Learning method, server control device, air conditioning control device, and learning device

Abstract

To solve the problem that a prediction model for predicting power consumption of an air conditioner installed in a server room in a facility has to be generated for each server room.SOLUTION: The air conditioner 2 is installed in a server room 92 in the facility 91. The air conditioning learning device 1 generates the prediction model M1 by learning the air conditioning feature values and the power consumption of the air conditioner 2 in association with each other. The prediction model M1 predicts the power consumption of the air conditioner 2 from the air conditioning feature amount. The air conditioner 2 includes an indoor fan 22, an air inlet 20a, and an air outlet 20b. The air conditioning characteristic amount includes the difference between the static pressure of the air in the vicinity of the air inlet 20a and the static pressure of the air in the vicinity of the air outlet 20b.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a learning method, a server control device, an air conditioning control device, and a learning device. [Background technology]

[0002] As shown in Patent Document 1 (JP 2014-234938 A), there is a technology that predicts the power consumption of air conditioners installed in a server room within a facility using a prediction model generated using information specific to the server room, such as the number of servers, the number of racks on which the servers are mounted, and the relative positions of the servers and air conditioners. Summary of the Invention [Problem to be solved by the invention]

[0003] Patent Document 1 has a problem in that a prediction model must be generated for each server room. [Means for solving the problem]

[0004] The learning method of the first aspect is performed by a learning device. The learning method is a method for learning a prediction model. The prediction model predicts the power consumption of a device. The device is installed in a server room in a facility. The learning device generates a prediction model by associating and learning a first feature amount with the power consumption of the device. The prediction model predicts the power consumption of the device from the first feature amount. The device has a first fan, an air inlet, and an air outlet. The air inlet is an opening for drawing air into the device using the first fan. The air outlet is an opening for blowing air out of the device using the first fan. The first feature amount includes the difference between the static pressure of air near the air inlet and the static pressure of air near the air outlet. The device is an air conditioner or a server. The air conditioner provides air conditioning for the server room.

[0005] In the learning method of the first aspect, the prediction model predicts the power consumption of an air conditioner or a server from a first feature amount. The first feature amount includes the difference between the static pressure of the air near the air inlet and the static pressure of the air near the air outlet. Therefore, the learning method can generate a prediction model that predicts the power consumption of an air conditioner or a server without using information specific to the server room. As a result, the learning method can generate a general-purpose prediction model specific to an air conditioner or a server that can be used regardless of the server room in which the air conditioner or server is installed.

[0006] A second aspect of the learning method is the first aspect of the learning method, wherein the device is a server, and the first feature amount further includes at least one of a processor utilization rate of the server and a temperature of air drawn in by the first fan.

[0007] With this configuration, the learning method according to the second aspect can generate a prediction model that predicts the power consumption of a server using information that can be obtained by the server owner.

[0008] A learning method according to a third aspect is the learning method according to the first aspect, wherein the device is an air conditioner. The device includes an indoor unit and an outdoor unit. The indoor unit has a first fan. The outdoor unit has a second fan. The first feature amount further includes at least one of a temperature of air blown out by the first fan, a rotation speed of the first fan, a rotation speed of the second fan, an outdoor temperature of the facility, and an outdoor humidity of the facility.

[0009] With this configuration, the learning method of the third aspect can generate a prediction model that predicts the power consumption of air conditioners using information that can be obtained by the facility owner.

[0010] The learning method of a fourth aspect is the learning method of any one of the first aspect to the third aspect, in which the facility is a data center.

[0011] A server control device according to a fifth aspect includes a first control unit. The first control unit uses a prediction model to predict power consumption of each of a plurality of servers. The prediction model is generated by the learning method according to the first, second, or fourth aspect. The first control unit preferentially allocates processing to a server with a lower predicted power consumption.

[0012] With such a configuration, the server control device according to the fifth aspect can reduce the power consumption of the server.

[0013] An air conditioning control device according to a sixth aspect includes a second control unit. The second control unit predicts the power consumption of the air conditioner using a prediction model. The prediction model is generated by the learning method according to the first, third, or fourth aspect. The second control unit controls the air conditioner based on the predicted power consumption.

[0014] With such a configuration, the air conditioning control device of the sixth aspect can reduce the power consumption of the air conditioner.

[0015] A control device according to a seventh aspect includes a control unit. The control unit uses a prediction model to predict the power consumption of each of a plurality of servers. The prediction model is generated by the learning method according to the first, second, or fourth aspect. The control unit preferentially assigns processing to a server with lower predicted power consumption. The control unit uses the prediction model to predict the power consumption of an air conditioner. The prediction model is generated by the learning method according to the first, third, or fourth aspect. The control unit controls the air conditioner based on the predicted power consumption.

[0016] With such a configuration, the control device according to the seventh aspect can reduce the power consumption of both the server and the air conditioner.

[0017] A learning device according to an eighth aspect includes a control unit. The control unit generates a prediction model by learning by associating a first feature amount with the power consumption of the device. The device is installed in a server room within a facility. The prediction model predicts the power consumption of the device from the first feature amount. The device has a first fan, an air inlet, and an air outlet. The air inlet is an opening for drawing air into the device using the first fan. The air outlet is an opening for blowing air out of the device using the first fan. The first feature amount includes a difference between the static pressure of the air near the air inlet and the static pressure of the air near the air outlet. The device is a server or an air conditioner. The air conditioner provides air conditioning for the server room.

[0018] In the learning device of the eighth aspect, the prediction model predicts the power consumption of an air conditioner or a server from a first feature amount. The first feature amount includes the difference between the static pressure of the air near the air inlet and the static pressure of the air near the air outlet. Therefore, the learning device can generate a prediction model that predicts the power consumption of an air conditioner or a server without using information specific to the server room. As a result, the learning device can generate a general-purpose prediction model specific to an air conditioner or a server that can be used regardless of the server room in which the air conditioner or server is installed. [Brief explanation of the drawings]

[0019] [Figure 1] FIG. 1 is a schematic diagram of a server room within a facility. [Figure 2] FIG. 2 is a functional block diagram of an air conditioner and an air conditioning control device. [Figure 3] FIG. 2 is a functional block diagram of the air conditioning learning device. [Figure 4] FIG. 2 is a functional block diagram of a server and a server control device. [Figure 5] FIG. 2 is a functional block diagram of a server learning device. [Figure 6] 4 is a flowchart illustrating an example of processing by an air conditioning control device and an air conditioning learning device. [Figure 7]10 is a flowchart illustrating an example of processing by a server control device and a server learning device. [Figure 8] 10 is a flowchart illustrating an example of processing by a control device and a learning device. DETAILED DESCRIPTION OF THE INVENTION

[0020] (1) Background Recently, the spread of generative AI has led to a rapid increase in demand for facilities such as data centers, and it is predicted that the amount of power consumed at these facilities will also increase sharply in the future. As a result, there is a need to efficiently control the equipment installed at these facilities and reduce the amount of power consumed at these facilities.

