Vehicle control device

Abstract

To execute avoidance processing by acquiring a change in a load acting on a power train.SOLUTION: A system control unit 100 comprises a process circuit 110 and a storage device 120. The storage device 120 stores data on learned model with a long short-term memory neural network trained using training data including data obtained by measuring torque on a second input shaft 26. The process circuit 110 executes a calculation process to calculate the torque on the second input shaft 26 by input data made up of time-series data of explanatory variables into the learned model. The explanatory variables include rotation speeds of respective wheels and a second motor generator 12, acceleration of a vehicle 10 and a torque command value. Also, the process circuit 110 executes an avoidance process to reduce the torque to be generated when determining an action of an excessive load based on an estimated value obtained through the calculation.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a vehicle control device. [Background technology]

[0002] When a vehicle travels on a rough road such as a wavy road, the rotational speed of the drive wheels fluctuates periodically, which may cause torsional resonance in the shaft that constitutes the power transmission path in the vehicle's powertrain. When such torsional resonance occurs, there is a risk that an excessive load will be applied to the powertrain.

[0003] Patent Document 1 discloses that a rotation speed change amount is calculated based on the rotation angle of a driving motor, and whether or not resonance is occurring is determined based on the rotation speed change amount. Patent Document 1 also discloses that a torque command value is reduced when it is determined that resonance is occurring. By reducing the torque of the motor, the load acting on the powertrain can be reduced. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2008-072868 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, when it is determined that resonance is occurring, resonance has already occurred and the amplitude has become large to a certain extent, so that a considerable load is acting on the shaft. In contrast, if the threshold value used for the determination is reduced in order to output the determination result at an earlier point in time, the number of false positive determinations will increase. In other words, even if resonance is not actually occurring, it will be determined that resonance is occurring and the torque will be reduced. [Means for solving the problem]

[0006] The means for solving the above problems and their effects will be described below. A vehicle control device for solving the above problem includes a storage device in which learned model data is stored, the learned model being obtained by training a long-term short-term memory neural network that calculates the torque of the shaft based on input data consisting of time-series data of explanatory variables during a predetermined period of time using training data including data of actual torque in a shaft constituting a power transmission path that transmits driving force from a driving force source mounted on a vehicle to driving wheels, and time-series data of explanatory variables including the rotation speed of each wheel of the vehicle, the rotation speed of a rotating shaft of the driving force source, the acceleration of the vehicle, and a torque command value for the driving force source during a predetermined period of time prior to the time point of measuring the actual torque. The vehicle control device also includes a processing circuit that executes a calculation process of inputting the input data into the learned model stored in the storage device to calculate the torque of the shaft, and an avoidance process of reducing the torque generated by the driving force source when it is determined that an excessive load is acting on the powertrain of the vehicle based on an estimated value of the torque of the shaft calculated through the calculation process.

[0007] When a load is applied to the powertrain due to torsional resonance, a large torsional torque is applied to the shaft. In other words, the magnitude of the load on the powertrain is more directly reflected in the torque of the shaft than in the amount of change in rotation speed, etc.

[0008] In the above configuration, a trained model trained by supervised learning, which is a type of machine learning, is used. Then, whether or not an excessive load is acting is determined based on the estimated torque value calculated using the trained model. In short, according to the above configuration, since the avoidance process is performed based on the estimated torque value, it is possible to more directly grasp the change in the load acting on the powertrain based on the torque and more accurately perform the avoidance process.

[0009] In one aspect of the vehicle control device, the explanatory variables include data on the accelerator operation amount. The torque generated in the powertrain changes according to the magnitude of the required output that is determined according to the accelerator operation amount. In other words, the accelerator operation amount is an index value that indicates the magnitude of the required output to the vehicle. Therefore, with the above configuration, it is possible to calculate the torque by reflecting information that is correlated with the magnitude of the required output that determines the magnitude of the torque.

[0010] In one aspect of the vehicle control device, the explanatory variables include data on a vehicle speed, which is the speed of the vehicle. If the vehicle speed is different, the torque acting on the shaft will be different. In other words, the vehicle speed data also affects the torque. With the above configuration, the torque can be estimated more accurately by reflecting the vehicle speed data.

[0011] In the case where the vehicle is equipped with a motor as the driving force source, the rotation speed of the rotating shaft included in the explanatory variables may be the rotation speed of the rotating shaft of the motor. In one aspect of the vehicle control device, the processing circuit calculates the torque of the rotating shaft of the motor in the calculation process.

[0012] When resonance occurs, the torque caused by the torsional vibration is superimposed on the torque of the torque command value. As the resonance grows, the torque amplitude gradually increases. Therefore, the deviation between the torque estimate value and the torque command value increases until an excessive load is applied due to the resonance.

