Artificial intelligence-based data center energy consumption optimization method and system

Abstract

The application relates to the technical field of artificial intelligence, in particular to a data center energy consumption optimization method and system based on artificial intelligence. The method comprises the following steps: sequentially obtaining energy consumption data of a data center generated on each monitoring date in chronological order; checking a plurality of monitoring dates as a plurality of key dates from a monitoring date interval by different date checking methods; for each checking method, generating different long short-term memory network models based on the energy consumption data of each date in the monitoring date interval and an important numerical value corresponding to the energy consumption data; and obtaining the energy consumption data of a predicted date by using the recorded different long short-term memory network models to perform energy consumption optimization processing. The application can accurately estimate the energy consumption data of the data center and timely perform energy consumption optimization processing.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a data center energy consumption optimization method and system based on artificial intelligence. Background Technology

[0002] Currently, in order to take timely measures to optimize the energy consumption of data centers and achieve the goal of energy conservation and emission reduction, it is necessary to predict the energy consumption demand of data centers in advance.

[0003] Chinese patent application CN118013459A discloses a data center energy consumption prediction method: continuously collecting real-time data from data center servers; using a time-series model to augment the real-time data, performing Fourier transform to obtain the periodic and intra-periodic patterns of the real-time data, converting the augmented data into a two-dimensional tensor using the periodic and intra-periodic patterns, reducing the dimensionality of the two-dimensional tensor to obtain a one-dimensional tensor; using a trend-period prediction model to decompose and autocorrelate the real-time data, combining the periodic and intra-periodic patterns of the real-time data to obtain the energy consumption periodic characteristics and energy consumption trend characteristics of the real-time data, fusing the one-dimensional tensor with the energy consumption periodic characteristics and energy consumption trend characteristics, and performing linear regression on the fused characteristics to obtain the energy consumption data for the next time step. Furthermore, Chinese patent application CN119988859A discloses a method for predicting the cooling energy consumption of a data center. The method includes acquiring multiple first monitoring data points within a first historical time period; preprocessing the multiple first monitoring data points to obtain multiple target monitoring data points; sequentially inputting each target monitoring data point into a pre-determined target prediction model; and outputting the first cooling energy consumption for a future time period that corresponds to the first historical time period through the target prediction model. However, the accuracy of the energy consumption prediction methods in the aforementioned two patent applications needs improvement. Summary of the Invention

[0004] This application divides the monitoring date interval into several sub-intervals using different inspection methods. For each inspection method, important values ​​are set for all energy consumption data in each sub-interval to train and generate different Long Short-Term Memory (LSTM) network models. Furthermore, this application uses the LTM network model corresponding to the inspection method with the highest prediction accuracy to obtain the energy consumption data for the predicted date to perform energy consumption optimization processing. This application aims to accurately predict energy consumption data, thereby facilitating the implementation of energy consumption optimization processing.

[0005] This application provides an artificial intelligence-based method for optimizing data center energy consumption, including the following steps:

[0006] The preparation module determines several consecutive monitoring dates in the past, obtains the energy consumption data generated by the data center on each monitoring date in chronological order, and sets the later monitoring dates among all the monitoring dates as target dates.

[0007] The analysis module organizes all monitoring dates, except for a number of target dates, into monitoring date intervals according to date order. Then, it uses different date checking methods to check out a number of monitoring dates from the monitoring date intervals and treats them as a number of key dates.

[0008] For each inspection method, the analysis module divides the monitoring date interval into several monitoring date sub-intervals based on several key dates corresponding to the inspection method, sets important values ​​for all energy consumption data in each monitoring date sub-interval, and trains and generates different long short-term memory network models based on the energy consumption data of each date in the monitoring date interval and the important values ​​corresponding to the energy consumption data.

[0009] The analysis module uses different long short-term memory network models corresponding to each inspection method to predict the final energy consumption data for several target dates, records the different long short-term memory network models corresponding to the inspection method with the highest prediction accuracy, and the optimization module uses the recorded different long short-term memory network models to obtain the energy consumption data for the predicted dates to perform energy consumption optimization processing.

