Power consumption prediction system and power consumption prediction method

Abstract

To accurately predict power consumption of a server.SOLUTION: A power consumption prediction system that predicts power consumption of a server comprises: a constraint condition definition unit that defines a constraint condition associated with a dependence relation between power consumption of an information processing instrument and power consumption of a cooling apparatus; a power consumption prediction unit that predicts time-lag from change in power consumption of an information processing apparatus to change in power consumption of a cooling apparatus; a correction unit that corrects a predicted value of power consumption using time-lag; and an output unit that outputs a result of the correction. When a direct dependence relation between the power consumption of the information processing apparatus and the power consumption of the cooling apparatus is strong, the power consumption prediction system predicts time-lag from the change in the power consumption of the information processing apparatus to the change in the power consumption of the cooling apparatus on the basis of the dependence relation, and when the direct dependence relation is weak, the power consumption prediction system predicts time-lag from the change in the power consumption of the information processing apparatus to the change in the power consumption of the cooling apparatus through a mediation parameter.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a power consumption prediction system and a power consumption prediction method. [Background technology]

[0002] When predicting the power consumption of a server, it is necessary to take into consideration the time dependency of the power consumption of the information processing device and the power consumption of a cooling device that cools the information processing device in response to heat generation. Patent Document 1 discloses a technology for determining time dependency and inter-time-series dependency in multivariate time series data. It discloses that multiple time series data are encoded into respective recurrent neural networks (RNNs), and then decoded into the RNN based on the calculated time context vector to determine the time dependency relationship between multiple time series data. In addition, as an application example of the disclosed technology, it discloses prediction of a time lag occurring between multiple time series data. Specifically, the Japanese Abstract of the US Patent Publication 1 states that "the system of the present invention includes a memory storing computer-executable components, and a processor operatively coupled to the memory and executing the computer-executable components stored in the memory, where the computer-executable components include a computation component that encodes at least two recurrent neural networks (RNNs) having respective time series data and determines at least two decoded RNNs based on at least two temporal context vectors to determine time dependency of the at least two time series data, a combination component that combines the at least two decoded RNNs and a component that determines an inter-time series dependent context vector and an RNN dependent decoder, and an analysis component that determines time series dependencies in the at least two time series data and predicted values ​​for one or more time series data based on an RNN encoder and an RNN dependent decoder using a neural network based on an attention mechanism." [Prior art documents] [Patent documents]

[0003] [Patent Document 1] US2019 / 0354836A1 Summary of the Invention [Problem to be solved by the invention]

[0004] When predicting the power consumption of a server, the dependency between the power consumption of information processing equipment and cooling equipment is not uniform. Because the usage conditions of each server room are different, there are cases where the direct dependency can be observed to be strong and cases where the direct dependency is weak. In particular, when the direct dependency between the power consumption of information processing equipment and cooling equipment is weak, accurate prediction is difficult even with the above-mentioned conventional technology. Furthermore, if the amount of power available in the future is specified, the power consumption of information processing equipment and cooling equipment must be covered within that range. Because there is a time lag between changes in the power consumption of information processing equipment and that of cooling equipment, highly accurate prediction of the time lag is important in order to align the total power consumption at a certain point in time with the value specified in the plan. An example of a case where the amount of electricity that can be used in the future is specified is when surplus renewable energy is consumed by a server. The amount of electricity generated by renewable energy is not constant, so it does not necessarily match demand, resulting in a surplus. Therefore, rather than supplying electricity according to the amount of operation of the server, there is a demand to advance the decarbonization of servers by determining the amount of operation of the server in relation to the amount of electricity that can be supplied. An object of the present invention is to predict the power consumption of a server with high accuracy. [Means for solving the problem]

[0005] In order to achieve the above object, one representative power consumption prediction system of the present invention is a power consumption prediction system that predicts the power consumption of a server having an information processing device that generates heat as it operates and a cooling device that cools the information processing device in response to heat generated by the information processing device, and includes a constraint condition definition unit that defines constraint conditions relating to a dependency between the power consumption of the information processing device and the power consumption of the cooling device, a power consumption prediction unit that predicts a time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device, a correction unit that corrects predicted values ​​of power consumption of the information processing device and the cooling device using the time lag, and outputs a result of the correction made by the correction unit. and an output unit, wherein the constraint condition definition unit defines a direct dependency between the power consumption of the information processing device and the power consumption of the cooling device, a dependency between the power consumption of the information processing device and an intermediary parameter, and a dependency between the intermediary parameter and the power consumption of the cooling device, and the power consumption prediction unit predicts a time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device based on the dependency when the direct dependency is strong, and predicts the time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device via the intermediary parameter when the direct dependency is weak. Furthermore, one representative power consumption prediction method of the present invention is a power consumption prediction method for predicting the power consumption of a server having an information processing device that generates heat as it operates and a cooling device that cools the server in response to the heat generated by the information processing device, wherein the power consumption prediction system includes the steps of: defining constraint conditions related to a dependency between the power consumption of the information processing device and the power consumption of the cooling device; predicting a time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device; correcting predicted values ​​of the power consumption of the information processing device and the cooling device using the time lag; and outputting a result of the correction, wherein the step of defining constraint conditions defines a direct dependency between the power consumption of the information processing device and the power consumption of the cooling device, a dependency between the power consumption of the information processing device and an intermediate parameter, and a dependency between the intermediate parameter and the power consumption of the cooling device, and the step of predicting the time lag predicts a time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device based on the dependency when the direct dependency is strong, and predicts the time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device via the intermediate parameter when the direct dependency is weak. Effect of the Invention