[0013] In one aspect of the vehicle control device, the processing circuit determines that an excessive load is acting on the power train of the vehicle when a deviation between a torque estimation value, which is a torque value calculated in the calculation process, and a torque command value for the motor is greater than a threshold value. With this configuration, the timing to execute the avoidance process can also be determined. [Brief description of the drawings]

[0014] [Figure 1] FIG. 1 is a schematic diagram showing the relationship between a system control unit, which is a vehicle control device according to one embodiment, and a power train of the vehicle. [Diagram 2] FIG. 2 is a schematic diagram showing the structure of a neural network used to calculate torque. [Diagram 3] FIG. 3 is a schematic diagram showing the structure of a recurrent neural network. [Figure 4] FIG. 4 is a flowchart showing a series of steps in a routine for training a learned model. [Diagram 5] FIG. 5 is a flowchart showing a series of processing steps in a routine related to the calculation processing executed by the above-mentioned system control unit. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, an embodiment of a vehicle control device mounted on a vehicle will be described with reference to Figs. 1 to 5. <Vehicle configuration> As shown in Fig. 1, a vehicle 10 to be controlled includes an engine 50. The vehicle 10 also includes a battery 30 that stores electric power. The vehicle 10 further includes a first motor generator 11 and a second motor generator 12. The first motor generator 11 and the second motor generator 12 are motors that generate driving force in response to power supplied from the battery 30, and also function as generators that receive external power and generate electric power to charge the battery 30.

[0016] The vehicle 10 is further provided with a planetary gear mechanism 13 having three rotating elements, a sun gear 14, a planetary carrier 15, and a ring gear 16. A crankshaft 59, which is a rotating shaft of the engine 50, is connected to the planetary carrier 15 of the planetary gear mechanism 13, and a first input shaft 25, which is connected to a rotating shaft of the first motor generator 11, is connected to the sun gear 14 of the planetary gear mechanism 13. A counter drive gear 17 is integrally provided to the ring gear 16 of the planetary gear mechanism 13. A counter driven gear 18 is meshed with the counter drive gear 17. A reduction gear 19 is meshed with the counter driven gear 18. A second input shaft 26, which is connected to the rotating shaft of the second motor generator 12, is connected to the reduction gear 19.

[0017] A final drive gear 20 is connected to the counter driven gear 18 so as to be rotatable together with the counter driven gear 18. A final driven gear 21 is meshed with the final drive gear 20. A drive shaft 24 of drive wheels 23 is connected to the final driven gear 21 via a differential mechanism 22.

[0018] <About the System Control Unit 100> The system control unit 100 includes a storage device 120 in which programs are stored, and a processing circuit 110 that executes various controls by executing the programs stored in the storage device 120. The system control unit 100 is connected to a power control unit 200 and an engine control unit 300.

[0019] <About the power control unit 200> The first motor generator 11 and the second motor generator 12 are connected to the battery 30 via a power control unit 200. The power control unit 200 includes a control unit, an inverter, and a converter. The power control unit 200 operates based on a command from the system control unit 100. The power control unit 200 adjusts the amount of power supplied from the battery 30 to the first motor generator 11 and the second motor generator 12, and the amount of charge from the first motor generator 11 and the second motor generator 12 to the battery 30. The vehicle 10 is provided with a connector 31 that can be connected to an external power source 40. Therefore, the battery 30 can also be charged by the power supplied from the external power source 40. That is, the vehicle 10 is a plug-in hybrid vehicle.

[0020] <About the engine control unit 300> The engine control unit 300 controls the engine 50 based on commands from the system control unit 100 .

[0021] Detection signals from various sensors that detect the operating state of engine 50 are input to engine control unit 300. The sensors that input detection signals to engine control unit 300 include a crank position sensor 133 that detects the rotation angle of crankshaft 59. The crank position sensor 133 outputs a crank angle signal corresponding to a change in the rotation phase of crankshaft 59.

[0022] Detection signals from these sensors are input to the engine control unit 300. The engine control unit 300 calculates the engine speed, which is the rotation speed of the crankshaft 59, based on the detection signal of the rotation angle of the crankshaft 59 input from the crank position sensor 133.