[0010] As a preferred technical solution of this application, the date checking method includes: for each monitoring date in the monitoring date interval except for the first monitoring date and the last monitoring date, calculating the average value of all energy consumption data before the monitoring date and the average value of all energy consumption data after the monitoring date; when the difference between the two is greater than a preset threshold, the monitoring date is determined to be a critical date.

[0011] As a preferred technical solution of this application, the date checking method includes: for each monitoring date in the monitoring date interval except for the first monitoring date, calculating the fluctuation value of the energy consumption data of the monitoring date; and for each monitoring date in the monitoring date interval except for the first monitoring date and the last monitoring date, calculating the average of all fluctuation values ​​before the monitoring date and the average of all fluctuation values ​​after the monitoring date; when the difference between the two is greater than a preset threshold, the monitoring date is determined to be a critical date.

[0012] As a preferred technical solution of this application, the analysis module sets important values ​​for all energy consumption data in each monitoring date sub-interval, including: setting the same important value for all energy consumption data in each monitoring date sub-interval, and setting the important value for the later monitoring date sub-intervals to be larger.

[0013] As a preferred technical solution of this application, before the analysis module generates different long short-term memory network models based on the energy consumption data and the important values ​​corresponding to the energy consumption data for each date in the monitoring date interval, the following steps are included:

[0014] The analysis module uses the least squares method to perform data fitting on the energy consumption data for all dates within the monitoring date range to obtain a fitting formula;

[0015] For each date within the monitoring date range, the analysis module uses the date's energy consumption data to subtract the fitted data output by the fitted formula corresponding to the date to correct the date's energy consumption data.

[0016] As a preferred technical solution of this application, the analysis module generates different long short-term memory network models based on the energy consumption data of each date in the monitoring date interval and the important values ​​corresponding to the energy consumption data, including the following steps:

[0017] For each date within the monitoring date range, the analysis module uses the date, the energy consumption data for that date, and the important values ​​corresponding to the energy consumption data to form a data record. The analysis module then classifies all the data records according to the preset date classification rules.

[0018] For each category, the analysis module sorts all data records in the category according to date, and based on all sorted data records in each category, the analysis module trains and generates a long short-term memory network model.

[0019] As a preferred technical solution of this application, the analysis module trains and generates a long short-term memory network model based on all sorted data records in each category, including the following steps:

[0020] For each category, the analysis module normalizes the energy consumption data and the corresponding important values ​​of all sorted data records in the category.

[0021] For each category, the analysis module sets a time step and generates different practice data using all sorted data records in the category according to the set time step. Each practice data includes energy consumption data for the number of time steps and a target energy consumption data. The significance value of each practice data is set to the mean of the significance values ​​corresponding to all the energy consumption data contained therein.

[0022] For each category, the analysis module sequentially inputs the energy consumption data of the number of time steps in all the training data into the Long Short-Term Memory network model, calculates the result value of the weighted loss function, updates the parameters of the Long Short-Term Memory network model through backpropagation, and repeats this step until the stopping condition is met.

[0023] As a preferred technical solution of this application, the analysis module uses different long short-term memory network models corresponding to various inspection methods to predict the final energy consumption data for several target dates, including the following steps:

[0024] The analysis module classifies several target dates according to the date classification rules. For each target date in each category, the analysis module uses a long short-term memory network model with the same category as each inspection method to estimate the energy consumption data of each target date.

[0025] For each inspection method, the analysis module sorts the estimated energy consumption data for all target dates in chronological order. For each target date, the estimated energy consumption data is added to the fitted data output by the fitting formula corresponding to the target date to obtain the final energy consumption data for the target date.

