles Fp and Fa of each waveform and the amplitudes Ap and Aa in the waveforms of the predicted acceleration Gpre and the actual acceleration Gact. In addition, the amplitude Ap is the same as the amplitude obtained in step S202.

[0089] In step S206 following step S204, the ECU 30 determines whether there is an error in the half cycle Fp of the predicted acceleration Gpre relative to the half cycle Fa of the actual acceleration Gact (i.e., a time delay in the waveform of the predicted acceleration Gpre relative to the waveform of the actual acceleration Gact). This determination can be made based on, for example, whether the absolute value of the difference between the half cycle Fa and the half cycle Fp is equal to or greater than a specified threshold. As a result, when there is an error in the half cycle Fp, the process proceeds to step S208. On the other hand, when there is no error in the half cycle Fp, the process proceeds to step S210.

[0090] In step S208, the ECU 30 calculates a correction gain Gf for correcting the time delay of the waveform of the predicted acceleration Gpre relative to the waveform of the actual acceleration Gact, and corrects the time delay of the predicted acceleration Gpre. The correction gain Gf is calculated based on the half-cycle Fa. For example, the ECU 30 uses the values ​​of the half-cycle Fa and Fp obtained in step S204 to calculate a value obtained by dividing the half-cycle Fa by the half-cycle Fp as the correction gain Gf.

[0091] Fig. 9A and Fig. 9B Is used to describe Figure 6 The diagram shown is an example of a method in which the model correction unit 52 corrects the time delay and amplitude of the predicted acceleration Gpre. Fig. 9A The waveform of the predicted acceleration Gpre which is the output of the acceleration prediction model 50 (ie, the waveform before correction by the model correction unit 52) ​​is shown. Fig. 9B 2 shows the waveform of the predicted acceleration Gpre' after correction by the model correction unit 52. Fig. 9B In the example shown, both time delay and amplitude are corrected.

[0092] In step S208, the ECU 30 calculates the correction gain Gf as described above, and then corrects the time delay of the waveform of the predicted acceleration Gpre by reflecting the calculated correction gain Gf to the waveform (time series data) of the predicted acceleration Gpre before correction. Specifically, the correction gain Gf is multiplied by the predicted acceleration Gpre before correction and output from the acceleration prediction model 50. As a result, the half period Fp of the waveform of the predicted acceleration Gpre as the output of the acceleration prediction model 50 is corrected to Fig. 9B After the correction gain Gf is calculated, the correction of the time delay using this correction gain Gf is continuously reflected to the predicted acceleration Gpre output from the acceleration prediction model 50.

[0093] In step S210 following step S208, the ECU 30 determines whether there is an error in the amplitude Ap of the predicted acceleration Gpre relative to the amplitude Aa of the actual acceleration Gact. This determination can be made based on, for example, whether the absolute value of the difference between the amplitude Aa and the amplitude Ap is equal to or greater than a specified threshold. As a result, when there is an error in the amplitude Ap, the process proceeds to step S212. On the other hand, when there is no error in the amplitude Ap, the ECU 30 ends the current processing cycle.

[0094] In step S212, the ECU 30 calculates a correction gain Ga for correcting the error of the amplitude Ap of the waveform of the predicted acceleration Gpre relative to the waveform of the actual acceleration Gact, and then corrects the error of the amplitude Ap of the predicted acceleration Gpre. The correction gain Ga is calculated based on the amplitude Aa. For example, the ECU 30 uses the values ​​of the amplitudes Aa and Ap acquired in step S204 to calculate a value obtained by dividing the amplitude Aa by the amplitude Ap as the correction gain Ga.

[0095] In step S212, the ECU 30 calculates the correction gain Ga as described above, and then corrects the amplitude Ap by reflecting the calculated correction gain Ga to the waveform (time series data) of the predicted acceleration Gpre before correction. Specifically, the correction gain Ga is multiplied by the predicted acceleration Gpre before correction and output from the acceleration prediction model 50. As a result, the amplitude Ap of the waveform of the predicted acceleration Gpre as the output of the acceleration prediction model 50 is corrected to Fig. 9B After the correction gain Ga is calculated, the correction of the amplitude Ap using this correction gain Ga is continuously reflected to the predicted acceleration Gpre output from the acceleration prediction model 50.

[0096] in addition, Figure 7 The processing of steps S200 to S212 in corresponds to an example of “acceleration correction processing” according to the present disclosure.

