Vehicle control device

Abstract

To provide a control device of a vehicle which can suppress vibration due to torsional resonance of a drive shaft.SOLUTION: A control device 10 includes: a processor 11; and a storage device 12 which stores a learned model learned by machine learning. A processor 11 enforces vibration control for suppressing vibration due to torsional resonance of a drive shaft 50 when a vehicle travels on a wavelike passage. The learned model is a model for predicting a ground load of a driving wheel 52 after a predetermined time. In the vibration control, the processor 11 determines whether speed increase correction is enforced or speed reduction correction is enforced using a prediction result by the learned model and switches the speed increase correction and the speed reduction correction, and thereby rotational speed of a motor 30 is increased when the driving wheel 52 is slipped and the rotational speed of the motor 30 is reduced when the driving wheel 52 is gripped.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] This invention relates to a vehicle control device.

Background Art

[0002] Patent Document 1 discloses that in an electric vehicle, vibration of the vehicle body occurs due to torsional resonance of the drive shaft. Therefore, in an electric vehicle, vibration suppression control for suppressing vehicle body vibration is implemented. In the vibration suppression control, the torque of the motor is controlled so as to suppress the vibration caused by torsional resonance. Patent Document 1 discloses adapting the control gain of the vibration suppression control based on the data of the rotational speed of the motor acquired in a state where the wheels of the electric vehicle are grounded.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] When the vehicle travels on a wavy road, the ground load of each wheel may fluctuate and the wheels may repeat slipping and gripping. Therefore, in the vibration suppression control adapted on the premise that the wheels are grounded, the vibration caused by torsional resonance cannot be effectively suppressed.

Means for Solving the Problems

[0005] Hereinafter, means for solving the above problems and their operational effects will be described. The vehicle control device for solving the above problems is applied to an electric vehicle that drives drive wheels with a motor. This control device includes a processor and a storage device in which a learned model learned by machine learning is stored. In this control device, when traveling on a wavy road, the processor performs vibration suppression control to suppress vibration caused by torsional resonance of the drive shaft. The learned model is a model that predicts the ground load of the drive wheels after a predetermined time by inputting time-series data for a predetermined period of explanatory variables including the rotational speed of the motor, the torque generated by the motor, and the rotational speed of the drive wheels and performing calculations. In the vibration suppression control, the processor determines whether to perform speed increase correction to generate positive torque in the motor or speed decrease correction to generate negative torque in the motor using the prediction result by the learned model. Then, the processing circuit increases the rotational speed of the motor when the drive wheels are slipping and decreases the rotational speed of the motor when the drive wheels are gripping by switching between the speed increase correction and the speed decrease correction.

[0006] Also, in one aspect of the vehicle control device, the learned model is a model that predicts whether the drive wheels are slipping or gripping after a predetermined time by inputting time-series data for a predetermined period of explanatory variables including the rotational speed of the motor, the torque generated by the motor, and the rotational speed of the drive wheels and performing calculations.

Advantages of the Invention

[0007] This control device can suppress vibration caused by torsional resonance by switching between speed increase correction and speed decrease correction at appropriate timing using the prediction result.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Mode for Carrying Out the Invention

[0009] Hereinafter, a control device 10 which is an embodiment of a vehicle control device will be described with reference to FIGS. 1 to 7. <Configuration of Vehicle> As shown in FIG. 1, a vehicle equipped with the control device 10 is an electric vehicle having a motor 30 as a driving power source. The rotating shaft of the motor 30 is connected to the left and right drive wheels 52 via a differential 51 and a drive shaft 50.

[0010] The motor 30 is connected to an inverter 20. The inverter 20 is connected to a battery 40. Further, the inverter 20 is connected to the control device 10. The control device 10 includes a processor 11 and a storage device 12. The processor 11 includes a CPU and memories such as a RAM and a ROM, and the memories store program codes or instructions configured to cause the CPU to execute processes. The processor 11 generates a signal for controlling the inverter 20. The inverter 20 converts the direct current supplied from the battery 40 into an alternating current based on the signal received from the control device 10 and adjusts the current supplied to the motor 30.

[0011] The motor 30 is, for example, a three-phase AC motor, and generates a driving force by the alternating current supplied from the inverter 20. The driving force generated by the motor 30 is transmitted to each of the left and right drive wheels 52 via the differential 51 and the drive shaft 50. Also, when the motor 30 is rotated by being carried around by the drive wheels 52, it generates a regenerative braking force. Thereby, the kinetic energy of the vehicle is converted into electrical energy and charged to the battery 40.

