Power amount control device

Abstract

A power amount control device (10) has a storage unit (180) that stores situation information as training data (550). The power amount control device (10) also comprises a training data reuse unit (161) that, if new situation information including an unencountered outside air temperature range that is not stored as the training data (550) is generated in an operation stage, identifies situation information having the same in-room initial temperature level, load arrangement pattern, and air-conditioning control level in an outside air temperature range closest to the outside air temperature of the unencountered outside air temperature range, reuses an air-conditioning power function and a temperature function indicated by the identified situation information, and sets the air-conditioning control level indicated by the identified situation information to the new situation information after a prescribed adjustment process has been carried out.

Description

Power consumption control device

[0001] The present invention relates to a power control device for controlling the power consumption of servers and air conditioners in each room of a data center (hereinafter sometimes referred to as "DC").

[0002] With the goal of reducing greenhouse gas emissions, efforts are being made to expand the use of renewable energy sources such as solar and wind power, promoting the adoption of facilities, equipment, and technologies. However, because the supply of renewable energy is affected by changes in the natural environment and it cannot be stored, it is necessary to maximize the efficiency of renewable energy utilization by increasing demand in line with peak electricity supply. Consequently, it is anticipated that electricity prices will tend to fluctuate throughout the day, for example, by making daytime electricity, which is expected to be generated from solar power, cheaper than nighttime electricity.

[0003] Furthermore, the load within the data center (DC) fluctuates throughout the day, and the system controlling the DC needs to consider both the load arriving at the DC and fluctuations in electricity prices. In addition, the amount of data processed (load) in the DC is increasing year by year, and in this respect as well, it is necessary to improve the overall power consumption efficiency of the DC (the amount of power consumed by the DC as a whole for a certain amount of data processing). In a DC, in addition to the power consumption of servers, the power consumption of air conditioning accounts for a large proportion, and a reduction in overall power consumption is required for the DC as a whole.

[0004] Non-Patent Document 1 describes a technology that optimizes the overall power consumption of a data center by considering the power consumption of air conditioning and servers (IT equipment). The data center air conditioning-linked IT load allocation optimization method described in Non-Patent Document 1 predicts the future load trends of IT equipment by collecting operational and monitoring information of IT equipment in the data center, and calculates the power increase of the air conditioning system in accordance with the power increase of the IT equipment. Then, it solves an optimization problem that minimizes the objective function, which is the power consumption of the data center, so that the load concentration rate on IT equipment increases over time, that is, so that the number of operating IT equipment is reduced. This calculates the allocation of IT load (virtual machines) to IT equipment that minimizes the power consumption of the data center.

[0005] However, in the technology described in Non-Patent Document 1, in the air-conditioning power model used to calculate the power of the air-conditioning equipment, a general rule-based criterion that does not depend on facility conditions different for each DC is adopted. Therefore, it has been difficult to perform optimization for reducing the power consumption of the DC in consideration of individual facility conditions such as the arrangement position of the air-conditioning equipment, the air flow, the server arrangement configuration in the DC, and the heat cooling efficiency.

[0006] Also, in the technology described in Non-Patent Document 1, optimization of power cost efficiency has not been considered in consideration of the daily fluctuation of the electricity charge, the fluctuation of the load flowing into the DC, the load arrangement in the DC, and the air-conditioning control. Regarding these points, the inventors have developed the technology described in Patent Document 1 based on the following technical idea.

[0007] For a certain amount of load arriving in the DC, factors affecting the power consumption of air-conditioning cooling include the initial temperature in each room in the DC (control parameter <1>), the air-conditioning control level of the air conditioner (control parameter <2>), and in addition, the load arrangement in the DC. The load arrangement in the DC includes how much load is allocated to each room in the DC (load allocation pattern between DC rooms (allocation pattern): control parameter <3>) and how the load allocated to each room is further arranged on the servers in that room (load arrangement pattern in the DC room (arrangement pattern): control parameter <4>).

[0008] Here, a predetermined control time for updating the air-conditioning control and the load arrangement in each room is defined as one turn (for example, one hour). Consider the "load allocation amount per room and load arrangement pattern" (hereinafter referred to as the "allocation arrangement pattern") in which the load allocation amount to each room for the total load amount of the DC in each turn is changed, and further, the load allocation amount to each room is changed at the allocation ratio for each area in the room (the "arrangement control area 30" in FIG. 17 described later).

[0009] By fixing one of these distribution patterns, that is, by fixing the load distribution amount (load distribution pattern) for each room in each turn and the load distribution pattern within the DC room, the power consumption cost efficiency (E) that maximizes the air conditioning power cost efficiency for the DC as a whole, and the air conditioning control pattern for each room at that time are determined. As a result, the power consumption control device described in Patent Document 1 can maximize the power consumption cost efficiency of the entire DC by optimizing the air conditioning control level for each room, the load distribution amount between rooms, and the load distribution pattern within rooms when the load amount and electricity charges flowing into the DC fluctuate during a certain continuous time period.

[0010] Jun Okitsu et al., "Optimization Method for IT Load Allocation Linked to Air Conditioning for Environmentally Friendly Data Centers," FIT (Forum on Information Technology) 2010, 9th Forum on Information Science and Technology, RC-009.

[0011] International Publication No. 2024 / 171365

[0012] However, even with the same server load, the amount of power consumed by the air conditioner is affected by the ambient temperature (temperature outside the DC). For example, as shown in Figure 15, even when processing the same load on the DC, if the ambient temperature increases from 21°C to 31°C, the amount of power consumed by the air conditioner increases by 17%, from 87 kW to 107 kW. Therefore, the training data required to start optimal control of the air conditioner needs to cover ambient temperatures for all seasons (at least all temperature ranges of the DC's ambient temperature).

[0013] However, the training data required to initiate optimal control is expected to be collected only during a limited period of initial training, making it difficult to collect training data for ambient temperature across all seasons. Therefore, conventional techniques use data collected during a limited period of initial training, which presents a challenge in that the optimization accuracy may be biased when ambient temperature is used as a control parameter.

[0014] The present invention was made in consideration of these points, and aims to improve the optimization accuracy for reducing the overall power consumption of the DC by accumulating training data during operation, without collecting training data for ambient temperature in untrained areas.

[0015] The power consumption control device according to the present invention is a power consumption control device that is communicated to a room-specific control device that controls a plurality of servers and a plurality of air conditioners in a room in a data center, and generates learning data by controlling the load distribution pattern between the rooms, the server load distribution pattern indicating the distribution pattern of the loads distributed to each room to the servers, and the air conditioning control level of the air conditioners via the room-specific control device, for each turn indicating a predetermined control time, and determines the air conditioning control level of the air conditioners in the distribution distribution pattern indicating the distribution pattern and the server load distribution pattern that maximizes the power consumption cost efficiency of the entire data center, and in the initial learning stage of the learning data, the power consumption control device uses the outside temperature level indicating the range of outside temperatures outside the data center, the inside initial temperature level indicating the range of initial temperatures inside the rooms, the load amount distributed to each room in the distribution pattern, the server load distribution pattern, and the air conditioning control level as Situation components, and calculates the power consumption cost efficiency, The system has a storage unit that generates an air conditioning power function showing the air conditioning power consumption for the load allocated to the rooms, and a temperature function showing the room temperature after the end of each turn for the load allocated to each room, and stores the situation information associated with each of the situation components as learning data. In the operational phase of power control of the data center, when generating new situation information including an unencountered outside temperature range that is not stored as learning data, the system identifies situation information with the same initial room temperature level, server load arrangement pattern, and air conditioning control level in the outside temperature range closest to the unencountered outside temperature range, and reuses the air conditioning power function and temperature function shown in the identified situation information as the air conditioning power function and temperature function of the new situation information. The system also sets the air conditioning control level shown in the identified situation information to the new situation information after performing a predetermined adjustment process according to the high / low relationship between the unencountered outside temperature range and the outside temperature range of the identified situation information.It is characterized by having a learning data repurposing unit that performs repurposing learning that shows the following.

[0016] According to the present invention, it is possible to improve the optimization accuracy for reducing the overall power consumption of the DC by accumulating training data during operation, without having to collect training data for ambient temperature in untrained areas.

[0017] This is a functional block diagram showing an example configuration of the in-room control device according to this embodiment. This is a functional block diagram showing an example configuration of the power consumption control device according to this embodiment. This is a diagram showing an example of learning data in the initial learning stage according to this embodiment. This is a diagram showing a function for each situation information and an example of its use when the outside temperature range to which the unencountered outside temperature belongs is lower than the learned outside temperature range. This is a diagram showing a function for each situation information and an example of its use when the outside temperature range to which the unencountered outside temperature belongs is higher than the learned outside temperature range. This is a diagram showing the processing result when the in-room temperature at the end of a control turn using settings learned by repurposed learning is t°C or more lower than the temperature calculated from the learning data of the source. This is a diagram showing the processing result when the in-room temperature at the end of a control turn using settings learned by repurposed learning is t°C or more higher than the temperature calculated from the learning data of the source. This is a diagram showing an example of generating a new function due to the addition of situation information of a new load level. This is a diagram showing an example of generating a new function and updating it due to the addition of situation information of a new load level. This is a diagram showing an example of dividing the range. This is a diagram showing an example of dividing the range. This is a flowchart showing the flow of data reuse processing and other operations performed by the power consumption control device according to this embodiment. This is a flowchart showing the flow of range division processing performed by the power consumption control device according to this embodiment. This is a hardware configuration diagram showing an example of a computer that implements the functions of the power consumption control device and the in-room control device according to this embodiment. This is a diagram showing that the amount of power consumed by the air conditioner differs depending on the difference in outside temperature. This is a diagram showing the overall configuration of a power consumption control system including the power consumption control device according to the prior art and this embodiment. This is a diagram for explaining the areas within a room according to the prior art and this embodiment. This is a diagram for explaining the learning pattern for acquiring the learning history related to the prior art. This is a diagram showing an example of the control stages of the air conditioner control level according to the prior art and this embodiment. This is a diagram illustrating the air conditioner power function according to this embodiment. This is a diagram illustrating the temperature function according to this embodiment.