[0006] According to the present invention, the power consumption of a server can be predicted with high accuracy. Problems, configurations and effects other than those described above will become apparent from the following description of the embodiments. [Brief description of the drawings]

[0007] [Figure 1] FIG. 1 is a functional block diagram of a power consumption prediction system in a data center. [Diagram 2] FIG. 2 is a functional block diagram of a data input unit of the renewable energy forecasting system. [Diagram 3] FIG. 2 is a functional block diagram of a constraint condition definition unit of the renewable energy prediction system. [Figure 4] 13 is an example of a data configuration of a renewable energy usage plan table. [Diagram 5] FIG. 2 is a functional block diagram of a power consumption prediction unit of the renewable energy prediction system. [Figure 6] FIG. 11 is a functional block diagram of a correction unit for prediction results of the renewable energy prediction system. [Figure 7] FIG. 2 is a functional block diagram of a data output unit of the renewable energy forecasting system. [Figure 8] This is a flowchart showing how the renewable energy forecasting system predicts and corrects the time lag in the power consumption of IT equipment and cooling facilities. [Figure 9] 13 is a flowchart for determining dependencies between time series data for each server room and calculating correlation coefficients using each data set. [Figure 10] 1 is a flowchart for creating a predictive model based on correlation coefficients by machine learning. [Figure 11] 13 is a flowchart for predicting a time lag from a predicted result of a correlation coefficient. [Figure 12] 11 is a flowchart for calculating a time lag correction value according to a renewable energy utilization plan. [Figure 13] 11 is a flowchart of a process for correcting and visualizing a predicted value of power consumption. [Figure 14] 13 is an example of a management screen displayed on a DC user terminal. [Figure 15] FIG. 1 is a diagram illustrating a first embodiment of a power consumption prediction system in a data center. [Figure 16] FIG. 13 is a diagram illustrating a second embodiment of a power consumption prediction system in a data center. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] In recent years, the decarbonization of data centers (DCs) has been attracting a great deal of attention due to the explosive increase in power consumption at DCs. As renewable green energy sources such as solar power generation, wind power generation, and biopower generation are increasingly being introduced, the amount of surplus power that exceeds the supply and demand balance of the power grid has become non-negligible. Therefore, it is necessary to adjust the amount of work in the DC according to the surplus power. In this case, a consumption plan for electricity generated by renewable energy is given, and work is assigned according to the consumption plan, so it is important to accurately predict the power consumption of the DC. The power consumption of a DC is mainly composed of information processing equipment (IT equipment) and cooling equipment (cooling facilities), and there is a time lag between an increase in the power consumption of the former and an increase in the power consumption of the latter. The dependency between this IT equipment and cooling facilities varies greatly depending on the environment of the server room. In the system disclosed in the embodiment, if the dependency between the IT equipment and the cooling equipment is strong, the system predicts the time lag from the dependency. If the dependency between the IT equipment and the cooling equipment is weak, the system predicts the time lag via an intermediary parameter. The intermediary parameter is a parameter that depends on both the IT equipment and the cooling equipment, and the room temperature and humidity in the server room can be used. The system disclosed in the embodiments can accurately predict the time lag in power consumption between IT equipment and cooling equipment, thereby accurately predicting server power consumption and enabling effective use of renewable energy. For example, in a locally produced and consumed container-type DC, it is possible to control IT workloads according to the amount of renewable energy supplied. In addition, IT operations currently being operated in urban DCs can be transferred to regional DCs with abundant natural resources, reducing the pressure on electricity in urban areas while also contributing to the revitalization of regional areas. EXAMPLES

[0009] Figure 1 shows a functional block diagram of a power consumption prediction system in a data center. Here, three types of DCs are assumed as DC operators 3: a hosting DC 10a, a container-type DC 10b, and a hybrid DC 10c, and each DC 10 can collect data on IT equipment power consumption 11, cooling equipment power consumption 12, sensor group control 13, and management and monitoring of renewable energy production volume 14. The data collected from each DC is transferred to a renewable energy prediction system 20 via a network 5 and accumulated in a database 30. Note that DC 10 is just an example of a DC, and is not limited to these three types of DCs.

[0010] IT equipment power consumption 11 is collected for each server or rack in the DC. Cooling equipment power consumption 12 is not limited to air conditioners that perform demand control, but data is collected for each number of units. Sensor group control 13 is data collected by installing room temperature sensors, humidity sensors, etc. Renewable energy production management and monitoring 14 is data that collects the production volume of green energy generated from power generation devices such as solar and wind power installed in the DC in chronological order. Each data is managed collectively, and the generated dataset can be used as a learning dataset for machine learning.

