JP2017033546A - Machine learning device that learns cooling device operating condition, motor control device having machine learning device, motor control system, and machine learning method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To control the temperature of a motor and a motor control device by machine learning to keep it below a prescribed temperature, and reduce losses in the motor, the motor control device, and a cooling device.SOLUTION: Provided is a machine learning device 1 that learns a cooling device operating condition for cooling a motor or a motor control device, the machine learning device comprising: a state observation unit 2 for observing a state variable including at least one of temperature data at a specific place of each of the motor and the motor control device while the cooling device is operating; a determination data acquisition unit 3 for acquiring determination data for determining an allowance for a loss in each of the motor, the motor control device, and the cooling device and for a permissible value for the temperature at the specific place of the motor and the motor control device; and a learning unit 4 for learning a cooling device operating condition in accordance with a training data set constituted with a combination of the state variable and the determination data.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a machine learning device, an electric motor control device, an electric motor control system, and a machine learning method, and in particular, a machine learning device that learns operating conditions of a cooling device, an electric motor control device and an electric motor control system that include the machine learning device, and It relates to a machine learning method.

  As the motor (motor) is driven, the temperature of the motor rises due to the heat generated by the iron loss of the stator core and the copper loss of the winding, and the motor loss may increase or the motor may be damaged. Therefore, a method of cooling the electric motor has been proposed in order to absorb and cool the generated heat (for example, Patent Document 1).

  Further, in the control device that drives the electric motor, the power element inside the control device generates heat and the temperature rises as the electric motor is driven, and the power element may be damaged. Therefore, a method for cooling the control device has been proposed in order to maintain the life of the control device.

  The motor cooling device described in Patent Document 1 is a winding ear formed by a cooling coolant supply path provided along the axial direction in a rotating shaft of a motor, and a winding wound around a stator core. A cooling refrigerant jet hole provided so as to be opposed to the cooling section and to eject the cooling refrigerant from the cooling refrigerant supply path to the winding ear, a pump for supplying the cooling refrigerant to the cooling refrigerant supply path, Pump control means is provided for varying the amount of cooling refrigerant discharged from the pump in accordance with the driving state of the motor. According to the motor cooling device described in Patent Document 1, it is possible to efficiently cool the stator core, which makes it possible to efficiently cool the entire motor.

  However, since the conventional motor control device only changes the operating rate of the cooling device in accordance with the temperature of the electric motor, the electric motor and the electric motor control device are controlled to be equal to or lower than the specified temperature while the electric motor is controlled. There is a problem that it is difficult to reduce the loss of the motor control device and the cooling device.

JP-A-5-236704

  The present invention provides a machine learning device capable of reducing the loss of the electric motor, the electric motor control device, and the cooling device while controlling the temperature of the electric motor and the electric motor control device to be equal to or lower than a prescribed temperature by machine learning. An object of the present invention is to provide an electric motor control device, an electric motor control system, and a machine learning method provided with a machine learning device.

  A machine learning device according to an embodiment of the present invention is a machine learning device that learns operating conditions of a cooling device for cooling an electric motor or an electric motor control device, and is a temperature at each specific location of the electric motor and the electric motor control device. A state observation unit for observing a state variable including at least one of the data during operation of the cooling device, a total loss of each of the motor, the motor control device, and the cooling device, and identification of each of the motor and the motor control device A determination data acquisition unit for acquiring determination data for determining a margin for the allowable value of the temperature of the location, a learning unit for learning the operating condition of the cooling device according to a training data set constituted by a combination of the state variable and the determination data, It is characterized by providing.

  An electric motor control device according to an embodiment of the present invention is an electric motor control device provided with the machine learning device, wherein the learning unit learns the rotational speed and operating time of the cooling device based on the result learned according to the training data set. It further comprises a decision-making unit that determines at least one command value of the set or a set of refrigerant temperature and refrigerant flow rate of the cooling device.

  An electric motor control system according to an embodiment of the present invention includes the electric motor control device and a temperature detection element that outputs temperature data.