[0018] <Prior Art and its Problems> First, the prior art that forms the basis of this invention and its problems will be explained in detail.

[0019] The power consumption control device described in Patent Document 1 calculates the power consumption cost efficiency (E) of the entire DC for N turns, after changing the pattern of the air conditioning control level for each room for each room, based on the load distribution amount and load placement pattern (details of the "distribution placement pattern" are described later). Specifically, in calculating this power consumption cost efficiency (E), one pattern from each distribution placement pattern is sequentially fixed (selected), and the power consumption cost efficiency (E) is calculated. This power consumption cost efficiency (E) is defined as the value obtained by dividing the sum of the electricity charges based on the air conditioning power consumption amount for each consecutive turn and the electricity charges based on the server load amount for each consecutive turn by the total load amount for each consecutive turn, as shown in the following formula (1).

[0020]

[0021] Here, the coefficients and parameters for the j-th turn (tj turn) are as follows: "m" is the number of rooms, and "n" is the number of turns. "c[tj]" represents the power consumption cost efficiency coefficient for the turn, specifically the electricity rate coefficient. For example, (7 yen / kWh), it is a coefficient that converts power consumption into electricity rates. "k" is a coefficient that converts server load into server power consumption (server load amount / power coefficient). The parameters of equation (1) are as follows.

[0022]

[0023] Here, the load amount and arrangement pattern within the room refer to the "distribution arrangement pattern". Also, the total load amount flowing into DC1000 from turn "1" to turn "n", indicated by X[tj] in the lower part of equation (1), is given by the following equation (2).

[0024]

[0025] The power control device then fixes one of the distribution arrangement patterns, calculates the power consumption cost efficiency (E) for the DC total, and determines the distribution arrangement pattern that maximizes the power consumption cost efficiency (E) for the DC total, as well as the air conditioning control pattern for each room at that time.

[0026] Figure 16 is a diagram showing the overall configuration of a power consumption control system 1A, including a conventional power consumption control device 10A. As shown in Figure 16, the power consumption control system 1A is configured to include a DC (data center) 1000 having multiple rooms 500 (rooms "A", "B", and "C" in Figure 16) each equipped with multiple servers 3 and one or more air conditioners 4, an in-room control device 20A that is connected to the multiple servers 3 and one or more air conditioners 4 in each room 500 and provided corresponding to each room 500, and a power consumption control device 10A that is connected to each in-room control device 20A and controls the total power consumption of the DC 1000.

[0027] In DC1000, a room control device 20A is provided in each of the rooms 500. The room control device 20A corresponding to each room 500 may acquire status information (such as air conditioning power consumption) of the air conditioners 4 installed in the room 500 or transmit air conditioning control information (such as set temperature and airflow) via an air conditioning management device (not shown), or it may be directly connected to each air conditioner 4 without going through the air conditioning management device.

[0028] Furthermore, the in-room control device 20A is connected to each server 3 in room 500. The in-room control device 20A may be connected to the server 3 via a server management device (not shown) to acquire status information (server load, etc.) and transmit control information (load allocation, etc.) from the servers 3 located in room 500, or it may be directly connected to the servers 3.

[0029] In this description, it is assumed that a virtualization infrastructure is built and operated on each server 3 within DC1000. Known open-source virtualization infrastructures include OpenStack®, software for building cloud environments, and Kubernetes®, software for managing and operating containerized workloads and services. OpenStack is primarily used for managing and operating physical machines and virtual machines (VMs). Kubernetes is primarily used for managing and operating containers. In this specification, a virtualized application (consisting of one or more containers, or one or more VMs, etc.) on a virtualization infrastructure is referred to as a virtual resource. In Kubernetes, the smallest execution unit of an application is a Pod, which consists of one or more containers.

[0030] In the conventional DC1000, the entire server 3 to be accommodated is divided into areas where multiple servers 3 (server groups) are located, as shown in Figure 17, and controlled as a "placement control area" (server area). This placement control area 30 is an area that accommodates a group of servers where virtual resources are placed, and represents a consolidated area for processing load. Figure 17 shows an example in which placement control areas "1" to "4" are provided.

[0031] Furthermore, corresponding to the server group placement control area 30 in the conventional technology, an "air conditioning control area" is provided as shown in Figure 17. The air conditioning control area 40 is a consolidated area for measuring the room temperature effect of air conditioning control, and it faces either the intake side or the exhaust side of the server 3. The air supplied from the air conditioner 4 is blown out from air conditioning control areas "1", "2", "5", and "6" in the intake side air conditioning control area 40, for example, via piping installed under the floor of the DC1000. Then, in the exhaust side air conditioning control area 40, air whose temperature has risen due to the heat from each server 3 is taken in from the intake ports of the piping installed in air conditioning control areas "3" and "4", creating an airflow that returns to the air conditioner 4.

[0032] Each of these air conditioning control zones 40 is equipped with multiple sensors (temperature sensors, etc.). In addition, each of the placement control zones 30 is equipped with temperature sensors at the intake and exhaust ports of each server 3 and the servers 3 selected within the area. Information obtained from these sensors (sensor information) can be acquired by the room control device 20A and the power consumption control device 10A associated with that room via a communication line or the like.

[0033] Furthermore, in conventional technology, in each room within the DC, the room control device 20A controls a total of l × m × n × o patterns (hereinafter referred to as "predetermined learning patterns") by changing the initial room temperature (l), server load (m), server load arrangement pattern (n), and air conditioning control level (o), and acquires information on the air conditioning power consumption, the room temperature at the end of the turn, and the room temperature reward pass / fail judgment as learning history (learning data).

[0034] The in-room control device 20A sets the initial temperature (initial room temperature) to three levels: "18°C", "24°C", and "30°C", and sets the server load to three levels: "20kW", "50kW", and "90kW", as shown in Figure 18. Regarding the load distribution pattern, the example shown in Figure 18 indicates that the server load of the room 500 is distributed as "25%-25%-25%-25%", "50%-50%-0%-0%", and "0%-0%-50%-50%" in each of the four distribution control areas "1", "2", "3", and "4" (Figure 17).

[0035] Regarding air conditioning control, the output value of the air conditioning control, which indicates the air conditioning control capacity, is divided into multiple stages according to its intensity, and these are defined as air conditioning control levels. For example, as shown in Figure 19, air conditioning control level "1" corresponds to a set temperature of 32°C. Air conditioning control level "2" corresponds to a set temperature of 24°C. And air conditioning control level "3" corresponds to a set temperature of 16°C. Thus, the higher the intensity of the air conditioning control level, the higher the cooling capacity. On the other hand, the amount of air conditioning power consumed tends to increase. The example shown in Figure 18 shows that the air conditioning control is set to these three levels: "1", "2", and "3".

[0036] The room control device 20A acquires information on the amount of air conditioning power consumed during a turn as a result of executing a predetermined learning pattern. The room control device 20A also acquires information on the room temperature at the end of the turn from a temperature sensor. The room temperature reward pass / fail judgment is information indicating the result (pass / fail) of whether the temperature after the turn exceeded a predetermined room temperature threshold (room temperature threshold). The room temperature threshold is a threshold (limit temperature) set by the system management side in each room 500 of DC1000, and is set on the premise that control will not be performed when the temperature exceeds the temperature threshold. Patterns that fail the room temperature reward pass / fail judgment are not executed in actual operation and are therefore not stored as learning data. This reduces the computational load on the power consumption control device 10A. The room control device 20A then transmits the learning history (learning data) executed with the predetermined learning pattern to the power consumption control device 10.

[0037] In calculating the above-mentioned power consumption cost efficiency (E), the power consumption control device 10 approximates the air conditioning power function and temperature function for each room 500 distribution arrangement pattern. For each room "ri" (where i is a positive integer), the power consumption control device 10 defines the air conditioning power function as the amount of air conditioning power consumed relative to the load in room 500. The power consumption control device 10 also defines the temperature function as the temperature at the end of the turn relative to the load in room 500.

[0038] Here, let "A[ri][tj]" be the air conditioning control level for the j-th turn "tj" in room "ri" (where j is a positive integer). Note that the air conditioning control level for turn "tj" in room "ri" is uniquely determined, for example, "A(1)" (air conditioning control level "1" as shown in Figure 19). Also, let "p[ri][tj]" be the load distribution pattern for turn "tj" in room "ri".

[0039] The power control device 10 defines the air conditioning power function as the air conditioning power consumption for the load (server load) allocated to room 500, and approximates it using the learning history (learning data) of room 500. Figure 20 shows an example where the load distribution pattern is "p[r1][t1]", and in air conditioning control when a certain initial temperature in the room and a certain air conditioning control level are selected, the corresponding air conditioning power consumption is obtained by acquiring learning data for loads of "60kW", "80kW", and "100kW" from the learning history information of the range that includes the initial temperature in the room. As a result, the equation indicated by the symbol a in Figure 20 (power consumption prediction function p(x)) can be obtained as an approximated equation of the air conditioning power function.

[0040] The power consumption control device 10 defines the temperature function as the temperature after the turn for the load allocated to room 500, and approximates the temperature function using the learning history (learning data) of room 500. Figure 21 shows an example where the load distribution pattern is "p[r1][t1]", and in air conditioning control when a certain initial temperature and air conditioning control level are selected, the learning data for load amounts of "60kW", "80kW", and "100kW" is obtained from the learning history information of the range that includes the initial temperature in the room, thereby obtaining the corresponding temperature after the turn. As a result, the equation indicated by the sign b in Figure 21 (temperature prediction function u(x)) can be obtained as an approximated equation of the temperature function.