[0011] In this example, the renewable energy prediction system 20 performs correlation analysis on two time series data of power consumption of IT equipment and cooling equipment of a DC, and predicts the time lag between the two. In order to effectively utilize surplus renewable energy, the renewable energy prediction system 20 flexibly adjusts the IT workload according to a renewable energy operation plan formulated in advance. In this adjustment, the renewable energy prediction system 20 can predict the power consumption of the cooling equipment with high accuracy by correcting with the predicted time lag. To perform such processing, first, data collected from each DC 10 is introduced into the user data input unit 40, and each influencing factor is determined for each server room of the DC based on the constraint conditions defined in advance by the constraint condition definition unit 50. The constraints that are defined in advance are a first constraint regarding the dependency of factors that affect the IT and cooling equipment in each server room, and a second constraint that corrects the predicted time lag in accordance with the dependency of the first constraint. When there is a strong correlation between the power consumption of IT equipment and cooling equipment, the method of calculating the difference from the original data of each power consumption is often used in statistics. The reason for introducing the calculation of the difference data this time is to remove the effect of the outside temperature on the power consumption of IT equipment and cooling equipment. Next, the power consumption prediction unit 60 uses the difference data to calculate a correlation coefficient from each piece of data using the rolling window time-lagged cross correlations (RWTLCC) method. The RWTLCC method is often used in data analysis. Using the calculated correlation coefficient, the power consumption prediction unit 60 creates a learning model for predicting the time lag between the power consumption of the IT equipment and the cooling equipment by machine learning, predicts the time lag with each common influence parameter from the learning model, and calculates the time lag between the IT equipment and the cooling equipment by adding up the predicted time lags. Then, the prediction result correction unit 70 calculates a similarity from the intercept of the power consumption requirement of the renewable energy implementation plan and the change in the power consumption of the IT equipment in the server room based on the renewable energy operation plan defined in advance by the constraint condition definition unit 50, calculates a correction value of the time lag between the IT equipment and the cooling equipment according to the calculated similarity, and corrects the predicted time lag. By performing such processing, it is expected that the accuracy of the predicted value of the time lag between the IT equipment and the cooling equipment can be improved. The power consumption of each IT equipment, the power consumption of the cooling equipment, the predicted result of the time lag between them, and the remaining amount of renewable energy can be provided to the DC business operator 3 and general users 90 by the prediction data output unit 80 as the output of the renewable energy prediction system 20.

[0012] FIG. 2 shows a functional block diagram of the user data input unit 40 of the renewable energy prediction system. The raw data collected from each type of DC 10 is accumulated in the data accumulation DC 21. When prediction starts, the data reading program 100 accesses the data accumulation DB and reads the raw data. If a learning period needs to be specified, the data learning period can be specified by the data use period specification program 110. After various data are read, it is desirable to format the data required for learning by the metrics extraction program 120 and perform data cleansing. In addition, the center data processing program 130 can perform the above series of processes (data reading, specification of the use period, extraction of metrics, and data cleansing process) for the data collected for the sensor group control 13. In order to make various data easy to use, it is desirable to process the raw data in a well-organized form after shaping and data cleansing.

[0013] FIG. 3 shows a functional block diagram of the constraint condition definition unit 50 of the renewable energy prediction system. The constraint condition definition unit 50 is mainly configured to have a dependency relationship extraction unit 31 of the influencing factors of the power consumption of the IT equipment and the cooling equipment and a renewable energy utilization planning unit 32 in order to define the first constraint condition and the second constraint condition in advance. In the former, in order to deal with the difference in the correlation between the IT equipment and the cooling equipment for each server room of the DC, the strength of the correlation between the two is analyzed by a correlation analysis program 200 of the power consumption of the IT equipment and the cooling equipment. If the correlation is strong, the program is terminated. On the other hand, in the case where the correlation between the power consumption of the IT equipment and the cooling equipment is not strong, the present invention analyzes the correlation with a parameter that is commonly influenced by both. For example, there is a positive correlation characteristic in which the room temperature increases due to an increase in the power consumption of the IT equipment, and the power consumption of the cooling equipment increases due to the increase in the room temperature. In addition, as with the room temperature, a negative correlation characteristic can be captured in which the humidity in the room decreases due to an increase in the power consumption of the IT equipment, and the power consumption of the cooling equipment decreases due to the decrease in the humidity. In order to determine the influence of each room temperature element and humidity element, it is desirable to perform calculations using a correlation analysis program 210 between the power consumption of IT equipment and room temperature, a correlation analysis program 220 between the power consumption of IT equipment and humidity, and a correlation analysis program 230 between the power consumption of IT equipment and humidity, and a correlation analysis program 240 between humidity and power consumption of cooling equipment. Note that the room temperature and humidity listed in the dependency relationship extraction unit 31 of the influence elements of the power consumption of IT equipment and cooling equipment are only examples, and are not limited to these two types. In addition, the renewable energy utilization planning unit 32 stores a renewable energy operation plan table 300, and is indispensable for collectively carrying out utilization planning of surplus renewable energy, management of DC, job management, and the like.

[0014] FIG. 4 shows an example of the data configuration of the renewable energy operation plan table 300. The renewable energy usage plan is essential for making a plan while simulating which DC and which location are targeted, and how much renewable energy should be adjusted for how long, based on the surplus amount of renewable energy at each DC. The DC name 301 is a symbol of a building specially made for installing servers and network equipment, and is mainly composed of symbols and numbers. The location 302 is written as a symbol for indicating the location of the DC. In addition, the job ID 303 is a unique ID for recording a job running on the system. The implementation plan date and time 304 allows the user to set a planned time point or time period when introducing renewable energy into the server room of the DC. The planned increase amount 305 of IT equipment power consumption indicates the amount of change in power consumption due to the adjustment of IT workload based on the implementation plan date of the specified renewable energy. The predicted surplus amount of renewable energy 306 indicates how much renewable energy is currently remaining.

