Machine learning device, cooling temperature estimation device, cooling circuit control device, low-temperature circuit, on-board temperature control device, and computer program

Abstract

To provide a vehicle which constructs a cooling liquid temperature estimation model with excellent accuracy on which the influence of a response lag is properly reflected.SOLUTION: A machine learning device 30 performs machine learning of a cooling liquid temperature estimation model by using teacher data with time-series data including a control target value to an electrically-driven power train component, a parameter that changes according to long-term deterioration of the electrically-driven power train component, a control target value to a cooling circuit component, and an inlet side cooling liquid temperature of the electrically-driven power train component for cooling liquid flowing into the electrically-driven power train component when the cooling circuit component is operated as an input data set, and time-series data of an outlet side cooling liquid temperature of the electrically-driven power train component for the cooling liquid flowing out from the electrically-driven power train component at the same time as the input data set as an output data set.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a machine learning device, a cooling temperature estimation device, a cooling circuit control device, a low-temperature circuit, an on-board temperature adjustment device, and a computer program. [Background technology]

[0002] It is known that various controls are performed on a cooling circuit based on the temperature of the coolant that cools electric powertrain components such as a motor, a battery, etc. For example, a cooling device for a vehicle drive system that is configured to cool hybrid system coolant using a heat pump is known (for example, see Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2019-189004 A Summary of the Invention [Problem to be solved by the invention]

[0004] Waste heat from electric powertrain components is effective as a heat source for warming up motors and batteries in winter and for heat pumps. In order to achieve both warming and utilizing waste heat, it is important to have technology that accurately estimates and controls the amount of heat transferred from the electric powertrain components to the coolant and the coolant temperature. In coolant control, the response delay of the actual coolant temperature (output) relative to the target value (input) is large and not constant. In the case of coolant control using feedback control, when the control target value for the electric powertrain components changes, a response delay occurs before the change in heat generation due to the change in the target value is reflected in the change in the coolant temperature, significantly reducing controllability. Because this response delay is not constant, it is difficult to accurately estimate and control the coolant that reflects the response delay. [Means for solving the problem]

[0005] The gist of the present disclosure is as follows.

[0006] (1) A machine learning device that performs machine learning of a coolant temperature estimation model using training data in which an input dataset is made up of time series data consisting of control target values ​​for an electric powertrain component device, parameters that change due to deterioration of the electric powertrain component device over time, control target values ​​for a cooling circuit component device, and a coolant temperature on the inlet side of the electric powertrain component device for coolant flowing into the electric powertrain component device as the cooling circuit component device is operated, and an output dataset is time series data of a coolant temperature on the outlet side of the electric powertrain component device for the coolant flowing out of the electric powertrain component device at the same time as the input dataset. (2) The machine learning device according to (1) above; an acquisition unit that acquires, as the input data set, time-series data including a control target value for the electric powertrain constituent device, a control target value for the cooling circuit constituent device, and an inlet side coolant temperature of the electric powertrain constituent device; an estimation unit that calculates an estimate of an outlet side coolant temperature of the electric powertrain component from the input data set acquired by the acquisition unit, using the coolant temperature estimation model that has been learned by the machine learning device; A cooling temperature estimation device comprising: (3) The cooling temperature estimation device according to (2) above, a feedforward control unit that outputs a first command value by feedforward control based on an estimated value of an outlet side coolant temperature of the electric powertrain constituent device calculated by the cooling temperature estimation device; a feedback control unit that outputs a second command value by feedback control based on a deviation between a target value of the outlet side coolant temperature of the electric powertrain constituent device and a sensor value of the outlet side coolant temperature of the electric powertrain constituent device; a cooling circuit control unit that controls the cooling circuit constituent devices based on a total command value obtained by adding the first command value and the second command value; A cooling circuit control device comprising: (4) The cooling circuit control device according to (3) above; A pump constituting the cooling circuit constituent equipment controlled by the cooling circuit control device; Chiller and The electric powertrain components; A radiator that exchanges heat with outside air; a coolant flow path that communicates with the pump, the chiller, the electric power train components, and the radiator, and through which the coolant circulates when the pump is operated, through the pump, the chiller, the electric power train components, and the radiator; A low temperature circuit comprising: (5) The low-temperature circuit according to (4) above; a refrigeration circuit including a compressor that compresses a refrigerant, an inter-medium heat exchanger that causes the refrigerant to release heat to a heat medium to condense the refrigerant, the chiller, and an evaporator that causes the refrigerant to absorb heat and evaporate the refrigerant, and in which the refrigerant circulates through the compressor, the inter-medium heat exchanger, and the chiller or the evaporator; a high-temperature circuit including a heater core used for heating a vehicle interior, an electric heater, and the intermediate heat exchanger, in which a heat medium circulates through the heater core, the electric heater, and the intermediate heat exchanger; An on-board temperature control device comprising: (6) an acquisition step of acquiring, as an input data set, time-series data including a control target value for an electric powertrain component device, a parameter that changes with aging of the electric powertrain component device, a control target value for a cooling circuit component device, and an inlet side coolant temperature of the electric powertrain component device for coolant that flows into the electric powertrain component device as a result of operation of the cooling circuit component device; a learning step of performing machine learning of a coolant temperature estimation model using teacher data including the input data set and an output data set which is time-series data of an outlet side coolant temperature of the coolant flowing out of the electric powertrain component device at the same time as the input data set; A computer program for causing a computer to execute a machine learning process comprising: Effect of the Invention

[0007] According to the present disclosure, by performing machine learning using time-series data for a continuous predetermined period accompanied by changes in the control target value and aging as teacher data, it is possible to construct a highly accurate coolant temperature estimation model that appropriately reflects both the influence of transient errors (response delays) associated with changes in the control target value of the electric powertrain components and the operating state of the cooling circuit components, and the influence of steady errors associated with output changes due to deterioration over time of the electric powertrain components. In addition, by using feedforward control using the estimated value of the coolant temperature calculated using the learned coolant temperature estimation model in combination with feedback control, it is possible to improve the accuracy and convergence of the coolant temperature control. By improving the accuracy and convergence of the coolant temperature control in this way, it is possible to optimize the temperature adjustment of the electric powertrain components and the heater core, and improve the air conditioning performance and system efficiency of the vehicle-mounted temperature adjustment device. [Brief description of the drawings]

[0008] [Figure 1] 1 is a diagram showing a schematic configuration of an in-vehicle temperature control device having a machine learning device according to an embodiment of the present disclosure, in which an electric powertrain component device is a motor. FIG. [Diagram 2] 2 is a schematic configuration diagram of the cooling circuit control device shown in FIG. 1 and devices connected to the cooling circuit control device. [Diagram 3] FIG. 4 is a diagram illustrating heat exchange between a motor and coolant in a low-temperature circuit. [Figure 4] 10 is a diagram illustrating an example of a relationship between a flow rate of cooling water flowing through a water pipe provided around a motor and a response delay of an outlet side cooling liquid temperature with respect to a controlled variable. [Diagram 5] FIG. 13 is a diagram illustrating machine learning of a coolant temperature estimation model using a recurrent neural network. [Figure 6]11 is a diagram illustrating a comparison result between a temperature estimated by a cooling temperature estimation device according to an embodiment of the present disclosure and a temperature estimated by a conventional temperature estimation device that does not take into account a response delay. FIG. [Figure 7] 1 is a diagram showing a schematic configuration of an in-vehicle temperature regulating device having a machine learning device according to an embodiment of the present disclosure, in which an electric powertrain component device is a battery. [Figure 8] 8 is a schematic configuration diagram of the cooling circuit control device shown in FIG. 7 and devices connected to the cooling circuit control device. FIG. [Figure 9] FIG. 4 is a diagram illustrating heat exchange between a battery and coolant in a low-temperature circuit. [Figure 10] 1 is a diagram showing a schematic configuration of an in-vehicle temperature control device having a machine learning device according to an embodiment of the present disclosure, in which the electric powertrain components are a motor and a battery. [Figure 11] FIG. 11 is a schematic configuration diagram of the cooling circuit control device shown in FIG. 10 and devices connected to the cooling circuit control device. [Figure 12] FIG. 4 is a diagram illustrating heat exchange between the motor and the battery and the coolant in the low-temperature circuit. [Figure 13] 1 is a diagram showing a schematic configuration of an in-vehicle temperature control device having a machine learning device according to an embodiment of the present disclosure, in which the low-temperature circuit is equipped with an oil cooler. [Figure 14] 3A and 3B are diagrams illustrating heat exchange between a motor and cooling oil, and heat exchange between cooling oil and cooling water in an oil cooler. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] The machine learning device, the cooling temperature estimation device, the cooling circuit control device, the low-temperature circuit, the vehicle-mounted temperature control device, and the computer program will be described below with reference to the drawings. In each drawing, similar components are given similar reference symbols. In addition, the scale of these drawings is appropriately changed to facilitate understanding. The illustrated form is one example for carrying out the embodiment, and the present invention is not limited to these forms.

[0010] <Overall configuration of on-board temperature control device> FIG. 1 is a diagram illustrating a schematic configuration of an in-vehicle temperature regulating device having a machine learning device according to an embodiment of the present disclosure, showing a case where an electric powertrain component device is a motor.

[0011] In this embodiment, the vehicle-mounted temperature control device 1000, which adjusts the temperature inside the vehicle, is mounted on, for example, an electric vehicle driven by a motor, a hybrid vehicle driven by a motor and an internal combustion engine, or an internal combustion-driven vehicle driven by an internal combustion engine.

[0012] The vehicle-mounted temperature adjustment device 1000 includes a low-temperature circuit 1, a refrigeration circuit 2, and a high-temperature circuit 3. The electric powertrain constituent devices include a motor and a battery, but here, a case will be described in which the electric powertrain constituent device is a motor 14-1.

[0013] First, the low temperature circuit 1 will be described.

[0014] The low-temperature circuit 1 includes a cooling circuit control device 10, a water pump 11 which is a pump constituting a cooling circuit component device, a chiller 12 having a refrigerant pipe 12a and a cooling water pipe 12b, a low-temperature radiator 13, a motor 14-1 which is an electric powertrain component device, and coolant flow paths 15a and 15b.

[0015] The water pump 11 pumps the cooling water circulating in the low-temperature circuit 1 under the control of the cooling circuit control device 10. The water pump 11 has a duty ratio of wp The discharge capacity is changed steplessly by adjusting the discharge amount.