[0041] The power consumption control device 10 calculates the value of the air conditioning power consumption corresponding to the load using an approximate formula for the air conditioning power function, and calculates the temperature after the end of a turn (initial temperature at the start of the next turn) for the load using an approximate formula for the temperature function, thereby calculating the overall power consumption cost efficiency (E) of the DC when the air conditioning control level in each room is changed for each distribution arrangement pattern. The power consumption control device 10 then extracts the air conditioning control level for each room that maximizes the overall power cost efficiency of the DC for each distribution arrangement pattern, and determines the distribution arrangement pattern and the air conditioning control level for each room in that pattern that maximize the overall power consumption cost efficiency of the DC from among all load distribution patterns.

[0042] However, as described above, the temperature outside the DC (external temperature) is assumed to be collected during a limited period at the time of initial learning, and although it is a control parameter that affects the air-conditioning power consumption, since the learning data is limited to a partial period, even if such learning data is adopted, there is a risk of causing a bias in the optimization accuracy. Therefore, when starting the learning and operation of the air-conditioning optimal control, even if there is learning data on the external temperature in the unlearned area, improving the optimization accuracy by accumulating the learning data becomes an issue.

[0043] <The Present Embodiment> Next, the power consumption control system 1 including the power consumption control device 10 according to the present embodiment will be described in detail. The overall configuration of the power consumption control system 1 according to the present embodiment is the same as that of the power consumption control system 1A according to the prior art shown in FIG. 16. As shown in FIG. 16, the power consumption control device 10 includes a DC 1000 having a plurality of rooms 500 (rooms "A", "B", and "C" in FIG. 16) provided with a plurality of servers 3 and one or more air conditioners 4, a room control device 20 that is communicatively connected to the plurality of servers 3 and one or more air conditioners 4 in the room 500 and is provided corresponding to each room 500, and a power consumption control device 10 that is communicatively connected to each room control device 20 and controls the total power consumption of the DC 1000.

[0044] The power consumption control device 10 according to this embodiment is characterized by having a learning data completion unit 160 (Figure 2, described later) that completes the learning history (learning data) for unencountered ambient temperatures in the initial learning stage. In this embodiment, "unencountered" means that the ambient temperature has not been learned up to that point when controlling with the predetermined learning pattern described above. The learning data completion unit 160 of the power consumption control device 10 adapts the temperature function and air conditioning power function of the learned data to new learning data for unencountered ambient temperatures, depending on the difference between the learned ambient temperature data and the unencountered ambient temperature, and also adjusts the air conditioning level of the learned data and adapts it to the new learning data. Furthermore, the learning data completion unit 160 feeds back the control results with the adapted learning data (hereinafter referred to as "adapted learning data") and further adjusts the air conditioning control level to modify the learning data. As a result, when the power consumption control device 10 adapts and utilizes learning data from existing data, it can derive a control solution for the air conditioner 4 that ensures indoor temperature quality while providing a higher power saving effect. Furthermore, the power consumption control device 10 can adjust subsequent control cycles based on feedback of the control results, thereby enabling it to converge to the optimal solution more quickly.

[0045] The following describes the details of the in-room control device 20 and the power consumption control device 10, which constitute the power consumption control system 1 according to this embodiment. Note that the processing of the learning data completion unit 160 of the power consumption control device 10 is performed in the operation phase, assuming that in the initial learning phase, the in-room control device 20 collects the learning history and the power consumption control device 10 acquires that learning history to generate learning data 550 (Figure 2). Therefore, first, the collection of the learning history by each in-room control device 20 and the calculation of the power consumption cost efficiency (E) using the learning data 550 by the power consumption control device 10 will be described.

[0046] <Room Interior Control Device> First, the room interior control device 20 will be described. Note that this room interior control device 20 is provided corresponding to each room 500 and has the same configuration. FIG. 1 is a functional block diagram showing a configuration example of the room interior control device 20 according to the present embodiment. When the room interior control device 20 receives a learning history collection instruction from the power consumption control device 10, it changes the initial temperature (room interior initial temperature), load amount (server load amount distributed to the room 500), load distribution pattern, and air conditioning control level at the outside temperature (outdoor air temperature) during measurement, executes control of the server 3 and the air conditioner 4, collects predetermined air conditioning cooling learning data (room interior temperature, air conditioning power consumption, etc.) as a learning history, and transmits it to the power consumption control device 10. Further, the room interior control device 20 causes the server 3 and the air conditioner 4 in the room 500 to execute control based on the load amount and load distribution pattern according to the distribution arrangement pattern that maximizes the power consumption cost efficiency determined by the power consumption control device 10, and control based on the determined air conditioning control level.

[0047] This room interior control device 20 is configured by a computer including a control unit 200, an input / output unit 250, and a storage unit 260.

[0048] The input / output unit 250 inputs and outputs information with the power consumption control device 10, each device (each server 3 and each air conditioner 4) in the room 500, etc. This input / output unit 250 is composed of a communication interface that transmits and receives information via a communication line, and an input / output interface that inputs and outputs information between an input device such as a keyboard (not shown) and an output device such as a monitor.

[0049] The storage unit 260 is composed of a hard disk, a flash memory, a RAM (Random Access Memory), etc. In this storage unit 260, a program for executing each function of the control unit 200 and information necessary for the processing of the control unit 200 are temporarily stored.

[0050] The control unit 200 oversees all processes performed by the in-room control device 20 and, as shown in Figure 1, includes a situation recognition unit 210, a learning history collection unit 220, an in-room server control unit 230, and an in-room air conditioning control unit 240.

[0051] The situation recognition unit 210 considers the following information in room 500 within DC 1000 during the initial learning stage (before control) as "Situation components," and acquires each piece of information from these Situation components as a predetermined learning pattern. Hereafter, the set of each Situation component will be referred to as "Situation information."

[0052] The situation recognition unit 210 acquires information from a temperature sensor set outside the DC1000 as "outside temperature (ambient air temperature)". This ambient temperature is classified into ambient temperature levels (low, medium, high) for each predetermined range. For example, ambient temperature level [low] is defined as "less than 10°C", ambient temperature level [medium] as "10°C or more and less than 20°C", and ambient temperature level [high] as "20°C or more and less than 30°C".

[0053] Furthermore, the situation recognition unit 210 acquires temperature information from multiple temperature sensors installed around the air intake ports of the servers 3 in each server area (arrangement control area 30 in Figure 17), calculates an average value, and calculates the average temperature of the air intake ports for each server area (average air intake temperature). The situation recognition unit 211 then averages the calculated average temperatures for each server area over the entire room 500, and defines the resulting temperature as the "initial temperature (initial room temperature)". This initial temperature (initial room temperature) is classified into initial temperature levels (low, medium, high) for each predetermined range. For example, the initial temperature level [low] is defined as "21℃±1℃" (20℃ or more and less than 22℃), the initial temperature level [medium] as "23℃±1℃" (22℃ or more and less than 24℃), and the initial temperature level [high] as "25℃±1℃" (24℃ or more and less than 26℃).

[0054] The server load is the load allocated to Room 500 and is classified into load levels (low, medium, high) for each predetermined range. For example, load level [low] is defined as "less than 30 kW", load level [medium] as "30 kW or more and less than 90 kW", and load level [high] as "90 kW or more".

[0055] The server load allocation pattern is, for example, a pattern in which the server load of the room 500 is allocated to each of the four server areas (allocation control areas 30 in Figure 17) "1", "2", "3", and "4" as "25%-25%-25%-25%", "50%-50%-0%-0%", and "0%-0%-50%-50%". This server load allocation pattern is set in advance by the system administrator or the like.

[0056] The air conditioning control level is set to three levels ("1", "2", and "3"), as shown in Figure 19, for example. Note that if multiple air conditioners 4 (for example, four units) are installed, the power efficiency of each air conditioner 4 can be examined to divide them into a power-efficient group and a power-inefficient group. The power-efficient group (two units) can then be set to the same air conditioning level, and the power-inefficient group can be set to the same air conditioning level. This grouping reduces the number of calculation terms and allows for more efficient calculation compared to finding the air conditioning control solution for each individual air conditioner 4.

[0057] Returning to Figure 1, when the learning history collection unit 220 receives learning history collection instruction information from the power consumption control device 10, it executes a predetermined learning pattern control that changes the initial room temperature, server load, load placement pattern, and air conditioning control level, respectively, at the ambient temperature (outside air temperature) at that time. The learning history collection unit 220 then acquires the air conditioning power consumption, the room temperature at the end of the turn, and the room temperature reward pass / fail judgment information as learning history as a result of this control.

[0058] Here, the learning history collection unit 220 obtains information on the amount of air conditioning power consumed during that turn from the air conditioning power measurement unit 242, which will be described later. The learning history collection unit 220 also obtains information on the room temperature at the end of the turn from the temperature sensor via the situation recognition unit 210. The room temperature at the end of the turn is set in advance as, for example, the average temperature of the server air intake at the end of the turn (average intake temperature).

[0059] The learning history collection unit 220 transmits the learning history, which was executed according to a predetermined learning pattern, to the power consumption control device 10, along with the identification information of the room control device 20 itself. The information of each parameter set in this predetermined learning pattern may be registered in advance in the learning history collection unit 220, or it may be obtained from the DC 1000 management device (not shown) or the power consumption control device 10. In addition, when generating the learning history of this predetermined pattern, the learning history collection unit 220 instructs the room server control unit 230 and the room air conditioning control unit 240 to execute load arrangement and air conditioning control within the room 500.

[0060] The in-room server control unit 230 (Figure 1) controls the server 3 based on the load distribution pattern for the server area (placement control area 30) (Figure 17) set up in the room 500 during the collection of learning history (initial learning stage) and during operation (operation stage), and also measures the server power consumption for each turn. This in-room server control unit 230 includes a load distribution pattern setting unit 231 and a server power measurement unit 232.