[0015] FIG. 5 shows a functional block diagram of the power consumption prediction unit 60 of the renewable energy prediction system. The power consumption prediction unit is mainly composed of a learning data creation unit 41, a learning model generation unit 42, and a time lag prediction unit 43 between the IT equipment and the cooling equipment. Regarding the learning data creation unit 41, using the processing data created in FIG. 2, first, based on the correlation strength of the dependency of each data set, the influential factors are determined, and then, using the difference calculation program 400, the difference from each original data set and a certain period is obtained, and the difference data is calculated. In addition, using the calculated difference data, a correlation coefficient calculation program 410 is used to determine the window size by the RWTLCC method, and the correlation coefficient of two time series data within each window size is calculated.

[0016] The calculated correlation coefficient is used as learning data, and the learning model generation unit 42 separates the training data and the test data, and creates a learning model by the learning model creation program 420 for each time lag difference using each learning data. The prediction unit 43 of the time lag between the IT equipment and the cooling equipment can predict the correlation coefficient for each future time lag difference by using the learning model obtained by the learning model creation program 420 and the future correlation coefficient prediction program 430. It is preferable to extract the time lag value when the correlation coefficient predicted for each time lag difference is maximum for each influencing element by the time lag extraction program 440 when the correlation coefficient is maximum, and calculate the sum of each time lag by the time lag processing program 450 according to the dependency relationship of each influencing element. For example, by using the above method, the time lag between the power consumption of the IT equipment and the cooling equipment can be calculated by adding both the time lag between the IT equipment and the room temperature and the time lag between the room temperature and the cooling equipment.

[0017] FIG. 6 shows a functional block diagram of the prediction result correction unit 70 of the renewable energy prediction system. The prediction result correction unit is mainly composed of a time lag correction unit 51 and a power consumption correction unit 52. In the former, first, a value in a predetermined minute time width (intercept) is acquired for each server room at a timing when the power consumption changes drastically, and an intercept analysis program 500 that analyzes the power consumption change amount for each intercept calculates the increase / decrease relationship between the power consumption change amount for each intercept and the time lag. Next, based on the renewable energy operation plan, a similarity calculation program 510 calculates the similarity between the power consumption increase amount of the IT equipment and the power consumption change amount for each intercept calculated by the intercept analysis program 500 for the set implementation date. Then, a time lag correction value calculation program 520 calculates the correction value of the time lag calculated from the change amount at the intercept that is most suitable for the actual power consumption increase amount of the IT equipment. By correctly calculating the time lag for the power consumption increase of the IT equipment, it is possible to make a correction corresponding to the predicted future change in the time lag on the scheduled implementation date of the renewable energy operation plan. The power consumption correction unit 52 has a correction program 530 for correcting the predicted value of the power consumption of the cooling equipment. The correction program 530 corrects the predicted value of the power consumption of the cooling equipment. Therefore, the power consumption correction unit 52 can highly accurately predict the future power consumption of the cooling equipment on the scheduled implementation date of the future renewable energy operation plan, based on the increase in the power consumption of the IT devices and the corrected value of the time lag.

[0018] FIG. 7 shows a functional block diagram of the data output unit 80 of the renewable energy prediction system. The data output unit 80 has a visualization program 600 for the prediction result after time lag correction, a visualization program 610 for the prediction result after correction of the power consumption of the cooling equipment, and a visualization program 620 for the prediction result of the surplus renewable energy amount after correction. The visualization program 600 visualizes the prediction result before and after the correction of the time lag calculated by the prediction result correction unit 70. The visualization program 610 visualizes the prediction result before and after the correction of the power consumption amount of the cooling equipment. The visualization program 620 visualizes the remaining amount of surplus renewable energy after correction. The data output unit 80 can provide the visualized information to a user via a network.

[0019] FIG. 8 shows a flowchart in which the renewable energy prediction system 20 predicts and corrects the time lag in the power consumption of IT devices and cooling facilities.

[0020] First, the DC operator 3 requests the renewable energy forecasting system (REPS) 20 to accumulate collected data (s1). The renewable energy forecasting system (REPS) 20 reads the power consumption data of IT devices and cooling facilities provided by the DC operator 3 and performs preprocessing of the raw data (s2) (see Fig. 2). Next, it performs processing to read sensor data collected from the sensor group for each server room of the DC (s3), and then performs processing to extract metrics required for the read sensor data (s4), combines the data processed in s2 and s4, and creates learning data (s5).

[0021] The renewable energy forecasting system (REPS) 20 uses the combined learning data to analyze and determine the correlation between the power consumption of the IT equipment and the cooling equipment for each server room, and the dependency between the parameters that are commonly affected by both (s6). If the correlation between the power consumption of the IT equipment and the cooling equipment is strong (s6; Yes), the process of calculating the correlation coefficient from the differential data (s9) is performed directly. On the other hand, if the correlation between the power consumption of the IT equipment and the cooling equipment is not strong (s6; No), the dependency of the intermediate parameters is established (s7). For example, the correlation between the IT equipment and the room temperature, and the correlation between the room temperature and the cooling equipment are established. Note that a weak correlation between the power consumption of the IT equipment and the cooling equipment includes no correlation. The established dependency parameter information is provided to the hosting user terminal 4a, the container-type DC user terminal 4b, the hybrid DC user terminal 4c, etc. (s8). Note that each user terminal is an example, and is not limited to these three types of user terminals. Furthermore, the difference between each of the data for which the dependency has been established and a certain time from the original data is obtained by calculating the correlation coefficient from the differential data (s9), and the differential data is calculated. As mentioned above, this is to eliminate the effect of outside temperature on the power consumption of IT equipment and cooling equipment.