  A machine learning method according to an embodiment of the present invention is a machine learning method for learning an operating condition of a cooling device for cooling an electric motor or an electric motor control device, and includes a specific part of each of the electric motor and the electric motor control device. Observe state variables including temperature data during operation of the cooling system, and obtain judgment data for determining the total loss of the motor, motor controller, and cooling system and the margin for the temperature tolerance at each specific location And learning the operating condition of the cooling device according to the training data set constituted by the combination of the state variable and the determination data.

  According to the present invention, a machine learning device capable of reducing the total loss of the electric motor, the motor control device, and the cooling device by machine learning, an electric motor control device including the machine learning device, an electric motor control system, and a machine learning method. Can be provided.

It is a block diagram of the machine learning apparatus which concerns on the Example of this invention. It is a block diagram of the electric motor control system which concerns on the Example of this invention. It is a block diagram of the electric motor control system which concerns on the 1st modification of the Example of this invention. It is a block diagram of the electric motor control system which concerns on the 2nd modification of the Example of this invention. It is a block diagram of the electric motor control system which concerns on the 3rd modification of the Example of this invention. It is a block diagram of the electric motor control apparatus which concerns on the Example of this invention. It is a flowchart for demonstrating the operation | movement procedure of the machine learning apparatus which concerns on the Example of this invention. It is a flowchart for demonstrating the operation | movement procedure of the electric motor control system which concerns on the Example of this invention.

  Hereinafter, a machine learning device, an electric motor control device, an electric motor control system, and a machine learning method according to the present invention will be described with reference to the drawings.

  FIG. 1 is a configuration diagram of a machine learning apparatus according to an embodiment of the present invention. FIG. 2 is a configuration diagram of an electric motor control system according to the embodiment of the present invention. FIG. 6 is a configuration diagram of an electric motor control device according to an embodiment of the present invention.

  A machine learning device 1 according to an embodiment of the present invention is a machine learning device that learns operating conditions of a cooling device 7 for cooling an electric motor 5 or an electric motor control device 6, and includes a state observation unit 2 and acquisition of determination data. A unit 3 and a learning unit 4 are provided. In FIG. 2, the machine learning device is included in the motor control device 6.

  The state observing unit 2 observes state variables including temperature data of specific locations of the electric motor 5 (see FIG. 2) and the electric motor control device 6 during the operation of the cooling device 7.

  The determination data acquisition unit 3 determines the determination data for determining the margin for the loss of each of the electric motor 5, the electric motor control device 6, and the cooling device 7 and the allowable value of the temperature at each specific location of the electric motor 5 and the electric motor control device 6. get.

  The learning unit 4 learns the operating conditions of the cooling device 7 in accordance with a training data set configured by a combination of state variables and determination data.

  The temperature data are the winding temperature of the motor 5 and the power element temperature of the motor control device 6 detected by the temperature detection elements (8, 9) provided in the motor 5 and the motor control device 6, respectively.

  The machine learning device 1 further includes a loss estimation unit 11 that estimates the loss of the electric motor 5 and the electric motor control device 6 from the temperature data, and a loss calculation unit 12 that calculates the loss of the cooling device 7 from the operating conditions of the cooling device 7. It is preferable to provide. The loss estimation unit 11 and the loss calculation unit 12 may be included in the determination data acquisition unit 3.

  The loss of the cooling device 7 is calculated using the rotation speed and the operation time of the cooling device 7. Further, when cooling with a liquid, the cooling flow rate and the refrigerant temperature detected by the temperature detection element (10, 10 ') may be used. The temperature detection element 10 detects the temperature of the refrigerant in the direction of flowing from the electric motor 5 to the cooling device 7, and the temperature detection element 10 ′ detects the temperature of the refrigerant in the direction of flowing from the cooling device 7 to the electric motor 5.

  In the configuration diagram of the motor control system 100 shown in FIG. 2, an example in which only one cooling device is provided is shown, but the present invention is not limited to such an example, and a plurality of cooling devices may be provided (FIG. 3). , FIG. 4 and FIG. 5).