[0023] As shown in FIG. 1, the system control unit 100 is connected to an accelerator position sensor 130 that detects an accelerator operation amount and a vehicle speed sensor 132 that detects the vehicle speed, which is the speed of the vehicle 10. The detection signal of the accelerator position sensor 130 and the detection signal of the vehicle speed sensor 132 are input to the system control unit 100. The system control unit 100 is also connected to a G sensor 134. The G sensor 134 is a three-axis acceleration sensor that detects acceleration in the front-rear, left-right, and up-down directions of the vehicle 10. Furthermore, the system control unit 100 is connected to wheel speed sensors 135 to 138 that detect the rotation speed of each of the front-rear, left-right, and up-down wheels of the vehicle 10, i.e., each of the four wheels. The wheel speed sensor 135 detects the rotation speed of the right front wheel. The wheel speed sensor 136 detects the rotation speed of the left front wheel. The wheel speed sensor 137 detects the rotation speed of the right rear wheel. The wheel speed sensor 138 detects the rotation speed of the left rear wheel. In the vehicle 10, the right and left front wheels are drive wheels 23.

[0024] Further, the current, voltage, and temperature of the battery 30 are input to the power control unit 200. The power control unit 200 calculates a state of charge index value SOC, which is the ratio of the remaining charge to the charge capacity of the battery 30, based on the current, voltage, and temperature.

[0025] The engine control unit 300 and the power control unit 200 are each connected to the system control unit 100. The system control unit 100, the power control unit 200, and the engine control unit 300 mutually exchange and share information based on detection signals input from sensors and calculated information.

[0026] Based on this information, the system control unit 100 outputs commands to the engine control unit 300 and controls the engine 50 through the engine control unit 300. The system control unit 100 also outputs commands to the power control unit 200 based on this information. As a result, the system control unit 100 controls the first motor generator 11 and the second motor generator 12 and controls the charging of the battery 30 through the power control unit 200. In this way, the system control unit 100 controls the vehicle 10 by outputting commands to the power control unit 200 and the engine control unit 300. In other words, the system control unit 100 is a control device that controls the vehicle 10.

[0027] <Control of Vehicle 10> Next, the control of the vehicle 10 performed by the system control unit 100 will be described in detail.

[0028] The system control unit 100 calculates a required output, which is a required value of the output of the vehicle 10, based on the accelerator operation amount and the vehicle speed. The system control unit 100 then determines the torque distribution of the engine 50, the first motor generator 11, and the second motor generator 12 according to the required output and a state of charge index value SOC of the battery 30. The system control unit 100 then controls the output of the engine 50 and power running / regeneration by the first motor generator 11 and the second motor generator 12. The system control unit 100 switches the driving mode of the vehicle 10 depending on the magnitude of the state of charge index value SOC.

[0029] When the state of charge index value SOC exceeds a certain level, the system control unit 100 selects a motor driving mode in which the vehicle runs using the driving force from the second motor generator 12 and the driving force from the first motor generator 11 without operating the engine 50. In other words, the system control unit 100 selects the motor driving mode when the remaining charge of the battery 30 is sufficient.

[0030] On the other hand, when the state of charge index value SOC falls below a certain level, the system control unit 100 selects a hybrid driving mode in which the engine 50 is used in addition to the first motor generator 11 and the second motor generator 12.

[0031] Even if the state of charge index value SOC exceeds a certain level, the system control unit 100 selects the hybrid driving mode in the following cases. When the vehicle speed exceeds the upper limit of the motor driving mode.

[0032] When a large amount of power is needed temporarily, such as during rapid acceleration with a large amount of accelerator pedal operation. When engine 50 needs to be started. When the hybrid driving mode is selected, the system control unit 100 causes the first motor generator 11 to function as a starter motor when starting the engine 50. Specifically, the system control unit 100 causes the first motor generator 11 to rotate the sun gear 14, thereby rotating the crankshaft 59 and starting the engine 50.

[0033] Furthermore, when the hybrid driving mode is selected, the system control unit 100 switches the control during vehicle stop according to the magnitude of the state of charge index value SOC. Specifically, when the state of charge index value SOC is equal to or greater than a threshold value, the system control unit 100 stops the operation of the engine 50 and does not drive the first motor generator 11 and the second motor generator 12. That is, the system control unit 100 stops the operation of the engine 50 during vehicle stop to suppress idling. Note that, when the state of charge index value SOC of the battery 30 is less than a threshold value, the system control unit 100 operates the engine 50. Then, the first motor generator 11 is driven by the output of the engine 50 to function as a generator.