[0022] Next, the details of the process shown in s9, that is, a flowchart for determining the dependency between time series data for each server room and calculating the correlation coefficient using each data set, is shown in FIG.

[0023] First, the renewable energy forecasting system (REPS) 20 establishes a dependency relationship s21 between the power consumption of the IT equipment and the cooling equipment using the learning data generated in s5. If the correlation between the power consumption of the IT equipment and the cooling equipment is strong (S21; Yes), the system directly calculates the difference data in s23. On the other hand, if the correlation between the power consumption of the IT equipment and the cooling equipment is not strong (S21; No), the system establishes a dependency relationship between the power consumption of the IT equipment and the cooling equipment and the parameters that are commonly affected by them in s22. From there, the system calculates the difference data between the parameters that are commonly affected by them in s23. Next, the window size is determined by the RWTLCC method, and the correlation coefficient is calculated from each difference data (s24) (see FIG. 5).

[0024] Returning to the explanation of Figure 8, the renewable energy forecasting system (REPS) 20 uses the difference data calculated in s9 to predict the correlation coefficient s10 (see Figure 5). From the predicted correlation coefficients for each time lag, the correlation coefficient s11 when the correlation coefficient is maximum within each time lag is obtained, and from the obtained correlation coefficient, the predicted time lag (time lag) from each data set can be obtained.

[0025] Next, the details of the process shown in s10, that is, a flowchart for creating a correlation coefficient prediction model by machine learning using differential data, is shown in FIG.

[0026] The renewable energy forecasting system (REPS) 20 introduces the time series correlation coefficient data calculated by the RWTLCC method as learning data into a learning device s31. The learning data itself is also divided into training data and test data (S32). Then, using the training data, a learning model is created based on the random forest learning method (S33). Finally, the future correlation coefficient predicted by the learning model created in S33 is output as output data (S34).

[0027] Next, the details of the process shown in s11, that is, a flowchart of the process of predicting the time lag from the predicted result of the correlation coefficient, is shown in FIG.

[0028] First, the renewable energy forecasting system (REPS) 20 uses each time series correlation coefficient data to extract the maximum correlation coefficient for each time lag from the predicted future correlation coefficients (s41). Next, it calculates the time lag value corresponding to when each extracted future correlation coefficient is at its maximum (s42). Finally, it adds up the time lags calculated from each influence parameter according to the established dependency (s43). As a result, it becomes possible to predict the time lag between the power consumption of IT equipment and cooling equipment.

[0029] Returning to the explanation of FIG. 8, the DC business operator 3 refers to the surplus amount of renewable energy and again requests the renewable energy prediction system 20 to request constraint conditions (s12). The renewable energy prediction system (REPS) 20 performs correction processing (s13) of the time lag value predicted in s11 (see FIG. 6). For the corrected time lag value, correction processing (s14) of the power consumption of the cooling equipment is performed according to the change. The renewable energy prediction system (REPS) 20 requests the DB to write the processing results such as the value before and after correction of the predicted time lag, the value before and after correction of the power consumption of the cooling equipment, and the remaining amount of renewable energy (s15). Finally, the result after correction of the predicted power consumption of the cooling equipment is transferred to the hosting user terminal 4a, the container-type DC user terminal 4b, and the hybrid DC user terminal 4c (s16), and it is desirable to perform visualization processing by the user terminal.

[0030] Next, the details of the process shown in s13, that is, a flowchart of the correction process of the predicted time lag, is shown in FIG.

[0031] The renewable energy forecasting system (REPS) 20 first extracts constraint requirements s51 from the renewable energy implementation plan. These constraint requirements include, for example, the implementation date and the increase in power consumption of IT equipment. Next, for each server room, the amount of change in power consumption is analyzed for each predetermined small time range (intercept) at a timing when the change in power consumption is drastic (s52). Then, based on the renewable energy operation plan, a similarity calculation is performed between the increase in power consumption of IT equipment and the change in power consumption for each intercept for the set implementation date (s53). Finally, a correction value for the time lag calculated from the change at the intercept that is most suitable for the actual increase in power consumption of the IT equipment is calculated according to the calculated similarity (s54). As a result, the time lag can be correctly calculated for the increase in power consumption of IT equipment, and it is possible to appropriately correct the predicted future change in time lag on the scheduled implementation date of the renewable energy operation plan.

[0032] Next, the details of the process shown in s14, that is, a flowchart of the correction process and visualization of the predicted value of power consumption, are shown in FIG.

[0033] First, the renewable energy forecasting system (REPS) 20 uses the calculated corrected time lag value to calculate a corrected predicted value of the power consumption of the cooling equipment (s61). Next, a visualization process is performed for the corrected predicted value of the power consumption of the cooling equipment (s62). Finally, a visualization process is performed for the predicted value of the surplus renewable energy amount (s63).