[0016] The coolant flow paths 15a and 15b are connected to the water pump 11, the chiller 12, the motor 14-1, and the low-temperature radiator 13. In the low-temperature circuit 1, the water pump 11 is operated to cause the coolant to flow through the coolant flow paths 15a and 15b, and the coolant circulates through the water pump 11, the chiller 12, the motor 14-1, and the low-temperature radiator 13. A water distribution pipe connected to the coolant flow paths 15a and 15b is provided around the motor 14-1, and heat exchange is performed between the coolant flowing through the water distribution pipe and the motor 14-1. The water distribution pipe provided around the motor 14-1 is provided with an inlet-side temperature sensor 16 for detecting an inlet-side coolant temperature of the coolant flowing into the motor 14-1 through the water distribution pipe, and an outlet-side temperature sensor 17 for detecting an outlet-side coolant temperature of the coolant flowing out of the motor 14-1 through the water distribution pipe.

[0017] The low-temperature radiator 13 is a heat exchanger that exchanges heat between the coolant in the low-temperature circuit 1 and the air (outside air) outside the vehicle 100. The low-temperature radiator 13 is configured to radiate heat from the coolant to the outside air when the temperature of the coolant is higher than the temperature of the outside air, and to absorb heat from the outside air to the coolant when the temperature of the coolant is lower than the temperature of the outside air. The low-temperature radiator 13 is provided with an electric fan 18 and a grill shutter 19 adjacent to the low-temperature radiator 13 that is disposed inside the front grill of the vehicle. When the vehicle is running, the low-temperature radiator 13 is exposed to the wind caused by the running of the vehicle, but the amount of the wind that hits the low-temperature radiator 13 can be adjusted by controlling the opening and closing of the grill shutter 19. In addition, even when the vehicle is not running, the wind generated by driving the electric fan 18 can be applied to the low-temperature radiator 13.

[0018] The above is the configuration of the low-temperature circuit 1. The details of the cooling circuit control device 10 will be described later. Next, the refrigeration circuit 2 will be described.

[0019] The refrigeration circuit 2 includes a compressor 41, a water-cooled condenser 42 which is an inter-medium heat exchanger, a chiller 12 having a refrigerant pipe 12a and a cooling water pipe 12b, a first electromagnetic expansion valve 43, a second electromagnetic expansion valve 44, an evaporator 45, and a refrigeration basic flow path 46a, an evaporator flow path 46b, and a chiller flow path 46c which are refrigerant flow paths.

[0020] The refrigeration circuit 2 is configured to realize a refrigeration cycle by circulating a refrigerant through the water-cooled condenser 42 and the chiller 12 or the evaporator 45 when the compressor 41 is driven. The refrigerant flow path in the refrigeration circuit 2 is divided into a refrigeration basic flow path 46a, an evaporator flow path 46b, and a chiller flow path 46c. The evaporator flow path 46b and the chiller flow path 46c are provided in parallel with each other and are each connected to the refrigeration basic flow path 46a. As the refrigerant, any substance generally used as a refrigerant in a refrigeration cycle, such as hydrofluorocarbon (e.g., HFC-134a), is used.

[0021] The compressor 41 functions as a compressor that compresses the refrigerant to raise its temperature. Note that a control unit that controls the compressor 41 is not shown. The compressor 41 is, for example, an electric type, and is configured so that its discharge capacity can be changed steplessly by adjusting the power supplied to the compressor 41 or the duty ratio. In the compressor 41, the refrigerant that is low temperature and low pressure and is mainly gaseous is adiabatically compressed to change into a refrigerant that is high temperature and high pressure and is mainly gaseous.

[0022] The water-cooled condenser 42, which is an inter-medium heat exchanger, has a heat medium pipe 42a and a refrigerant pipe 42b. The water-cooled condenser 42 functions as a heat exchanger that condenses the refrigerant by dissipating heat from the refrigerant to the cooling water of the high-temperature circuit 3. The refrigerant pipe 42b of the water-cooled condenser 42 functions as a condenser that condenses the refrigerant and dissipates heat in the refrigeration cycle. In the refrigerant pipe 42b of the water-cooled condenser 42, the high-temperature, high-pressure, mainly gaseous refrigerant flowing out from the compressor 41 is isobarically cooled and changes into a high-temperature, high-pressure, mainly liquid refrigerant.

[0023] The first electromagnetic expansion valve 43 and the second electromagnetic expansion valve 44 function as expanders that expand the refrigerant. The first electromagnetic expansion valve 43 and the second electromagnetic expansion valve 44 are provided with, for example, a narrow-diameter passage, and the pressure of the refrigerant is suddenly reduced by spraying the refrigerant from the narrow-diameter passage. In the first electromagnetic expansion valve 43 and the second electromagnetic expansion valve 44, the high-temperature, high-pressure liquid refrigerant flowing out of the water-cooled condenser 42 is reduced in pressure and partially vaporized, changing into a low-temperature, low-pressure mist of refrigerant.

[0024] The evaporator 45 functions as a heat exchanger that exchanges heat between the surrounding air (particularly, the air supplied to the vehicle interior) and the refrigerant. Therefore, if the temperature of the surrounding air is higher than the temperature of the refrigerant, the evaporator 45 absorbs heat from the surrounding air into the refrigerant, and if the refrigerant is in a liquid state, the evaporator 45 evaporates the low-temperature, low-pressure mist refrigerant that flows out of the first electromagnetic expansion valve 43, and changes it into a low-temperature, low-pressure gaseous refrigerant. As a result, the air around the evaporator 45 is cooled, and the vehicle interior can be cooled. Conversely, if the temperature of the surrounding air is lower than the temperature of the refrigerant, the evaporator 45 dissipates heat from the refrigerant to the surrounding air.

[0025] The chiller 12 has a refrigerant pipe 12a and a cooling water pipe 12b. The chiller 12 functions as a heat exchanger that exchanges heat between the cooling water of the low-temperature circuit 1 and the refrigerant. When the temperature of the refrigerant is lower than the cooling water of the low-temperature circuit 1, the refrigerant pipe 12a of the chiller 12 functions as an evaporator that evaporates the refrigerant to cool the cooling water. In the refrigerant pipe 12a of the chiller 12, the low-temperature, low-pressure mist refrigerant that flows out from the second electromagnetic expansion valve 44 evaporates and changes into a low-temperature, low-pressure gaseous refrigerant. As a result, the cooling water of the low-temperature circuit 1 is cooled.

[0026] Next, the high temperature circuit 3 will be described.

[0027] The high temperature circuit 3 includes a pump 51, a heater core 52, an electric heater 53, a water-cooled condenser 42, and coolant flow paths 54a and 54b. In the high temperature circuit 3, coolant, which is a heat medium, circulates through these components. Note that in the high temperature circuit 3, any other heat medium may be used instead of the coolant.

[0028] The pump 51 pumps the cooling water circulating in the high temperature circuit 3. Note that a control unit that controls the pump 51 is not shown. The pump 51 is an electric water pump similar to the water pump 11. Note that, although not shown here, the high temperature circuit 3 may include a high temperature radiator that performs heat exchange between the cooling water circulating in the high temperature circuit 3 and the outside air.

[0029] The heater core 52 is configured to perform heat exchange between the coolant circulating in the high-temperature circuit 3 and the air around the heater core 52 (particularly, the air supplied to the vehicle interior) to heat the vehicle interior. Specifically, the heater core 52 is configured to exhaust heat from the coolant to the air around the heater core 52. Therefore, when high-temperature coolant flows through the heater core 52, the temperature of the heat medium decreases and the air around the heater core 52 is warmed.

[0030] The electric heater 53 is configured to heat the vehicle interior by heat generated by passing an electric current through an electric conductor. Examples of the electric heater 53 include a sheath heater, a ceramic heater, and a carbon heater. Pipes connected to the cooling water flow paths 54a and 54b are provided around the electric heater 53, and heat exchange is performed between the electric heater 53 and the cooling water flowing through the pipes.

[0031] <Configuration of cooling circuit control device> The cooling circuit control device 10 controls a water pump 11 constituting a cooling circuit component device in the low temperature circuit 1. The cooling circuit control device 10 includes a cooling temperature estimation device 20, a feedforward control unit 21, a feedback control unit 22, a cooling circuit control unit 23, a subtractor 24, and an adder 25.

[0032] The cooling temperature estimation device 20 calculates an estimated value of the coolant temperature using a coolant temperature estimation model that has been trained by machine learning (AI: artificial intelligence). The cooling temperature estimation device 20 includes a machine learning device 30, an acquisition unit 31, and an estimation unit 32.

[0033] The acquisition unit 31 acquires, as an input data set, time-series data consisting of a control target value for the motor 14-1, which is an electric powertrain component device, a thermal resistance that is a parameter that changes with aging of the motor 14-1, a control target value for the water pump 11, which is a cooling circuit component device, and an inlet-side coolant temperature of the motor 14-1 for the coolant that flows into the motor 14-1 through a water distribution pipe as the water pump 11 operates, from among time-series data during a continuous predetermined period accompanied by changes in the control target value. The inlet-side coolant temperature of the coolant that flows into the motor 14-1 through the water distribution pipe is detected by the inlet-side temperature sensor 16. The input data set acquired by the acquisition unit 31, together with the output data set, is used for machine learning of the coolant temperature estimation model by the machine learning device 30, and is also used for calculation of an estimated value of the outlet-side coolant temperature of the motor 14-1 by the estimation unit 32 using the learned coolant temperature estimation model.

[0034] Furthermore, the acquisition unit 31 acquires, as an output data set, time series data of the outlet side coolant temperature of the motor 14-1 for the coolant flowing out from the motor 14-1 through the water pipe, out of the time series data during a continuous predetermined period accompanied by a change in the control target value. The output data set is acquired at the same time as the input data set. The outlet side coolant temperature for the coolant flowing out from the motor 14-1 through the water pipe is detected by the outlet side temperature sensor 17. The output data set acquired by the acquisition unit 31 is used in machine learning of the coolant temperature estimation model by the machine learning device 30, together with the input data set.