[0021] The power consumption of a facility is primarily composed of the power consumption of equipment such as air conditioners and lighting installed in the facility's server room, as well as the power consumption of IT equipment such as servers installed in the server room. In particular, the power consumption of air conditioners and servers accounts for the majority of the facility's power consumption. Therefore, to reduce the facility's power consumption, it is effective to generate prediction models for the power consumption of air conditioners and servers and then implement appropriate power-saving control for each based on the predicted power consumption. Meanwhile, air conditioners are controlled by the facility owner, and servers are controlled by the server owner. Furthermore, the information available to the facility owner and the information available to the server owner are generally different. Therefore, it is desirable to generate a prediction model for air conditioner power consumption using information available to the facility owner (so that the air conditioner power consumption is predicted using information available to the facility owner). Similarly, it is desirable to generate a prediction model for server power consumption using information available to the server owner (so that the server power consumption is predicted using information available to the server owner).

[0022] Furthermore, a predictive model generated using information specific to a server room, such as the number of servers, the number of racks on which the servers are mounted, and the relative positions of the servers and air conditioners, poses the problem that a predictive model must be generated for each server room.

[0023] Therefore, it is desirable that the prediction model be a general-purpose model specific to an air conditioner or a server, which can be used regardless of the type of server room in which the air conditioner or server is installed.

[0024] (2) Overall structure 1 is a schematic diagram of a server room 92 in a facility 91. In this embodiment, the facility 91 is a data center. The facility 91 may also be a school, an office, or the like.

[0025] As shown in FIG. 1, in a server room 92 in a facility 91, an indoor unit 20 constituting an air conditioner 2 (device), a plurality of servers 7 (devices), and a server control device 8 are installed.

[0026] The air conditioner 2 configures a vapor compression refrigeration cycle and conditions the air of one or more target spaces, including a server room 92, within the facility 91. In this embodiment, the air conditioner 2 is a multi-type air conditioning system for a building. The air conditioner 2 and the air conditioning control device 4 are communicatively connected via a communication line 94. The air conditioning control device 4 and the air conditioning learning device 1 are communicatively connected via a network NW1, such as the Internet.

[0027] The multiple servers 7 are installed in multiple server racks 93. In each server rack 93, multiple servers 7 are installed lined up vertically. As shown by the arrows in FIG. 1, air (cool air) blown out from the indoor unit 20 is drawn into each server 7. The air that is heated in each server 7 and blown out from each server 7 passes through the ceiling space and is drawn into the indoor unit 20.

[0028] The server control device 8 relays communications between each server 7 and external users. The server control device 8 and each server 7 are communicatively connected via a network NW2 such as a LAN. The server control device 8 is communicatively connected to the server learning device 5 and external users via a network NW3 such as the Internet.

[0029] (3) Detailed configuration (3-1) Air conditioner The air conditioner 2 has one or more refrigerant systems. Each refrigerant system has one outdoor unit 30 and one or more indoor units 20. The outdoor units 30 and indoor units 20 belonging to the same refrigerant system are connected by a liquid refrigerant communication pipe and a gas refrigerant communication pipe to form a refrigerant circuit. The outdoor units 30 and indoor units 20 belonging to the same refrigerant system are connected to each other via a communication line 95 so that they can communicate with each other.

[0030] Figure 2 is a functional block diagram of the air conditioner 2 and the air conditioning control device 4. Figures 1 and 2 show, as representatives, one outdoor unit 30 and one indoor unit 20 belonging to one refrigerant system.

[0031] (3-1-1) Indoor unit As shown in FIGS. 1 and 2, the indoor unit 20 is installed on the floor of the server room 92. The indoor unit 20 mainly has an indoor heat exchanger, an indoor fan 22 (first fan), an indoor expansion valve 23, and an indoor control unit 29. The indoor unit 20 also has an air inlet 20a and an air outlet 20b. The air inlet 20a is an opening for drawing air into the indoor unit 20 using the indoor fan 22. The air outlet 20b is an opening for blowing air out of the indoor unit 20 using the indoor fan 22. The air inlet 20a is formed on the top surface of the indoor unit 20. The air outlet 20b is formed on the side surface of the indoor unit 20. The indoor unit 20 also has various sensors, such as an indoor intake temperature sensor 61, an indoor outlet temperature sensor 62, an intake side static pressure sensor 63, and an outlet side static pressure sensor 64.

[0032] The indoor heat exchanger exchanges heat between the refrigerant flowing through it and the air in the server room 92. The indoor fan 22 draws air from the server room 92 into the indoor unit 20 through the air inlet 20a, exchanges heat between the drawn air and the refrigerant in the indoor heat exchanger, and blows the air out into the server room 92 through the air outlet 20b. The indoor expansion valve 23 is a mechanism for adjusting the pressure and flow rate of the refrigerant flowing through the refrigerant circuit. The indoor air inlet temperature sensor 61 measures the temperature of the air drawn in by the indoor fan 22. The indoor air outlet temperature sensor 62 measures the temperature of the air blown out by the indoor fan 22. The air inlet static pressure sensor 63 measures the static pressure of the air near the air inlet 20a. The air inlet static pressure sensor 63 is installed near the outside of the air inlet 20a. The air outlet static pressure sensor 64 measures the static pressure of the air near the air outlet 20b. The outlet-side static pressure sensor 64 is installed near the outside of the outlet 20b.