[0034] When the hybrid driving mode is selected, the system control unit 100 switches the control according to the state of charge index value SOC even during driving. When the state of charge index value SOC of the battery 30 is equal to or greater than the threshold value when starting or during light load driving, the system control unit 100 starts and drives the vehicle 10 only by the driving force of the second motor generator 12. In this case, the engine 50 is stopped, and the first motor generator 11 does not generate power. On the other hand, when the state of charge index value SOC of the battery 30 is less than the threshold value when starting or during light load driving, the system control unit 100 starts the engine 50, causes the first motor generator 11 to generate power, and charges the generated power to the battery 30. In this case, the vehicle 10 runs by part of the driving force of the engine 50 and the driving force of the second motor generator 12. During steady running, when the state of charge index value SOC of the battery 30 is equal to or higher than the threshold value, the system control unit 100 operates the engine 50 in a state of high operating efficiency, and runs the vehicle 10 mainly with the output of the engine 50. At this time, the power of the engine 50 is divided between the drive wheels 23 side and the first motor generator 11 side via the planetary gear mechanism 13. As a result, the vehicle 10 runs while generating power with the first motor generator 11. Then, the system control unit 100 drives the second motor generator 12 with the generated power, and the power of the second motor generator 12 assists the power of the engine 50. On the other hand, during steady running, when the state of charge index value SOC of the battery 30 is less than the threshold value, the system control unit 100 increases the engine rotation speed. Then, the power generated by the first motor generator 11 is used to drive the second motor generator 12, and the surplus power is charged to the battery 30. During acceleration, the system control unit 100 increases the engine rotation speed and uses the electric power generated by the first motor generator 11 to drive the second motor generator 12. This causes the vehicle 10 to accelerate using the power of the engine 50 and the power of the second motor generator 12. During deceleration, the system control unit 100 stops the operation of the engine 50.The system control unit 100 then causes the second motor generator 12 to function as a generator, and the generated electric power is charged to the battery 30. In the vehicle 10, the resistance caused by such power generation is used as a brake. Such power generation control during deceleration is called regenerative control.

[0035] <About the effect of excessive torque due to resonance> When traveling on a rough road such as an undulating road, the rotation speed of the drive wheels 23 may periodically fluctuate, causing torsional resonance in the powertrain. When such torsional resonance occurs, a torsion is repeatedly input to the rotating shaft of the powertrain while gradually increasing in amplitude. As a result, there is a risk of an excessive load being applied to the powertrain.

[0036] The first motor generator 11 and the second motor generator 12 are connected to the drive wheels 23 without a torsional damper or a torque converter. Therefore, if such resonance occurs while the vehicle is traveling using only the driving force of the second motor generator 12, a particularly large load is likely to act on the second input shaft 26, which is the power transmission path.

[0037] Therefore, the system control unit 100 calculates an estimated value y of the torque in the second input shaft 26, and judges whether or not an excessive load due to such resonance will act based on the estimated torque value y. If it is judged that an excessive load will act, the system control unit 100 reduces the torque generated by the second motor generator 12 before the resonance grows and an excessive load acts, and executes an avoidance process to avoid the action of the excessive load.

[0038] <About trained models> Next, a description will be given of calculation of the estimated value y of the torque in the second input shaft 26. The storage device 120 of the system control unit 100 stores data of a trained model used for calculating the estimated value y.

[0039] In this system control unit 100, 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 as the trained model. The long short-term memory neural network is a so-called LSTM (Long Short-Term Memory) neural network. The LSTM neural network is a type of recurrent neural network.

[0040] The neural network that constitutes the LSTM neural network is shown in Figure 2. Note that in Figure 2, the transmission lines connecting the nodes of adjacent layers are omitted. As shown on the left side of Figure 2, this neural network has an input layer with multiple 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 contents of the input data X will be described later.

[0041] 2, this neural network has an output layer consisting of one node. This output layer outputs an estimated value y of the torque at the second input shaft 26. This neural network also has a hidden layer between the input layer and the output layer, which is made up of multiple layers with fewer nodes than the input layer.

[0042] The activation function in the hidden layer of this neural network is, for example, a hyperbolic tangent. The activation function may be a sigmoid function. The activation function may be a ReLU function. The number of layers in the hidden layer and the number of nodes in each hidden layer are hyperparameters that are adjusted and set in the design stage so that the estimated value y can be appropriately estimated.

[0043] In this neural network, explanatory variables that make up the input data X, which is the time series data of explanatory variables, are input to the input layer, and the sum of values ​​weighted according to each transmission path is input to the activation function. The output value of the activation function is then input to the next layer. This calculation is repeated, and finally the estimated value y is output from the output layer.

[0044] Torsional resonance occurs when the rotational speed of each wheel, including the drive wheels 23, periodically fluctuates while traveling on a rough road such as an undulating road, and a specific frequency range including the resonance frequency is repeatedly input to the rotating shaft of the powertrain. Therefore, the system control unit 100 uses, as explanatory variables, the rotational speed of each wheel of the vehicle 10, the rotational speed of the rotating shaft of the second motor generator 12, the longitudinal acceleration of the vehicle 10, and the torque command value for the second motor generator 12. The explanatory variables also include the accelerator operation amount and the vehicle speed.