  FIG. 3 is a configuration diagram of an electric motor control system 100 ′ according to a first modification of the embodiment of the present invention. As shown in FIG. 3, two cooling devices, that is, a first cooling device 71 and a second cooling device 72 may be provided, and each may cool the electric motor 5 and the electric motor control device 6.

  4 is a configuration diagram of an electric motor control system 100 ″ according to a second modification of the embodiment of the present invention. As shown in FIG. 4, two cooling devices, that is, a first cooling device 71 ′ and A second cooling device 72 ′ may be provided, the first cooling layer device 71 ′ may cool the electric motor 5, and the second cooling device 72 ′ may cool the electric motor control device 6.

  FIG. 5 is a configuration diagram of an electric motor control system 100 ′ ″ according to a third modification of the embodiment of the present invention. As shown in FIG. 5, four cooling devices, that is, a first cooling device 71 ″, a second cooling device 72 ″, a third cooling device 73 ″, and a fourth cooling device 74 ″ are provided, and the first cooling device is provided. The device 71 ″ and the second cooling device 72 ″ may cool the motor 5, and the third cooling device 73 ″ and the fourth cooling device 74 ″ may cool the motor control device 6. In addition, the number of cooling devices is not limited to the examples shown in FIGS. 3 to 5, and three or more cooling devices may be provided.

  The learning unit 4 may be configured to learn operating conditions according to a training data set acquired for a plurality of cooling devices.

  The motor control device 6 includes a reward calculation unit 14 that calculates a reward based on the determination data. The learning unit 4 determines an appropriate operating condition (rotation speed, operation based on the current state variable based on the reward from the current state variable. A function updating unit 15 for updating a function for determining time).

  The reward calculation unit 14 determines determination data, that is, the total loss of the electric motor 5, the electric motor control device 6, and the cooling device 7, and the margin for the allowable value of the temperature of the electric motor 5 and the electric motor control device 6, each specific location. The reward is calculated based on the result.

  Specifically, the remuneration calculation unit 14 sets the respective losses of the cooling device 7, the electric motor 5, and the electric motor control device 6, and the temperatures of the specific portions of the cooling device 7, the electric motor 5, and the electric motor control device 6 to an allowable value. On the other hand, the reward can be determined based on whether or not there is room. For example, the total loss of each of the cooling device 7, the motor 5, and the motor control device 6 is decreased from the previous value (one trial before), and each of the cooling device 7, the motor 5, and the motor control device 6 is specified. If the temperature at the location is less than the allowable value, the reward is increased according to the margin, and the total loss of each of the cooling device 7, the motor 5, and the motor control device 6 is increased from the previous value (before one trial). Alternatively, the reward may be decreased when the temperatures of the specific portions of the cooling device 7, the electric motor 5, and the electric motor control device 6 are equal to or higher than an allowable value. In this embodiment, the remuneration is calculated from the total loss of the electric motor 5 and the electric motor control device 6 in addition to the loss of the cooling device 7, but the loss of the electric motor 5 and the electric motor control device 6 is small. May not be used to calculate rewards.

The function updating unit 15 preferably performs reinforcement learning using so-called Q learning. Q learning is a method of learning a value (action value) Q (s, a) for selecting an action a under a certain environment s. In a certain state s, the action a having the highest Q (s, a) is selected as the optimum action. The function updating unit 15 updates the function (behavior value function Q (s _t , a _t )) using the following equation (1).

_{Here, Q (s t, a t} ) is action-value function, s _t the state at time t (environment), a _t the behavior in time t, α is the learning coefficient, r t _{+ 1} reward, γ is discount Rate. The behavior value function means an expected value of reward. The term with max is the Q value multiplied by γ when the action a having the highest Q value is selected under the environment _{st + 1} .

  It is known that the learning coefficient and the discount rate are set as 0 <α and γ ≦ 1, but here, for the sake of simplicity, if the learning coefficient and the discount rate are set to 1, the learning coefficient and the discount rate can be expressed as the following equation (2). .