[0035] The machine learning device 30 performs machine learning of the coolant temperature estimation model using teacher data in which the input data set includes time series data of the control target value for the motor 14-1, the thermal resistance of the motor 14-1, the control target value for the water pump 11, and the inlet side coolant temperature of the motor 14-1 acquired by the acquisition unit 31, and the output data set includes time series data of the outlet side coolant temperature of the motor 14-1 for the coolant flowing out from the motor 14-1 through the water pipe at the same time as the input data set. The machine learning device 30 performs machine learning of the coolant temperature estimation model using a neural network. For example, a recurrent neural network (RNN), a long short term memory (LSTM), a gated recurrent unit (GRU), or the like can be used as the neural network. Details of the machine learning device 30 will be described later.

[0036] The estimation unit 32 uses the coolant temperature estimation model trained by the machine learning device 30 to calculate an estimate of the outlet side coolant temperature of the motor 14-1 from the input data set acquired by the acquisition unit 31.

[0037] The feedforward control unit 21 outputs a first command value by feedforward control based on the estimated value of the outlet side coolant temperature of the motor 14-1 calculated by the cooling temperature estimation device 20.

[0038] A subtractor 24 calculates the deviation between the target value of the outlet side coolant temperature of the motor 14-1 and the sensor value of the outlet side coolant temperature of the motor 14-1 detected by the outlet side temperature sensor 17, and a feedback control unit 22 outputs a second command value by feedback control based on this deviation. As the feedback control, for example, PID control or PI control is used.

[0039] The adder 25 adds the first command value and the second command value to generate a total command value, and the cooling circuit control unit 23 controls the water pump 11, which is a cooling circuit component device, based on this total command value.

[0040] FIG. 2 is a schematic configuration diagram of the cooling circuit control device shown in FIG. 1 and devices connected to the cooling circuit control device.

[0041] The cooling circuit control device 10 is a computer including a communication interface 101, a memory 102, and a processor 103, and is configured, for example, in an ECU (electronic control unit) provided in a vehicle equipped with an in-vehicle temperature adjustment device 1000. The overall configuration of the ECU is not shown in the figure.

[0042] In the cooling circuit control device 10, the communication interface 101, the memory 102, and the processor 103 are electrically connected to each other via signal lines.

[0043] The communication interface 101 has an interface circuit for connecting the cooling circuit control device 10 to various devices (e.g., the motor 14-1, the water pump 11, the inlet side temperature sensor 16, and the outlet side temperature sensor 17) that constitute the low-temperature circuit 1. The cooling circuit control device 10 communicates with other devices via the communication interface 101.

[0044] The memory 102 includes, for example, a volatile semiconductor memory (for example, a RAM), a non-volatile semiconductor memory (for example, a ROM), etc. The memory 102 stores computer programs for executing various processes in the processor 103, various data used when the processor 103 executes various processes, etc.

[0045] The memory 102 stores, for example, a computer program for making a computer execute machine learning processing. The computer program for making a computer execute machine learning processing includes an acquisition step and a learning step. In the acquisition step, time series data consisting of a control target value for the motor 14-1, which is an electric power train component device, a thermal resistance, which is a parameter that changes with aging of the motor 14-1, a control target value for the water pump 11, which is a cooling circuit component device, and an inlet side coolant temperature of the motor 14-1 for the coolant flowing into the motor 14-1 by the operation of the water pump 11, is acquired as an input data set. In the learning step, machine learning of a coolant temperature estimation model is performed using teacher data consisting of the input data set and an output data set, which is time series data of the outlet side coolant temperature of the motor 14-1 for the coolant flowing out from the motor 14-1 at the same time as the input data set. A neural network that performs machine learning of the coolant temperature estimation model is implemented in the computer program for making a computer execute machine learning processing. For example, a recurrent neural network, LSTM, GRU, etc. can be used as the neural network.

[0046] The processor 103 includes one or more central processing units (CPUs) and their peripheral circuits. The processor 103 may further include an arithmetic circuit such as a logical arithmetic unit or a numerical arithmetic unit. The processor 103 executes various processes based on computer programs stored in the memory 102.

[0047] In the above embodiment, the machine learning device 30 is included in the vehicle-mounted temperature adjustment device 1000. However, as a modified example, the machine learning device 30 may be provided in a computer (not shown) outside the vehicle-mounted temperature adjustment device 1000. In this case, the acquisition unit 31 and the estimation unit 32 are configured to be included in the vehicle-mounted temperature adjustment device 1000. The acquisition unit 31 sequentially acquires an input data set, which is time-series data consisting of a control target value for the motor 14-1, the thermal resistance of the motor 14-1, a control target value for the water pump 11, and the inlet-side coolant temperature of the motor 14-1, and an output data set, which is time-series data of the outlet-side coolant temperature of the motor 14-1 for the coolant flowing out from the motor 14-1 through the water pipe at the same time as the input data set, and stores them in the memory 102. Then, when a certain amount of input data sets and output data sets are stored in the memory 102, the input data sets and the output data sets are read out to an external computer via the communication interface 101, and the machine learning device 30 in the external computer performs machine learning processing of the coolant temperature estimation model. Then, the learned coolant temperature estimation model is written to the memory 102 via the communication interface 101. Thereafter, the estimation unit 32 can calculate an estimate of the outlet side coolant temperature of the motor 14-1 from the input data set acquired by the acquisition unit 31 using the coolant temperature estimation model learned by the machine learning device 30. According to this modification, for example, if multiple copies of the coolant temperature estimation model learned by the machine learning device 30 are applied to the in-vehicle temperature adjustment device 1000 having the same configuration of the low-temperature circuit 1, a temperature estimation process with a certain degree of accuracy can be realized at low cost.

[0048] <Heat exchange between the motor and the cooling water> FIG. 3 is a diagram illustrating heat exchange between the motor and the coolant in the low-temperature circuit.

[0049] As shown in FIG. 3, the water pump 11 pumps the cooling water circulating in the low-temperature circuit 1 under the control of the cooling circuit control device 10 shown in FIG. 1 and FIG. 2. The water pump 11 has a duty ratio of wpThe motor 14-1 is configured so that its discharge capacity can be changed steplessly by adjusting the amount of cooling liquid flowing through the motor 14-1. A water pipe 15 connected to the cooling liquid flow paths 15a and 15b shown in FIG. 1 is provided around the motor 14-1, and heat exchange is performed between the motor 14-1 and the cooling water flowing through the water pipe 15. Of the control target values ​​for the motor 14-1, the amount of regenerative power of the motor 14-1 is set to W in The motor shaft output of motor 14-1 is set to W out The amount of regenerative power W of the motor 14-1 in and motor shaft output W out The information on is acquired, for example, from a control unit (not shown) that controls the motor 14-1 or various sensors (not shown) provided on the motor 14-1. The thermal resistance of the motor 14-1, which is a parameter that changes with the aging of the motor 14-1, is R mot The inlet temperature sensor 16 detects the inlet coolant temperature Tw in The outlet temperature sensor 17 detects the outlet coolant temperature Tw out Detect.

[0050] <Recurrent Neural Network> 4 is a diagram illustrating the relationship between the flow rate of the cooling water flowing through the water pipe provided around the motor and the response delay of the outlet side cooling liquid temperature to the control amount. As shown in FIG. 4, the duty ratio of the water pump 11 is wp When the flow rate of the cooling water flowing through the water pipe 15 provided around the motor 14-1 increases due to an increase in the amount of heat and a rise in the water temperature, the control amount (Duty wp , W in , W out ) for the outlet coolant temperature Tw of motor 14-1 out In this way, the response delay of the control target value (W in , W out ) and the operation state (Duty) of the water pump 11 wp When the control amount (Duty wp , W in , Wout ) for the outlet coolant temperature Tw of motor 14-1 out The response delay of the motor 14-1 changes and appears as a transient error. mot changes, and the outlet coolant temperature Tw out This phenomenon appears as a steady error in the form of Equation 1.

[0051]

number

[0052] Based on the formula 1, in this embodiment, the control amount (Duty wp , W in , W out ) for the outlet coolant temperature Tw of motor 14-1 out The transient error caused by the response delay of the motor 14-1 and the heat loss R of the motor 14-1 due to the aging of the motor 14-1 are mot In order to construct a coolant temperature estimation model that appropriately reflects both the steady error caused by the change in the temperature and the steady error caused by the change in the temperature, machine learning processing is performed using a neural network that is suitable for learning time-dependent features in a situation where signals at each time are input sequentially. As a neural network designed to recognize such numerical time-series data, for example, a recurrent neural network, LSTM, GRU, etc. can be used.

[0053] FIG. 5 is a diagram for explaining machine learning of a coolant temperature estimation model using a recurrent neural network.

[0054] A recurrent neural network has a hidden layer (hidden state) connected to itself, and uses the state at a given time as the input value for the next state. Each node in a layer is connected in one direction to all nodes in the next layer. The input data set at a given time t is x t , the output data set is y t , RNN model activation function σ h and σ yA hidden layer with h t The input data set x at a certain point in time t is expressed as t and the hidden layer h at the previous time t-1 t-1 Based on this, a new hidden layer h t and the hidden layer h t from the output data set y t That is, the output data set y t is the current input dataset x t and the hidden state h t Based on this hidden state h t is the previous input data set x t-1 For example, the input data set value x at time t=-1 -1 and the hidden layer h at the previous time t-2 -2 Based on this, the hidden layer h0 transitions to the new hidden layer h0 at time t=0, and the output data set y0 is output from the hidden layer h0. The output data set y0 at time t=0 is based on the current input data set x0 and hidden state h0, and this hidden state h0 is based on the previous input data set x -1 It is based on.

[0055] In this embodiment, the control target value (W in , W out ) and the thermal resistance R of motor 14-1 mot and the control target value (Duty) for the water pump 11, which is a component of the cooling circuit. wp ) and the inlet side coolant temperature Tw of the motor 14-1 for the coolant flowing into the motor 14-1 through the water pipe 15 by the operation of the water pump 11. inThe input data set is made up of time series data. The output data set is made up of time series data of the outlet coolant temperature of the motor 14-1 for the coolant flowing out of the motor 14-1 through the drain pipe 15 at the same time as the input data set. The input data set and the output data set are acquired by the acquisition unit 31. The machine learning device 30 performs machine learning of a coolant temperature estimation model using a recurrent neural network with teacher data consisting of the input data set and the output data set. Since the recurrent neural network is designed to recognize numerical time series data, the control amount (Duty wp , W in , W out ) for the outlet coolant temperature Tw of motor 14-1 out The transient error caused by the response delay of the motor 14-1 and the heat loss R of the motor 14-1 due to the aging of the motor 14-1 are mot The estimation unit 32 uses the coolant temperature estimation model learned by the machine learning device 30 to calculate an estimate of the outlet side coolant temperature of the motor 14-1 from the input data set acquired by the acquisition unit 31.