[0033] The indoor control unit 29 controls the operation of each component constituting the indoor unit 20. As shown in FIG. 2 , the indoor control unit 29 is communicatively connected to the indoor fan 22 and the indoor expansion valve 23. The indoor control unit 29 is also communicatively connected to various sensors, such as an indoor intake temperature sensor 61, an indoor discharge temperature sensor 62, an intake-side static pressure sensor 63, and a discharge-side static pressure sensor 64. The indoor control unit 29 has a control and arithmetic device and a storage device. The control and arithmetic device is a processor such as a CPU or a GPU. The storage device is a storage medium such as RAM, ROM, or flash memory. The control and arithmetic device reads programs stored in the storage device and performs predetermined arithmetic processing in accordance with the programs, thereby controlling the operation of each component constituting the indoor unit 20. The control and arithmetic device can also write arithmetic results to the storage device and read information stored in the storage device in accordance with the programs. The indoor control unit 29 is configured to receive various signals transmitted from an operation remote control corresponding to the indoor unit 20. Furthermore, the indoor control unit 29 exchanges various information such as control signals, signals relating to measurements by various sensors, and signals relating to various settings with the outdoor control unit 39 of the outdoor unit 30 via a communication line 95.

[0034] (3-1-2) Outdoor unit 1 and 2, the outdoor unit 30 is installed outdoors, such as on the roof of a facility 91. The outdoor unit 30 mainly has a compressor 31, a flow path switching valve 32, an outdoor heat exchanger, an outdoor expansion valve 34, an outdoor fan 36 (second fan), and an outdoor control unit 39. The outdoor unit 30 also has various sensors, such as an outdoor temperature sensor 66 and an outdoor humidity sensor 67.

[0035] The compressor 31 draws low-pressure refrigerant through the suction pipe, compresses the refrigerant using a compression mechanism, and discharges the compressed refrigerant to a discharge pipe. The flow path switching valve 32 is a mechanism that switches the refrigerant flow path between a first state and a second state. During cooling operation, the flow path switching valve 32 sets the refrigerant flow path to the first state. At this time, the refrigerant discharged from the compressor 31 flows through the refrigerant circuit in the following order: the outdoor heat exchanger, the outdoor expansion valve 34, the indoor expansion valve 23, and the indoor heat exchanger, before returning to the compressor 31. In the first state, the outdoor heat exchanger functions as a condenser, and the indoor heat exchanger functions as an evaporator. During heating operation, the flow path switching valve 32 sets the refrigerant flow path to the second state. At this time, the refrigerant discharged from the compressor 31 flows through the refrigerant circuit in the following order: the indoor heat exchanger, the indoor expansion valve 23, the outdoor expansion valve 34, and the outdoor heat exchanger, before returning to the compressor 31. In the second state, the outdoor heat exchanger functions as an evaporator, and the indoor heat exchanger functions as a condenser. The outdoor heat exchanger exchanges heat between the refrigerant flowing therethrough and the outdoor air of the facility 91. The outdoor expansion valve 34 is a mechanism for adjusting the pressure and flow rate of the refrigerant flowing through the refrigerant circuit. The outdoor fan 36 draws outdoor air into the outdoor unit 30, exchanges heat between the drawn air and the refrigerant in the outdoor heat exchanger, and then blows the air outdoors. The outdoor temperature sensor 66 measures the temperature of the air drawn in by the outdoor fan 36 (the outdoor temperature of the facility 91). The outdoor humidity sensor 67 measures the humidity of the air drawn in by the outdoor fan 36 (the outdoor humidity of the facility 91).

[0036] The outdoor control unit 39 controls the operation of each component constituting the outdoor unit 30. As shown in FIG. 2 , the outdoor control unit 39 is communicatively connected to the compressor 31, the flow path switching valve 32, the outdoor expansion valve 34, and the outdoor fan 36. The outdoor control unit 39 is also communicatively connected to various sensors, such as the outdoor temperature sensor 66 and the outdoor humidity sensor 67. The outdoor control unit 39 has a control and arithmetic device and a storage device. The control and arithmetic device is a processor such as a CPU or a GPU. The storage device is a storage medium such as RAM, ROM, or flash memory. The control and arithmetic device reads programs stored in the storage device and performs predetermined arithmetic processing in accordance with the programs, thereby controlling the operation of each component constituting the outdoor unit 30. The control and arithmetic device can also write arithmetic results to the storage device and read information stored in the storage device in accordance with the programs. The outdoor control unit 39 exchanges various information, such as control signals, signals related to measurements by various sensors, and signals related to various settings, with the indoor control unit 29 of the indoor unit 20 via a communication line 95. The outdoor control unit 39 also exchanges various types of information with the air conditioning control device 4 via a communication line 94, such as control signals, signals related to measurements by various sensors, and signals related to various settings.

[0037] (3-1-3) Air conditioning control device The air conditioning control device 4 centrally controls one or more refrigerant systems. The air conditioning control device 4 is, for example, a device called an Edge. The air conditioning control device 4 is installed, for example, in a computer room within the facility 91. As shown in FIG. 2, the air conditioning control device 4 mainly has a memory unit 41, a communication unit 44, and a control unit 49 (second control unit).

[0038] The storage unit 41 is a storage medium such as RAM, ROM, or flash memory. The storage unit 41 stores programs executed by the control unit 49, data necessary for executing the programs, etc. The communication unit 44 includes a network interface device for communicating with the air conditioning learning device 1 via the network NW1, and a network interface device for communicating with the air conditioner 2 via the communication line 94.

[0039] The control unit 49 is a processor such as a CPU or GPU. The control unit 49 centrally controls one or more refrigerant systems by reading out programs stored in the storage unit 41 and performing predetermined calculations in accordance with the programs. The control unit 49 can also write calculation results to the storage unit 41 and read out information stored in the storage unit 41 in accordance with the programs. The control unit 49 exchanges various types of information, such as control signals, signals related to measurements by various sensors, and signals related to various settings, with the air conditioner 2 via a communication line 94. The control unit 49 also exchanges various types of information, such as control signals, signals related to measurements by various sensors, and signals related to various settings, with the air conditioning learning device 1 via a network NW1.

[0040] The control unit 49 periodically (for example, once an hour) acquires from the air conditioner 2, as operating data D1, the rotation speed of the indoor fan 22, the opening degree of the indoor expansion valve 23, the indoor intake temperature (measured value of the indoor intake temperature sensor 61), the indoor discharge temperature (measured value of the indoor discharge temperature sensor 62), the rotation speed of the compressor 31, the opening degree of the outdoor expansion valve 34, the rotation speed of the outdoor fan 36, the outdoor temperature (measured value of the outdoor temperature sensor 66), and the outdoor humidity (measured value of the outdoor humidity sensor 67). The control unit 49 also periodically calculates, as operating data D1, the static pressure difference of the indoor unit 20. The static pressure difference of the indoor unit 20 is the difference between the measurement value of the suction side static pressure sensor 63 and the measurement value of the discharge side static pressure sensor 64. The control unit 49 also periodically acquires the power consumption of the air conditioner 2 as operating data D1. The power consumption of the air conditioner 2 may be obtained from an electric power company via the facility owner, for example, or may be calculated from the rotation speed of the compressor 31. The control unit 49 stores the obtained operating data D1 in the memory unit 41.