[0045] The rotation speed of each wheel is used in order to reflect the fluctuation state of the rotation speed of each wheel in the calculation of the estimated value y. The rotation speed of each wheel is detected by each of the four wheel speed sensors 135 to 138.

[0046] The rotation speed of the rotating shaft of the second motor generator 12 is used because the torque acting on the second input shaft 26 varies depending on the rotation speed of the rotating shaft. In other words, the rotation speed of the rotating shaft of the second motor generator 12 is a parameter that has a correlation with the torque at the second input shaft 26. Therefore, in this model, the rotation speed of the rotating shaft of the second motor generator 12 obtained from the power control unit 200 is included in the explanatory variables, and is reflected in the calculation of the estimated value y.

[0047] The longitudinal acceleration of the vehicle 10 reflects the state of fluctuation in the driving force transmitted to the drive wheels 23 via the second input shaft 26 and the state of undulations on the road surface. In this model, in order to reflect such information in the calculation of the estimated value y, the longitudinal acceleration of the vehicle 10 detected by the G sensor 134 is included in the explanatory variables.

[0048] In this model, in order to reflect the magnitude of the torque generated by the second motor generator 12 in the calculation of the estimated value y, the torque command value to the second motor generator 12 is also included in the explanatory variables. The torque command value is obtained from the power control unit 200.

[0049] As described above, the torque generated in the powertrain changes according to the magnitude of the required output that is determined according to the accelerator operation amount. That is, the accelerator operation amount is an index value that indicates the magnitude of the required output to the vehicle 10. Therefore, in this model, the accelerator operation amount detected by the accelerator position sensor 130 is included in the explanatory variables in order to calculate the torque by reflecting information that correlates with the magnitude of the required output that determines the magnitude of the torque.

[0050] Furthermore, if the vehicle speed is different, the magnitude of the required output will be different. Also, the torque acting on the second input shaft 26 will be different. In other words, the vehicle speed data is also data that affects the torque. Therefore, in this model, the vehicle speed data detected by the vehicle speed sensor 132 is also included in the explanatory variables.

[0051] To calculate the torque of the second input shaft 26 at a certain point in time t, the processing circuit 110 of the system control unit 100 creates input data X from data acquired during a predetermined period of time T prior to the point in time t.

[0052] Specifically, the processing circuit 110 of the system control unit 100 acquires the above data constituting the explanatory variables at regular time intervals when the vehicle 10 is running only by the driving force of the second motor generator 12. Then, these values ​​are stored in the storage device 120. Then, the processing circuit 110 sets all values ​​of the explanatory variables acquired during the predetermined period T as the input data X. That is, the time series data of the explanatory variables acquired during the predetermined period T is set as the input data X. For example, when explanatory variables are collected 15 times during the predetermined period T, the input data X is a set of explanatory variables collected 15 times. Specifically, the input data X is a set from collected data X(1) consisting of explanatory variables collected first during the predetermined period T to collected data X(15) consisting of explanatory variables collected last during the predetermined period T.

[0053] As described above, each explanatory variable includes the rotation speeds detected by the four wheel speed sensors 135-138, the rotation speed of the rotating shaft of the second motor generator 12, the longitudinal acceleration of the vehicle 10, and the torque command value for the second motor generator 12. The explanatory variables also include the accelerator operation amount and the vehicle speed. Therefore, each collected data is a nine-dimensional vector consisting of these nine values. Therefore, in this case, the input layer of the neural network shown in FIG. 2 has nine nodes.

[0054] Figure 3 shows a schematic diagram of a recurrent neural network. The vertical arrows in Figure 3 indicate the direction of forward propagation of the neural network shown in Figure 2, where explanatory variables are input. "n" in Figure 3 indicates the chronological order of explanatory variables in the input data X.

[0055] The neural network to which collected data X(n) shown at the right end in Fig. 3 is input receives a nine-dimensional vector, which is collected data X(n), and outputs an estimated value y of the torque at the second input shaft 26. That is, in this case, collected data X(15) is input to the neural network. This neural network is a fully connected neural network that propagates forward to the output layer, which outputs the estimated value y.

[0056] As shown in FIG. 3, the output of the hidden layer of the neural network to which the collected data X(14) collected at the immediately previous timing is input is reflected in the hidden layer of the neural network to which the collected data X(15) collected last during a given period T is input.

[0057] 3, the output of the hidden layer of the neural network to which collected data X(13) collected at the immediately previous timing during the predetermined period T is input is reflected in the hidden layer of the neural network to which collected data X(14) is input. In this way, the output of the hidden layer of the neural network to which collected data collected at the immediately previous timing is input is reflected in each neural network to which collected data is input.