[0012] An accelerator position sensor 100, a first wheel speed sensor 101, and a second wheel speed sensor 102 are connected to the control device 10. Also, an acceleration sensor 103, a current sensor 104, a rotation sensor 105, and a torque sensor 106 are connected to the control device 10.

[0013] The accelerator position sensor 100 detects the accelerator operation amount ACCP. The first wheel speed sensor 101 detects the rotational speed ωt1 of the right drive wheel 52. The second wheel speed sensor 102 detects the rotational speed ωt2 of the left drive wheel 52. The control device 10 calculates the average value of the rotational speed ωt1 and the rotational speed ωt2 as the rotational speed ωt. Also, the control device 10 calculates the vehicle speed V, which is the speed of the vehicle, based on the rotational speed ωt.

[0014] The acceleration sensor 103 detects the acceleration in the longitudinal, lateral, and vertical directions of the vehicle and the inclination of the vehicle. The current sensor 104 detects the current flowing through the motor 30. The rotation sensor 105 detects the rotational speed ωmg of the motor 30. The torque sensor 106 detects the torque generated by the motor 30.

[0015] <Regarding the control device 10> For example, the processor 11 calculates a target torque value Tm based on the accelerator operation amount ACCP and the vehicle speed V. Then, the processor 11 calculates a target value of the current value to be supplied to the motor 30 based on the target value Tm and the rotational speed ωmg. The control device 10 drives the inverter 20 so as to realize the calculated target value of the current value. In this way, the control device 10 controls the motor 30 so that the torque of the motor 30 approaches the target value Tm.

[0016] <Regarding torsional resonance> When the vehicle is traveling on a wavy road, the ground load of the drive wheel 52 may fluctuate, and the drive wheel 52 may repeat slipping and gripping. As a result, torsion may repeatedly occur in the drive shaft 50, and torsional resonance may occur.

[0017] This torsional resonance will be described with reference to FIG. 2. When the drive wheel 52 slips, the rotational speed ωt of the drive wheel 52 increases. Then, as shown by the solid arrow in FIG. 2, the rotational speed ωt of the drive wheel 52 becomes larger than the rotational speed ωmg of the motor 30. Due to the difference between this rotational speed ωt and the rotational speed ωmg, torsion occurs in the drive shaft 50.

[0018] As shown by the broken arrow in FIG. 2, as the torsion is eliminated, the rotational speed ωmg of the motor 30 rises with a delay. However, when the drive wheel 52 grips, the rotational speed ωt of the drive wheel 52 decreases this time. Then, as shown by the broken arrow in FIG. 2, the rotational speed ωt of the drive wheel 52 becomes smaller than the rotational speed ωmg of the motor 30. Due to the difference between this rotational speed ωt and the rotational speed ωmg, torsion in the opposite direction to when slipping occurred occurs in the drive shaft 50.

[0019] When traveling on a wavy road and the drive wheel 52 repeats slipping and gripping, such torsion may repeatedly occur, and torsional resonance may occur. When torsional resonance occurs, the amplitude increases, and a large load is applied to the drive shaft 50. Also, the vibration of the vehicle increases.

[0020] Therefore, when the vehicle is traveling on a wavy road, the control device 10 executes torque-down control and vibration damping control. The torque-down control is control to reduce the torque of the motor 30 compared to when the vehicle is not traveling on a wavy road. And the vibration damping control is control to increase the rotational speed ωmg of the motor 30 when the drive wheels 52 are slipping and to decrease the rotational speed ωmg of the motor 30 when the drive wheels 52 are gripping by switching between speed increase correction and speed decrease correction.

[0021] <Regarding Torque-Down Control> FIG. 3 is a flowchart showing the flow of processing in a routine related to wavy road determination and torque-down control. This routine is repeatedly executed by the processor 11 of the control device 10 while the vehicle is traveling.

[0022] When this routine is started, in the process of step S100, the processor 11 calculates the change amount Δωmg of the rotational speed ωmg of the motor 30. Specifically, the processor 11 calculates the difference obtained by subtracting the rotational speed ωmg when this routine was last executed from the rotational speed ωmg when this routine is being executed this time. The calculated difference is the change amount Δωmg.