[0026] This application also provides an artificial intelligence-based data center energy consumption optimization system, including the following modules:

[0027] The preparation module is used to determine several consecutive monitoring dates in the past, and to obtain the energy consumption data generated by the data center on each monitoring date in chronological order, and to set the later monitoring dates among all the monitoring dates as target dates;

[0028] The analysis module is used to organize all monitoring dates, except for a number of target dates, into monitoring date intervals in chronological order. It then uses different date checking methods to identify several monitoring dates from these intervals as key dates. For each checking method, it divides the monitoring date interval into several monitoring date sub-intervals based on the corresponding key dates. It sets important values ​​for all energy consumption data within each sub-interval. Based on the energy consumption data and corresponding important values ​​for each date within the monitoring date interval, it trains and generates different Long Short-Term Memory (LSTM) network models. Furthermore, it uses these different LTM network models corresponding to various checking methods to predict the final energy consumption data for a number of target dates, recording the LTM network model corresponding to the checking method with the highest prediction accuracy.

[0029] The optimization module is used to obtain energy consumption data for the estimated date using different recorded long short-term memory network models in order to perform energy consumption optimization processing.

[0030] Compared with the prior art, the beneficial effects of this application are at least as follows:

[0031] In the technical solution provided in this application, firstly, several consecutive monitoring dates are determined, and energy consumption data generated by the data center on each monitoring date is obtained sequentially according to the date order. Several later monitoring dates are designated as target dates. Secondly, all monitoring dates except the target dates are grouped into monitoring date intervals according to date order. Several monitoring dates are identified from each monitoring date interval using different date checking methods and designated as key dates. Thirdly, for each checking method, the monitoring date interval is divided into several monitoring date sub-intervals based on the corresponding key dates. Important values ​​are assigned to all energy consumption data in each monitoring date sub-interval. Different Long Short-Term Memory (LSTM) network models are generated based on the energy consumption data and corresponding important values ​​for each date within the monitoring date interval. Finally, the different LTM network models corresponding to each checking method are used to predict the final energy consumption data for the target dates. The different LTM network models corresponding to the checking method with the highest prediction accuracy are recorded. The recorded different LTM network models are used to obtain the energy consumption data for the predicted dates to perform energy consumption optimization processing. This application divides the monitoring date interval into several monitoring date sub-intervals using different inspection methods. For each inspection method, different important values ​​are set for the different monitoring date sub-intervals corresponding to the inspection method. Based on the different energy consumption patterns exhibited on different types of dates, different long short-term memory network models are generated. Therefore, this application can accurately predict the energy consumption data of the data center, thereby facilitating timely energy consumption optimization. Attached Figure Description

[0032] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a flowchart of an AI-based data center energy consumption optimization method in an embodiment of this application;

[0034] Figure 2 This is a schematic diagram of an AI-based data center energy consumption optimization system in an embodiment of this application. Detailed Implementation

[0035] This application provides an artificial intelligence-based data center energy consumption optimization method and system. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data used can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0036] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 The data center energy consumption optimization method based on artificial intelligence in this application includes the following main steps:

[0037] The preparation module determines several consecutive monitoring dates in the past, obtains the energy consumption data generated by the data center on each monitoring date in chronological order, and sets the later monitoring dates among all the monitoring dates as target dates.

[0038] The analysis module organizes all monitoring dates, except for a number of target dates, into monitoring date intervals according to date order. Then, it uses different date checking methods to check out a number of monitoring dates from the monitoring date intervals and treats them as a number of key dates.

[0039] For each inspection method, the analysis module divides the monitoring date interval into several monitoring date sub-intervals based on several key dates corresponding to the inspection method, sets important values ​​for all energy consumption data in each monitoring date sub-interval, and trains and generates different long short-term memory network models based on the energy consumption data of each date in the monitoring date interval and the important values ​​corresponding to the energy consumption data.

[0040] The analysis module uses different long short-term memory network models corresponding to each inspection method to predict the final energy consumption data for several target dates, records the different long short-term memory network models corresponding to the inspection method with the highest prediction accuracy, and the optimization module uses the recorded different long short-term memory network models to obtain the energy consumption data for the predicted dates to perform energy consumption optimization processing.