[0056] FIG. 6 is a diagram illustrating a comparison result between a temperature estimated by a cooling temperature estimation device according to an embodiment of the present disclosure and a temperature estimated by a conventional temperature estimation device that does not take into account a response delay.

[0057] 6, for the outlet side coolant temperature, the actual value of the outlet side coolant temperature of the motor 14-1 is indicated by a solid line, the estimated value by the cooling temperature estimation device 20 according to one embodiment of the present disclosure (i.e., the estimated value taking into account the response delay) is indicated by a dotted line, and the estimated value by a conventional temperature estimation device that does not take into account the response delay is indicated by a dashed line. wpis also shown. When compared at time 0 [sec], it can be seen that the error of the estimated value by the cooling temperature estimation device 20 according to an embodiment of the present disclosure with respect to the actual value is smaller than the error of the estimated value by a conventional temperature estimation device that does not consider the response delay with respect to the actual value. In this way, according to an embodiment of the present disclosure, it is possible to construct a highly accurate coolant temperature estimation model that appropriately reflects both the influence of the transient error (response delay) associated with the change in the control target value of the motor 14-1 and the operating state of the water pump 11, and the influence of the steady error associated with the output change due to the deterioration over time of the motor 14-1. In addition, by using feedforward control using the estimated value of the coolant temperature calculated using the learned coolant temperature estimation model in combination with feedback control, it is possible to improve the accuracy and convergence of the coolant temperature control. By improving the accuracy and convergence of the coolant temperature control in this way, it is possible to optimize the temperature adjustment of the motor 14-1 and the heater core 52, and improve the air conditioning performance and system efficiency of the in-vehicle temperature adjustment device 1000.

[0058] In the above embodiment, a recurrent neural network is used as the neural network, but LSTM, GRU, etc. may be used instead.

[0059] <Heat exchange between battery and coolant> FIG. 7 is a diagram illustrating a schematic configuration of an in-vehicle temperature regulating device having a machine learning device according to an embodiment of the present disclosure, showing a case where an electric powertrain component device is a battery.

[0060] As shown in Fig. 7, even when the electric powertrain component in the low-temperature circuit 1 is a battery, it is possible to configure an in-vehicle temperature adjustment device similar to the in-vehicle temperature adjustment device shown in Fig. 1. That is, in the in-vehicle temperature adjustment device shown in Fig. 7, a battery 14-2 is provided instead of the motor 14-1 in the low-temperature circuit 1 in the in-vehicle temperature adjustment device shown in Fig. 1.

[0061] The low-temperature circuit 1 includes a cooling circuit control device 10, a water pump 11 constituting a cooling circuit component device, a chiller 12 having a refrigerant pipe 12a and a cooling water pipe 12b, a low-temperature radiator 13, a battery 14-2 constituting an electric power train component device, and coolant flow paths 15a and 15b. The water pump 11, the chiller 12, and the low-temperature radiator 13 are as described with reference to Fig. 1. The configurations of the refrigeration circuit 2 and the high-temperature circuit 3 are as described with reference to Fig. 1.

[0062] The coolant flow paths 15a and 15b are connected to the water pump 11, the chiller 12, the battery 14-2, and the low-temperature radiator 13. In the low-temperature circuit 1, the water pump 11 is operated to cause the coolant to flow through the coolant flow paths 15a and 15b, and the coolant circulates through the water pump 11, the chiller 12, the battery 14-2, and the low-temperature radiator 13. A water distribution pipe connected to the coolant flow paths 15a and 15b is provided around the battery 14-2, and heat exchange is performed between the coolant flowing through the water distribution pipe and the battery 14-2. The water distribution pipe provided around the battery 14-2 is provided with an inlet-side temperature sensor 16 for detecting an inlet-side coolant temperature of the coolant flowing into the battery 14-2 through the water distribution pipe, and an outlet-side temperature sensor 17 for detecting an outlet-side coolant temperature of the coolant flowing out of the battery 14-2 through the water distribution pipe.

[0063] The cooling circuit control device 10 controls a water pump 11 constituting a cooling circuit component device in the low temperature circuit 1. The cooling circuit control device 10 includes a cooling temperature estimation device 20, a feedforward control unit 21, a feedback control unit 22, a cooling circuit control unit 23, a subtractor 24, and an adder 25.

[0064] The cooling temperature estimation device 20 calculates an estimate of the outlet side coolant temperature of the battery 14-2 using a coolant temperature estimation model trained by machine learning (AI: artificial intelligence). The cooling temperature estimation device 20 includes a machine learning device 30, an acquisition unit 31, and an estimation unit 32.

[0065] The acquisition unit 31 acquires, as an input data set, time series data consisting of a control target value for the battery 14-2, a thermal resistance which is a parameter that changes with aging of the battery 14-2, a control target value for the water pump 11, and an inlet side coolant temperature of the battery 14-2 for the coolant flowing into the battery 14-2 through a water distribution pipe as the water pump 11 operates. The inlet side coolant temperature for the coolant flowing into the battery 14-2 through the water distribution pipe is detected by the inlet side temperature sensor 16. The input data set acquired by the acquisition unit 31, together with the output data set, is used for machine learning of a coolant temperature estimation model by the machine learning device 30, and is also used for calculation of an estimated value of an outlet side coolant temperature of the battery 14-2 by the estimation unit 32 using the learned coolant temperature estimation model.

[0066] The acquisition unit 31 also acquires, as an output dataset, time series data of the outlet coolant temperature of the battery 14-2 for the coolant flowing out from the battery 14-2 through the drainage pipe at the same time as the input dataset. The outlet coolant temperature for the coolant flowing out from the battery 14-2 through the drainage pipe is detected by the outlet temperature sensor 17. The output dataset acquired by the acquisition unit 31 is used, together with the input dataset, for machine learning of a coolant temperature estimation model by the machine learning device 30.

[0067] The machine learning device 30 performs machine learning of the coolant temperature estimation model using teacher data in which the input data set is time series data including the control target value for the battery 14-2 acquired by the acquisition unit 31, the thermal resistance of the battery 14-2, the control target value for the water pump 11, and the inlet side coolant temperature of the battery 14-2, and the output data set is time series data of the outlet side coolant temperature of the battery 14-2 for the coolant flowing out from the battery 14-2 through the water pipe at the same time as the input data set. The machine learning device 30 performs machine learning of the coolant temperature estimation model using a neural network. The neural network is as described with reference to FIG. 5 and FIG. 6.

[0068] The estimation unit 32 uses the coolant temperature estimation model trained by the machine learning device 30 to calculate an estimate of the outlet side coolant temperature of the battery 14-2 from the input data set acquired by the acquisition unit 31.

[0069] The feedforward control unit 21 outputs a first command value by feedforward control based on the estimated value of the outlet side coolant temperature of the battery 14-2 calculated by the cooling temperature estimation device 20.

[0070] The subtractor 24 calculates the deviation between the target value of the outlet side coolant temperature of the battery 14-2 and the sensor value of the outlet side coolant temperature of the battery 14-2 detected by the outlet side temperature sensor 17, and the feedback control unit 22 outputs a second command value by feedback control based on this deviation. As the feedback control, for example, PID control or PI control is used.

[0071] The adder 25 adds the first command value and the second command value to generate a total command value, and the cooling circuit control unit 23 controls the water pump 11, which is a cooling circuit component device, based on this total command value.

[0072] FIG. 8 is a schematic configuration diagram of the cooling circuit control device shown in FIG. 7 and devices connected to the cooling circuit control device.

[0073] The cooling circuit control device 10 is a computer including a communication interface 101, a memory 102, and a processor 103, and is configured, for example, in an ECU provided in a vehicle equipped with an on-board temperature adjustment device 1000. The overall configuration of the ECU is not shown in the figure.

[0074] In the cooling circuit control device 10, the communication interface 101, the memory 102, and the processor 103 are electrically connected to each other via signal lines.

[0075] The communication interface 101 has an interface circuit for connecting the cooling circuit control device 10 to various devices (e.g., the battery 14-2, the water pump 11, the inlet side temperature sensor 16, and the outlet side temperature sensor 17) that constitute the low-temperature circuit 1. The cooling circuit control device 10 communicates with other devices via the communication interface 101.

[0076] The memory 102 has been described with reference to FIG. 2. In this embodiment, the computer program stored in the memory 102 for causing a computer to execute the machine learning process includes an acquisition step and a learning step as follows. That is, in the acquisition step, time series data consisting of a control target value for the battery 14-2, a thermal resistance of the battery 14-2, a control target value for the water pump 11, and an inlet side coolant temperature of the battery 14-2 for the coolant flowing into the battery 14-2 by the operation of the water pump 11 is acquired as an input data set. In the learning step, machine learning of the coolant temperature estimation model is performed using teacher data consisting of the input data set and an output data set which is time series data of the outlet side coolant temperature of the battery 14-2 for the coolant flowing out from the battery 14-2 at the same time as the input data set. A neural network for performing machine learning of the coolant temperature estimation model is implemented in the computer program for causing a computer to execute the machine learning process. For example, a recurrent neural network, LSTM, GRU, or the like can be used as the neural network.

[0077] The processor 103 is as described with reference to FIG.

[0078] In the above-described embodiment, the machine learning device 30 is included in the vehicle-mounted temperature control device 1000. However, as a variation of this, the machine learning device 30 may be provided in a computer (not shown) external to the vehicle-mounted temperature control device 1000, as described with reference to FIG. 2.

[0079] FIG. 9 is a diagram illustrating heat exchange between the battery and the coolant in the low-temperature circuit.