[0023] In the process of the next step S110, the processor 11 calculates an integrated value ΣΔωmg obtained by integrating the change amount Δωmg calculated during the most recent regular period. Specifically, the absolute value of the change amount Δωmg calculated through the process of step S100 is added to the integrated value ΣΔωmg calculated when this routine was last executed. Then, the change amount Δωmg added to the integrated value ΣΔωmg when this routine was executed a predetermined period before is subtracted from the calculated sum. The difference thus calculated is the new integrated value ΣΔωmg. That is, the processor 11 calculates the integrated value ΣΔωmg by integrating the absolute value of the change amount Δωmg calculated during the most recent regular period by repeatedly executing this routine. When traveling on a wavy road, as described with reference to FIG. 2, the rotational speed ωmg of the motor 30 repeats increases and decreases. Therefore, the integrated value ΣΔωmg increases. That is, the integrated value ΣΔωmg is a value that becomes an index value for determining whether the vehicle is traveling on a wavy road.

[0024] In the process of the next step S120, the processor 11 determines whether the integrated value ΣΔωmg is equal to or greater than a first threshold value. The first threshold value is a threshold value for determining that the vehicle is traveling on a wavy road based on the integrated value ΣΔωmg being equal to or greater than the first threshold value. The magnitude of the first threshold value is set to a magnitude that can appropriately determine that the vehicle is traveling on a wavy road based on the results of experiments performed in advance.

[0025] In the process of step S120, if it is determined that the integrated value ΣΔωmg is equal to or greater than the first threshold value (step S120: YES), the processor 11 advances the process to step S130.

[0026] In the process of step S130, the processor 11 sets the flag Fb to "1" and performs torque-down control. The flag Fb has an initial value of "0" and is a flag that is set to either "0" or "1". The flag Fb indicates that the vehicle is traveling on a wavy road when it is "1". The flag Fb indicates that the vehicle is not traveling on a wavy road when it is "0". The processor 11 performs torque-down control when the flag Fb is "1".

[0027] Torque-down control is control that reduces the torque of the motor 30 compared to when not driving on a wavy road. Specifically, in torque-down control, the processor 11 sets an upper limit value for the target torque value Tm. And the upper limit value is a value smaller than the target value Tm when torque-down control is not being executed. The processor 11 reduces the torque of the motor 30 compared to when not driving on a wavy road by limiting the target value Tm to the upper limit value. When the process of step S130 is thus executed, the processor 11 temporarily ends this routine.

[0028] In the process of step S120, when it is determined that the integrated value ΣΔωmg is less than the first threshold (step S120: NO), the processor 11 advances the process to step S140.

[0029] In the process of step S140, the processor 11 determines whether the flag Fb is "1". In the process of step S140, when it is not determined that the flag Fb is "1" (step S140: NO), the processor 11 ends this routine as it is. That is, when not driving on a wavy road, the processor 11 ends this routine without performing torque-down control.

[0030] On the other hand, in the process of step S140, when it is determined that the flag Fb is "1" (step S140: YES), the processor 11 advances the process to step S150. And in the process of step S150, the processor 11 determines whether the integrated value ΣΔωmg is less than or equal to the second threshold. The second threshold is a threshold for determining that the vehicle is no longer driving on a wavy road based on the integrated value ΣΔωmg being less than or equal to the second threshold. The second threshold is a value smaller than the first threshold. The magnitude of the second threshold is set to a magnitude that can appropriately determine that the vehicle is not driving on a wavy road based on the results of experiments performed in advance.

[0031] In the process of step S150, when it is determined that the integrated value ΣΔωmg is greater than the second threshold value (step S150: NO), the processor 11 ends this routine as it is. In this case, since the flag Fb remains "1", the processor 11 continues to perform torque-down control.

[0032] On the other hand, in the process of step S150, when it is determined that the integrated value ΣΔωmg is less than or equal to the second threshold value (step S150: YES), the processor 11 advances the process to step S160. In the process of step S160, the processor 11 sets the flag Fb to "0" and releases the torque-down control. Then, the processor 11 temporarily ends this routine. In this way, when the integrated value ΣΔωmg becomes less than or equal to the second threshold value, the control device 10 determines that the wavy road running has ended and ends the implementation of the torque-down control.