[0041] Specifically, to accurately predict data center energy consumption, the preparation module first identifies several consecutive monitoring dates and sequentially acquires energy consumption data generated by the data center on each monitoring date, for example, acquiring energy consumption data generated by the data center from May 1st to May 31st. From all the monitoring dates, several later monitoring dates are selected and set as target dates. The number of target dates is set according to actual application needs; for example, May 29th, May 30th, and May 31st are set as target dates. Secondly, the analysis module groups all monitoring dates except the target dates into monitoring date intervals in chronological order. For example, the monitoring date interval includes May 1st to May 28th. The analysis module uses different date checking methods to identify several monitoring dates from the monitoring date interval as key dates. Key dates refer to dates where the corresponding energy consumption data shows significant fluctuations; energy consumption data after key dates may represent a new energy consumption pattern. Furthermore, for each inspection method, the analysis module divides the monitoring date interval into several monitoring date sub-intervals based on several key dates determined by the inspection method. For example, if the key date is May 11, then two monitoring date sub-intervals can be obtained, namely May 1 to May 11 and May 12 to May 28. The analysis module sets important values ​​for all energy consumption data in each monitoring date sub-interval. Based on the energy consumption data and the corresponding important values ​​of the energy consumption data for each date in the monitoring date interval, different long short-term memory network models are generated. The dates here include the monitoring dates and key dates mentioned above. Finally, for each inspection method, the analysis module uses different long short-term memory network models corresponding to the inspection method to predict the final energy consumption data for each target date. First, it calculates the deviation between the final energy consumption data for each target date and its corresponding actual energy consumption data. Then, it calculates the average of all deviations. Based on this, it stores the different long short-term memory network models corresponding to the inspection method with the highest prediction accuracy, i.e., the smallest calculated average. Thus, the optimization module can use the stored different long short-term memory network models to obtain the energy consumption data for the predicted date and then perform energy consumption optimization processing.

[0042] Furthermore, the date checking method includes: for each monitoring date in the monitoring date interval except for the first and last monitoring dates, calculating the average of all energy consumption data before the monitoring date and the average of all energy consumption data after the monitoring date; when the difference between the two is greater than a preset threshold, the monitoring date is determined to be a critical date.

[0043] Specifically, a date checking method is introduced: For each monitoring date in the monitoring date interval, excluding the first and last monitoring dates, calculate the average of all energy consumption data before the monitoring date and the average of all energy consumption data after the monitoring date. It should be noted that "before the monitoring date" here includes the monitoring date itself. If the difference between the two is greater than a preset threshold, it is determined that the energy consumption data of the monitoring date has fluctuated significantly, and the monitoring date is identified as a critical date. The threshold is set according to the actual application.

[0044] Furthermore, the date checking method includes: for each monitoring date in the monitoring date interval except for the first monitoring date, calculating the fluctuation value of the energy consumption data for the monitoring date; and for each monitoring date in the monitoring date interval except for the first and last monitoring dates, calculating the average of all fluctuation values ​​before the monitoring date and the average of all fluctuation values ​​after the monitoring date; when the difference between the two is greater than a preset threshold, the monitoring date is determined to be a critical date.

[0045] Specifically, another date checking method is introduced: For each monitoring date in the monitoring date interval except for the first monitoring date, the fluctuation value of the energy consumption data of the monitoring date is calculated, that is, the difference between the energy consumption data of the monitoring date and the energy consumption data of the previous one is calculated. Based on this, for each monitoring date in the monitoring date interval except for the first and last monitoring dates, the average of all fluctuation values ​​before the monitoring date and the average of all fluctuation values ​​after the monitoring date are calculated. It should be noted that "before the monitoring date" here includes the monitoring date itself. If the difference between the two is greater than a preset threshold, the energy consumption data of the monitoring date is considered to have fluctuated significantly, and the monitoring date is determined to be a critical date. The threshold is set according to the actual application situation.

[0046] Furthermore, other date checking methods are introduced: First, identify the key factors that can affect the energy consumption data generated by the data center. For example, the number of servers included in the data center will directly affect the generation of energy consumption data, and the ambient temperature will directly affect the cooling system included in the data center, thus indirectly affecting the generation of energy consumption data. Second, the monitoring date on which the key factors change significantly is regarded as the key date.