[0061] The load allocation pattern setting unit 231 sets an allocation pattern in which the server load amount set by the learning history collection unit 220 is allocated to each server area (allocation control area 30 (Figure 17)) within Room 500. Then, according to the set allocation pattern, the load allocation pattern setting unit 231 generates and places virtual resources (VMs, containers, etc.) in each location control area 30 and executes the load processing set during the collection of the learning history. Furthermore, during operation (operational phase), the load allocation pattern setting unit 231 sets an allocation pattern for the server groups in each server area (allocation control area 30 (Figure 17)) within Room 500 that is shown as a pattern that maximizes the power consumption cost efficiency of the entire DC, as determined by the power consumption control device 10.

[0062] The server power measurement unit 232 measures the amount of server power (server power consumption) when each server 3 performs processing in the load allocation pattern setting unit 231.

[0063] The in-room air conditioning control unit 240 causes the air conditioner 4 to perform air conditioning control at a preset air conditioning control level (for example, air conditioning control levels "1", "2", and "3"), and also measures the amount of air conditioning power consumed. This in-room air conditioning control unit 240 includes an air conditioning control execution unit 241 and an air conditioning power measurement unit 242.

[0064] The air conditioning control execution unit 241 controls each air conditioner 4 in the room 500 at the air conditioning control level set by the learning history collection unit 220. In addition, during operation (operation phase), the air conditioning control execution unit 241 performs air conditioning control of each air conditioner 4 in the room 500 at the air conditioning control level of the pattern determined by the power consumption control device 10 (a pattern that maximizes the power consumption cost efficiency (E) of the entire DC, or a pattern set by the repurposed learning described later).

[0065] The air conditioning power measurement unit 242 measures the amount of air conditioning power consumed by the air conditioner 4 in the room 500 for each turn, both when collecting learning history (initial learning stage) and during operation (operation stage).

[0066] <Power Consumption Control Device> Next, the power consumption control device 10 according to this embodiment will be described. Figure 2 is a functional block diagram showing an example of the configuration of the power consumption control device 10 according to this embodiment. The power consumption control device 10 collects learning history from each of the in-room control devices 20 corresponding to each room 500 and generates learning data for each room 500. Then, based on the learning data, the power consumption control device 10 approximates the "air conditioning power function" and "temperature function" for each room. The power consumption control device 10 generates a distribution arrangement pattern for each room in each turn from the input information of the total load amount in the DC over N turns. Then, for each distribution arrangement pattern, the power consumption control device 10 calculates the power cost efficiency of the entire DC when the air conditioning control level in each room is changed, and extracts the air conditioning control level for each room that maximizes the power cost efficiency of the entire DC in the corresponding pattern. From all distribution arrangement patterns, the power consumption control device 10 determines the distribution arrangement pattern and the air conditioning control level in the corresponding pattern that maximize the power consumption cost efficiency of the entire DC.

[0067] Furthermore, during operation (operation phase), if the external temperature (outside air temperature) is one that has not been encountered in the initial learning phase and the temperature function and air conditioning power function have not been calculated from the learning data, the power consumption control device 10, in that Situation information, will repurpose the temperature function and air conditioning power function from the learned data as new learning data for the unencountered external temperature, according to the difference between the learned external temperature and the unencountered external temperature, and will also adjust the air conditioning level of the learned data and repurpose it as new learning data. The power consumption control device 10 will then feed back the control results from the repurposed learning data (repurposed learning data) to further adjust the air conditioning control level and modify the learning data. Furthermore, during operation (operation phase), the power consumption control device 10 will save the control results for all turns in which control was performed for each Situation information and will generate or update the temperature function and air conditioning power function. In addition, the power consumption control device 10 will subdivide the range ranges of the outside air temperature range and the room temperature range according to the accumulation of learning data related to outside air temperature during operation.

[0068] This power consumption control device 10 is composed of a computer that includes a control unit 100, an input / output unit 170, and a storage unit 180.

[0069] The input / output unit 170 performs input and output of information between the in-room control device 20 and an external system management device (not shown), etc. This input / output unit 170 consists of a communication interface that sends and receives information via a communication line and an input / output interface that performs input and output of information between the input device such as a keyboard and an output device such as a monitor (not shown).

[0070] The storage unit 180 is composed of a hard disk, flash memory, RAM (Random Access Memory), etc. The learning data 550 generated by the learning data generation unit 120 is stored in this storage unit 180. In addition, the storage unit 180 temporarily stores programs for executing each function of the control unit 100 and information necessary for the processing of the control unit 100. For example, the power consumption cost efficiency "c[tj]" for the relevant turn (power consumption cost efficiency coefficient (electricity rate coefficient) at turn j) (i.e., information on the change in the electricity rate coefficient) and various coefficients such as the load-server power proportionality coefficient "k" are stored in the storage unit 180. Furthermore, the storage unit 180 stores the total load amount in DC 1000 for N turns (information on the load amount flowing into DC in each turn) from the management device of the power consumption control system 1 (Figure 16), etc.

[0071] The control unit 100 oversees the overall processing performed by the power consumption control device 10 and, as shown in Figure 2, includes a room definition unit 110, a learning data generation unit 120, an air conditioning power / temperature function approximation unit 130, a distribution and arrangement pattern generation unit 140, an optimal control calculation unit 150 (power cost efficiency calculation unit 151, control pattern determination unit 152), and a learning data completion unit 160.

[0072] The room definition unit 110 defines room 500 within DC 1000 as a cohesive logical area where a specific group of air conditioners exerts a cooling effect on the group of servers. Specifically, when predetermined triggering conditions are met, for example, during the initial learning phase, at predetermined time intervals (e.g., 3 months, 6 months), when a new service process is started, or when maintenance or equipment upgrades are performed on the air conditioner 4, the room definition unit 110 transmits air conditioning control information to the room control device 20 to operate the group of air conditioners located in the cohesive area. As a result, the room definition unit 110 identifies the server 3 that will be affected by the cooling effect of that group of air conditioners (for example, when the server intake temperature is lowered by 2°C or more) using temperature information obtained from temperature sensors, and sets the installation area of ​​the server 3 as room 500.

[0073] The learning data generation unit 120 acquires the learning history executed in the predetermined learning pattern from each of the in-room control devices 20 corresponding to each room 500. The learning data generation unit 120 then classifies the learning history executed in the predetermined learning pattern according to the combination of the outside temperature range and the initial temperature range, and generates learning data for each situation, including information on server load, load placement pattern and air conditioning control level, and the control results, namely air conditioning power consumption (power consumption), in-room temperature at the end of the turn (average temperature of the air intake at the end), and in-room temperature reward pass / fail judgment (pass / fail judgment). The learning data generation unit 120 then stores the generated learning data 550 for each room 500 in the storage unit 180.

[0074] Figure 3 shows an example of the learning data 550 in the initial learning stage according to this embodiment. Here, the outside temperature level is set to three stages: outside temperature level [low: less than 10℃], outside temperature level [medium: 10℃ or more and less than 20℃], and outside temperature level [high: 20℃ or more and less than 30℃]. The initial temperature level is set to three stages: initial temperature level [low: 21℃±1℃] (20℃ or more and less than 22℃), initial temperature level [medium: 23℃±1℃] (22℃ or more and less than 24℃), and initial temperature level [high: 25℃±1℃] (24℃ or more and less than 26℃). The load level is set to three stages: load level [low: less than 30Kw], load level [medium: 30Kw or more and less than 90Kw], and load level [high: 90Kw or more]. The air conditioning level is set to three stages: air conditioning control level "1", "2", and "3", as shown in Figure 19. Furthermore, units 1 and 2 of air conditioner 4 are grouped together as a high-power-efficiency group, while units 3 and 4 are grouped together as a low-power-efficiency group. The same air conditioning control level is set for each unit within this same power efficiency group. To simplify the explanation, the load distribution pattern is assumed to be uniform in each server area, and its depiction in the learning data diagrams from Figure 3 onward is omitted.

[0075] In the first row of training data (Situation information) shown in Figure 3, the outside temperature is "8°C," which falls under the "Low" outside temperature level. The initial temperature is "20.8°C," which falls under the "Low" initial temperature level. The load is "20 kW," which falls under the "Low" load level. The air conditioning control level is "3" for units 1 and 2, and "3" for units 3 and 4.

[0076] Then, the measured value of the air conditioning power consumption (power consumption) during this control is stored as "30kW", the intake air temperature at the end is "23℃", the "Measurement Status" indicating whether the measurement has been taken is "Done", and the room temperature reward pass / fail judgment (pass / fail judgment) is stored as "OK".

[0077] In this initial learning stage, due to the limited data collection period, there is learning data for Situation information where the measurement status is "not yet encountered" (unlearned region) for ambient temperatures that have not yet been encountered. The power consumption control device 10 (learning data supplementation unit 160) according to this embodiment supplements this unencountered ambient temperature learning data from already learned data (details will be described later).

[0078] The air conditioning power / temperature function approximation unit 130 defines the air conditioning power function as the amount of air conditioning power consumed relative to the load in room 500 for each room "ri" (where i is a positive integer) set by the room definition unit 110. The air conditioning power / temperature function approximation unit 130 also defines the temperature function as the temperature at the end of the turn (in this case, the average temperature of the air intake at the end) relative to the load in room 500.

[0079] The air conditioning power and temperature function approximation unit 130 uses the training data 550 to calculate a power consumption prediction function p(x) (see Figure 20) that approximates the air conditioning power function and a temperature prediction function u(x) (see Figure 21) that approximates the temperature function.

[0080] Specifically, the air conditioning power / temperature function approximation unit 130 defines the air conditioning power function as the amount of air conditioning power consumed for the load (server load) allocated to the room 500, and approximates it using the learning data 550 for the room 500 to calculate the power consumption prediction function p(x). For example, when the load arrangement pattern is "p[r1][t1]", and the outside temperature (outside air temperature), the initial temperature inside the room, and the air conditioning control level are selected, the air conditioning power / temperature function approximation unit 130 can obtain the power consumption prediction function p(x) by acquiring information on the corresponding air conditioning power consumption when the load is "20kW", "50kW", and "90kW" from the learning data 550 in the range that includes the outside temperature level, the initial temperature inside the room level, and the air conditioning control level.