[0058] An LSTM neural network is a recurrent neural network that has a mechanism called a Long Short-Term Memory (LSTM) block in each hidden layer of the recurrent neural network, making it possible to adjust the propagation of time-series information.

[0059] The LSTM block consists of a cell that keeps the error internal and prevents the gradient from disappearing, an input gate that controls the input to the cell, an output gate that controls the output from the cell, and a forget gate that prevents the error from remaining in the cell for too long.

[0060] The trained model stored in the storage device 120 has been trained in advance using training data X_tr, which is created based on the results of a previously conducted driving experiment and includes information on actual measurement data of the torque of the second input shaft 26. To collect data for creating the training data X_tr, an experimental vehicle equipped with a sensor for measuring the torque at the second input shaft 26 is used. A large amount of data is collected by repeating driving experiments in which the experimental vehicle travels in various driving conditions on roads with various road surface conditions, including rough roads where various torsional resonances occur.

[0061] Next, the training process for obtaining a trained model used to calculate the estimated value y will be described with reference to FIG. The training data X_tr is generated based on a large amount of data collected from the results of experiments conducted in advance, etc. A computer that executes the training process of the neural network generates the training data X_tr based on the collected data, and trains the neural network based on the training data X_tr.

[0062] <About the training process> 4 is a flowchart showing the flow of processing in a routine related to the training processing. In the training processing, the computer first reads time-series data of explanatory variables in the processing of step S100.

[0063] Then, in the process of the next step S110, the computer generates a plurality of training data X_tr from the read time series data. The training data X_tr here is data that includes information on the actual measurement data of the torque of the second input shaft 26 in a set of the above-mentioned 15 collected data. Specifically, the collected data that are explanatory variables collected during the predefined period T are substituted into X(1) to X(15) in the order of the time series. That is, the data substituted into X(1) is the oldest collected data in the predefined period T, and the data substituted into X(15) is the latest collected data in the predefined period T. In this way, time series data in the same format as the input data X is generated. Then, one training data X_tr is generated by combining the time series data corresponding to the time t that corresponds to the measurement time of the actual torque with the actual torque as the correct label.

[0064] After generating an infinite number of training data X_tr from the read data in this way, the computer proceeds to step S120. In step S120, data X(1) to X(15) corresponding to input data X from the training data X_tr are input to the LSTM neural network to calculate an estimated torque value y.

[0065] Then, the computer performs learning in the process of the next step S130. Specifically, the weights in the neural network are adjusted so as to reduce the error between the estimated value y calculated through the process of step S120 and the actual torque, which is the correct label in the training data X_tr used for the estimation.

[0066] The computer then repeats the calculation of the estimated value y in the process of step S120 and the adjustment of the weight in the process of step S130. When the error of the estimated value y calculated using the LSTM neural network becomes sufficiently small, the computer determines that the learning is complete. Then, the computer stores the data of the trained LSTM neural network in the storage device 120 and ends the training process.

[0067] The data of the trained LSTM neural network whose weights have been adjusted through the training process is stored in the storage device 120 of the system control unit 100. That is, the trained model is stored in the storage device 120 of the system control unit 100.

[0068] <About calculation and avoidance processes> Next, the torque calculation process and avoidance process executed by the processing circuit 110 of the system control unit 100 will be described. The processing circuit 110 calculates the torque at the second input shaft 26 by repeatedly executing the routine shown in Fig. 5 when the vehicle is traveling only with the driving force of the second motor generator 12. Then, based on the torque estimate value y, it is determined whether or not an excessive load due to torsional resonance is applied. If it is determined that an excessive load is applied, the torque generated by the second motor generator 12 is reduced before the resonance grows and an excessive load is applied, and an avoidance process is executed to avoid the application of the excessive load.

[0069] While the system control unit 100 is operating, it collects detection values ​​from various sensors and constantly stores the time series data of the explanatory variables for a predetermined period T in the memory device 120, updating the data each time new data is collected.

[0070] 5 starts, first in step S200, the processing circuit 110 reads the time-series data of the explanatory variables stored in the storage device 120. Then, in step S210, the processing circuit 110 shapes the input data X from the read data.

[0071] Specifically, the processing circuit 110 substitutes the 15 collected data read during the predetermined period T in chronological order into X(1) to X(15). The computer thus formats the input data X in which the 15 collected data are arranged in chronological order.

[0072] Next, in step S220, the processing circuit 110 inputs the input data X to the trained model stored in the storage device 120, and calculates an estimated torque value y. Having calculated the estimated value y in this manner, the processing circuit 110 stores the value in the storage device 120 .