[0097] Then, Fig.10 is shown in Figure 6 The model relearning unit 50a shown in FIG. 1 is a time chart showing an example of waveforms of the predicted acceleration Gpre and the actual acceleration Gact in the case where the predicted acceleration Gpre is relearned. Fig.10 In the waveform of the predicted acceleration Gpre in the example shown, it is similar to Figure 8The waveform of the predicted acceleration Gpre shown in FIG. 2 is a waveform of the predicted acceleration Gpre, in which the error (i.e., the prediction error) of the predicted acceleration Gpre relative to the actual acceleration Gact in the determination period td is equal to or greater than the threshold value (see step S200). However, no vibration is generated in the waveform of the predicted acceleration Gpre. When no vibration is generated in this way, the correction of the model correction unit 52 cannot be performed. As a result, the determination result of step S202 becomes negative, and the processing proceeds to step S214.

[0098] In step S214, the ECU 30 acquires learning data for relearning the acceleration prediction model 50 at the time of acceleration (and deceleration) that comes after the program proceeds to step S214. Specifically, the learning data is a combination of "time series data of command torque Tpcm and one or more variables x (which are explanatory variables (i.e., inputs))" and "time series data of vehicle longitudinal acceleration G (more specifically, actual vehicle longitudinal acceleration Gact) (which is a target variable (i.e., output)")", similar to the data used in the learning performed in advance.

[0099] In step S216 following step S214, the ECU 30 determines whether or not acquisition of a designated number of learning data sufficient to relearn the acceleration prediction model 50 has been completed. As a result, when acquisition of the designated number of learning data has been completed, the process proceeds to step S218.

[0100] In step S218, the ECU 30 uses the acquired learning data to perform re-learning of the acceleration prediction model 50. Specifically, in step S218, the ECU 30 creates a copy of the acceleration prediction model 50 before re-learning, and then stores the copy in the storage device 30b. Then, the ECU 30 feeds the learning data to the copied acceleration prediction model (referred to as "acceleration prediction model C" for convenience of explanation), and then performs learning of the acceleration prediction model C. It should be noted that the learning of the acceleration prediction model C can be performed using the learning data acquired in step S214 and the old learning data previously used for learning, or can be performed without using the old learning data.

[0101] In step S220 after step S218, ECU 30 determines whether the calculation accuracy of the predicted acceleration Gpre of the acceleration prediction model C is higher than the calculation accuracy of the current acceleration prediction model 50. Specifically, the determination can be performed, for example, as follows. That is, at the time of acceleration of the vehicle 100 that comes after the learning of the acceleration prediction model C is completed, ECU 30 calculates the prediction error of the predicted acceleration Gpre using each of the two acceleration prediction models 50 and C through the same processing as in step S200. Then, ECU 30 determines whether the prediction error of the acceleration prediction model C is less than the prediction error of the current acceleration prediction model 50.

[0102] When the prediction accuracy of the acceleration prediction model C is better than the prediction accuracy of the current acceleration prediction model 50 in step S220, the process proceeds to step S222. In step S222, the ECU 30 deletes the current acceleration prediction model 50 and then updates the acceleration prediction model 50 by the acceleration prediction model C. On the other hand, when the prediction accuracy of the current acceleration prediction model 50 is better than the prediction accuracy of the acceleration prediction model C in step S220, the ECU 30 ends the current processing cycle while deleting the acceleration prediction model C (i.e., without updating the acceleration prediction model 50).

[0103] As described above, the control structure of the vehicle driving torque according to the second embodiment also includes the model correction unit 52. As a result, for example, when the accuracy of the predicted acceleration Gpre is not good due to individual differences of the vehicles 100 on which the acceleration prediction model 50 is installed, the prediction accuracy can be ensured by using the model correction unit 52.

[0104] Furthermore, the acceleration prediction model 50 according to the second embodiment has the model relearning unit 50a. Therefore, even when the correction of the model correction unit 52 cannot be performed because no vibration is generated in the waveform of the predicted acceleration Gpre, the prediction accuracy can be ensured by using the model relearning unit 50a.

[0105] In addition, when Figure 7 When the prediction error of the predicted acceleration Gpre obtained continuously for a specified number of times in the process of step S200 shown is equal to or greater than the specified threshold value, the ECU 30 can proceed to the process of step S202. As a result, when continuous prediction errors are generated in the predicted acceleration Gpre, the ECU 30 can ensure the prediction accuracy by using the model correction unit 52 or the model relearning unit 50a.