This update formula is the value Q (s _{t + 1} , max a _{t + 1} ) of the best action in the next environmental state by the action a, rather than the value Q (s _t , a _t ) of the action a in the environment s. If the direction is larger, Q (s _t , a _t ) is increased, and if it is smaller, Q (s _t , a _t ) is decreased. That is, the value of a certain action in a certain state is brought close to the value of the best action in the next state. The state used in this update formula corresponds to a state variable that can be acquired from the training data set. The reward can be acquired from the reward calculation unit 14. The action is to change the operating condition of the cooling device 7, that is, the rotational speed of the cooling device 7 or the like. The action value Q (s _t , a _t ) may be stored as a table for each environment s and action a (hereinafter referred to as an action value table).

  As shown in FIG. 6, the state in the motor control device 6 includes a state that changes indirectly by action and a state that changes directly by action. The state indirectly changed by the action includes the temperature of a specific portion of the motor control device 6 (winding temperature, power element temperature, etc.), the cooling device 7, the motor 5, and the loss of the power element. The state that changes directly by action includes the rotational speed of the cooling device 7 and the operating time of the cooling device 7. In addition, in the case of cooling with liquid, the state that indirectly changes by action further includes the temperature of the refrigerant in the cooling device 7 (in the direction of flowing from the electric motor 5 to the cooling device 7). In addition, the state that directly changes depending on the behavior further includes the refrigerant flow rate of the cooling device 7 and the temperature of the refrigerant of the cooling device 7 (in the direction of flowing from the cooling device 7 to the electric motor 5).

  The learning unit 4 updates the behavior value corresponding to the current state variable and possible actions from the behavior value table based on the update formula and the reward.

  The learning unit 4 may update the action value table based on state variables and rewards of other electric motors having the same configuration as the electric motor 5 and the electric motor control device 6 and an electric motor control device (not shown).

  Next, the motor control device 6 according to the embodiment of the present invention will be described. Note that this embodiment describes the case of cooling with a liquid. The electric motor control device 6 according to the embodiment of the present invention is the electric motor control device 6 including the machine learning device 1 described above, and the rotation of the cooling device 7 based on the result learned by the learning unit 4 according to the training data set. It further includes a decision making unit 16 that gives an instruction to change operating conditions such as speed.

  The learning unit 4 is configured to relearn and update the operating condition of the cooling device 7 according to an additional training data set configured by a combination of the current state variable and the determination data.

  The machine learning device 1 may be connected to the motor control device 6 via a network, and the state observation unit 2 may be configured to acquire the current state variable via the network.

  The machine learning device 1 is preferably present in the cloud server.

  The machine learning device 1 may be incorporated in an electric motor control device 6 that controls the electric motor 5.

  The electric motor control system 100 includes the electric motor control device 6, a cooling device 7 for cooling the electric motor 5 and the electric motor control device 6, and temperature detection elements (8, 9, 10, 10 ′) that output temperature data. . The motor control device 6 receives AC power from the AC power source 20 and drives the motor 5.

  The cooling device 7 further includes a cooling device control unit 13 that controls the refrigerant flow rate and the refrigerant temperature of the cooling device 7, and the state observation unit 2 observes the refrigerant flow rate and the refrigerant temperature of the cooling device 7. The cooling device 7 further includes a rotational speed meter 21 that detects the rotational speed of the cooling device 7 and a flow meter 22 that detects the flow rate of the refrigerant in the cooling device.

  Next, a machine learning method according to an embodiment of the present invention will be described. A machine learning method according to an embodiment of the present invention is a machine learning method for learning an operating condition of a cooling device 7 for cooling an electric motor 5 or an electric motor control device 6, and includes an electric motor 5 and an electric motor control device 6. A state variable including at least one of the temperature data of the specific location is observed during the operation of the cooling device 7, and the loss of the electric motor 5, the electric motor control device 6, and the cooling device 7, and the electric motor 5 and the electric motor control device 6 are detected. Including obtaining determination data for determining a margin for the allowable value of the temperature at each specific location, and learning the operating condition of the cooling device 7 according to a training data set constituted by a combination of the state variable and the determination data It is characterized by.