[0073] Then, in the process of the next step S230, the processing circuit 110 determines whether or not the magnitude of the deviation between the torque estimation value Tin_pred and the torque command value Tmg to the second motor generator 12 is greater than the threshold value A. Specifically, the processing circuit 110 substitutes the estimation value y for the torque estimation value Tin_pred, and compares the absolute value of the difference obtained by subtracting the torque command value Tmg to the second motor generator 12 from the torque estimation value Tin_pred with the threshold value A. If the absolute value is greater than the threshold value A (step S230: YES), it is determined that an excessive load due to torsional resonance is acting, and the process proceeds to step S240. The magnitude of the threshold value A is set to a magnitude that allows it to be determined that an excessive load due to torsional resonance is acting based on the magnitude of the absolute value being greater than the threshold value A.

[0074] Then, in step S240, the avoidance process is executed. Specifically, the processing circuit 110 reduces the torque of the second motor generator 12, and reduces the torque generated by the power train. By reducing the torque in this manner, the vehicle speed decreases and the period of fluctuation in the rotation speed of the drive wheels 23 becomes longer. This suppresses an increase in amplitude, converges the torsional vibration generated in the rotating shaft, and prevents resonance from occurring.

[0075] On the other hand, if it is determined in the process of step S230 that the absolute value is equal to or less than the threshold value A (step S230: NO), the processing circuit 110 ends this routine without executing the process of step S240. That is, in this case, the processing circuit 110 does not execute the avoidance process.

[0076] When it is determined that an excessive load due to torsional resonance will occur, the system control unit 100 determines that avoidance processing is necessary and executes the avoidance processing, thereby making it possible to prevent problematic loads from acting on the powertrain due to torsional resonance.

[0077] <Action of this embodiment> In the system control unit 100, input data X, which is obtained by arranging explanatory variables collected during a predetermined period T in chronological order, is used in the calculation process of the estimated value y.

[0078] In the calculation process, the input data X is input to an LSTM neural network, which is a learned model that has been trained using training data X_tr that includes the measured torque, to calculate an estimated value y.

[0079] That is, according to the system control unit 100, the estimated value y is calculated using a recurrent neural network that can handle time-series data while maintaining the information on the transition along the time axis. Therefore, the estimated value y is calculated in a form that reflects the influence of the transition along the time axis of the input on the output. Then, the avoidance process of reducing the torque is executed based on the estimated value y of the torque.

[0080] <Effects of this embodiment> (1) When a load is applied to the powertrain due to torsional resonance, a large torsional torque is applied to the second input shaft 26. That is, the magnitude of the load on the powertrain is more directly reflected in the torque of the second input shaft 26 than in the amount of change in rotation speed, etc.

[0081] The system control unit 100 uses a learned model trained by supervised learning, which is a type of machine learning. Then, it is determined whether or not an excessive load is acting based on the torque estimate value y calculated using the learned model. In short, according to the system control unit 100, since the avoidance process is performed based on the torque estimate value y, it is possible to accurately perform the avoidance process by capturing a change in the load acting directly on the powertrain based on the torque.

[0082] (2) It is a recurrent neural network that can handle time-series data while retaining the information of the transition along the time axis, and uses an LSTM neural network that can adjust the propagation of time-series information. Therefore, it is possible to calculate torque by reflecting the time-series transition of multiple data in a predetermined period T contained in the input data X. Therefore, it is possible to estimate torque with high accuracy.

[0083] (3) The torque generated in the powertrain changes according to the magnitude of the required output that is determined according to the accelerator operation amount. That is, the accelerator operation amount is an index value that indicates the magnitude of the required output to the vehicle 10. Therefore, the system control unit 100 can calculate the torque by reflecting information that is correlated with the magnitude of the required output that determines the magnitude of the torque.

[0084] (4) If the vehicle speed is different, the torque acting on the second input shaft 26 will be different. In other words, the vehicle speed data also affects the torque. According to the system control unit 100, it is possible to estimate the torque more accurately by reflecting the vehicle speed data.

[0085] This embodiment can be modified as follows: This embodiment and the following modifications can be combined with each other to the extent that there is no technical contradiction. <Example of change> In the above embodiment, an example of estimating the torque in the second input shaft 26 has been described, but the system control unit 100 may estimate the torque in another shaft of the powertrain. For example, the system control unit 100 may estimate the torque in the first input shaft 25 to which the rotating shaft of the first motor generator 11 is connected. In this case, training data X_tr including data on the measured torque in the first input shaft 25 is required. In other words, if training data X_tr including data on the measured torque in the shaft corresponding to the part for which the torque is to be estimated is prepared, the torque of any shaft can be calculated.