[0041] The control unit 49 periodically transmits the acquired operating data D1 to the air conditioning learning device 1. For example, the control unit 49 transmits the acquired operating data D1 to the air conditioning learning device 1 every time it acquires operating data D1.

[0042] The control unit 49 acquires the prediction model M1 generated by the air conditioning learning device 1 from the air conditioning learning device 1. Each time a prediction model M1 is generated by the air conditioning learning device 1, the control unit 49 acquires the generated prediction model M1 from the air conditioning learning device 1. The control unit 49 stores the acquired prediction model M1 in the memory unit 41.

[0043] The control unit 49 uses the prediction model M1 to predict the current or future power consumption of the air conditioner 2. The control unit 49 controls the air conditioner 2 based on the predicted power consumption. For example, the control unit 49 adjusts the rotation speed of the indoor fan 22, the opening of the indoor expansion valve 23, the rotation speed of the compressor 31, the opening of the outdoor expansion valve 34, etc., by using MPC (Model Predictive Control) or regression predictive control using the prediction model M1, so that the indoor discharge temperature falls within a predetermined temperature range while suppressing the power consumption of the air conditioner 2. The control unit 49 uses, for example, the value of the most recent operating data D1 or the value of future operating data D1 predicted from the operating data D1 using machine learning or the like, as the air conditioning feature quantity to be input to the prediction model M1 (described later).

[0044] (3-2) Air conditioning learning device The air conditioning learning device 1 is installed, for example, on the cloud. FIG. 3 is a functional block diagram of the air conditioning learning device 1. As shown in FIG. 3, the air conditioning learning device 1 mainly has a memory unit 11, a communication unit 14, and a control unit 19. In this embodiment, the functions of the air conditioning learning device 1 described below are realized by a single device. However, the functions of the air conditioning learning device 1 may also be realized in a distributed manner by multiple devices.

[0045] The storage unit 11 is a storage medium such as RAM, ROM, or flash memory. The storage unit 11 stores programs executed by the control unit 19, data necessary for executing the programs, etc. The communication unit 14 includes a network interface device for communicating with the air conditioning control device 4 via the network NW1.

[0046] (3-2-1) Control Unit The control unit 19 is a processor such as a CPU or GPU. The control unit 19 reads and executes programs stored in the storage unit 11 to realize various functions of the air conditioning learning device 1. The control unit 19 can also write calculation results to the storage unit 11 and read information stored in the storage unit 11 according to the programs.

[0047] As shown in FIG. 3, the control unit 19 has an acquisition unit 191, a generation unit 192, and a transmission unit 193 as functional blocks.

[0048] (3-2-1-1) Acquisition department The acquisition unit 191 periodically acquires the operating data D1 from the air conditioning control device 4. The acquisition unit 191 stores the acquired operating data D1 in the storage unit 11.

[0049] (3-2-1-2) Generation part The generation unit 192 generates a prediction model M1 by periodically (for example, once a week) associating and learning the air conditioning feature (first feature) with the power consumption of the air conditioner 2. The prediction model M1 predicts the power consumption of the air conditioner 2 from the air conditioning feature.

[0050] The air conditioning feature quantity includes the static pressure difference of the indoor unit 20. The air conditioning feature quantity may further include at least one of the indoor discharge temperature, the rotation speed of the indoor fan 22, the rotation speed of the outdoor fan 36, the outdoor temperature, and the outdoor humidity.

[0051] The generation unit 192 acquires the air conditioning feature amount and the power consumption of the air conditioner 2 from the operation data D1.

[0052] The predictive model M1 is, for example, a regression, machine learning model, or a statistical model.

[0053] The generation unit 192 stores the generated prediction model M1 in the storage unit 11.

[0054] (3-2-1-3) Transmitter The transmission unit 193 transmits the generated prediction model M1 to the air conditioning control device 4 every time the generation unit 192 generates a prediction model M1.

[0055] (3-3) Server FIG. 4 is a functional block diagram of a server 7 and a server control device 8. FIG. 4 shows one server 7 as a representative. As shown in FIG. 4, the server 7 mainly has a memory unit 71, a server fan 72 (first fan), a communication unit 74, and a control unit 79. The server 7 also has an intake port 70a and an outlet port 70b. The intake port 70a is an opening for drawing air into the server 7 using the server fan 72. The outlet port 70b is an opening for blowing air out of the server 7 using the server fan 72. The server 7 also has various sensors, such as an intake temperature sensor 65, an intake side static pressure sensor 68, and an outlet side static pressure sensor 69.

[0056] The memory unit 71 is a storage medium such as RAM, ROM, or flash memory. The memory unit 71 stores programs executed by the control unit 79, data necessary for executing the programs, and the like. The communication unit 74 includes a network interface device for communicating with the server control device 8 via the network NW2. The server fan 72 draws air from the server room 92 into the server 7 through the air inlet 70a, cools the inside of the server 7 with the drawn air, and blows the air out into the server room 92 through the air outlet 70b. The air inlet temperature sensor 65 measures the temperature of the air drawn in by the server fan 72. The air inlet side static pressure sensor 68 measures the static pressure of the air near the air inlet 70a. The air inlet side static pressure sensor 68 is installed near the outside of the air inlet 70a. The air outlet side static pressure sensor 69 measures the static pressure of the air near the air outlet 70b. The air outlet side static pressure sensor 69 is installed near the outside of the air outlet 70b.

[0057] The control unit 79 is a processor such as a CPU or a GPU. The control unit 79 reads and executes programs stored in the storage unit 71 to realize various functions of the server 7. The control unit 79 can also write calculation results to the storage unit 71 and read information stored in the storage unit 71 according to the programs. The control unit 79 exchanges various information, such as control signals, signals related to measurements by various sensors, and signals related to various settings, with the server control device 8 via the network NW2. For example, the control unit 79 receives a task from an external user via the server control device 8. The control unit 79 processes the received task and transmits the processing results to the external user via the server control device 8.