[0080] As shown in FIG. 9, the water pump 11 pumps the cooling water circulating in the low-temperature circuit 1 under the control of the cooling circuit control device 10 shown in FIG. 7 and FIG. 8. The water pump 11 has a duty ratio of wp The discharge capacity is changed steplessly by adjusting the discharge amount of the battery 14-2. A water pipe 15 connected to the coolant flow paths 15a and 15b shown in FIG. 7 is provided around the battery 14-2, and heat exchange is performed between the battery 14-2 and the coolant flowing through the water pipe 15. Of the control target values ​​for the battery 14-2, the battery charge amount of the battery 14-2 is set to W in The battery discharge amount of the battery 14-2 is W out The battery charge amount W of the battery 14-2 is in and battery discharge amount W out The information on the thermal resistance of the battery 14-2, which is a parameter that changes with the aging of the battery 14-2, is obtained from, for example, a control unit (not shown) that controls the battery 14-2 or various sensors (not shown) provided on the battery 14-2. bat The inlet temperature sensor 16 detects the inlet coolant temperature Tw in The outlet temperature sensor 17 detects the outlet coolant temperature Tw out Detect.

[0081] The relationship between the flow rate of the cooling water flowing through the water pipe 15 provided around the motor 14-1 and the response delay of the outlet side cooling liquid temperature to the controlled variable, which has been described with reference to FIG. 4, also holds true for the relationship between the flow rate of the cooling water flowing through the water pipe 15 provided around the battery 14-2 and the response delay of the outlet side cooling liquid temperature to the controlled variable. That is, the duty ratio of the water pump 11 wp When the flow rate of the cooling water flowing through the water pipe 15 provided around the battery 14-2 increases due to an increase in the amount of heat and a rise in the water temperature, the control amount (Duty wp , W in , W out ) for the outlet coolant temperature Tw of battery 14-2 outIn this way, the response delay of the control target value (W in , W out ) and the operation state (Duty) of the water pump 11 wp When the control amount (Duty wp , W in , W out ) for the outlet coolant temperature Tw of battery 14-2 out The response delay of the battery 14-2 changes and appears as a transient error. bat changes, and the outlet coolant temperature Tw out This phenomenon appears as a steady error in the form of Equation 2.

[0082]

number

[0083] Based on Equation 2, in this embodiment, the control amount (Duty wp , W in , W out ) for the outlet coolant temperature Tw of battery 14-2 out and the thermal loss R of the battery 14-2 due to the aging of the battery 14-2. bat In order to construct a coolant temperature estimation model that appropriately reflects both the steady error caused by the change in the motor speed and the steady error caused by the change in the motor speed, a machine learning process is performed using a neural network that is suitable for learning time-dependent features in a situation where signals at each time are input sequentially. As a neural network designed to recognize such numerical time-series data, for example, a recurrent neural network, LSTM, GRU, etc. can be used. For the machine learning of the coolant temperature estimation model using a recurrent neural network, for example, the content described with reference to FIG. 5 is applied in which the part "motor 14-1" is replaced with "battery 14-2".

[0084] In this embodiment, the control target value (W in , Wout ) and the thermal resistance R of battery 14-2 bat and the control target value (Duty wp ) and the inlet side coolant temperature Tw of the battery 14-2 for the coolant flowing into the battery 14-2 through the water pipe 15 by the operation of the water pump 11. in The time series data consisting of the above is used as an input data set. Also, the outlet side coolant temperature Tw of the battery 14-2 for the coolant flowing out of the battery 14-2 through the water pipe 15 at the same time as the input data set is out The time series data of the above is used as an output data set. The input data set and the output data set are acquired by an acquisition unit 31. The machine learning device 30 performs machine learning of a coolant temperature estimation model using a recurrent neural network with teacher data consisting of the input data set and the output data set. Since the recurrent neural network is designed to recognize numerical time series data, the control amount (Duty wp , W in , W out ) for the outlet coolant temperature Tw of battery 14-2 out and the thermal loss R of the battery 14-2 due to the aging of the battery 14-2. bat The estimation unit 32 uses the coolant temperature estimation model learned by the machine learning device 30 to calculate an estimate of the outlet side coolant temperature of the battery 14-2 from the input data set acquired by the acquisition unit 31.

[0085] In this way, according to this embodiment, it is possible to construct a highly accurate coolant temperature estimation model that appropriately reflects both the influence of a transient error (response delay) associated with a change in the control target value of the battery 14-2 and the operating state of the water pump 11, and the influence of a steady error associated with an output change due to deterioration over time of the battery 14-2. In addition, by using feedforward control using an estimated value of the coolant temperature calculated using the learned coolant temperature estimation model in combination with feedback control, it is possible to improve the accuracy and convergence of the coolant temperature control. By improving the accuracy and convergence of the coolant temperature control in this way, it is possible to optimize the temperature adjustment of the battery 14-2 and the heater core 52, and improve the air conditioning performance and system efficiency of the vehicle-mounted temperature adjustment device 1000.

[0086] In the above embodiment, a recurrent neural network is used as the neural network, but LSTM, GRU, etc. may be used instead.

[0087] <Heat exchange between the motor / battery and the coolant> FIG. 10 is a diagram showing a schematic configuration of an in-vehicle temperature regulating device having a machine learning device according to an embodiment of the present disclosure, showing a case where the electric powertrain components are a motor and a battery.

[0088] As shown in Fig. 10, even when both a motor and a battery are provided as electric powertrain components in the low-temperature circuit 1, it is possible to configure an in-vehicle temperature control device similar to the in-vehicle temperature control devices shown in Fig. 1 and Fig. 7. That is, the in-vehicle temperature control device shown in Fig. 10 is provided with both the motor 14-1 in the low-temperature circuit 1 in the in-vehicle temperature control device shown in Fig. 1 and the battery 14-2 in the low-temperature circuit 1 in the in-vehicle temperature control device shown in Fig. 7.

[0089] The low-temperature circuit 1 includes a cooling circuit control device 10, a water pump 11 constituting cooling circuit components, a chiller 12 having a refrigerant pipe 12a and a cooling water pipe 12b, a low-temperature radiator 13, a battery 14-2 and a motor 14-1 constituting electric power train components, and coolant flow paths 15a and 15b. The water pump 11, the chiller 12, and the low-temperature radiator 13 are as described with reference to Fig. 1. The configurations of the refrigeration circuit 2 and the high-temperature circuit 3 are as described with reference to Fig. 1.

[0090] The coolant flow paths 15a and 15b are connected to the water pump 11, the chiller 12, the battery 14-2, the motor 14-1, and the low-temperature radiator 13. In the example shown in FIG. 10, the battery 14-2 and the motor 14-1 are arranged in this order along the direction in which the coolant flows, but the motor 14-1 and the battery 14-2 may be arranged in this order. In the low-temperature circuit 1, the coolant flows through the coolant flow paths 15a and 15b when the water pump 11 is operated, and the coolant circulates through the water pump 11, the chiller 12, the battery 14-2, the motor 14-1, and the low-temperature radiator 13. Around each of the battery 14-2 and the motor 14-1, water pipes connected to the coolant flow paths 15a and 15b are provided, and heat exchange is performed between the coolant flowing through the water pipes and each of the battery 14-2 and the motor 14-1. In the water distribution pipes provided around the battery 14-2 and the motor 14-1, an inlet temperature sensor 16 is provided to detect the inlet coolant temperature of the coolant flowing into the battery 14-2 through the water distribution pipe, and an outlet temperature sensor 17 is provided to detect the outlet coolant temperature of the coolant flowing out from the motor 14-1 through the water distribution pipe. When the motor 14-1 and the battery 14-2 are provided in this order along the direction in which the coolant flows, the inlet temperature sensor 16 is provided to detect the inlet coolant temperature of the coolant flowing into the motor 14-1 through the water distribution pipe, and the outlet temperature sensor 17 is provided to detect the outlet coolant temperature of the coolant flowing out from the battery 14-2 through the water distribution pipe.

[0091] The cooling circuit control device 10 controls a water pump 11 constituting a cooling circuit component device in the low temperature circuit 1. The cooling circuit control device 10 includes a cooling temperature estimation device 20, a feedforward control unit 21, a feedback control unit 22, a cooling circuit control unit 23, a subtractor 24, and an adder 25.

[0092] The cooling temperature estimation device 20 calculates an estimate of the outlet side coolant temperature of the motor 14-1 using a coolant temperature estimation model that has been trained by machine learning (AI: artificial intelligence). The cooling temperature estimation device 20 includes a machine learning device 30, an acquisition unit 31, and an estimation unit 32.

[0093] The acquisition unit 31 acquires, as an input data set, time series data consisting of control target values ​​for each of the battery 14-2 and the motor 14-1, the thermal resistances of each of the battery 14-2 and the motor 14-1, a control target value for the water pump 11, and an inlet side coolant temperature of the battery 14-2 for the coolant flowing into the battery 14-2 through the water pipe as the water pump 11 operates. The inlet side coolant temperature for the coolant flowing into the battery 14-2 through the water pipe is detected by the inlet side temperature sensor 16. The input data set acquired by the acquisition unit 31, together with the output data set, is used for machine learning of a coolant temperature estimation model by the machine learning device 30, and is also used for calculation of an estimated value of the outlet side coolant temperature of the motor 14-1 by the estimation unit 32 using the learned coolant temperature estimation model.

[0094] The acquisition unit 31 also acquires, as an output data set, time series data of the outlet side coolant temperature of the battery 14-2 for the coolant flowing out from the motor 14-1 through the water pipe at the same time as the input data set. The outlet side coolant temperature for the coolant flowing out from the motor 14-1 through the water pipe is detected by the outlet side temperature sensor 17. The output data set acquired by the acquisition unit 31 is used in machine learning of a coolant temperature estimation model by the machine learning device 30 together with the input data set.

[0095] The machine learning device 30 performs machine learning of the coolant temperature estimation model using teacher data in which the input data set includes time series data consisting of the control target values ​​for the battery 14-2 and the motor 14-1, the thermal resistances of the battery 14-2 and the motor 14-1, the control target value for the water pump 11, and the inlet side coolant temperature of the battery 14-2 acquired by the acquisition unit 31, and the output data set includes time series data of the outlet side coolant temperature of the motor 14-1 for the coolant flowing out from the motor 14-1 through the water pipe at the same time as the input data set. The machine learning device 30 performs machine learning of the coolant temperature estimation model using a neural network. The neural network is as described with reference to FIG. 5 and FIG. 6.

[0096] The estimation unit 32 uses the coolant temperature estimation model trained by the machine learning device 30 to calculate an estimate of the outlet side coolant temperature of the motor 14-1 from the input data set acquired by the acquisition unit 31.

[0097] The feedforward control unit 21 outputs a first command value by feedforward control based on the estimated value of the outlet side coolant temperature of the motor 14-1 calculated by the cooling temperature estimation device 20.