[0033] <Regarding Vibration Control> Next, vibration control will be described with reference to FIGS. 4 to 7. FIG. 4 is a flowchart showing the flow of the process of a routine related to vibration control. This routine is repeatedly executed by the processor 11 when the flag Fb is set to "1".

[0034] As shown in FIG. 4, when starting this routine, the processor 11 performs a determination process in the process of step S200. The determination process is a process of determining whether it is a period for performing acceleration correction or a period for performing deceleration correction.

[0035] In the control device 10, in the determination process, the processor 11 predicts the ground load of the drive wheel 52 using the learned model. Then, based on the predicted ground load, the processor 11 predicts whether the drive wheel 52 is slipping or gripping after a predetermined time. Further, based on this prediction result, the processor 11 determines whether it is an acceleration correction period for performing acceleration correction or a deceleration correction period for performing deceleration correction. Note that the predetermined time is several milliseconds.

[0036] The storage device 12 of the control device 10 stores data of a learned model that predicts the ground load of the drive wheels 52. In the control device 10, as the learned model, a long short-term memory neural network that can handle time-series data while retaining information on the transition along the time axis is used. The long short-term memory neural network is a so-called LSTM neural network. The LSTM neural network is a type of recurrent neural network.

[0037] Fig. 5 shows the neural network that constitutes the LSTM neural network. In Fig. 5, the notation of the transmission paths connecting the nodes of adjacent layers is omitted. As shown at the left end of Fig. 5, this neural network includes an input layer having a plurality of nodes. The number of nodes in the input layer is equal to the number of explanatory variables that make up the input data X. The content of the input data X will be described later.

[0038] As shown at the right end of Fig. 5, this neural network includes an output layer consisting of one node. This output layer outputs the ground load y. And this neural network includes a hidden layer consisting of a plurality of layers with fewer nodes than the input layer between the input layer and the output layer.

[0039] The activation function in the hidden layer of this neural network is, for example, the hyperbolic tangent. Also, the number of layers in the hidden layer and the number of nodes in each layer of the hidden layer are hyperparameters that are set after being adjusted in the design stage so that the ground load y can be appropriately estimated.

[0040] In this neural network, by inputting the explanatory variables that make up the input data X, which is time-series data of the explanatory variables, into the input layer, the sum of the values multiplied by the weights corresponding to each transmission path is input into the activation function. Then, the output value of the activation function is input into the next layer. Such operations are repeated, and finally, the ground load y is output from the output layer.

[0041] The control device 10 uses the torque of the motor 30, the rotational speed ωmg of the motor 30, the acceleration of the vehicle in the longitudinal direction, and the rotational speed ωt of the drive wheel 52 as explanatory variables for predicting the ground load y of the drive wheel 52. The processor 11 creates the input data X from the data acquired during a predetermined period until this routine is started in order to calculate the ground load y of the drive wheel 52 after a predetermined time. That is, the processor 11 sets all the values of the explanatory variables acquired during the predetermined period as the input data X. Note that the length of the predetermined period is, for example, several tens of milliseconds. When the explanatory variables are acquired 10 times during the predetermined period, the input data X is a set of 10 consecutive explanatory variables. Specifically, the input data X is a set from the collected data X(1) consisting of the explanatory variable collected first in the predetermined period to the collected data X(10) consisting of the explanatory variable collected last in the predetermined period.

[0042] Each explanatory variable contains four types of information as described above. Therefore, each collected data is a four-dimensional vector consisting of these four values. Therefore, in this case, the input layer of the neural network shown in FIG. 5 has four nodes.

[0043] FIG. 5 schematically shows the configuration of a recurrent neural network. Note that the arrow extending in the vertical direction in FIG. 6 indicates the forward propagation direction of the neural network shown in FIG. 5 to which the explanatory variable is input. Note that "n" in FIG. 6 indicates the chronological order of the explanatory variables in the input data X.

[0044] The neural network into which the collected data X(n) shown at the right end in FIG. 6 is input takes as input a four-dimensional vector that is the collected data X(n) and outputs the ground load y. That is, in this case, it is the neural network into which the collected data X(10) is input. This neural network is a fully connected neural network that propagates forward to the output layer that outputs the ground load y.

[0045] As shown in FIG. 6, the output of the hidden layer of the neural network into which the collection data X(10) collected last in a predetermined period is input reflects the output of the hidden layer of the neural network into which the collection data X(9) collected at the previous timing is input.