[0086] Also, an example has been shown in which the vehicle control device is embodied as a system control unit 100 that controls the vehicle 10, which is a hybrid vehicle. The control device that calculates torque using a trained model as in the above embodiment can be applied to vehicles other than hybrid vehicles. For example, it can be applied to a vehicle that has only an engine 50 as a driving force source. It can also be applied to an electric vehicle that has only a motor as a driving force source and does not have an engine 50.

[0087] The contents of the explanatory variables can be changed as appropriate. For example, information on the rotation speed of the crankshaft 59, the rotation speed of the rotating shaft of the first motor-generator 11, and the rotation speed of the rotating shaft of the second motor-generator 12 may be included in the explanatory variables.

[0088] In the above embodiment, the accelerator operation amount data is included in the explanatory variables, but the accelerator operation amount data may be omitted. In the above embodiment, the vehicle speed data is included in the explanatory variables, but the vehicle speed data may be omitted.

[0089] The method of determining that an excessive load is acting on the power train of the vehicle 10 based on the torque estimate y may be modified as appropriate. For example, when the amplitude of the torque estimate y is greater than a threshold value, it may be determined that an excessive load is acting due to torsional resonance.

[0090] In the above embodiment, the system control unit 100 includes a processing circuit 110 and a storage device 120 to execute software processing. However, this is merely an example. For example, the system control unit 100 may include a dedicated hardware circuit (e.g., ASIC, etc.) that processes at least a part of the software processing executed in the above embodiment. That is, the system control unit 100 may have any of the following configurations (a) to (c). (a) The system control unit 100 includes a processing device that executes all processing according to a program, and a storage device that stores the program. That is, the system control unit 100 includes a software execution device. (b) The system control unit 100 includes a processing device that executes a part of the processing according to a program, and a storage device. Furthermore, the system control unit 100 includes a dedicated hardware circuit that executes the remaining processing. (c) The system control unit 100 includes a dedicated hardware circuit that executes all processing. Here, the software execution device and / or the dedicated hardware circuit may be multiple. That is, the above processing may be executed by a processing circuitry that includes at least one of one or more software execution devices and one or more dedicated hardware circuits. The storage devices or computer-readable media for storing the programs include any available media that can be accessed by a general purpose or special purpose computer. [Explanation of symbols]

[0091] 10. Vehicle 11…First motor generator 12...Second motor generator 23...Drive wheel 24…Drive shaft 25…First input shaft 26…Second input shaft 50…Engine 59…Crankshaft 100...System control unit 110...Processing circuit 120...Storage device 130...Accelerator position sensor 132...Vehicle speed sensor 133...Crank position sensor 134…G sensor 135...Wheel speed sensor 136...Wheel speed sensor 137...Wheel speed sensor 138...Wheel speed sensor 200…Power control unit 300…Engine control unit

Claims

1. a storage device storing data of a learned model obtained by training a long short-term memory neural network that calculates a torque of the shaft based on input data consisting of time series data of explanatory variables during a predetermined period of time using training data including data of actual torque in a shaft that constitutes a power transmission path that transmits driving force from a driving force source mounted on a vehicle to driving wheels, and time series data of explanatory variables during a predetermined period of time prior to the measurement of the actual torque, the explanatory variables including the rotation speed of each wheel of the vehicle, the rotation speed of a rotating shaft of the driving force source, the acceleration of the vehicle, and a torque command value for the driving force source; a processing circuit that executes a calculation process of inputting the input data into the trained model stored in the storage device to calculate a torque in the shaft, and an avoidance process of reducing the torque generated by the driving force source when it is determined that an excessive load is acting on a power train of the vehicle based on an estimated value of the torque in the shaft calculated through the calculation process; A vehicle control device comprising:

2. The explanatory variables include data on the accelerator operation amount. The vehicle control device according to claim 1 .

3. The explanatory variables include data on the vehicle speed, which is the speed of the vehicle. The vehicle control device according to claim 1 or 2.

4. the vehicle is a vehicle equipped with a motor as the driving force source, The rotation speed of the rotating shaft included in the explanatory variable is the rotation speed of the rotating shaft of the motor. The vehicle control device according to any one of claims 1 to 3.

5. The processing circuit calculates the torque of the rotating shaft of the motor in the calculation process. The vehicle control device according to claim 4.

6. The processing circuit determines that an excessive load is acting on a power train of the vehicle when a deviation between a torque estimation value, which is a torque value calculated in the calculation process, and a torque command value for the motor is greater than a threshold value. The vehicle control device according to claim 5.

Images