[0034] Returning to the explanation of FIG. 8, based on the above, metrics are extracted from the data accumulated by the data center operator 3, and a learning dataset is created (s5), and a dependency relationship corresponding to the difference in correlation between IT equipment and cooling equipment is established for each server room (s7). Then, by calculating the difference data for each dataset and calculating the correlation coefficient from the difference data (s9), it becomes possible to predict the future correlation coefficient (s10). In addition, the time lag when the correlation coefficient is maximum for each time difference can be calculated from each predicted future correlation coefficient (s11). Finally, by correcting the predicted value of the time lag according to the renewable energy operation plan, the predicted value of the shifted time lag according to the implementation date of the renewable energy can also be corrected (s13), and the predicted value of the future power consumption of the cooling equipment can also be corrected (s14), and it is expected that the prediction accuracy of the future cooling equipment corresponding to the adjustment of the renewable energy that is implemented irregularly can be improved.

[0035] FIG. 14 shows an example of a management screen displayed on a DC user terminal. A selection box 1010 is arranged on the left side of the management screen 1000. A user of the renewable energy prediction system can use the renewable energy prediction system mainly in the following two steps. First, in step 1, the renewable energy prediction system sets the implementation time and constraint conditions related to the formulation of the operation plan of the DC operator. Next, in step 2, for the selected DC (including location information, etc.) and the selected server room 1030, the dependency of the influencing factors on the IT equipment and the cooling equipment, the change status before and after correction of the time lag between the predicted IT equipment and the cooling equipment, and the change status before and after correction of the power consumption of the cooling equipment can be confirmed. FIG. 14 shows the prediction results.

[0036] A user can select the renewable energy plan generation to be displayed in a selection box 1010. Here, description items are provided for a requirement definition screen 1020 of a renewable energy usage plan. When a user makes a plan for utilizing surplus renewable energy, first, the user checks the remaining amount of surplus renewable energy, and enters in 1021 the implementation time constraints on the date on which the renewable energy usage is desired to be implemented for the IT workload and the specific increase in power consumption of the IT equipment. The renewable energy forecasting system displays the details of the renewable energy usage plan for the described conditions as an output result in 1022 on the user's screen. If the user wants to edit the described conditions again, the user can modify the description of the date to be implemented again and the constraint conditions in the selection box 1010 for editing the renewable energy requirements. The renewable energy forecasting system displays the modified renewable energy usage plan again as an output result in 1022 on the user's screen.

[0037] After the user has finished defining the requirements for the renewable energy usage plan in Step 1, they can click the "Next" selection box on the screen to move on to the Step 2 screen.

[0038] In step 2, first, the user selects a DC that he / she wishes to use on a selection panel 1031. The renewable energy prediction system shows the user a list of information about each server room managed by the selected DC, and prompts the user to select a server room on a selection panel 1032. Then, the system shows the user a graph 1033 showing the correlation between the influential factors of the IT equipment and cooling facilities in the selected server room.

[0039] Next, the renewable energy prediction system presents to the user in a simulator 1034 the results of predicting the time lag between the IT equipment and the cooling equipment for the constraint conditions defined by the user in step 1 (including both before and after correction). The system also presents the impact of the correction for the portion where the predicted time lag is shifted from the defined renewable energy usage plan. Specifically, by correcting the predicted value of the power consumption of the cooling equipment using the corrected predicted time lag value, it becomes possible to present to the user in 1035 the simulation result of how much the power consumption of the cooling equipment changes in response to the change in the power consumption of the IT equipment.

[0040] By predicting the time lag between IT equipment and cooling equipment obtained through the above, or based on the renewable energy usage plan, it will be possible to correct the predicted time lag and the predicted power consumption of the cooling equipment, which is expected to make it easier to create plans to use surplus renewable energy and promote the decarbonization of data centers.

[0041] FIG. 15 shows the first form of the power consumption prediction system in the data center described above. In the first form, the DC business operator is, for example, a hosting DC or a container type DC. The DC business operator predicts and corrects the utilization of surplus renewable energy by each DC itself. First, in the formulation of a renewable energy utilization plan 2001, information such as the amount of adjustment of renewable energy and the implementation date is described for the DC and server room selected by the user. The details are the same as those explained in the constraint condition definition part of the renewable energy prediction system in FIG. 3. The renewable energy prediction system (REPS) 2002 inputs learning data and performs prediction and correction processing of the time lag of the power consumption of the IT equipment and the cooling equipment, and correction processing of the time lag based on the input constraint conditions. It also predicts and corrects the power consumption of the cooling equipment before and after the correction. The output 2003 of each prediction result is provided to the user terminal via a network. It is desirable to build the renewable energy prediction system in each DC. It is also possible to build the renewable energy prediction system itself on the cloud system side that has made a contract.