[0046] As shown in FIG. 6, the output of the hidden layer of the neural network into which the collection data X(8) collected at one more previous timing in a predetermined period is input is reflected in the hidden layer of the neural network into which the collection data X(9) is input. In this way, the output of the hidden layer in the neural network into which the collection data collected at the previous timing is input is reflected in each neural network into which each collection data is input.

[0047] The LSTM neural network is a neural network that provides a mechanism called an LSTM block in each hidden layer of such a recurrent neural network so that the propagation of time-series information can be adjusted.

[0048] Note that the learned model stored in the storage device 12 has been pre-trained with teacher supervision using training data X_tr including information on the measured data obtained by detecting the ground load of the drive wheel 52 with a sensor, which was created from the results of running experiments performed in advance. Note that, to collect the data for creating the training data X_tr, an experimental vehicle equipped with a sensor for measuring the ground load of the drive wheel 52 is used. Then, while repeating a running experiment in various running states including running on a wavy road with this experimental vehicle, a large amount of data is collected.

[0049] Learning for updating the weights of the neural network is performed using the large amount of data collected in this way. The computer that performs the learning generates training data X_tr based on the collected data and performs the learning.

[0050] Note that the training data X_tr is data that includes the measured ground contact load data in the set of the above-mentioned 10 collected data. The measured ground contact load data included in the training data X_tr is the measured ground contact load data at a given time after the end of a predetermined period during which the collected data included in the same training data X_tr was collected. The computer generates one training data X_tr by combining the measured ground contact load as the correct label with the collected data.

[0051] When the computer generates a large number of training data X_tr from the collected large amount of data, the computer inputs the data X(1) to X(10) corresponding to the input data X in the training data X_tr into the LSTM neural network to calculate the ground contact load y.

[0052] Then, the computer performs learning. Specifically, the computer adjusts the weights in the neural network so that the error between the calculated ground contact load y and the measured ground contact load, which is the correct label in the training data X_tr used for the calculation, becomes small.

[0053] Then, the computer repeats the calculation of the ground contact load y and the adjustment of the weights using a large number of training data X_tr. When the error of the ground contact load y calculated using the LSTM neural network becomes sufficiently small, the computer determines that the learning is completed. Then, the computer stores the data of the learned LSTM neural network in the storage device 12 as the data of the learned model.

[0054] The storage device 12 of the control device 10 stores the data of the learned LSTM neural network whose weights have been adjusted in this way. The processor 11 of the control device 10 repeatedly performs a process of inputting the input data X into the learned model and calculating the ground contact load y at a given time after the vehicle is running.

[0055] Therefore, as shown in FIG. 7, the control device 10 has information on the transition of the ground load y until after a predetermined time. In the determination process of step S200, the processor 11 determines whether it is an acceleration correction period for performing acceleration correction or a deceleration correction period for performing deceleration correction based on the information on the transition of the ground load y until after this predetermined time.

[0056] Specifically, the processor 11 determines that the period during which the ground load y is equal to or less than the third threshold value is a slip period in which the drive wheel 52 is slipping. And the processor 11 determines that the period during which the ground load y is greater than the third threshold value is a grip period in which the drive wheel 52 is gripping.

[0057] The processor 11 sets the acceleration correction period to a time slightly earlier than the slip period so that positive torque can be generated in the motor 30 from the start point of the slip period. Note that the amount by which the start time of the acceleration correction period is advanced from the start point of the slip period is determined based on the results of experiments performed in advance. For example, the amount by which the start time of the acceleration correction period is advanced from the start point of the slip period is determined based on the length of the period from when a signal is transmitted from the control device 10 until torque is actually generated in the motor 30.

[0058] Similarly, the processor 11 sets the deceleration correction period to a time slightly earlier than the grip period so that negative torque can be generated in the motor 30 from the start point of the grip period. Thus, as shown in FIG. 7, the acceleration correction period and the deceleration correction period are set to appear alternately.