  FIG. 7 shows a flowchart for explaining the operation procedure of the machine learning apparatus according to the embodiment of the present invention. First, in step S <b> 101, a state variable including at least one of temperature data of specific locations of the electric motor 5 and the electric motor control device 6 is observed during the operation of the cooling device 7.

  Next, in step S <b> 102, determination data for determining each loss of the electric motor 5, the electric motor control device 6, and the cooling device 7, and a margin for the allowable value of the temperature at each specific location of the electric motor 5 and the electric motor control device 6. To get.

  Next, in step S103, the operating conditions of the cooling device 7 are learned in accordance with a training data set configured by a combination of state variables and determination data.

  Next, a machine learning method using the motor control system according to the embodiment of the present invention will be described. FIG. 8 is a flowchart for explaining the operation procedure of the motor control system according to the embodiment of the present invention. First, in step S201, learning is started.

  Next, in step S202, operating conditions (rotation speed, operating time, etc.) of the cooling device 7 are set.

  Next, in step S203, the electric motor 5 is driven for a certain time.

  Next, in step S204, the temperature (winding temperature, etc.) of the electric motor 5 is measured, and the loss estimation unit 11 estimates the loss. Furthermore, the temperature of the power element of the motor control device 6 is measured, and the loss estimation unit 11 estimates the loss. Further, the state (rotation speed, operating time, etc.) of the cooling device 7 is measured, and the loss calculation unit 12 calculates the loss. Here, it is assumed that the loss estimation unit 11 has loss data at each temperature in advance.

  Next, in step S205, a reward is calculated based on the total of each loss (the electric motor 5, the cooling device 7, and the power element) and the temperature of each part.

  If the total loss has increased from the previous time or the temperature of each part has exceeded the allowable value, the value of the changed action is deducted in step S206. Thereafter, in step S208, the behavior value table is updated.

  On the other hand, if the total loss is reduced from the previous time and the temperature of each part is less than the allowable value, the value of the changed action is added in step S207. Thereafter, in step S208, the behavior value table is updated.

  However, the behavior value table is updated in step S208 without adjusting the value only for the first time.

  Next, in step S209, an item for changing the operating condition of the cooling device 7 is determined from the behavior value table with priority given to an item having a large behavior value score.

  Based on the operating condition of the cooling device 7 determined in step S209, the process returns to step S202 to operate the cooling device 7 so that the action value becomes the best.

  As described above, according to the machine learning device, the motor control device including the machine learning device, the motor control system, and the machine learning method according to the embodiment of the present invention, the temperature of the motor and the motor control device is determined by machine learning. Loss of the electric motor, the electric motor control device, and the cooling device can be reduced while controlling the temperature to be equal to or lower than the specified temperature.

DESCRIPTION OF SYMBOLS 1 Machine learning apparatus 2 State observation part 3 Judgment data acquisition part 4 Learning part 5 Electric motor 6 Electric motor control apparatus 7 Cooling device 71,71 ', 71 "1st cooling device 72,72', 72" 2nd cooling device 73 ', 73 ″ Third cooling device 74 ′, 74 ″ Fourth cooling device 8, 9, 10, 10 ′ Temperature detection element 11 Loss estimation unit 12 Loss calculation unit 13 Cooling device control unit 14 Reward calculation unit 15 Function update unit 16 Decision making Part 21 Tachometer 22 Flowmeter

Claims (17)