Claims

1. A control device for controlling a vehicle drive unit based on a running state of the vehicle, the vehicle drive unit having one or more power sources, the control device comprising: processor; and a storage device configured to store a vehicle fore-and-aft acceleration prediction model, the vehicle fore-and-aft acceleration prediction model being a machine learning model that receives a command torque as an input and outputs a predicted acceleration, wherein the command torque is a vehicle drive torque commanded to the vehicle drive unit, and the predicted acceleration is a predicted value of the vehicle fore-and-aft acceleration, wherein The processor is configured to: executing a predicted acceleration calculation process to calculate the predicted acceleration using the vehicle longitudinal acceleration prediction model; and A command torque calculation process is performed to calculate the command torque that minimizes an evaluation function, wherein minimizing the evaluation function means minimizing a deviation of the predicted acceleration relative to a target vehicle front and rear acceleration according to the target torque while reducing a deviation of the command torque relative to a target torque, wherein the target torque is a target value of the vehicle driving torque based on the operating state.

2. The control device according to claim 1, wherein The processor is configured to further perform acceleration correction processing to correct the predicted acceleration output from the vehicle fore-and-aft acceleration prediction model based on a difference between the predicted acceleration and an actual vehicle fore-and-aft acceleration generated when a vehicle drive torque according to the command torque is output from the vehicle drive unit.

3. The control device according to claim 1 or 2, wherein The processor is configured to perform the predicted acceleration calculation processing and the command torque calculation processing for each time step for data of the target torque, the command torque, the target vehicle front and rear acceleration, and the predicted acceleration for multiple time steps included in a prediction period, wherein the prediction period is a specified period from a current time step.

4. The control device according to claim 3, wherein The prediction period is a minimum period for predicting at least one of the vibration of the actual vehicle fore-and-aft acceleration caused by the torsion of the drive shaft of the vehicle and the vibration of the actual vehicle fore-and-aft acceleration caused by the elimination of the backlash of gears located in the power transmission path from the one or more power sources to the drive shaft.

5. The control device according to claim 4, wherein The processor is configured to keep the target torque at each time step of the plurality of time steps included in the prediction period constant at a value of the target torque at the current time step.

6. A method for controlling a vehicle drive unit based on a running state of the vehicle, the vehicle drive unit having one or more power sources, the control method comprising: a predicted acceleration calculation process for calculating a predicted acceleration using a vehicle fore-and-aft acceleration prediction model, the predicted acceleration being a predicted value of the vehicle fore-and-aft acceleration, the vehicle fore-and-aft acceleration prediction model being a machine learning model that receives as input a command torque and outputs the predicted acceleration, wherein the command torque is a vehicle drive torque commanded to the vehicle drive unit; and A command torque calculation process is used to calculate the command torque that minimizes an evaluation function, wherein minimizing the evaluation function means minimizing a deviation of the predicted acceleration from a target vehicle front and rear acceleration based on the target torque while reducing a deviation of the command torque from a target torque, wherein the target torque is a target value of the vehicle driving torque based on the operating state.

7. The control method according to claim 6 further includes an acceleration correction process for correcting the predicted acceleration output from the vehicle fore-and-aft acceleration prediction model based on a difference between the predicted acceleration and an actual vehicle fore-and-aft acceleration generated when a vehicle driving torque according to the command torque is output from the vehicle drive unit.

8. The control method according to claim 6 or 7, further comprising: The predicted acceleration calculation process and the command torque calculation process are performed for each time step for the data of the target torque, the command torque, the target vehicle longitudinal acceleration and the predicted acceleration for multiple time steps included in a prediction period, and the prediction period is a specified period from the current time step.

9. The control method according to claim 8, wherein The prediction period is a minimum period for predicting at least one of the vibration of the actual vehicle fore-and-aft acceleration caused by the torsion of the drive shaft of the vehicle and the vibration of the actual vehicle fore-and-aft acceleration caused by the elimination of the backlash of gears located in the power transmission path from the one or more power sources to the drive shaft.

10. The control method according to claim 9, further comprising: The target torque at each time step of the plurality of time steps included in the prediction period is made constant at a value of the target torque at the current time step.

Images