[0047] Furthermore, the analysis module sets important values ​​for all energy consumption data in each monitoring date sub-interval, including setting the same important value for all energy consumption data in each monitoring date sub-interval, and setting the important value for the later monitoring date sub-intervals to be larger.

[0048] Specifically, this section explains how to set important values ​​for all energy consumption data in each monitoring date sub-interval. The analysis module sets the important values ​​for all energy consumption data in each monitoring date sub-interval to the same value. At the same time, the analysis module sets the important values ​​for the later monitoring date sub-intervals to be larger. This is because the later monitoring date sub-intervals are more likely to represent new energy consumption patterns. Continuing with the example above, if there are two monitoring date sub-intervals, namely monitoring date sub-interval one: 05.01 to 05.11, and monitoring date sub-interval two: 05.12 to 05.28, then the important values ​​for all energy consumption data in monitoring date sub-interval one can be set to 1, and the important values ​​for all energy consumption data in monitoring date sub-interval two can be set to 3.

[0049] Furthermore, before the analysis module generates different long short-term memory network models based on energy consumption data and corresponding important values ​​for each date within the monitoring date interval, the following steps are included:

[0050] The analysis module uses the least squares method to perform data fitting on the energy consumption data for all dates within the monitoring date range to obtain a fitting formula;

[0051] For each date within the monitoring date range, the analysis module uses the date's energy consumption data to subtract the fitted data output by the fitted formula corresponding to the date to correct the date's energy consumption data.

[0052] Specifically, this section describes the processing performed by the analysis module before generating different Long Short-Term Memory (LSTM) network models based on energy consumption data and corresponding key values ​​for each date within the monitoring date interval. These dates may be monitoring dates or key dates. First, the analysis module uses the least squares method to fit the energy consumption data for all dates within the monitoring date interval to obtain a fitting formula. This formula expresses the overall fluctuation trend of energy consumption data across the entire monitoring date interval. It's important to note that the date itself is not a factor influencing the overall fluctuation trend; for example, the number of servers in a data center might have an impact. Second, for each date within the monitoring date interval, the analysis module subtracts the corresponding fitted data from the fitted formula's output to correct the date's energy consumption data. Based on this, the corrected energy consumption data for all dates within the monitoring date interval is obtained, reflecting the true fluctuation trend.

[0053] Furthermore, the analysis module generates different long short-term memory network models based on the energy consumption data for each date within the monitoring date interval and the corresponding important values ​​of the energy consumption data, including the following steps:

[0054] For each date within the monitoring date range, the analysis module uses the date, the energy consumption data for that date, and the important values ​​corresponding to the energy consumption data to form a data record. The analysis module then classifies all the data records according to the preset date classification rules.

[0055] For each category, the analysis module sorts all data records in the category according to date, and based on all sorted data records in each category, the analysis module trains and generates a long short-term memory network model.

[0056] Specifically, this section describes the process of generating different Long Short-Term Memory (LSTM) network models based on energy consumption data and corresponding key values ​​for each date within a monitoring date range. Here, "date" can refer to either the monitoring date or a critical date. First, for each date within the monitoring date range, the analysis module uses the date, its energy consumption data, and corresponding key values ​​to create a data record. Based on this, the analysis module categorizes all data records according to preset date classification rules. These rules are set according to the actual application scenario; for example, they might classify dates by workdays and holidays, taking into account that energy consumption patterns may differ across different date categories. Secondly, for each category, the analysis module sorts all data records within that category according to date. This allows the analysis module to practice generating a Long Short-Term Memory (LSTM) network model based on all the sorted data records in each category. It's important to note that dates within different data records belonging to the same category may not be adjacent. However, since the overall fluctuations in energy consumption data for these dates have been removed above, reducing the differences between these dates, they can be considered adjacent in this embodiment. The process of practicing generating the LTM network model will be described in detail below. By using the above methods, the accuracy of the generated LTM network model can be improved as much as possible.

[0057] Furthermore, based on all sorted data records in each category, the analysis module trains to generate a Long Short-Term Memory (LSTM) network model, including the following steps:

[0058] For each category, the analysis module normalizes the energy consumption data and the corresponding important values ​​of all sorted data records in the category.