A machine learning device for learning operating conditions of a cooling device for cooling an electric motor or an electric motor control device,
A state observing unit for observing a state variable including at least one of temperature data of each specific location of the electric motor and the electric motor control device during operation of the cooling device;
A determination data acquisition unit for acquiring determination data for determining each of the loss of the electric motor, the electric motor control device, and the cooling device and a margin for an allowable value of a temperature of a specific location of each of the electric motor and the electric motor control device;
A learning unit that learns operating conditions of the cooling device according to a training data set configured by a combination of the state variable and the determination data;
A machine learning device comprising:   The temperature data includes at least one of a winding temperature of the electric motor and a power element temperature of the electric motor control device detected by a temperature detection element provided in each of the electric motor and the electric motor control device. Item 2. The machine learning device according to Item 1. A loss estimation unit that estimates the loss of the electric motor and the electric motor control device from the temperature data;
A loss calculation unit for calculating the loss of the cooling device from the operating conditions of the cooling device;
The machine learning device according to claim 1, further comprising:   The loss of the cooling device is calculated using at least one of a set of rotational speed and operating time of the cooling device, or a set of refrigerant temperature and refrigerant flow rate of the cooling device. The machine learning device according to any one of the above.   The machine learning device according to any one of claims 1 to 4, wherein the learning unit is configured to learn the operation condition according to a training data set acquired for a plurality of cooling devices. The learning unit
A reward calculation unit for calculating a reward based on the determination data;
Based on the reward, an appropriate set of rotational speed and operating time of the cooling device that reduces the sum of losses of the motor, at least one of the motor control devices and the cooling device from the current state variable, or A function updater for updating a function for estimating at least one of a set of refrigerant temperature and refrigerant flow rate of the cooling device;
The machine learning device according to claim 1, comprising:   The learning unit, based on at least one state variable of the electric motor and the electric motor control device and the reward, sets of the rotational speed and operating time of the cooling device, or sets of the refrigerant temperature and the refrigerant flow rate of the cooling device The machine learning device according to claim 6, wherein an action value table corresponding to at least one of the behavior value tables is updated.   The learning unit rotates a cooling device for cooling the other motor or the motor control device based on the state variable and the reward of another motor or the motor control device having the same configuration as the motor or the motor control device. The machine learning device according to claim 6, wherein an action value table corresponding to at least one of a set of speed and operating time or a set of refrigerant temperature and refrigerant flow rate of the cooling device is updated.   The reward calculation unit is a total of a copper loss and an iron loss of the motor, a loss of a power element of the motor control device and a loss of the cooling device, or a specific part of the motor and the motor control device. The machine learning device according to claim 7, wherein the reward is calculated based on a margin for the allowable temperature value. An electric motor control device comprising the machine learning device according to any one of claims 1 to 9,
Based on a result learned by the learning unit according to the training data set, at least one command value is determined from a set of the rotation speed and operating time of the cooling device or a set of refrigerant temperature and flow rate of the cooling device. An electric motor control device further comprising a decision making unit.   The motor control according to claim 10, wherein the learning unit is configured to relearn and update the operating condition according to an additional training data set configured by a combination of a current state variable and the determination data. apparatus. The machine learning device is connected to the motor control device via a network;
The motor control device according to claim 10 or 11, wherein the state observation unit is configured to acquire a current state variable via the network.   The motor control device according to claim 12, wherein the machine learning device exists in a cloud server.   The motor control device according to any one of claims 10 to 12, wherein the machine learning device is built in the motor control device that controls the motor. The motor control device according to any one of claims 10 to 14,
A cooling device for cooling the electric motor or the electric motor control device;
A temperature detecting element for outputting the temperature data;
An electric motor control system comprising: The cooling device further includes a cooling device control unit that controls the flow rate of the cooling device and the refrigerant temperature,
The motor control system according to claim 15, wherein the state observation unit observes a refrigerant flow rate and a refrigerant temperature of the cooling device. A machine learning method for learning operating conditions of a cooling device for cooling an electric motor or an electric motor control device,
Observing a state variable including at least one of temperature data of each specific location of the electric motor and the electric motor control device during operation of the cooling device;
Obtaining determination data for determining the margin for the loss of each of the electric motor, the electric motor control device, and the cooling device and the allowable value of the temperature of each specific location,
Learning operating conditions of the cooling device according to a training data set constituted by a combination of the state variable and the determination data,
A machine learning method characterized by including:

Images