[0098] A subtractor 24 calculates the deviation between the target value of the outlet side coolant temperature of the motor 14-1 and the sensor value of the outlet side coolant temperature of the motor 14-1 detected by the outlet side temperature sensor 17, and a feedback control unit 22 outputs a second command value by feedback control based on this deviation. As the feedback control, for example, PID control or PI control is used.

[0099] The adder 25 adds the first command value and the second command value to generate a total command value, and the cooling circuit control unit 23 controls the water pump 11, which is a cooling circuit component device, based on this total command value.

[0100] FIG. 11 is a schematic configuration diagram of the cooling circuit control device shown in FIG. 10 and devices connected to the cooling circuit control device.

[0101] The cooling circuit control device 10 is a computer including a communication interface 101, a memory 102, and a processor 103, and is configured, for example, in an ECU provided in a vehicle equipped with an on-board temperature adjustment device 1000. The overall configuration of the ECU is not shown in the figure.

[0102] In the cooling circuit control device 10, the communication interface 101, the memory 102, and the processor 103 are electrically connected to each other via signal lines.

[0103] The communication interface 101 has an interface circuit for connecting the cooling circuit control device 10 to various devices (e.g., the motor 14-1, the battery 14-2, the water pump 11, the inlet side temperature sensor 16, and the outlet side temperature sensor 17) that constitute the low-temperature circuit 1. The cooling circuit control device 10 communicates with other devices via the communication interface 101.

[0104] The memory 102 has been described with reference to FIG. 2. In this embodiment, the computer program stored in the memory 102 for causing a computer to execute the machine learning process includes the following acquisition step and learning step. That is, in the acquisition step, time series data consisting of the control target values ​​for each of the battery 14-2 and the motor 14-1, the thermal resistances of each of the battery 14-2 and the motor 14-1, the control target value for the water pump 11, and the inlet side coolant temperature of the battery 14-2 for the coolant flowing into the battery 14-2 by the operation of the water pump 11 is acquired as an input data set. In the learning step, machine learning of the coolant temperature estimation model is performed using teacher data consisting of the input data set and an output data set which is time series data of the outlet side coolant temperature of the motor 14-1 for the coolant flowing out of the motor 14-1 at the same time as the input data set. A neural network for performing machine learning of the coolant temperature estimation model is implemented in the computer program for causing a computer to execute the machine learning process. For example, a recurrent neural network, LSTM, GRU, etc. can be used as the neural network.

[0105] The processor 103 is as described with reference to FIG.

[0106] In the above-described embodiment, the machine learning device 30 is included in the vehicle-mounted temperature control device 1000. However, as a variation of this, the machine learning device 30 may be provided in a computer (not shown) external to the vehicle-mounted temperature control device 1000, as described with reference to FIG. 2.

[0107] FIG. 12 is a diagram illustrating heat exchange between the motor and the battery and the coolant in the low-temperature circuit.

[0108] As shown in FIG. 12, the water pump 11 pumps the cooling water circulating in the low-temperature circuit 1 under the control of the cooling circuit control device 10 shown in FIG. 10 and FIG. 11. The water pump 11 has a duty ratio of wpThe discharge capacity is changed steplessly by adjusting the amount of cooling fluid flowing through the battery 14-2 and the motor 14-1. A water pipe 15 connected to the cooling fluid flow paths 15a and 15b shown in Fig. 10 is provided around each of the battery 14-2 and the motor 14-1, and heat exchange is performed between the cooling water flowing through the water pipe 15 and each of the battery 14-2 and the motor 14-1. Of the control target values ​​for the battery 14-2, the battery charge amount of the battery 14-2 is set to W batin The battery discharge amount of the battery 14-2 is W batout The thermal resistance of the battery 14-2, which is a parameter that changes with the aging of the battery 14-2, is R bat Of the control target values ​​for the motor 14-1, the amount of regenerative power of the motor 14-1 is set as W motin The motor shaft output of motor 14-1 is set to W motout The thermal resistance of the motor 14-1, which is a parameter that changes with the aging of the motor 14-1, is R mot The inlet temperature sensor 16 detects the inlet coolant temperature Tw in The outlet temperature sensor 17 detects the outlet coolant temperature Tw out Detect.

[0109] The relationship between the flow rate of the cooling water flowing through the water pipe 15 provided around the motor 14-1 and the response delay of the outlet side cooling liquid temperature to the controlled variable, which has been described with reference to Fig. 4, also holds true for the relationship between the flow rate of the cooling water flowing through the water pipes 15 provided around the battery 14-2 and the motor 14-1, respectively, and the response delay of the outlet side cooling liquid temperature to the controlled variable. That is, the duty ratio of the water pump 11 wp When the flow rate of the cooling water flowing through the water pipes 15 provided around the battery 14-2 and the motor 14-1 increases due to an increase in the amount of heat and a rise in the water temperature, the control amount (Duty wp , W batin , W batout , W motin , W motout ) for the outlet coolant temperature Tw of motor 14-1 outIn this way, the response delay of the temperature sensor 17 (detected by the outlet side temperature sensor 17) is reduced. batin , W batout , W motin , W motout ) and the operation state (Duty) of the water pump 11 wp When the control amount (Duty wp , W batin , W batout , W motin , W motout ) for the outlet coolant temperature Tw of battery 14-2 out The response delay of the battery 14-2 changes and appears as a transient error. bat changes and the heat loss R of the motor 14-1 decreases due to the deterioration of the motor 14-1 with time. mot changes, and the outlet coolant temperature Tw out This phenomenon appears as a steady error in the form of Equation 3.

[0110]

number

[0111] Based on Equation 3, in this embodiment, the control amount (Duty wp , W batin , W batout , W motin , W motout ) for the outlet coolant temperature Tw of battery 14-2 out and the thermal loss R of the battery 14-2 due to the aging of the battery 14-2. bat and the heat loss R of the motor 14-1 due to the aging of the motor 14-1. motIn order to construct a coolant temperature estimation model that appropriately reflects both the steady error caused by the change in the motor speed and the steady error caused by the change in the motor speed, a machine learning process is performed using a neural network that is suitable for learning time-dependent features in a situation where signals at each time are input sequentially. As a neural network designed to recognize such numerical time-series data, for example, a recurrent neural network, LSTM, GRU, etc. can be used. For the machine learning of the coolant temperature estimation model using a recurrent neural network, for example, the content described with reference to FIG. 5 is applied in which the part "motor 14-1" is replaced with "battery 14-2" and "motor 14-1".

[0112] In this embodiment, the control target value (W batin , W batout , R bat ) and the control target value (W motin , W motout , R mot ) and the thermal resistance R of battery 14-2 bat and the thermal resistance R of motor 14-1 mot and the control target value (Duty) for the water pump 11, which is a component of the cooling circuit. wp ) and the inlet side coolant temperature Tw of the battery 14-2 for the coolant flowing into the battery 14-2 through the water pipe 15 by the operation of the water pump 11. in The time series data consisting of the above is used as an input data set. Also, the outlet side coolant temperature Tw of the motor 14-1 for the coolant flowing out of the motor 14-1 through the water pipe 15 at the same time as the input data set is out The time series data of the above is used as an output data set. The input data set and the output data set are acquired by an acquisition unit 31. The machine learning device 30 performs machine learning of a coolant temperature estimation model using a recurrent neural network with teacher data consisting of the input data set and the output data set. Since the recurrent neural network is designed to recognize numerical time series data, the control amount (Dutywp , W batin , W batout , W motin , W motout ) for the outlet coolant temperature Tw of motor 14-1 out and the thermal loss R of the battery 14-2 due to the aging of the battery 14-2. bat and the heat loss R of the motor 14-1 due to the aging of the motor 14-1. mot It is possible to construct a coolant temperature estimation model that appropriately reflects both the steady error caused by the change in the temperature and the temperature of the outlet side of the motor 14-1 from the input data set acquired by the acquisition unit 31. The estimation unit 32 calculates an estimate of the coolant temperature of the outlet side of the motor 14-1 from the input data set acquired by the acquisition unit 31, using the coolant temperature estimation model that has been learned by the machine learning device 30. When the motor 14-1 and the battery 14-2 are provided in this order along the direction in which the coolant flows, the estimation unit 32 calculates an estimate of the coolant temperature of the outlet side of the battery 14-2 from the input data set acquired by the acquisition unit 31, using the coolant temperature estimation model that has been learned by the machine learning device 30.

[0113] In this way, according to the present embodiment, it is possible to construct a highly accurate coolant temperature estimation model that appropriately reflects both the influence of a transient error (response delay) associated with changes in the control target values ​​of the battery 14-2 and the motor 14-1 and the operating state of the water pump 11, and the influence of a steady error associated with output changes due to deterioration over time of the battery 14-2 and the motor 14-1. In addition, by using feedforward control using an estimated value of the coolant temperature calculated using the learned coolant temperature estimation model in combination with feedback control, it is possible to improve the accuracy and convergence of the coolant temperature control. By improving the accuracy and convergence of the coolant temperature control in this way, it is possible to optimize the temperature adjustment of the battery 14-2, the motor 14-1, and the heater core 52, and improve the air conditioning performance and system efficiency of the vehicle-mounted temperature adjustment device 1000.

[0114] In the above embodiment, a recurrent neural network is used as the neural network, but LSTM, GRU, etc. may be used instead.

[0115] <On-vehicle temperature control device equipped with an oil cooler> Fig. 13 is a diagram showing a schematic configuration of an in-vehicle temperature control device having a machine learning device according to an embodiment of the present disclosure, showing a case where the low-temperature circuit includes an oil cooler. Note that the configurations of the refrigeration circuit 2 and the high-temperature circuit 3 are as described with reference to Fig. 1, and are therefore omitted from Fig. 13. Here, a case where the electric powertrain component is the motor 14-1 will be described, but the same explanation applies when the electric powertrain component is the battery 14-2 or when the electric powertrain component is both the motor 14-1 and the battery 14-2.