[0081] Furthermore, the air conditioning power / temperature function approximation unit 130 defines the temperature function as the temperature after the turn for the load amount allocated to the room 500, and approximates the temperature function using the learning data 550 of the room 500 to calculate the temperature prediction function u(x). For example, when the load arrangement pattern is "p[r1][t1]", and the outside temperature (outside air temperature), the initial temperature inside the room, and the air conditioning control level are selected, the air conditioning power / temperature function approximation unit 130 can obtain the temperature prediction function u(x) by acquiring information on the temperature after the turn (average temperature of the intake port at the end) for load amounts of "20kW", "50kW", and "90kW" from the learning data 550 in the range that includes the outside temperature level, the initial temperature level inside the room, and the air conditioning control level.

[0082] The air conditioning power and temperature function approximation unit 130 stores the calculated power consumption prediction function p(x) and temperature prediction function u(x) information as training data 550, linked to each situation information.

[0083] As described above, the distribution and placement pattern generation unit 140 sets a predetermined time interval for updating the air conditioning control and load placement within the room 500 as one turn, changes the load distribution amount to each room 500 for the total load amount of DC 1000 in each turn, and further changes the load distribution amount to each room 500 according to the placement ratio for each server area (placement control area 30) within the room 500, thereby generating a "room-specific load distribution amount / load placement pattern" (distribution and placement pattern).

[0084] For example, when given "X[t1] = 50" (total load for turn "1": 50), the load distribution pattern generation unit 140 generates various patterns as load distribution patterns for turn "1", such as a pattern that distributes 0 kW to room "1" and 50 kW to room "2" (x[r1][t1] = 0, x[r2][t1] = 50), and a pattern that distributes 10 kW to room "1" and 40 kW to room "2" (x[r1][t1] = 10, x[r2][t1] = 40). Furthermore, when given "X[t2] = 40" (total negative amount for turn "2": 40), the distribution pattern generation unit 140 generates various patterns as load distribution patterns for turn "2," such as a pattern that distributes 0 kW to room "1" and 40 kW to room "2" (x[r1][t2] = 0, x[r2][t2] = 40), and a pattern that distributes 10 kW to room "1" and 30 kW to room "2" (x[r1][t2] = 10, x[r2][t2] = 30).

[0085] Furthermore, the distribution and allocation pattern generation unit 140 generates load allocation patterns for each server area (allocation control area 30) within the room 500, for example, patterns such as "25%-25%-25%-25%", "50%-50%-0%-0%", and "0%-0%-50%-50%" for four server areas. The distribution and allocation pattern generation unit 140 then generates a distribution and allocation pattern by combining this load distribution pattern and load allocation pattern.

[0086] The optimal control calculation unit 150 calculates the power consumption cost efficiency (E) of the entire DC1000 over N turns by changing the pattern of the air conditioning control level for each distribution arrangement pattern. The optimal control calculation unit 150 then calculates the distribution arrangement pattern and the air conditioning control level for each room 500 as the optimal solution that maximizes the power consumption cost efficiency (E) for the entire DC1000. This optimal control calculation unit 150 includes a power cost efficiency calculation unit 151 and a control pattern determination unit 152.

[0087] The power cost efficiency calculation unit 151 calculates the power consumption cost efficiency (E) of the entire DC 1000 over N turns when the air conditioning control level pattern is changed for each room 500 in each distribution arrangement pattern. The power cost efficiency calculation unit 151 may, when calculating the power consumption cost efficiency (E) of the entire DC 1000, exclude data from the learning dataset stored as learning data 550 where the room temperature reward pass / fail judgment was "failed". By doing so, the power cost efficiency calculation unit 151 can speed up the calculation of power consumption cost efficiency (E).

[0088] Specifically, the power cost efficiency calculation unit 151 uses the above-mentioned equation (1) to calculate the power consumption cost efficiency (E) for each air conditioning control level pattern. The power cost efficiency calculation unit 151 also uses the temperature prediction function u(x) calculated by the air conditioning power / temperature function approximation unit 130 to calculate the temperature after the end of the turn (average temperature of the air intake at the end), that is, the initial temperature at the start of the next turn. Furthermore, the power cost efficiency calculation unit 151 uses the power consumption prediction function p(x) calculated by the air conditioning power / temperature function approximation unit 130 to calculate the air conditioning power consumption for the load amount (server load amount) allocated to room 500.

[0089] In calculating the power consumption cost efficiency (E), the power cost efficiency calculation unit 151 fixes (selects) one of the generated distribution patterns. In this way, the amount of load to be distributed to each room and the load distribution pattern can be uniquely determined for each turn. The power cost efficiency calculation unit 151 then calculates the power consumption cost efficiency (E) when the air conditioning control level pattern is changed for each room 500 in the fixed (selected) load distribution pattern. The power cost efficiency calculation unit 151 sequentially fixes (selects) each load distribution pattern and calculates the power consumption cost efficiency (E).

[0090] The control pattern determination unit 152 determines Eall = min(E{p}) which maximizes the power consumption cost efficiency (E) for the entire DC1000 among the total load distribution patterns calculated by the power cost efficiency calculation unit 151. In other words, the control pattern determination unit 152 further determines the optimal distribution pattern and air conditioning control level for each room 500 which maximizes the power consumption cost efficiency (E) for the entire DC1000 from among the power consumption cost efficiency (E) that maximizes the cost efficiency calculated for each selected distribution pattern. Then, the control pattern determination unit 152 sets the load distribution in each room 500 (load distribution pattern between rooms, load distribution pattern within rooms) based on the determined optimal distribution pattern and air conditioning control level, and executes air conditioning control at the determined air conditioning control level for each room 500.

[0091] The learning data completion unit 160 completes the learning data for Situation information to which unencountered external temperatures belong by adjusting and repurposing already learned learning data regarding external temperatures not encountered in the initial learning stage. The learning data completion unit 160 also feeds back the control results from the repurposed learning data (repurposed learning data) to further adjust the air conditioning control level and modify the learning data. This learning data completion unit 160 comprises a learning data repurposing unit 161, a learning data update unit 162, a function update unit 163, and a range division unit 164.

[0092] When the learning data repurposing unit 161 detects Situation information that includes an unencountered outdoor temperature range during operation (i.e., when attempting to perform control at an unencountered outdoor temperature), it repurposes the power consumption prediction function and temperature prediction function (hereinafter simply referred to as "functions") from the learned outdoor temperature range learning data. The learning data repurposing unit 161 then fine-tunes the control result (air conditioning control level) derived from the existing function (a predetermined adjustment process) according to the high-low relationship between the unencountered outdoor temperature range and the learned outdoor temperature range. This will be explained in detail below. The predetermined adjustment process includes, as shown below, a process that sets the repurposed air conditioning control level as the new Situation information without changing it, and a process that adjusts the air conditioning control level to be raised by one level to be set as the control solution for the air conditioning control level in the new Situation information. The repurposing of functions and the predetermined adjustment process of the air conditioning control level by the learning data repurposing unit 161 are referred to as "repurposing learning."

[0093] (Case 1: When the unencountered ambient temperature range is lower than the learned ambient temperature range) If the unencountered ambient temperature range is lower than the learned ambient temperature range, the learning data repurposing unit 161 repurposes a function of Situation information with the same initial room temperature level (initial temperature), load placement pattern, and air conditioning control level for the closest learned ambient temperature range that is higher than the unencountered ambient temperature, as a component of new Situation information for the unencountered ambient temperature range. The learning data repurposing unit 161 then uses the air conditioning control level from the source as the air conditioning control level for the new Situation information.

[0094] Figure 4 shows the functions for each Situation information and examples of their application when the outdoor temperature range to which an unencountered outdoor temperature belongs is lower than the learned outdoor temperature range. As shown in Figure 4, when the outdoor temperature range to which an unencountered outdoor temperature belongs is "low" and is lower than the outdoor temperature range for which existing functions have been calculated (e.g., "medium" and "high"), the learning data application unit 161 extracts the function for the closest range where the initial temperature, load placement pattern, and air conditioning control level are the same, and the outdoor temperature is higher than the outdoor temperature of the new Situation information, and applies that function as the function for the new Situation information. Here, an example is shown in which functions (power consumption prediction function p(x), temperature prediction function u(x)) are applied from Situation information with the same air conditioning control level [2,2,2,2] as the new Situation information, but with a higher outdoor temperature and a value that is 1x closer (indicated by c in Figure 4). Note that, as mentioned above, the load distribution pattern is uniform across each server area, so the diagram is omitted here (the same applies to the following diagrams). Also, the repurposed functions and the newly set air conditioning control levels after repurposing (repurposed learning data) are registered with a repurposing history of "Yes".

[0095] (If the unencountered outdoor temperature range is higher than the learned outdoor temperature range) If the unencountered outdoor temperature range is higher than the learned outdoor temperature range, the learned data repurposing unit 161 repurposes a function of Situation information with the same initial room temperature level (initial temperature), load placement pattern, and air conditioning control level for the closest learned outdoor temperature range that is lower than the unencountered outdoor temperature, as a component of the new Situation information for the unencountered outdoor temperature range. At that time, the learned data repurposing unit 161 sets the air conditioning control level as the control solution to be one level higher than the air conditioning control level of the source from which the repurposing was made.