[0058] (3-4) Server control device As shown in FIG. 4, the server control device 8 mainly includes a storage unit 81, a communication unit 84, and a control unit 89 (first control unit).

[0059] The storage unit 81 is a storage medium such as RAM, ROM, or flash memory. The storage unit 81 stores programs executed by the control unit 89, data necessary for executing the programs, etc. The communication unit 84 includes a network interface device for communicating with each server 7 via the network NW2, and a network interface device for communicating with the server learning device 5 and external users via the network NW3.

[0060] The control unit 89 is a processor such as a CPU or a GPU. The control unit 89 reads and executes programs stored in the storage unit 81 to realize various functions of the server control device 8. The control unit 89 can also write calculation results to the storage unit 81 and read information stored in the storage unit 81 according to the programs. The control unit 89 exchanges various information, such as control signals, signals related to measurements by various sensors, and signals related to various settings, with each server 7 via the network NW2. The control unit 89 also exchanges various information, such as control signals, signals related to measurements by various sensors, and signals related to various settings, with the server learning device 5 via the network NW3.

[0061] The control unit 89 periodically (for example, once an hour) acquires the processor utilization rate of each server 7 and the server suction temperature (measurement value of the suction temperature sensor 65) from each server 7 as operating data D2. The control unit 89 also periodically calculates the static pressure difference of each server 7 as operating data D2. The static pressure difference of a server 7 is the difference between the measurement value of the suction side static pressure sensor 68 and the measurement value of the discharge side static pressure sensor 69. The control unit 89 also periodically acquires the power consumption of each server 7 as operating data D2. The power consumption of each server 7 is acquired from a power company, for example, via the owner of each server 7. The control unit 89 stores the acquired operating data D2 in the memory unit 81.

[0062] The control unit 89 periodically transmits the acquired driving data D2 to the server learning device 5. For example, the control unit 89 transmits the acquired driving data D2 to the server learning device 5 every time it acquires driving data D2.

[0063] The control unit 89 acquires the prediction model M2 generated by the server learning device 5 from the server learning device 5. For example, each time a prediction model M2 is generated by the server learning device 5, the control unit 89 acquires the generated prediction model M2 from the server learning device 5. The control unit 89 stores the acquired prediction model M2 in the memory unit 81.

[0064] The control unit 89 uses the prediction model M2 to predict the power consumption of each of the multiple servers 7. The control unit 89 controls each server 7 based on the predicted power consumption. For example, when the control unit 89 receives a task for multiple servers 7 owned by the same person, it predicts the current or future power consumption of each server 7 and preferentially allocates task processing to the server 7 with the lowest predicted power consumption. The control unit 89 uses, for example, the value of the most recent operating data D2 or the value of future operating data D1 predicted from the operating data D2 by machine learning or the like as the server feature to be input into the prediction model M2 (described later).

[0065] (3-5) Server learning device The server learning device 5 is installed, for example, on a cloud. FIG. 5 is a functional block diagram of the server learning device 5. As shown in FIG. 5, the server learning device 5 mainly includes a memory unit 51, a communication unit 54, and a control unit 59. In this embodiment, the functions of the server learning device 5 described below are realized by a single device. However, the functions of the server learning device 5 may also be realized in a distributed manner by multiple devices.

[0066] The storage unit 51 is a storage medium such as RAM, ROM, or flash memory. The storage unit 51 stores programs executed by the control unit 59, data necessary for executing the programs, etc. The communication unit 54 includes a network interface device for communicating with the server control device 8 via the network NW3.

[0067] (3-5-1) Control Unit The control unit 59 is a processor such as a CPU or a GPU. The control unit 59 reads and executes programs stored in the storage unit 51 to realize various functions of the server learning device 5. The control unit 59 can also write calculation results to the storage unit 51 and read information stored in the storage unit 51 according to the programs.

[0068] As shown in FIG. 5, the control unit 59 has, as functional blocks, an acquisition unit 591, a generation unit 592, and a transmission unit 593.

[0069] (3-5-1-1) Acquisition department The acquisition unit 591 periodically acquires the driving data D2 from the server control device 8. The acquisition unit 591 stores the acquired driving data D2 in the storage unit 51.

[0070] (3-5-1-2) Generation part The generation unit 592 periodically (for example, once a week) generates a prediction model M2 for each server 7 by learning the server feature (first feature) in association with the power consumption of the server 7. The prediction model M2 predicts the power consumption of the server 7 from the server feature.

[0071] The server feature amount includes a static pressure difference of the server 7. The server feature amount may further include at least one of a processor utilization rate of the server 7 and a server intake temperature.

[0072] The generation unit 592 acquires the server feature amount and the power consumption of the server 7 from the operation data D2.

[0073] The predictive model M2 is, for example, a regression-type, machine learning model, or a statistical model.

[0074] The generation unit 592 stores the generated prediction model M2 in the storage unit 51.

[0075] (3-5-1-3) Transmitter The transmission unit 593 transmits the generated prediction model M2 to the server control device 8 every time the generation unit 592 generates the prediction model M2.

[0076] (4) Processing (4-1) Processing of air conditioning control device and air conditioning learning device An example of the processing of the air conditioning control device 4 and the air conditioning learning device 1 will be described using the flowchart in FIG.

[0077] As shown in step S11, the air conditioning control device 4 acquires operation data D1 from the air conditioner 2 and the like.

[0078] After step S11 is completed, the air conditioning control device 4 transmits the acquired operating data D1 to the air conditioning learning device 1, as shown in step S12.

[0079] When proceeding from step S12 to step S13, the air conditioning learning device 1 determines whether or not a first period (for example, one week) has passed since the previous generation of the prediction model M1. If the first period has passed, proceed to step S14. If the first period has not passed, end the processing.

[0080] When proceeding from step S13 to step S14, the air conditioning learning device 1 generates a prediction model M1 by learning the air conditioning features and the power consumption of the air conditioner 2 in association with each other from the time the previous prediction model M1 was generated until a first period has elapsed.

[0081] After completing step S14, the air conditioning learning device 1 transmits the generated prediction model M1 to the air conditioning control device 4, as shown in step S15.

[0082] After step S15 is completed, as shown in step S16, the air conditioning control device 4 updates the prediction model M1 that controls the air conditioner 2. Specifically, the air conditioning control device 4 uses the latest prediction model M1 to predict the power consumption of the air conditioner 2. The air conditioning control device 4 controls the air conditioner 2 based on the predicted power consumption.