[0116] As shown in FIG. 13, even when an oil-cooled circuit having an oil cooler 60 is further provided in the low-temperature circuit 1, an in-vehicle temperature control device 1000 similar to the in-vehicle temperature control device shown in FIG. 1 can be configured. The low-temperature circuit 1 includes a cooling circuit control device 10, a water pump 11 constituting a cooling circuit component device, a chiller 12 having a refrigerant pipe 12a and a cooling water pipe 12b, a low-temperature radiator 13, coolant flow paths 15a and 15b, an oil cooler 60, an oil pump 61 constituting a cooling circuit component device, a motor 14-1 which is an electric power train component device, and coolant flow paths 64a and 64b. The water pump 11, the chiller 12, and the low-temperature radiator 13 are as described with reference to FIG. 1. The configurations of the refrigeration circuit 2 and the high-temperature circuit 3 are also as described with reference to FIG. 1.

[0117] The oil cooler 60 is a heat exchanger in which a water distribution pipe connected to the cooling liquid flow paths 15a and 15b and an oil distribution pipe connected to the cooling liquid flow paths 64a and 64b are provided so as to pass through it, and which exchanges heat between the cooling water flowing through the water distribution pipe and the cooling oil flowing through the oil distribution pipe. The oil cooler 60 is configured to dissipate heat from the cooling water to the cooling oil when the temperature of the cooling water is higher than that of the cooling oil, and to absorb heat from the cooling oil to the cooling water when the temperature of the cooling water is lower than that of the cooling oil. In the water distribution pipe passing through the oil cooler 60, an inlet side temperature sensor 16 is provided for detecting the inlet side cooling liquid temperature of the cooling water flowing into the oil cooler 60 through the water distribution pipe, and an outlet side temperature sensor 17 is provided for detecting the outlet side cooling liquid temperature of the cooling water flowing out of the oil cooler 60 through the water distribution pipe.

[0118] The oil pump 61 pumps the cooling oil circulating through the cooling liquid flow paths 64a and 64b under the control of the cooling circuit control device 10. The oil pump 61 has a duty ratio of wp The discharge capacity is changed steplessly by adjusting the discharge amount.

[0119] The cooling liquid flow paths 64a and 64b are connected to the oil pump 61, the oil cooler 60, and the motor 14-1. In the oil cooling circuit, the oil pump 61 is operated to cause the cooling oil to flow through the cooling liquid flow paths 64a and 64b, and the cooling oil circulates through the oil pump 61, the motor 14-1, and the oil cooler 60. An oil distribution pipe connected to the cooling liquid flow paths 64a and 64b is provided around the motor 14-1, and heat exchange is performed between the cooling oil flowing through the oil distribution pipe and the motor 14-1. In the oil distribution pipe provided around the motor 14-1, an inlet side temperature sensor 62 is provided to detect an inlet side cooling liquid temperature of the cooling oil flowing into the motor 14-1 through the oil distribution pipe, and an outlet side temperature sensor 63 is provided to detect an outlet side cooling liquid temperature of the cooling oil flowing out of the motor 14-1 through the oil distribution pipe.

[0120] The cooling circuit control device 10 controls the water pump 11 and the oil pump 61 which constitute the cooling circuit components in the low temperature circuit 1. The cooling circuit control device 10 includes a cooling temperature estimation device 20, a feedforward control unit 21, a feedback control unit 22, a cooling circuit control unit 23, a subtractor 24, and an adder 25.

[0121] The cooling temperature estimation device 20 uses a coolant temperature estimation model that has been trained by machine learning (AI: artificial intelligence) to calculate an estimate of the outlet side coolant temperature of the oil cooler 60. The cooling temperature estimation device 20 includes a machine learning device 30, an acquisition unit 31, and an estimation unit 32.

[0122] The acquisition unit 31 acquires, as an input data set, time series data consisting of a control target value for the motor 14-1, a thermal resistance which is a parameter that changes with aging of the motor 14-1, a control target value for the oil pump 61, an inlet side coolant temperature of the motor 14-1 for the cooling oil flowing into the motor 14-1 through an oil distribution pipe as the oil pump 61 is operated, a control target value for the water pump 11, a thermal transmittance which is a parameter that changes with aging of the oil cooler 60, and an inlet side coolant temperature of the oil cooler 60 for the cooling water flowing into the oil cooler 60 through a water distribution pipe as the water pump 11 is operated. The data is used for machine learning of the coolant temperature estimation model by the machine learning device 30 together with the output data set, and is used for calculation of an estimated value of the outlet side coolant temperature of the motor 14-1 by the estimation unit 32 using the learned coolant temperature estimation model.

[0123] The acquisition unit 31 also acquires, as an output data set, time series data on the outlet side coolant temperature of the oil cooler 60 for the cooling water flowing out of the oil cooler 60 through the water pipe at the same time as the input data set.

[0124] The machine learning device 30 performs machine learning of the coolant temperature estimation model using teacher data consisting of an input data set acquired by the acquisition unit 31 and an output data set at the same time as the input data set. The machine learning device 30 performs machine learning of the coolant temperature estimation model using a neural network. The neural network is as described with reference to Figs. 5 and 6.

[0125] The estimation unit 32 uses the coolant temperature estimation model trained by the machine learning device 30 to calculate an estimate of the outlet side coolant temperature of the oil cooler 60 from the input data set acquired by the acquisition unit 31.

[0126] The feedforward control unit 21 outputs a first command value by feedforward control based on the estimated value of the outlet side coolant temperature of the oil cooler 60 calculated by the cooling temperature estimation device 20.

[0127] The subtractor 24 calculates the deviation between the target value of the outlet side coolant temperature of the oil cooler 60 and the sensor value of the outlet side coolant temperature of the oil cooler 60 detected by the outlet side temperature sensor 17, and the feedback control unit 22 outputs a second command value by feedback control based on this deviation. As the feedback control, for example, PID control or PI control is used.

[0128] The adder 25 adds the first command value and the second command value to generate a total command value, and the cooling circuit control unit 23 controls the water pump 11 and the oil pump 61, which are cooling circuit components, based on this total command value.

[0129] Although not shown, the cooling circuit control device 10 shown in FIG. 13 is a computer including a communication interface, a memory, and a processor, and is configured, for example, within an ECU provided in a vehicle equipped with the vehicle-mounted temperature adjustment device 1000.

[0130] Of these, the communication interface within the cooling circuit control device 10 has an interface circuit for connecting the cooling circuit control device 10 to various devices that constitute the low-temperature circuit 1 (e.g., motor 14-1, water pump 11, inlet side temperature sensor 16, outlet side temperature sensor 17, oil cooler 60, oil pump 61, inlet side temperature sensor 62, and outlet side temperature sensor 63, etc.).

[0131] The memory in the cooling circuit control device 10 has been described with reference to FIG. 2, FIG. 8, and FIG. 11. In this embodiment, the computer program for making a computer execute the machine learning process stored in the memory includes the following acquisition step and learning step. That is, in the acquisition step, time series data consisting of a control target value for the motor 14-1, the thermal resistance of the motor 14-1, a control target value for the oil pump 61, the inlet side coolant temperature of the motor 14-1, the control target value for the water pump 11, the thermal transmittance of the oil cooler 60, and the inlet side coolant temperature of the oil cooler 60 is acquired as an input data set, and time series data of the outlet side coolant temperature of the oil cooler 60 for the coolant flowing out of the oil cooler 60 through the water pipe at the same time as the input data set is acquired as an output data set. In the learning step, machine learning of the coolant temperature estimation model is performed using teacher data consisting of the input data set and the output data set at the same time as the input data set. A neural network for machine learning the coolant temperature estimation model is implemented in the computer program for making a computer execute the machine learning process. As the neural network, for example, a recurrent neural network, LSTM, GRU, etc. can be used.

[0132] The processor has been described with reference to FIGS.

[0133] In the above-described embodiment, the machine learning device 30 is included in the vehicle-mounted temperature control device 1000. However, as a variation of this, the machine learning device 30 may be provided in a computer (not shown) external to the vehicle-mounted temperature control device 1000, as described with reference to FIG. 2.

[0134] FIG. 14 is a diagram for explaining heat exchange between the motor and the cooling oil, and heat exchange between the cooling oil and the cooling water in the oil cooler.

[0135] As shown in Fig. 12, the oil pump 61 pumps the cooling oil circulating through the cooling liquid flow paths 64a and 64b under the control of the cooling circuit control device 10 shown in Fig. 13. The oil pump 61 has a duty ratio op The motor 14-1 is configured so that its discharge capacity can be changed steplessly by adjusting the amount of cooling oil. A cooling liquid flow path 64b shown in FIG. 10 is provided around the motor 14-1, and heat exchange takes place between the motor 14-1 and the cooling oil flowing through the cooling liquid flow path 64b. Of the control target values ​​for the motor 14-1, the amount of regenerative power of the motor 14-1 is set to W in The motor shaft output of motor 14-1 is set to W out The thermal resistance of the motor 14-1, which is a parameter that changes with the aging of the motor 14-1, is R mot The inlet temperature sensor 62 detects the inlet cooling oil temperature To in The outlet temperature sensor 17 detects the outlet coolant temperature To out Detect.

[0136] The relationship between the flow rate of the cooling water flowing through the water pipe 15 provided around the motor 14-1 and the response delay of the outlet side cooling liquid temperature to the controlled variable, which has been described with reference to FIG. 4, also holds true for the relationship between the flow rate of the cooling oil flowing through the cooling liquid flow passage 64b provided around the motor 14-1 and the response delay of the outlet side cooling liquid temperature to the controlled variable. That is, the duty ratio of the oil pump 61 opWhen the flow rate of the cooling oil flowing through the cooling liquid flow passage 64b provided around the motor 14-1 increases due to an increase in the oil temperature and the oil pressure, the control amount (Duty op , W in , W out ) for the outlet coolant temperature To of motor 14-1 out In this way, the response delay of the control target value (W in , W out ) and the operating state (Duty) of the oil pump 61 op When the control amount (Duty wp , W in , W out ) for the outlet coolant temperature To of motor 14-1 out The response delay of the motor 14-1 changes and appears as a transient error. mot changes, and the outlet coolant temperature To out This phenomenon appears as a steady error in the form of Equation 4.