[0096] Figure 5 shows the functions for each Situation information and examples of their application when the outdoor temperature range to which the unencountered outdoor temperature belongs is higher than the learned outdoor temperature range. As shown in Figure 5, when the outdoor temperature range of the unencountered outdoor temperature is "high" and is higher than the outdoor temperature range for which existing functions have been calculated (e.g., "low" and "medium"), the learning data application unit 161 extracts the function for the closest range where the initial temperature, load placement pattern, and air conditioning control level are the same, and the outdoor temperature is lower than the outdoor temperature of the new Situation information, and applies that function as the function for the new Situation information. Here, an example is shown in which functions (power consumption prediction function p(x), temperature prediction function u(x)) are applied from Situation information with the same air conditioning control level [2,2,2,2] as the new Situation information, but with a lower outdoor temperature that is closer to one. Furthermore, the learning data repurposing unit 161 sets the air conditioning control level [3,3,3,3], which is one level higher than the air conditioning control level [2,2,2,2], as the control solution for the new Situation information (indicated as d in Figure 5). Note that if the air conditioning control level of the new Situation information was originally the highest level (maximum) air conditioning control level (in this case, air conditioning control level [3,3,3,3]), the learning data repurposing unit 161 cannot raise it by one level, so it sets that maximum air conditioning control level as the new Situation information. In addition, the learning data repurposing unit 161 registers the repurposed function and the newly set air conditioning control level (repurposed learning data) as having a "repurposing history".

[0097] Furthermore, the learning data repurposing unit 161 processes the data as follows when an unencountered ambient temperature range lies midway between learned ambient temperature ranges. If the ambient temperature in the nearest learned ambient temperature range is higher than the unencountered ambient temperature range, the learning data repurposing unit 161 processes the data in the same manner as in Case 1. On the other hand, if the ambient temperature in the nearest learned ambient temperature range is lower than the unencountered ambient temperature range, the learning data repurposing unit 161 processes the data in the same manner as in Case 2.

[0098] In this case, if the room temperature exceeds a threshold (room temperature threshold) during a turn controlled using the repurposed training data, forced cooling (for example, control to maximize the air conditioning control level) is performed, prioritizing the reduction of the room temperature. Furthermore, when performing control using the relevant situation information, if a repurposed history exists, air conditioning control is executed using the control solution based on the latest adjustment results.

[0099] Returning to Figure 2, the learning data update unit 162, during the operational phase, compares the difference between the room temperature (average intake temperature) indicating the cooling status after the turn ends and the temperature calculated from the original learning data, when it has executed control using settings derived from repurposed learning for an unencountered outside temperature range, and provides feedback. In other words, the learning data update unit 162 performs a process to update the air conditioning control level of the air conditioner 4 in the new Situation information by raising or lowering it according to the difference.

[0100] Specifically, the learning data update unit 162 executes the following (process 1) to (process 3) according to the difference between the actual room temperature (average intake temperature) after the end of the turn and the temperature calculated using the temperature function of the source learning data. (Process 1) If the room temperature at the end of the turn is within ±t°C (within a predetermined temperature range) of the temperature calculated using the source learning data, the control solution of the current air conditioner 4 is finalized as the learning data for the repurposed unit (repurposed learning data).

[0101] (Process 2) If the room temperature at the end of the turn is t°C or more (above a predetermined first threshold) lower than the temperature calculated from the learning data of the source, the air conditioning control level of the current control solution is lowered by one step and updated as the control solution for the air conditioner 4 of the learning data for the next control. Note that all power efficiency groups (low efficiency group, high efficiency group) may be lowered uniformly, but for example, the efficiency group may be lowered sequentially in the order of low efficiency group → high efficiency group → low efficiency group.

[0102] Figure 6 shows the processing result when the room temperature (average intake temperature) at the end of a turn controlled using settings learned through repurposing is t°C (here, 1°C) or more lower than the temperature calculated using the learning data from the source. Here, assume that the average intake temperature at the end of a turn predicted by the function from the source is "23.5°C", and the average intake temperature at the end of a turn controlled by the repurposed function is "22°C". In this case, the learning data update unit 162 finds that 23.5 - 22 = |1.5| ≥ 1, so it corrects the current control solution's air conditioning control level to a control solution that is one step lower. In Figure 6, an example is shown where the learning data update unit 162 corrects the control solution for each power efficiency group of each air conditioner 4, which was [2,2,2,2], to [2,2,1,1], which is one step lower, for the repurposed low-efficiency group (indicated as e in Figure 6).

[0103] (Process 3) If the room temperature at the end of the turn is t°C or more higher than the temperature calculated from the original learning data, the current control solution's air conditioning control level is increased by one level and updated as the control solution for air conditioner 4 of the repurposed learning data for the next control. Note that all power efficiency groups (low efficiency group, high efficiency group) may be increased uniformly, but for example, the power efficiency group may be increased sequentially in the order of low efficiency group → high efficiency group → low efficiency group.

[0104] Figure 7 shows the processing result when the room temperature (average intake temperature) at the end of a turn controlled using settings learned through repurposing is t°C (here, 1°C) or more higher than the temperature calculated using the learning data from the source. Here, assume that the average intake temperature at the end of a turn predicted by the function from the source was "23.5°C", and the actual average intake temperature at the end of a turn controlled by the repurposed function was "25°C". In this case, the learning data update unit 162 finds that 23.5 - 25 = |-1.5| ≥ 1, so it corrects the current control solution's air conditioning control level to a control solution one step higher. In Figure 7, the learning data update unit 162 corrects the control solution for each power efficiency group of each air conditioner 4, which was [2,2,2,2], to [3,3,3,3], which is one step higher, for both the low-efficiency group and the high-efficiency group after repurposing (indicated as f in Figure 7).

[0105] Furthermore, if the current air conditioning control level was originally the highest (maximum) air conditioning control level (in this case, air conditioning control level [3,3,3,3]), the learning data update unit 162 cannot raise it by one level, so it sets that maximum air conditioning control level as the control solution. Also, if forced cooling is performed in a turn where the repurposed learning data is used because the room temperature threshold is exceeded, the learning data update unit 162 performs the same processing as (process 3) above.

[0106] Returning to Figure 2, the function update unit 163 generates or updates functions (air conditioning power function, temperature function) based on new Situation information saved based on the control results for turns in which control was performed during operation. During the operation phase, if the room temperature conditions at the end of the turn are met and forced cooling or the like was not performed in that turn, the function update unit 163 appropriately adds learning data for Situation information consisting of a combination of outside temperature, initial room temperature, air conditioning control level, load placement pattern, and load amount. Then, when the control results for each Situation information are added, the function update unit 163 generates new functions (air conditioning power function, temperature function) for the first time if data for three different load amount conditions has been accumulated.

[0107] Figure 8 shows an example of generating a new function due to the addition of new Situation information for a new load level. As indicated by the symbol 81 in Figure 8, new Situation information is added for a load of "25 kW" (low) when the outside temperature is "9°C" (low), the initial room temperature is "21°C" (low), and the air conditioning control level is [2,2,3,3]. This results in three or more points (three points in this case) with different load levels being available as data for generating a function. In this case, the function update unit 163 uses the information from these three points to generate a new function (air conditioning power function, temperature function). Specifically, the function update unit 163 generates a power consumption prediction function p(x) and a temperature prediction function u(x) using the new three points (symbol g in Figure 8).

[0108] Furthermore, when control results are added in each Situation information, the function update unit 163 updates the functions (air conditioning power function, temperature function) with the current load conditions for n points (n > 3) if the functions (air conditioning power function, temperature function) have already been calculated.

[0109] Figure 9 shows an example of generating a new function due to the addition of new Situation information for a new load level. As indicated by the symbol 91 in Figure 9, new Situation information is added for a load of "10Kw" (very low) when the outside temperature is "9°C" (low), the initial room temperature is "21°C" (low), and the air conditioning control level is [3,3,3,3]. As a result, four points (n>3) with different load levels are available as data for generating a function. In this case, the function update unit 163 generates a function (air conditioning power function, temperature function) using the information from these four points and updates the function generated from the previous three points. Specifically, the function update unit 163 generates and updates the power consumption prediction function p(x) and the temperature prediction function u(x) using these four points (symbol h in Figure 9).

[0110] Returning to Figure 2, the range division unit 164 divides the range of the ambient temperature level and the initial room temperature level, which are presupposed environmental factors during cooling, among the Situation components, according to the accumulation of learning data during operation. Specifically, during the operation phase, at predetermined intervals, the range division unit 164 divides the range of the ambient temperature level and / or the room temperature level if the load amount exceeds three points for the same Situation information, and the load amount in any range after division by a predetermined number of divisions is three points or more. As a result, the power consumption control device 10 improves the accuracy of each function by regenerating the function (air conditioning power function, temperature function) in the divided range. This will be explained in detail below.

[0111] The range division unit 164 performs range division when the following division conditions are met. The range division unit 164 divides ranges where data has been accumulated and there are loads exceeding 3 points (m > 3) (division requirement 1). In this case, the range division unit 164 may assume that there is at least one load in each of the three load ranges where there are loads exceeding 3 points (m > 3). Furthermore, the range division unit 164 does not perform division or function regeneration if the loads in any of the divided ranges are less than 3 points, that is, if there are no ranges with loads exceeding 3 points (division requirement 2). In other words, the range division unit 164 performs range division when the loads in any of the ranges after division by a predetermined number of divisions are 3 points or more.

[0112] The range division unit 164, if there is a range range with fewer than three load points after division, fills in the missing points in order from the points closest to the range, regenerates the function (air conditioning power function, temperature function) including the filled-in data, and sets it as the function for the range. The range division unit 164 divides the range range at a predetermined period (interval), and repeats this process until each range approaches the minimum range range.

[0113] Figures 10 and 11 show examples of range divisions. The division of the ambient temperature range will be explained below, but the initial temperature range can also be divided using a similar method.

[0114] Here, we will explain an example in which the temperature function U1 in the outdoor temperature range "0-10℃", initial temperature range "20-22℃", and air conditioning control level "3" is divided into the temperature function U2 in the outdoor temperature range "0-5℃", initial temperature range "20-22℃", and air conditioning control level "3", and the temperature function U3 in the outdoor temperature range "5-10℃", initial temperature range "20-22℃", and air conditioning control level "3".