[0059] In the determination process of step S200, the processor 11 determines whether the current point in time when this routine is being executed corresponds to the acceleration correction period or the deceleration correction period. Then, as shown in FIG. 4, in the process of step S210, the processor 11 determines whether the result determined through such determination processing is an acceleration correction period. If it is determined in the process of step S210 that it is an acceleration correction period (step S210: YES), the processor 11 advances the process to step S220. Then, in the process of step S220, the processor 11 performs acceleration correction. Specifically, the processor 11 corrects the target value of the current value so as to output a positive torque greater than the target value Tm to the motor 30 and controls the inverter 20. When the acceleration correction is thus performed, the processor 11 temporarily ends this routine.

[0060] On the other hand, if it is not determined in the process of step S210 that it is an acceleration correction period (step S210: NO), the processor 11 advances the process to step S230. Then, in the process of step S230, the processor 11 performs deceleration correction. Specifically, the processor 11 controls the inverter 20 so as to output a negative torque to the motor 30. When the deceleration correction is thus performed, the processor 11 temporarily ends this routine.

[0061] In this way, when the control device 10 is traveling on a wavy road, it performs vibration control together with torque-down control. <Operation of this Embodiment> When the drive wheel 52 slips, as indicated by the solid arrow in FIG. 2, the rotational speed ωt of the drive wheel 52 increases and becomes higher than the rotational speed ωmg of the motor 30. As a result, torsion occurs in the drive shaft 50. Then, when the drive wheel 52 grips after slipping, as indicated by the broken arrow in FIG. 2, the rotational speed ωt of the drive wheel 52 decreases and becomes lower than the rotational speed ωmg of the motor 30. As a result, torsion occurs in the drive shaft 50. By such repeated occurrence of torsion alternately, torsional resonance occurs.

[0062] In the control device 10, the processor 11 alternately performs acceleration correction and deceleration correction while switching between them. As a result, when the drive wheels 52 are slipping, the rotational speed ωmg of the motor 30 is increased, and when the drive wheels 52 are gripping, the rotational speed ωmg of the motor 30 is decreased. That is, in the vibration damping control, the processor 11 corrects the torque in a direction to bring the rotational speed ωmg of the motor 30 closer to the rotational speed ωt of the drive wheels 52. Therefore, the torsion of the drive shaft 50 can be suppressed, and the vibration caused by torsional resonance can be suppressed.

[0063] Note that depending on the switching timing between the acceleration correction and the deceleration correction, there is a possibility that the torsional resonance may be rather promoted. In contrast, in the control device 10, the processor 11 switches between the acceleration correction and the deceleration correction using the prediction result by the learned model. Therefore, the processor 11 can perform the acceleration correction so as to increase the rotational speed ωmg of the motor 30 in accordance with the timing when the drive wheels 52 slip based on the prediction result. And the processor 11 can perform the deceleration correction so as to decrease the rotational speed ωmg of the motor 30 in accordance with the timing when the drive wheels 52 grip.

[0064] <Effects of the present embodiment> (1) The control device 10 can suppress the vibration caused by torsional resonance by switching between the acceleration correction and the deceleration correction at an appropriate timing using the prediction result.

[0065] (2) The lower the friction coefficient of the road surface, the easier it is for the drive wheels 52 to slip. And when the drive wheels 52 are slipping, even if the rotational speed ωt of the drive wheels 52 increases, the vehicle is difficult to accelerate. That is, the relationship between the rotational speed ωt of the drive wheels 52 and the acceleration in the longitudinal direction of the vehicle is data including information on the friction coefficient of the road surface. The explanatory variable includes data on the acceleration in the longitudinal direction of the vehicle. Therefore, the control device 10 can make a more accurate prediction reflecting the difference in the friction coefficient of the road surface by the learned model.

[0066] (3) The smaller the torque of the motor 30, the smaller the twist of the drive shaft 50 when traveling on the corrugated road. While the processor 11 is performing vibration control while performing torque-down control to make the torque of the motor 30 smaller when traveling on the corrugated road than when not traveling on the corrugated road. Therefore, the control device 10 can further suppress the vibration due to torsional resonance in combination with the effect of the vibration control.

[0067] <Modified Example> This embodiment can be implemented with the following modifications. This embodiment and the following modified examples can be implemented in combination with each other as long as they are not technically contradictory.