[0083] When the process proceeds from step S12 to step S17, the air conditioning control device 4 determines whether a second period (for example, one hour) has elapsed since the previous acquisition of the operating data D1. If the second period has elapsed, the process returns to step S11. If the second period has not elapsed, the process waits until the second period has elapsed.

[0084] (4-2) Processing of the server control device and the server learning device An example of the processing of the server control device 8 and the server learning device 5 will be described with reference to the flowchart of FIG.

[0085] As shown in step S21, the server control device 8 acquires the operating data D2 from each server 7 and the like.

[0086] After completing step S21, the server control device 8 transmits the acquired driving data D2 to the server learning device 5 as shown in step S22.

[0087] When the process proceeds from step S22 to step S23, the server learning device 5 determines whether a third period (e.g., one week) has passed since the previous generation of the prediction model M2. If the third period has passed, the process proceeds to step S24. If the third period has not passed, the process ends.

[0088] When proceeding from step S23 to step S24, the server learning device 5 generates a prediction model M2 by associating and learning the server features with the power consumption of the server 7 from the time the previous prediction model M2 was generated until the third period has elapsed.

[0089] After completing step S24, the server learning device 5 transmits the generated prediction model M2 to the server control device 8, as shown in step S25.

[0090] After step S25 is completed, as shown in step S26, the server control device 8 updates the prediction model M2 that controls each server 7. Specifically, the server control device 8 uses the latest prediction model M2 to predict the power consumption of each of the multiple servers 7. The server control device 8 preferentially assigns processing to a server 7 with the lowest predicted power consumption.

[0091] When the process proceeds from step S22 to step S27, the server control device 8 determines whether a fourth period (for example, one hour) has elapsed since the previous acquisition of the operating data D2. If the fourth period has elapsed, the process returns to step S21. If the fourth period has not elapsed, the process waits until the fourth period has elapsed.

[0092] (5) Features (5-1) Conventionally, there is a technology that predicts the power consumption of air conditioners installed in server rooms within a facility using a prediction model generated using information specific to the server room, such as the number of servers, the number of racks on which the servers are mounted, and the relative positions of the servers and air conditioners.The problem with this conventional technology is that a prediction model must be generated for each server room.

[0093] Furthermore, in order to reduce power consumption in a facility, it is effective to generate a prediction model for the power consumption of each air conditioner and server, and to perform appropriate power-saving control for each air conditioner and server based on the predicted power consumption. In this case, since the information available to the facility owner and the information available to the server owner generally differ, it is desirable that the prediction model for predicting the power consumption of the air conditioner be generated using information available to the facility owner, and the prediction model for predicting the power consumption of the server be generated using information available to the server owner.

[0094] The learning method of this embodiment is performed by the air conditioning learning device 1 or the server learning device 5 (learning device). The learning method is a learning method for prediction models M1 and M2. The prediction models M1 and M2 predict the power consumption of the equipment. The equipment is installed in a server room 92 within a facility 91. The air conditioning learning device 1 or the server learning device 5 (learning device) generates the prediction models M1 and M2 by associating and learning air conditioning features or server features (first features) with the power consumption of the equipment. The prediction models M1 and M2 predict the power consumption of the equipment from the air conditioning features or server features (first features). The equipment has an indoor fan 22 or a server fan 72 (first fan), air inlets 20a and 70a, and air outlets 20b and 70b. The air inlets 20a and 70a are openings for drawing air into the equipment using the indoor fan 22 or the server fan 72 (first fan). The air outlets 20b, 70b are openings for blowing air out from inside the equipment using the indoor fan 22 or the server fan 72 (first fan). The air conditioning feature or server feature (first feature) includes the difference between the static pressure of the air near the air inlets 20a, 70a and the static pressure of the air near the air outlets 20b, 70b. The equipment is an air conditioner 2 or a server 7. The air conditioner 2 conditions the air in the server room 92.

[0095] In the learning method of this embodiment, the prediction models M1, M2 predict the power consumption of the server 7 or the air conditioner 2 from air conditioning features or server features. The air conditioning features or server features include the difference between the static pressure of the air near the air inlets 20a, 70a and the static pressure of the air near the air outlets 20b, 70b. Therefore, the learning method can generate prediction models M1, M2 that predict the power consumption of the air conditioner 2 or the server 7 without using information specific to the server room 92. As a result, the learning method can generate general-purpose prediction models M1, M2 specific to the air conditioner 2 or the server 7 that can be used regardless of the server room 92 in which the air conditioner 2 or the server 7 is installed.

[0096] Furthermore, the facility owner can obtain information about the difference between the static pressure of the air near the air inlet 20a and the static pressure of the air near the air outlet 20b, and the power consumption of the air conditioner 2. As a result, the learning method can generate a prediction model M1 that predicts the power consumption of the air conditioner 2 using information that the facility owner can obtain.

[0097] Furthermore, the difference between the static pressure of the air near the air inlet 70a and the static pressure of the air near the air outlet 70b, and the power consumption of the server 7 are information that can be obtained by the server owner. As a result, the learning method can generate a prediction model M2 that predicts the power consumption of the server 7 using the information that can be obtained by the server owner.

[0098] (5-2) In the learning method of this embodiment, the device is a server 7. The server feature (first feature) further includes at least one of the processor usage rate of the server 7 and the temperature of the air drawn in by the server fan 72 (first fan).

[0099] As a result, the learning method can generate a generic prediction model M2 specific to the server 7 that can be used no matter what server room 92 the server 7 is installed in.

[0100] Furthermore, the processor utilization rate of the server 7 and the temperature of the air drawn in by the server fan 72 are information that can be obtained by the server owner. As a result, the learning method can generate a more accurate prediction model M2 using information that can be obtained by the server owner.

[0101] (5-3) In the learning method of this embodiment, the equipment is an air conditioner 2. The equipment includes an indoor unit 20 and an outdoor unit 30. The indoor unit 20 has an indoor fan 22 (first fan). The outdoor unit 30 has an outdoor fan 36 (second fan). The air conditioning feature (first feature) further includes at least one of the temperature of the air blown out by the indoor fan 22 (first fan), the rotation speed of the indoor fan 22 (first fan), the rotation speed of the outdoor fan 36 (second fan), the outdoor temperature of the facility 91, and the outdoor humidity of the facility 91.