[0042] FIG. 16 shows a second embodiment of the power consumption prediction system in a data center. In the second embodiment, the DC operator operates, for example, a hybrid DC. The DC operator can support the use of renewable energy at the source and destination of the IT workload transfer in terms of the utilization of surplus renewable energy. A renewable energy utilization plan is formulated 2001 from the source DC 2004, and the transfer amount of IT workload, the remaining amount of surplus renewable energy at the destination, the amount of IT workload that can be accepted, and the constraint conditions of the implementation period are registered in the renewable energy prediction system 2005. Then, a transfer destination 2006 that satisfies the constraint conditions is found, and the renewable energy operation plan is implemented. Specifically, first, the time lag of the power consumption of the IT equipment and cooling equipment of the destination DC is predicted. Next, the constraint conditions are extracted from the renewable energy operation plan described by the source, and the similarity is calculated for each intercept of the change in the power consumption of the IT equipment from the destination based on the constraint conditions, and the correction calculation of the time lag between the IT equipment and the cooling equipment is performed according to the calculated similarity. Then, the power consumption of the cooling equipment at the relocation destination is predicted and corrected. The output 2003 of each prediction result is provided to the user terminal of the relocation source's planner via the network. In addition, the time lag of the power consumption of the IT equipment and cooling equipment after correction at the relocation destination and the data record of the power consumption of the cooling equipment after correction are accumulated in the DC of the relocation destination. Note that it is preferable to build the renewable energy prediction system on the cloud system side where the relocation source and relocation destination are managed in an integrated manner. However, it is not limited to being built on a cloud system.

[0043] As described above, the disclosed power consumption prediction system is a power consumption prediction system that predicts the power consumption of a server (10) having an information processing device that generates heat as it operates and a cooling device that cools the information processing device in response to the heat generated by the information processing device, and includes a constraint condition definition unit 50 that defines constraint conditions related to the dependency between the power consumption of the information processing device and the power consumption of the cooling device, a power consumption prediction unit 60 that predicts the time lag from a change in the power consumption of the information processing device to a change in the power consumption of the cooling device, a correction unit 70 that corrects the predicted values ​​of the power consumption of the information processing device and the cooling device using the time lag, and an output unit 80 that outputs the result of the correction by the correction unit 70. The constraint condition definition unit 50 defines a direct dependency between the power consumption of the information processing device and the power consumption of the cooling device, a dependency between the power consumption of the information processing device and an intermediary parameter, and a dependency between the intermediary parameter and the power consumption of the cooling device, and the power consumption prediction unit predicts a time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device based on the dependency if the direct dependency is strong, and predicts a time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device via the intermediary parameter if the direct dependency is weak. With this configuration and operation, the power consumption of the server can be predicted with high accuracy.

[0044] In addition, in the disclosed system, when the information processing equipment and the cooling equipment are installed in multiple server rooms, the constraint condition definition unit 50 defines constraint conditions regarding dependencies for each server room, the power consumption prediction unit 60 predicts the time lag for each server room, and the correction unit 70 performs the correction for each server room. This allows for highly accurate predictions tailored to the environment of each server room.

[0045] In addition, the correction unit 70 performs correction based on a plan indicating the future amount of available power so that the predicted values ​​of the power consumption of the information processing device and the cooling device match the amount of available power indicated in the plan. As an example, the plan indicating the amount of future available electricity is a plan to use surplus electricity generated by renewable energy sources. In this way, the disclosed system contributes to the effective use of renewable energy.

[0046] Furthermore, the disclosed system allocates to the information processing device an amount of operation corresponding to the power consumption of the information processing device based on the correction result by the correction unit 70. This allows the amount of operation to be adjusted according to the power supplied.

[0047] In addition, the disclosed system has the information processing equipment and the cooling equipment provided in each of a plurality of data centers, and allocates operation amounts among the plurality of data centers based on the plan and the corrected results of power consumption of the information processing equipment and the cooling equipment in the plurality of data centers. This allows for optimization of operations across multiple data centers.

[0048] In addition, the power consumption prediction unit 60 classifies data on changes in power consumption of the information processing device and the cooling device according to the strength of the direct dependency, generates differential data for each classification, and performs machine learning using the changes in power consumption of the information processing device and the changes in power consumption of the cooling device over a specified time span as training data, thereby generating a learning model for determining the time lag. With this configuration and operation, the time lag can be easily determined.

[0049] As an example, the intermediate parameter is the room temperature in a server room, and the constraint condition definition unit defines a dependency between the power consumption of the information processing equipment and the room temperature, and a dependency between the room temperature and the power consumption of the cooling equipment. As another example, the intermediate parameter is humidity in a server room, and the constraint condition definition unit defines a dependency relationship between the power consumption of the information processing device and the humidity, and a dependency relationship between the humidity and the power consumption of the cooling device. In this way, by utilizing any parameter that has a dependency between the information processing device and the cooling device, the power consumption can be calculated with high accuracy.

[0050] Although the embodiments of the present invention have been described above, the present invention is not limited to the above-mentioned embodiments, and various modifications are included. In addition, the above-mentioned embodiments are described in detail to clearly explain the present invention, and the present invention is not necessarily limited to those having all of the described configurations. In addition, it is possible to add, delete, or replace a part of the configuration of each embodiment with another configuration. [Explanation of symbols]