[0137]

number

[0138] As shown in FIG. 12, the water pump 11 pumps the cooling water circulating through the cooling liquid passages 15a and 15b under the control of the cooling circuit control device 10 shown in FIG. 13. The water pump 11 has a duty ratio of wp The oil cooler 60 is configured so that its discharge capacity can be changed steplessly by adjusting the cooling liquid flow paths 15a and 15b shown in Fig. 13. The oil cooler 60 is provided with a water distribution pipe 15 connected to the cooling liquid flow paths 15a and 15b shown in Fig. 13, and is configured so that heat exchange occurs between the cooling water flowing through the water distribution pipe 15 and the cooling oil flowing through the cooling liquid flow path 64a. The thermal transmittance of the oil cooler 60, which is a parameter that changes with the aging of the oil cooler 60, is defined as R hx The inlet temperature sensor 16 detects the inlet coolant temperature Tw inThe outlet temperature sensor 17 detects the outlet coolant temperature Tw out Detect.

[0139] The relationship between the flow rate of the cooling water flowing through the water pipe 15 provided around the motor 14-1 and the response delay of the outlet side cooling liquid temperature to the controlled variable, which has been described with reference to FIG. 4, also holds true for the relationship between the flow rate of the cooling water flowing through the water pipe 15 provided in the oil cooler 60 and the response delay of the outlet side cooling liquid temperature to the controlled variable. That is, the duty ratio of the water pump 11 wp When the flow rate of the cooling water flowing through the water pipe 15 provided in the oil cooler 60 increases due to an increase in the amount of heat and a rise in the water temperature, the control amount (Duty wp , Duty op ) Coolant temperature Tw at the outlet of the oil cooler 60 out The response delay of the temperature sensor 17 (detected by the outlet side temperature sensor 17) is reduced. op ) operation status, water pump 11 operation status (Duty wp ), and the outlet coolant temperature To of the motor 14-1 out When the control amount (Duty wp , Duty op ) for the outlet coolant temperature Tw of motor 14-1 out The response delay of the oil cooler 60 changes and appears as a transient error. hx The temperature of the coolant at the outlet of the oil cooler 60, Tw out This phenomenon appears as a steady error in the form of Equation 5.

[0140]

number

[0141] Substituting Equation 4 into Equation 5 gives Equation 6.

[0142]

number

[0143] Based on Equation 6, in this embodiment, the control amount (Duty op , Duty wp , W min , W out ) Coolant temperature Tw at the outlet of the oil cooler 60 out The transient error caused by the response delay of the oil cooler 60 and the thermal transmittance R of the oil cooler 60 caused by the aging of the oil cooler 60 are hx and the heat loss R of the motor 14-1 due to the aging of the motor 14-1. mot In order to construct a coolant temperature estimation model that appropriately reflects both the steady error caused by the change in the temperature and the steady error caused by the change in the temperature, a machine learning process is performed using a neural network that is suitable for learning time-dependent features in a situation where signals at each time are input sequentially. As a neural network designed to recognize such numerical time-series data, for example, a recurrent neural network, LSTM, GRU, etc. can be used. For the machine learning of the coolant temperature estimation model using a recurrent neural network, for example, the content described with reference to FIG. 5 is applied in which the part "motor 14-1" is replaced with "oil cooler 60" and "motor 14-1".

[0144] In this embodiment, the control target value (W in , W out ) and the thermal resistance R of motor 14-1 mot and the control target value (Duty wp ) and the control target value (Duty op ) and the thermal transmittance R of the oil pump 61 hx The cooling oil flowing into the motor 14-1 through the cooling oil flow passage 64a by the operation of the oil pump 61 is the cooling oil temperature To in and the outlet coolant temperature To of the motor 14-1 for the coolant flowing out of the motor 14-1 through the coolant flow passage 64b. outand the inlet side coolant temperature Tw of the oil cooler 60 for the cooling water flowing into the oil cooler 60 through the water pipe 15 by the operation of the water pump 11. in The time series data consisting of the above is used as an input data set. In addition, the outlet side coolant temperature Tw of the oil cooler 60 for the coolant flowing out of the oil cooler 60 through the water pipe 15 at the same time as the input data set is out The time series data of the input data set and the output data set are acquired by an acquisition unit 31. The machine learning device 30 performs machine learning of a coolant temperature estimation model using a recurrent neural network with teacher data consisting of the input data set and the output data set. Since the recurrent neural network is designed to recognize numerical time series data, the control amount (Duty op , Duty wp , W min , W out ) Coolant temperature Tw at the outlet of the oil cooler 60 out The transient error caused by the response delay of the oil cooler 60 and the thermal transmittance R of the oil cooler 60 caused by the aging of the oil cooler 60 are hx and the heat loss R of the motor 14-1 due to the aging of the motor 14-1. mot The estimation unit 32 uses the coolant temperature estimation model learned by the machine learning device 30 to calculate an estimate of the outlet side coolant temperature of the oil cooler 60 from the input data set acquired by the acquisition unit 31.

[0145] In this way, according to the present embodiment, it is possible to construct a highly accurate coolant temperature estimation model that appropriately reflects both the influence of a transient error (response delay) associated with changes in the control target value of the motor 14-1 and the operating states of the water pump 11 and the oil pump 61, and the influence of a steady error associated with output changes due to deterioration over time of the motor 14-1 and the oil cooler 60. In addition, by using feedforward control using an estimated value of the coolant temperature calculated using the learned coolant temperature estimation model in combination with feedback control, it is possible to improve the accuracy and convergence of the coolant temperature control. By improving the accuracy and convergence of the coolant temperature control in this way, it is possible to optimize the temperature adjustment of the motor 14-1 and the heater core 52, and improve the air conditioning performance and system efficiency of the vehicle-mounted temperature adjustment device 1000.

[0146] In the above embodiment, a recurrent neural network is used as the neural network, but LSTM, GRU, etc. may be used instead.

[0147] As a modification of the above embodiment, the outlet side coolant temperature To of the motor 14-1 is calculated using a coolant temperature estimation model that has been trained by machine learning based on Equation 4. out The coolant temperature Tw at the outlet of the oil cooler 60 is estimated using this estimated value and the coolant temperature estimation model trained by machine learning based on Equation 5. out may be estimated. [Explanation of symbols]

[0148] 1 Low temperature circuit 2 Refrigeration circuit 3 High temperature circuit 10 Cooling circuit control device 11 Water Pump 12 Chiller 12a Refrigerant piping 12b Cooling water piping 13 Low temperature radiator 14-1 Motor 14-2 Batteries 15a, 15b Coolant flow path 16 Inlet temperature sensor 17 Outlet temperature sensor 18 Electric Fan 19 Grill Shutter 20 Cooling temperature estimation device 21 Feedforward control section 22 Feedback control section 22 23 Cooling circuit control unit 24 Subtractor 25 Adder 30 Machine Learning Device 31 Acquisition Department 32 Estimation part 41 Compressor 42 Water-cooled condenser 42a Heat medium piping 42b Refrigerant piping 43 First electromagnetic expansion valve 44 Second electromagnetic expansion valve 45 Evaporator 46a Refrigeration basic flow path 46b Evaporator flow path 46c Chiller flow path 51 Pump 52 Heater core 53 Electric heater 54a, 54b Cooling water flow path 60 Oil cooler 61 Oil pump 62 Inlet temperature sensor 63 Outlet temperature sensor 64a, 64b Coolant flow path 101 Communication Interface 102 Memory 103 Processors 1000 On-vehicle temperature control device

Claims

1. A machine learning device that performs machine learning of a coolant temperature estimation model using teacher data in which an input dataset is made up of time series data consisting of control target values ​​for an electric powertrain component device, parameters that change due to deterioration of the electric powertrain component device over time, control target values ​​for a cooling circuit component device, and a coolant temperature on the inlet side of the electric powertrain component device for coolant flowing into the electric powertrain component device as the cooling circuit component device operates, and an output dataset is time series data of a coolant temperature on the outlet side of the electric powertrain component device for the coolant flowing out of the electric powertrain component device at the same time as the input dataset.

2. The machine learning device according to claim 1 ; an acquisition unit that acquires, as the input data set, time-series data including a control target value for the electric powertrain constituent device, a control target value for the cooling circuit constituent device, and an inlet side coolant temperature of the electric powertrain constituent device; an estimation unit that calculates an estimate of an outlet side coolant temperature of the electric powertrain component from the input data set acquired by the acquisition unit, using the coolant temperature estimation model that has been learned by the machine learning device; A cooling temperature estimation device comprising:

3. The cooling temperature estimation device according to claim 2 ; a feedforward control unit that outputs a first command value by feedforward control based on an estimated value of an outlet side coolant temperature of the electric powertrain constituent device calculated by the cooling temperature estimation device; a feedback control unit that outputs a second command value by feedback control based on a deviation between a target value of a coolant temperature on an outlet side of the electric powertrain constituent device and a sensor value of the coolant temperature on an outlet side of the electric powertrain constituent device; a cooling circuit control unit that controls the cooling circuit components based on a total command value obtained by adding the first command value and the second command value; A cooling circuit control device comprising:

4. The cooling circuit control device according to claim 3 ; A pump constituting the cooling circuit constituent equipment controlled by the cooling circuit control device; Chiller and The electric powertrain components; A radiator that exchanges heat with outside air; a coolant flow path that communicates with the pump, the chiller, the electric power train components, and the radiator, and through which the coolant circulates when the pump is operated, through the pump, the chiller, the electric power train components, and the radiator; A low temperature circuit comprising:

5. A low temperature circuit according to claim 4; a refrigeration circuit including a compressor that compresses a refrigerant, an inter-medium heat exchanger that causes the refrigerant to release heat to a heat medium to condense the refrigerant, the chiller, and an evaporator that causes the refrigerant to absorb heat and evaporate the refrigerant, and in which the refrigerant circulates through the compressor, the inter-medium heat exchanger, and the chiller or the evaporator; a high-temperature circuit including a heater core used for heating a vehicle interior, an electric heater, and the intermediate heat exchanger, in which a heat medium circulates through the heater core, the electric heater, and the intermediate heat exchanger; An on-board temperature control device comprising:

6. an acquisition step of acquiring, as an input data set, time-series data including a control target value for an electric powertrain component device, a parameter that changes with aging of the electric powertrain component device, a control target value for a cooling circuit component device, and an inlet side coolant temperature of the electric powertrain component device for coolant that flows into the electric powertrain component device as a result of operation of the cooling circuit component device; a learning step of performing machine learning of a coolant temperature estimation model using teacher data including the input data set and an output data set which is time-series data of an outlet side coolant temperature of the coolant flowing out of the electric powertrain component device at the same time as the input data set; A computer program for causing a computer to execute a machine learning process comprising:

Images