[0102] As a result, the learning method can generate a general-purpose prediction model M1 specific to the air conditioner 2 that can be used regardless of the type of server room 92 the air conditioner 2 is installed in.

[0103] Furthermore, the temperature of the air blown out by the indoor fan 22, the rotation speed of the indoor fan 22, the rotation speed of the outdoor fan 36, the outdoor temperature of the facility 91, and the outdoor humidity of the facility 91 are information that can be obtained by the facility owner. As a result, the learning method can generate a more accurate prediction model M1 using information that can be obtained by the facility owner.

[0104] (5-4) In the learning method of this embodiment, the facility 91 is a data center.

[0105] (5-5) The server control device 8 of this embodiment includes a control unit 89 (first control unit). The control unit 89 (first control unit) uses a prediction model M2 to predict the power consumption of each of the multiple servers 7. The prediction model M2 is generated by the learning method of this embodiment. The control unit 89 (first control unit) preferentially allocates processing to a server 7 with low predicted power consumption.

[0106] As a result, the server control device 8 can reduce the power consumption of each server 7.

[0107] (5-6) The air conditioning control device 4 of this embodiment includes a control unit 49 (second control unit). The control unit 49 (second control unit) predicts the power consumption of the air conditioner 2 using a prediction model M1. The prediction model M1 is generated by the learning method of this embodiment. The control unit 49 (second control unit) controls the air conditioner 2 based on the predicted power consumption.

[0108] As a result, the air conditioning control device 4 can control the power consumption of the air conditioner 2.

[0109] (5-7) The air conditioning learning device 1 or server learning device 5 (learning device) of this embodiment includes a control unit 19, 59. The control unit 19, 59 generates prediction models M1, M2 by associating and learning air conditioning features or server features (first features) with the power consumption of the equipment. The equipment is installed in a server room 92 within a facility 91. The prediction models M1, M2 predict the power consumption of the equipment from the air conditioning features or server features (first features). The equipment has an indoor fan 22 or a server fan 72 (first fan), air inlets 20a, 70a, and air outlets 20b, 70b. The air inlets 20a, 70a are openings through which air is drawn into the equipment using the indoor fan 22 or the server fan 72 (first fan). The air outlets 20b, 70b are openings through which air is blown out of the equipment using the indoor fan 22 or the server fan 72 (first fan). The air conditioning feature or server feature (first feature) includes the difference between the static pressure of the air near the air inlets 20a, 70a and the static pressure of the air near the air outlets 20b, 70b. The device is an air conditioner 2 or a server 7. The air conditioner 2 conditions the air in the server room 92.

[0110] In the air conditioning learning device 1 or server learning device 5 of this embodiment, the prediction models M1, M2 predict the power consumption of the air conditioner 2 or server 7 from air conditioning features or server features (first features). The air conditioning features or server features (first features) include the difference between the static pressure of the air near the air inlets 20a, 70a and the static pressure of the air near the air outlets 20b, 70b. Therefore, the air conditioning learning device 1 or server learning device 5 can generate prediction models M1, M2 that predict the power consumption of the air conditioner 2 or server 7 without using information specific to the server room 92. As a result, the air conditioning learning device 1 or server learning device 5 can generate general-purpose prediction models M1, M2 specific to the air conditioner 2 or server 7 that can be used regardless of the server room 92 in which the air conditioner 2 or server 7 is installed.

[0111] Furthermore, the facility owner can obtain information about the difference between the static pressure of the air near the air inlet 20a and the static pressure of the air near the air outlet 20b, and the power consumption of the air conditioner 2. As a result, the air conditioning learning device 1 can generate a prediction model M1 that predicts the power consumption of the air conditioner 2 using information that the facility owner can obtain.

[0112] The server owner can obtain information about the difference between the static pressure of the air near the air inlet 70a and the static pressure of the air near the air outlet 70b, and the power consumption of the server 7. As a result, the server learning device 5 can generate a prediction model M2 that predicts the power consumption of the server 7 using the information that the server owner can obtain.

[0113] (6) Variations (6-1) Variation 1A If a server owner owns multiple servers 7, the air conditioning feature quantities in the prediction model M1 that predicts the power consumption of each of these multiple servers 7 may further include the power consumption of the server 7 on a server owner basis. As a result, the air conditioning learning device 1 can generate a general-purpose prediction model M1 specific to the air conditioner 2 that can be used regardless of the server room 92 in which the air conditioner 2 is installed.

[0114] Furthermore, the power consumption of the server 7 on a server owner basis is information that can be acquired by the facility owner. As a result, the air conditioning learning device 1 can generate a more accurate prediction model M1 using information that can be acquired by the facility owner.

[0115] (6-2) Variation 1B In this embodiment, the air conditioner 2 is a multi-type air conditioning system for a building. However, the air conditioner 2 may also be a central air conditioning system having a chiller. In this case, the air conditioning feature may further include at least one of the chiller water temperature and the chiller water flow rate. For example, the air conditioning control device 4 periodically acquires the chiller water temperature and the chiller water flow rate from the air conditioner 2 as the operating data D1. As a result, the air conditioning learning device 1 can generate a general-purpose prediction model M1 specific to the air conditioner 2 that can be used regardless of the server room 92 in which the air conditioner 2 is installed.

[0116] Furthermore, the chiller water temperature and chiller water flow rate are information that can be obtained by the facility owner. As a result, the air conditioning learning device 1 can generate a more accurate prediction model M1 using information that can be obtained by the facility owner.

[0117] (6-3) Variation 1C When the facility owner and the server owner are the same, such as an ASP (Application Service Provider) that owns its own facilities, the control unit 49 of the air conditioning control device 4 and the control unit 89 of the server control device 8 may work together to function as the control unit 209 of the control device 200. The control unit 19 of the air conditioning learning device 1 and the control unit 59 of the server learning device 5 may work together to function as the control unit 309 of the learning device 300. As a result, the control device 200 can reduce the power consumption of both the server 7 and the air conditioner 2.

[0118] An example of the processing of the control device 200 and the learning device 300 will be described with reference to the flowchart of FIG.

[0119] As shown in step S31, the control device 200 acquires operation data D1 from the air conditioner 2 or the like. Alternatively, the control device 200 acquires operation data D2 from each server 7 or the like.

[0120] After completing step S31, as shown in step S32, if the control device 200 acquires driving data D1, it transmits the acquired driving data D1 to the learning device 300. If the control device 200 acquires driving data D2, it transmits the acquired driving data D2 to the learning device 300.