[0115] First, as shown by reference numeral 101 in Figure 10, the initial state at the end of a turn is that there are "3" data points in the outside temperature range "0-10℃", with outside temperatures of 1℃, 4℃, and 9℃. Suppose that two new data points are acquired after the turn ends. These new data points have outside temperatures of 2℃ and 8℃. Therefore, since new data points have been acquired, the number of data points has become "5" (reference numeral 102 in Figure 10), thus satisfying (divide requirement 1).

[0116] Here, if we divide the outside temperature range into two, and look at the breakdown of the five data points, the outside temperature range "0-5°C" has "3" data points (1°C, 2°C, 4°C). Also, the outside temperature range "5-10°C" has "2" data points (8°C, 9°C). Since there is a range "0-5°C" with "3" data points, (division requirement 2) is satisfied.

[0117] Then, in the ambient temperature range "0-5°C" where there are "3" data points, the range division unit 164 regenerates the temperature function using these three data points and calculates the temperature function U2 (reference numeral 103 in Figure 10). On the other hand, in the ambient temperature range "5-10°C" where there are "2" data points (8°C, 9°C), the range division unit 164 cannot create a function because there are no three data points. Therefore, it searches for the closest data point from the training data from the lower limit (5°C in this range) or upper limit (10°C in this range) of the ambient temperature range. Here, as shown by reference numeral 104 in Figure 10, it is assumed that data for an ambient temperature of 11°C has been found. Therefore, in the ambient temperature range "5-10°C", the temperature function is regenerated using the three data points: 8°C, 9°C, and 11°C, and the temperature function U3 is calculated.

[0118] By doing so, the range division unit 164 can divide the ambient temperature range "0-10°C" into ambient temperature ranges "0-5°C" and "5-10°C," as shown in Figure 11, and generate new temperature functions (U2) and (U3) for the new ambient temperature ranges "0-5°C" and "5-10°C" in place of the temperature function (U1).

[0119] <Processing Flow> Next, we will explain the processing of repurposing the learning data and the range division processing, etc., which are mainly performed by the learning data supplementation unit 160 of the power consumption control device 10 according to this embodiment, during the operation phase.

[0120] Figure 12 is a flowchart illustrating the flow of the learning data reuse processing and other operations performed by the power consumption control device 10 according to this embodiment. Here, the power consumption control device 10 is in the operation stage after it has received the learning history from each room control device 20 in the initial learning stage and the learning data generation unit 120 has stored the learning data 550 in the storage unit 180.

[0121] ≪At the start of the turn≫ First, the learning data supplementation unit 160 (learning data repurposing unit 161) of the power consumption control device 10 acquires the current Situation information to be processed at the start of the turn and determines whether or not the outside temperature is one that has not been encountered by referring to the learning data 550 (step S10). Specifically, the learning data repurposing unit 161 determines, based on the current Situation information, which includes the outside temperature, initial room temperature, load amount, load arrangement pattern, and air conditioning control level, whether the measurement status is "completed" as shown in Figure 3 and whether or not a function (air conditioning power function and temperature function) has been generated in the Situation information. If a function has been generated and the outside temperature is not one that has not been encountered (step S10 → No), the process proceeds to step S11, and the optimal control of the air conditioner calculated by the optimal control calculation unit 150 of the power consumption control device 10 is executed. Then, the process proceeds to step S15.

[0122] On the other hand, if no function has been generated and the outside temperature is one that has not been encountered (step S10 → Yes), the learning data reuse unit 161 determines whether or not there is a reuse history for the learning data of the Situation information (step S12). If there is a reuse history (step S12 → Yes), the control is executed using the latest adjusted control solution for the air conditioner 4 (step S13). Then, the process proceeds to step S15.

[0123] On the other hand, if there is no history of reuse (step S12 → No), the learning data reuse unit 161 reuses the functions (power consumption prediction function, temperature prediction function) from the learning data of the learned outdoor temperature range (step S14). At that time, the learning data reuse unit 161 fine-tunes the control result (air conditioning control level) derived from the existing function (predetermined adjustment process) depending on whether the unencountered outdoor temperature range is lower than the learned outdoor temperature range (case 1) or whether the unencountered outdoor temperature range is higher than the learned outdoor temperature range (case 2), and then reuses the control solution of the air conditioning control level and the function. Then proceed to step S15.

[0124] <<During Turn Operation>> During turn operation, the learning data transfer unit 161 determines whether the room temperature exceeds a threshold (room temperature threshold) (step S15). If the room temperature threshold is exceeded (step S15 → Yes), forced cooling is performed (step S16). On the other hand, if the room temperature threshold is not exceeded (step S15 → No), control is continued as is.

[0125] ≪At the end of the turn≫ At the end of the turn, the learning data update unit 162 compares the difference between the room temperature (average intake temperature) indicating the cooling status after the end of the turn and the temperature calculated using the temperature function of the source learning data, for turns in which control was performed using settings from repurposed learning for an unencountered outside temperature range, and provides feedback (step S17). Specifically, the learning data update unit 162 performs one of the following (process 1) to (process 3) depending on the difference between the room temperature (average intake temperature) after the end of the turn and the temperature calculated using the temperature function of the source learning data. (process 1) If the room temperature at the end of the turn is within ±t℃ of the temperature calculated using the source learning data, the learning data update unit 162 confirms the current control solution of the air conditioner 4 as the learning data to be repurposed (repurposed learning data). (Process 2) If the room temperature at the end of the turn is t°C or more lower than the temperature calculated using the original learning data, the learning data update unit 162 updates the current control solution's air conditioning control level to a control solution lowered by one level as the learning data to be used for the next control. (Process 3) If the room temperature at the end of the turn is t°C or more higher than the temperature calculated using the original learning data, the learning data update unit 162 updates the current control solution's air conditioning control level to a control solution highered by one level as the learning data to be used for the next control.

[0126] Next, the function update unit 163 updates the functions (air conditioning power function, temperature function) based on the new Situation information saved based on the control results for each turn in which control was executed during operation. In addition, when control results are added in each Situation, the function update unit 163 generates new functions (air conditioning power function, temperature function) when data for three different load conditions is accumulated for the first time (step S18). Then, it returns to the processing at the start of the next turn.

[0127] Next, the range division process will be explained. Figure 13 is a flowchart showing the flow of the range division process performed by the power consumption control device 10 according to this embodiment. This range division process is performed at predetermined intervals when a predetermined amount of new learning data based on unencountered outside temperatures has been accumulated. This predetermined interval may be set in days, or it may be performed when a predetermined amount of learning data has been accumulated.

[0128] First, the range division unit 164 of the power consumption control device 10 determines for each range whether the division requirements are met for the range of ambient temperature (step S20). Here, the range division unit 164 determines whether the following conditions are met: whether there is a load amount of more than 3 points (m > 3) in the range (division requirement 1), and whether there is at least one range range with a load amount of 3 points in the divided range range (division requirement 2). If these division conditions are not met (step S20 → No), the range division for ambient temperature is not performed, and the process proceeds to step S30.

[0129] On the other hand, if the division requirements are met (step S20 → Yes), the range division unit 164 divides the temperature range into a predetermined number of divisions (for example, 2 divisions), and for the ranges that have been divided, it generates new functions (air conditioning power function, temperature function) based on the learned data after division (step S21). Here, for ranges with a load of 3 points or more, the function is generated using the learned data for that range. On the other hand, if there are ranges with a load of less than 3 points after division, the range division unit 164 fills in the missing points in order from the points closest to the range in question, and generates a function based on the learned data after the filling in.

[0130] Next, the range division unit 164 determines whether the range of room temperature satisfies the division requirements for each range (step S30). Here, the range division unit 164 determines whether the following conditions are met: whether there is a load of more than 3 points (m > 3) in the range (division requirement 1), and whether there is at least one range in the divided range where the load is 3 points (division requirement 2). If these division requirements are not met (step S30 → No), the range division for room temperature is not performed, and the process ends.

[0131] On the other hand, if the division requirements are met (step S30 → Yes), the range division unit 164 divides the room temperature range into a predetermined number of divisions (for example, 2 divisions), and for the ranges that have been divided, it generates new functions (air conditioning power function, temperature function) based on the learned data after division (step S31). Then the process ends.

[0132] By doing so, the power consumption control device 10 according to this embodiment can improve the optimization accuracy for reducing the overall power consumption of the DC by accumulating learning data during operation, without having to collect learning data for ambient temperature in unlearned areas.

[0133] <Hardware Configuration> The power consumption control device 10 and the in-room control device 20 according to this embodiment are implemented by a computer 900 having a configuration such as that shown in Figure 14. Figure 14 is a hardware configuration diagram showing an example of a computer 900 that implements the functions of the power consumption control device 10 and the in-room control device 20 according to this embodiment. The computer 900 has a CPU (Central Processing Unit) 901, ROM (Read Only Memory) 902, RAM 903, HDD (Hard Disk Drive) 904, input / output I / F (Interface) 905, communication I / F 906, and media I / F 907.

[0134] The CPU 901 operates based on programs stored in the ROM 902 or HDD 904 and is controlled by the control unit. The ROM 902 stores boot programs executed by the CPU 901 when the computer 900 starts up, as well as programs related to the computer 900's hardware.

[0135] The CPU 901 controls input devices 910, such as a mouse or keyboard, and output devices 911, such as a display or printer, via the input / output interface 905. The CPU 901 acquires data from the input devices 910 and outputs the generated data to the output devices 911 via the input / output interface 905. In addition to the CPU 901, a GPU (Graphics Processing Unit) or the like may also be used as a processor.

[0136] The HDD 904 stores programs executed by the CPU 901 and data used by those programs. The communication I / F 906 receives data from other devices via a communication network (e.g., NW (Network) 920) and outputs it to the CPU 901, and also transmits data generated by the CPU 901 to other devices via the communication network.