[0051] 1: Power consumption prediction system, 3: DC operator, 5, 6: Network, 10: DC, 11: IT equipment power consumption, 12: Cooling equipment power consumption, 13: Sensor group control, 14: Renewable energy production management and monitoring, 20: Renewable energy prediction system (REPS), 30: DB for storing collected data, 40: User data input unit (UDC), 50: Constraint condition definition unit (CC), 60: Power consumption prediction unit (PEP), 70: Prediction result correction unit (PC), 80: Data output unit (PDO), 21: Data storage DB, 100: Data reading program, 110: Data usage period designation program, 120: Metrics extraction program, 130: Sensor data processing program, 31: Dependency extraction unit for influencing factors of power consumption of IT equipment and cooling equipment, 32: Renewable energy usage planning unit, 200: Correlation analysis program for power consumption of IT equipment and cooling equipment, 210: Correlation analysis program for power consumption of IT equipment and room temperature, 220: Correlation analysis program for room temperature and power consumption of cooling equipment program, 230: correlation analysis program between power consumption and humidity of IT equipment, 240: correlation analysis program between humidity and power consumption of cooling equipment, 300: renewable energy operation plan table, 41: learning data creation unit, 42: learning model generation unit, 43: time lag prediction unit between IT equipment and cooling equipment, 400: difference calculation program, 410: correlation coefficient calculation program, 420: learning model generation program, 430: correlation coefficient prediction program, 440: time lag extraction program, 450: time lag processing program, 51: time lag correction unit, 52: power consumption correction unit, 500: power consumption change amount intercept analysis program, 510: similarity calculation program, 520: time lag correction value calculation program, 530: correction program for predicted value of power consumption of cooling equipment, 61: visualization result generation unit, 600: time lag visualization program after correction, 610: cooling equipment power consumption visualization program after correction, 620: visualization program for remaining amount of renewable energy after correction

Claims

1. A power consumption prediction system for predicting power consumption of a server having an information processing device that generates heat during operation and a cooling device that cools the information processing device in response to heat generated by the information processing device, comprising: a constraint condition definition unit that defines a constraint condition regarding a dependency relationship between a power consumption of the information processing device and a power consumption of the cooling device; a power consumption prediction unit that predicts a time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device; a correction unit that corrects predicted values ​​of power consumption of the information processing device and the cooling device by using the time lag; an output unit that outputs a result of the correction by the correction unit; Preparation, The constraint condition definition unit A direct dependency between the power consumption of the information processing device and the power consumption of the cooling device; A dependency relationship between the power consumption of the information processing device and an intermediate parameter; defining a dependency relationship between the intermediate parameter and the power consumption of the cooling device; The power consumption prediction unit is If the direct dependency is strong, predicting a time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device based on the direct dependency; When the direct dependency is weak, a time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device is predicted via the intermediate parameter. A power consumption prediction system comprising:

2. The power consumption prediction system according to claim 1 , When the information processing device and the cooling device are installed in each of a plurality of server rooms, the constraint definition unit defines a constraint regarding a dependency relationship for each of the server rooms; the power consumption prediction unit predicts the time lag for each of the server rooms; The correction unit performs the correction for each of the server rooms. A power consumption prediction system comprising:

3. The power consumption prediction system according to claim 1 , A power consumption prediction system characterized in that the correction unit corrects the predicted values ​​of power consumption of the information processing equipment and the cooling equipment based on a plan indicating future available power amount so that they match the available power amount indicated in the plan.

4. The power consumption prediction system according to claim 3, A power consumption prediction system, characterized in that the plan indicating the amount of electricity that can be used in the future is a plan to use surplus electricity generated by renewable energy.

5. The power consumption prediction system according to claim 3, a power consumption prediction system for allocating to said information processing device an amount of operation corresponding to a power consumption of said information processing device based on a correction result by said correction unit;

6. The power consumption prediction system according to claim 3, The information processing device and the cooling device are provided in each of a plurality of data centers; A power consumption prediction system comprising: a power consumption prediction unit that allocates operation amounts among the plurality of data centers based on the plan and correction results of the power consumption of the information processing equipment and the cooling equipment in the plurality of data centers.

7. The power consumption prediction system according to claim 1 , The power consumption prediction system is characterized in that the power consumption prediction unit classifies data on changes in power consumption of the information processing device and the cooling device according to the strength of the direct dependency, generates differential data for each classification, and performs machine learning using the changes in power consumption of the information processing device and the changes in power consumption of the cooling device over a predetermined time span as training data to generate a learning model that determines the time lag.

8. The power consumption prediction system according to claim 1 , the intermediate parameter being the room temperature of the server room; The constraint condition definition unit A dependency between the power consumption of the information processing device and the room temperature; A power consumption prediction system that defines a dependency between the room temperature and the power consumption of the cooling device.

9. The power consumption prediction system according to claim 1 , The mediation parameter is the humidity of the server room; The constraint condition definition unit A dependency between the power consumption of the information processing device and the humidity; A power consumption prediction system that defines a dependency between the humidity and the power consumption of the cooling device.

10. 1. A power consumption prediction method for predicting power consumption of a server having an information processing device that generates heat during operation and a cooling device that cools the information processing device in response to heat generated by the information processing device, comprising: The power consumption prediction system defining a constraint condition regarding a dependency relationship between a power consumption of the information processing device and a power consumption of the cooling device; predicting a time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device; correcting predicted values ​​of power consumption of the information processing device and the cooling device using the time lag; outputting a result of the correction; Including, The step of defining the constraints comprises: A direct dependency between the power consumption of the information processing device and the power consumption of the cooling device; A dependency relationship between the power consumption of the information processing device and an intermediate parameter; defining a dependency relationship between the intermediate parameter and the power consumption of the cooling device; The step of predicting a time lag includes: If the direct dependency is strong, predicting a time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device based on the direct dependency; When the direct dependency is weak, a time lag from a change in power consumption of the information processing device to a change in power consumption of the cooling device is predicted via the intermediate parameter. A power consumption prediction method comprising:

Images