[0068] · In the above embodiment, in the determination process, an example of using a learned model that predicts the ground load y of the drive wheel 52 after a predetermined time was shown. In contrast, the learned model may be a model that predicts whether the drive wheel 52 is slipping or gripping after a predetermined time. In that case, the training data X_tr may be data including data indicating whether the drive wheel 52 is slipping or gripping as the correct label. The data indicating whether the drive wheel 52 is slipping or gripping may be, for example, data of the result determined based on whether the ground load y is equal to or greater than the third threshold value. Also, it may be data of the result determined by analyzing the data of the wheel speed sensor by a person. Even when the determination process is performed using a learned model that predicts whether the drive wheel 52 is slipping or gripping after such a predetermined time, the same effect as the above embodiment can be obtained.

[0069] ·The learned model is not limited to the LSTM neural network. A learned model trained using other machine learning methods such as decision trees and regression models may also be used. In the above embodiment, an example was shown in which the input data X(1) to X(10) were input into the LSTM neural network in chronological order to calculate the ground contact load y. On the other hand, for example, a convolutional neural network can also be used as the learned model. In that case, the information of the input data X(1) to X(10) is made into data in the form of a single matrix of 4 rows and 10 columns, and input into the convolutional neural network to calculate the ground contact load y.

[0070] ·An example was shown in which the explanatory variable included data on the acceleration in the longitudinal direction of the vehicle, but the explanatory variable may not include data on the acceleration in the longitudinal direction of the vehicle. ·The explanatory variable may include data on the acceleration in the vertical direction of the vehicle. By including the data on the acceleration in the vertical direction of the vehicle in the explanatory variable, it becomes possible to more accurately predict the ground contact load y or more accurately predict whether the drive wheel 52 is slipping or gripping.

[0071] ·The explanatory variable may include data on the accelerator opening. By including the data on the accelerator opening in the explanatory variable, it becomes possible to more accurately predict the ground contact load y or more accurately predict whether the drive wheel 52 is slipping or gripping.

Explanation of Signs

[0072] 10... control device, 11... processor, 12... storage device, 20... inverter, 30... motor, 40... battery, 50... drive shaft, 51... differential, 52... drive wheel, 100... accelerator position sensor, 101... first wheel speed sensor, 102... second wheel speed sensor, 103... acceleration sensor, 104... current sensor, 105... rotation sensor, 106... torque sensor

Claims

1. Applied to an electric vehicle that drives drive wheels with a motor, comprising a processor and a storage device that stores a learned model learned by machine learning, When traveling on a wavy road, it is a vehicle control device in which the processor performs vibration suppression control to suppress vibration caused by torsional resonance of the drive shaft, The learned model is a model that predicts the ground load of the drive wheel after a predetermined time by inputting time series data of explanatory variables including the rotational speed of the motor, the torque generated by the motor, and the rotational speed of the drive wheel and performing calculations, In the vibration suppression control, the processor determines whether to perform speed increase correction to generate positive torque in the motor or deceleration correction to generate negative torque in the motor using the prediction result by the learned model, and switches between the speed increase correction and the deceleration correction, thereby increasing the rotational speed of the motor when the drive wheel is slipping and decreasing the rotational speed of the motor when the drive wheel is gripping Vehicle control device.

2. Applied to an electric vehicle that drives drive wheels with a motor, comprising a processor and a storage device that stores a learned model learned by machine learning, When traveling on a wavy road, it is a vehicle control device in which the processor performs vibration suppression control to suppress vibration caused by torsional resonance of the drive shaft, The learned model is a model that predicts whether the drive wheel is slipping or gripping after a predetermined time by inputting time series data of explanatory variables including the rotational speed of the motor, the torque generated by the motor, and the rotational speed of the drive wheel and performing calculations, In the vibration suppression control, the processor determines whether to perform speed increase correction to generate positive torque in the motor or deceleration correction to generate negative torque in the motor using the prediction result by the learned model, and switches between the speed increase correction and the deceleration correction, thereby increasing the rotational speed of the motor when the drive wheel is slipping and decreasing the rotational speed of the motor when the drive wheel is gripping Vehicle control device.

3. The explanatory variable includes data on the acceleration in the longitudinal direction of the vehicle The vehicle control device according to claim 1 or claim 2.

4. While the processor is traveling on the corrugated road, the processor performs the vibration control while performing torque-down control to reduce the torque of the motor compared to when not traveling on the corrugated road. The vehicle control device according to claim 1 or claim 2.

5. In the torque-down control, the processor reduces the torque of the motor by setting an upper limit value for the target value of the torque to limit the torque of the motor. The vehicle control device according to claim 4.

Images