[0137] The media interface 907 reads a program or data stored in the recording medium 912 and outputs it to the CPU 901 via the RAM 903. The CPU 901 loads the program related to the desired processing from the recording medium 912 onto the RAM 903 via the media interface 907 and executes the loaded program. The recording medium 912 can be an optical recording medium such as a DVD (Digital Versatile Disc) or PD (Phase Change Rewritable Disk), a magneto-optical recording medium such as an MO (Magneto Optical Disk), a magnetic recording medium, or a semiconductor memory.

[0138] For example, when computer 900 functions as the power consumption control device 10 and the in-room control device 20 of the present invention, the CPU 901 of computer 900 realizes the functions of the power consumption control device 10 and the in-room control device 20 by executing a program loaded on RAM 903. The HDD 904 stores the data in RAM 903. The CPU 901 reads and executes a program related to the desired processing from the recording medium 912. Alternatively, the CPU 901 may read a program related to the desired processing from another device via a communication network (NW 920).

[0139] <Effects> The effects of the power consumption control device 10, etc., according to the present invention will be described below. The power consumption control device according to the present invention is a power consumption control device 10 that is communicated to a room control device 20 for each room 500 that controls a plurality of servers 3 and a plurality of air conditioners 4 in a room 500 in a data center 1000. The power consumption control device 10 generates learning data 550 by controlling the load distribution pattern between rooms 500, the server load distribution pattern indicating the pattern of where the load distributed to each room is placed on the servers 3, and the air conditioning control level of the air conditioners 4 via the room control device 20 for each room 500, for each turn indicating a predetermined control time. The power consumption control device 10 determines the air conditioning control level of the air conditioners 4 in the distribution pattern indicating the distribution pattern and server load distribution pattern that maximize the power consumption cost efficiency of the entire data center 1000. In the initial learning stage of the learning data 550, the power consumption control device 10 generates learning data 550 by controlling the load distribution pattern between rooms 500, the server load distribution pattern indicating the initial temperature range inside the room 500, the load amount to be distributed to each room 500 in the distribution pattern, and the server load distribution pattern. The system has a storage unit 180 that uses patterns and air conditioning control levels as Situation components, generates an air conditioning power function that shows the air conditioning power consumption for the load allocated to each room 500, and a temperature function that shows the room temperature after the end of the turn for the load allocated to each room 500, for calculating power consumption cost efficiency, and stores Situation information associated with each Situation component as learning data 550. In the operational phase of power control of the data center 1000, when generating new Situation information including an unencountered outside temperature range that is not stored as learning data 550, the system identifies Situation information with the same initial room temperature level, server load arrangement pattern, and air conditioning control level in the outside temperature range closest to the unencountered outside temperature range, and reuses the air conditioning power function and temperature function shown in the identified Situation information as the air conditioning power function and temperature function of the new Situation information, and also uses the air conditioning control level shown in the identified Situation information.It is characterized by having a learning data repurposing unit 161 that performs repurposing learning, which involves performing predetermined adjustment processing according to the high / low relationship between the unencountered outside temperature range and the outside temperature range of the identified Situation information, and then setting it as new Situation information.

[0140] By doing so, the power consumption control device 10 can improve the optimization accuracy for reducing the overall power consumption of the DC and enhance the power saving effect by accumulating learning data during operation, without having to collect learning data for unlearned ambient temperatures. More specifically, the power consumption control device 10 only needs to perform learning in the initial learning stage, eliminating the need to set up another learning period during the operation stage, thus minimizing the operational impact on servers 3 and air conditioners 4 within the data center 1000. Furthermore, when the power consumption control device 10 reuses learning data for unencountered ambient temperature ranges from existing learning data, it identifies Situation information that is the ambient temperature range closest to the unencountered ambient temperature range, and has the same initial room temperature level, load placement pattern, and air conditioning control level, and sets new Situation information based on the learning data of the identified Situation information. As a result, the power consumption control device 10 can derive a control solution for the air conditioner 4 that is more power-saving while ensuring indoor temperature quality.

[0141] Furthermore, the power consumption control device 10 is characterized by having a learning data update unit 162 that, in turns in which control is performed using settings learned through repurposing for an unencountered outside temperature range, updates the air conditioning control level of the air conditioner 4 in the new situation information by raising or lowering it according to the difference between the room temperature showing the control result after the end of the turn and the temperature calculated by the temperature function of the source.

[0142] In this way, the power consumption control device 10 can further improve the accuracy of power consumption cost efficiency calculations by feeding back the difference between the room temperature at the end of a turn controlled using settings learned through repurposing and the temperature calculated from the learning data of the source of the repurposing, and making adjustments to the control in subsequent turns.

[0143] Furthermore, the power consumption control device 10 is characterized in that, during the operational phase of power consumption control of the data center 1000, when new Situation information is set, and Situation information is set that includes three or more load amounts with different load amounts, while the outside temperature level, initial room temperature level, server load placement pattern, and air conditioning control level are the same, the power consumption control device 10 includes a function update unit 163 that generates or updates the air conditioning power function and temperature function based on those three or more load amounts.

[0144] In this way, the power consumption control device 10 can further improve the accuracy of power consumption cost efficiency calculations by setting new situation information and generating or updating the air conditioning power function and temperature function.

[0145] Furthermore, the power consumption control device 10 is characterized in that, during the operational phase of power consumption control of the data center 1000, if the load amount exceeds three points for the same Situation information, and the load amount in any range after division by a predetermined number of divisions is three or more points, it includes a range range division unit 164 that divides the range of the outside temperature level and / or the initial temperature level inside the room.

[0146] In this way, the power consumption control device 10 can appropriately divide the range of the outside temperature level and the initial temperature level inside the room, thereby regenerating the air conditioning power function and temperature function within the divided range, and further improving the accuracy of the calculation of power consumption cost efficiency.

[0147] It should be noted that the present invention is not limited to the embodiments described above, and many modifications are possible within the technical concept of the present invention by those with ordinary skill in the art.

[0148] 1 Power Consumption Control System 3 Server 4 Air Conditioner 10 Power Consumption Control Device 20 In-Room Control Device 30 Placement Control Area 40 Air Conditioning Control Area 100, 200 Control Unit 110 Room Definition Unit 120 Learning Data Generation Unit 130 Air Conditioning Power / Temperature Function Approximation Unit 140 Distribution Placement Pattern Generation Unit 150 Optimal Control Calculation Unit 151 Power Cost Efficiency Calculation Unit 152 Control Pattern Determination Unit 160 Learning Data Complementary Unit 161 Learning Data Reuse Unit 162 Learning Data Update Unit 163 Function Update Unit 164 Range Division Unit 170, 250 Input / Output Unit 180, 260 Storage Unit 210 Situation Recognition Unit 220 Learning History Collection Unit 230 In-Room Server Control Unit 231 Load Placement Pattern Setting Unit 232 Server Power Measurement Unit 240 In-Room Air Conditioning Control Unit 241 Air conditioning control execution unit 242 Air conditioning power measurement unit 500 Room 550 Learning data 1000 Data center (DC)

Claims

1. A power consumption control device that is communicated to a room-specific control device for controlling multiple servers and multiple air conditioners in a room within a data center, wherein the power consumption control device generates learning data by controlling the load distribution pattern between the rooms, the server load distribution pattern indicating the distribution pattern of the loads distributed to each room to the servers, and the air conditioning control level of the air conditioners via the room-specific control device for each turn indicating a predetermined control time, and determines the air conditioning control level of the air conditioners in the distribution pattern and server load distribution pattern that maximizes the power consumption cost efficiency of the entire data center, and the power consumption control device, In the initial training phase of the training data, the system has a storage unit that stores Situation information associated with each of the Situation components, including an outside temperature level indicating the range of outside temperatures outside the data center, an initial room temperature level indicating the range of initial temperatures inside the room, the load amount distributed to each room according to the distribution pattern, the server load placement pattern, and the air conditioning control level, and generates an air conditioning power function indicating the air conditioning power consumption for the load amount distributed to each of the rooms, and a temperature function indicating the room temperature after the end of the turn for the load amount distributed to each of the rooms, for calculating the power consumption cost efficiency, and stores Situation information associated with each of the Situation components as the training data.A power consumption control device is characterized by comprising a learning data repurposing unit that performs repurposing learning, which, in the operational phase of power consumption control of the data center, generates new Situation information including an unencountered outside temperature range that is not stored as learning data, identifies Situation information with the same initial room temperature level, server load arrangement pattern, and air conditioning control level in the outside temperature range closest to the unencountered outside temperature range, repurposes the air conditioning power function and temperature function shown in the identified Situation information as the air conditioning power function and temperature function of the new Situation information, and sets the air conditioning control level shown in the identified Situation information to the new Situation information after performing a predetermined adjustment process according to the high / low relationship between the unencountered outside temperature range and the outside temperature range of the identified Situation information.

2. The power consumption control device according to claim 1, characterized in that, in a turn in which control is performed using the settings obtained by the transfer learning for an unencountered outside temperature range, the power consumption control device is further updated by raising or lowering the air conditioning control level of the air conditioner in the new situation information according to the difference between the room temperature showing the control result after the end of the turn and the temperature calculated by the temperature function of the transfer source.

3. The power control device according to claim 2, characterized in that, in the operational stage of power control of the data center, when the new Situation information is set, and Situation information is set that includes three or more load amounts with different load amounts in the same Situation information, with the same outside temperature level, the initial room temperature level, the server load arrangement pattern, and the air conditioning control level, the power control device according to claim 2 is further characterized in that it includes a function update unit that generates or updates the air conditioning power function and the temperature function based on the three or more load amounts.

4. The power control device according to claim 3, characterized in that, during the operation phase of power control of the data center, the load amount exceeds three points for the same Situation information, and the load amount in any range after division by a predetermined number of divisions is three or more points, the power control device is further characterized in that it includes a range division unit that divides the range of the outside temperature level and / or the initial temperature level in the room.

Images