[0121] When proceeding from step S32 to step S33, if the learning device 300 has acquired driving data D1, it determines whether a first period (e.g., one week) has elapsed since the previous generation of the prediction model M1. If the first period has elapsed, it proceeds to step S34. If the first period has not elapsed, it terminates the processing. If the learning device 300 has acquired driving data D2, it determines whether a third period (e.g., one week) has elapsed since the previous generation of the prediction model M2. If the third period has elapsed, it proceeds to step S34. If the third period has not elapsed, it terminates the processing.

[0122] Proceeding from step S33 to step S34, if a first period has passed since the generation of the previous prediction model M1, the learning device 300 generates a prediction model M1 by learning the air conditioning features and the power consumption of the air conditioner 2 in association with each other from the generation of the previous prediction model M1 until the first period has passed since the generation of the previous prediction model M1. If a third period has passed since the generation of the previous prediction model M2, the learning device 300 generates a prediction model M2 by learning the server features and the power consumption of the server 7 in association with each other from the generation of the previous prediction model M2 until the third period has passed since the generation of the previous prediction model M2.

[0123] After completing step S34, as shown in step S35, if the learning device 300 has generated a prediction model M1, it transmits the generated prediction model M1 to the control device 200. If the learning device 300 has generated a prediction model M2, it transmits the generated prediction model M2 to the control device 200.

[0124] After completing step S35, as shown in step S36, if the control device 200 acquires a prediction model M1 from the learning device 300, it updates the prediction model M1 that controls the air conditioner 2. Specifically, the control device 200 predicts the power consumption of the air conditioner 2 using the latest prediction model M1. The control device 200 controls the air conditioner 2 based on the predicted power consumption. If the control device 200 acquires a prediction model M2 from the learning device 300, it updates the prediction model M2 that controls each server 7. Specifically, the control device 200 predicts the power consumption of each of the multiple servers 7 using the latest prediction model M2. The control device 200 preferentially assigns processing to servers 7 with lower predicted power consumption.

[0125] When proceeding from step S32 to step S37, if the control device 200 has transmitted the operating data D1, it determines whether a second period (e.g., one hour) has elapsed since the previous acquisition of the operating data D1. If the second period has elapsed, it returns to step S31. If the second period has not elapsed, it waits until the second period has elapsed. If the control device 200 has transmitted the operating data D2, it determines whether a fourth period (e.g., one hour) has elapsed since the previous acquisition of the operating data D2. If the fourth period has elapsed, it returns to step S31. If the fourth period has not elapsed, it waits until the fourth period has elapsed.

[0126] (6-4) Although the embodiments of the present disclosure have been described above, it will be understood that various changes in form and details can be made without departing from the spirit and scope of the present disclosure as defined in the claims. [Explanation of symbols]

[0127] 1,300 Air conditioning learning device (learning device) 2. Air conditioner (equipment) 4,200 Air conditioning control device (control device) 5,300 Server Learning Devices (Learning Devices) 7 Server (device) 8,200 Server control unit (control unit) 19,59,309 Control section 20 Indoor unit 20a, 70a suction port 20b,70b outlet 22 Indoor fan 22 (first fan) 30 Outdoor unit 49,209 Control section (second control section) 72 Server fan 72 (first fan) 89,209 Control section (first control section) 91 facilities 92 Server Room M1,M2 forecast model [Prior art documents] [Patent documents]

[0128] [Patent Document 1] Japanese Patent Application Laid-Open No. 2014-234938

Claims

1. A learning method for a prediction model (M1, M2) for predicting power consumption of equipment (2, 7) installed in a server room (92) in a facility (91), performed by a learning device (1, 5, 300), comprising: the learning device generates the prediction model that predicts the power consumption of the device from the first feature amount by learning the first feature amount and the power consumption of the device in association with each other; The device has a first fan (22, 72), an intake port (20a, 70a) for drawing air into the device using the first fan, and an outlet port (20b, 70b) for blowing air out of the device using the first fan, the first feature amount includes a difference between a static pressure of air in the vicinity of the air inlet and a static pressure of air in the vicinity of the air outlet, The device is an air conditioner (2) that conditions the air in the server room or a server (7). How to learn.

2. The device is a server (7), The first feature amount further includes at least one of a processor utilization rate of the server (7) and a temperature of air drawn in by the first fan. The learning method according to claim 1 .

3. The device is an air conditioner (2), The device includes an indoor unit (20) having the first fan and an outdoor unit (30) having a second fan, the first feature amount further includes at least one of a temperature of air blown out by the first fan, a rotation speed of the first fan, a rotation speed of the second fan, an outdoor temperature of the facility, and an outdoor humidity of the facility; The learning method according to claim 1 .

4. the facility is a data center; A learning method according to any one of claims 1 to 3.

5. A first control unit (89, 209) is provided, The first control unit predicting the power consumption of each of a plurality of servers (7) using the prediction model generated by the learning method according to claim 1 or 2; Prioritize allocation of processing to a server (7) with a small predicted power consumption. Server control device (8, 200).

6. A second control unit (49, 209) is provided, The second control unit is predicting power consumption of an air conditioner (2) using the prediction model generated by the learning method according to claim 1 or 3; Controlling the air conditioner (2) based on the predicted power consumption. Air conditioning control device (4, 200).

7. A control unit (209) is provided, The control unit predicting the power consumption of each of a plurality of servers (7) using the prediction model generated by the learning method according to claim 1 or 2; Prioritize allocation of processing to a server (7) with the lowest predicted power consumption; predicting power consumption of an air conditioner (2) using the prediction model generated by the learning method according to claim 1 or 3; Controlling the air conditioner (2) based on the predicted power consumption. Control device (200).

8. A control unit (19, 59, 309) is provided, the control unit generates prediction models (M1, M2) that predict the power consumption of devices (2, 7) installed in a server room (92) in a facility (91) from the first feature amounts by learning the first feature amounts in association with the power consumption of the devices; The device has a first fan (22, 72), an intake port (20a, 70a) for drawing air into the device using the first fan, and an outlet port (20b, 70b) for blowing air out of the device using the first fan, the first feature amount includes a difference between a static pressure of air in the vicinity of the air inlet and a static pressure of air in the vicinity of the air outlet, The device is an air conditioner (2) that conditions the air in the server room or a server (7). Learning device (1,5,300).

Images