[0059] For each category, the analysis module sets a time step and generates different practice data using all sorted data records in the category according to the set time step. Each practice data includes energy consumption data for the number of time steps and a target energy consumption data. The significance value of each practice data is set to the mean of the significance values ​​corresponding to all the energy consumption data contained therein.

[0060] For each category, the analysis module sequentially inputs the energy consumption data of the number of time steps in all the training data into the Long Short-Term Memory network model, calculates the result value of the weighted loss function, updates the parameters of the Long Short-Term Memory network model through backpropagation, and repeats this step until the stopping condition is met.

[0061] Specifically, this section describes how to train and generate a Long Short-Term Memory (LSTM) network model based on all sorted data records in each category. First, to avoid the impact of differences in numerical ranges on the training and generation of the LSM network model, the analysis module normalizes the energy consumption data and its corresponding significance values ​​for all sorted data records in each category for each category. Second, for each category, the analysis module sets a time step, for example, a time step of 3. This means using energy consumption data from the past three days to predict the energy consumption data for the next day. Based on the set time step, the analysis module generates different training data using all sorted data records in the category. Each training data set includes energy consumption data for the specified number of time steps and a target energy consumption data set. For example, a training data set might include energy consumption data for 05.01, 05.02, 05.03, and 05.04, where the energy consumption data for 05.04 is the target energy consumption data set. The analysis module also sets the significance value of each training data set to the average of the significance values ​​of all the energy consumption data it contains. Finally, for each category, the analysis module sequentially inputs the energy consumption data of the number of time steps from all the training data into the Long Short-Term Memory network model. The purpose is to calculate the weighted loss function, which is: ,in, The total number of practice data. For the first Important values ​​for each practice data point For the first The target energy consumption data included in each training dataset For the first The analysis module updates the parameters of the Long Short-Term Memory network model using backpropagation based on the estimated energy consumption data corresponding to each training data point. This process is repeated until the loss function value no longer decreases or the number of executions reaches a preset threshold.

[0062] Furthermore, the analysis module uses different long short-term memory network models corresponding to each inspection method to predict the final energy consumption data for several target dates, including the following steps:

[0063] The analysis module classifies several target dates according to date classification rules. For each target date in each category, the analysis module uses a long short-term memory network model with the same category as each inspection method to estimate the energy consumption data of each target date.

[0064] For each inspection method, the analysis module sorts the estimated energy consumption data for all target dates in chronological order. For each target date, the estimated energy consumption data is added to the fitted data output by the fitting formula corresponding to the target date to obtain the final energy consumption data for the target date.

[0065] Specifically, this section describes how to use different Long Short-Term Memory (LSTM) network models corresponding to various inspection methods to predict the final energy consumption data for several target dates. First, the analysis module classifies the target dates according to the date classification rules mentioned above. For each target date in each category, the analysis module uses the LTM network model corresponding to the same category for each inspection method to predict the energy consumption data for each target date. For ease of understanding, for example, if the target dates are from June 1st to June 4th, according to the date classification rules mentioned above, the energy consumption data for each target date is... June 1st to June 2nd are classified into the same category, corresponding to holidays. June 3rd to June 4th are classified into the same category, corresponding to workdays. Check method 1 corresponds to model A and model B, with model A corresponding to workdays and model B corresponding to holidays. Check method 2 corresponds to model C and model D, with model C corresponding to workdays and model D corresponding to holidays. Therefore, model B and model D are used to predict the energy consumption data for June 1st and June 2nd, respectively, and model A and model C are used to predict the energy consumption data for June 3rd and June 4th, respectively. Secondly, for each inspection method, the analysis module sorts the estimated energy consumption data for all target dates according to date order. For example, for inspection method 1, the sorting process yields the energy consumption data for 06.01 (predicted by model C), 06.02 (predicted by model C), 06.03 (predicted by model A), and 06.04 (predicted by model A). For the estimated energy consumption data for each target date, the analysis module uses the estimated energy consumption data for the target date plus the fitted data corresponding to the target date output by the fitting formula to obtain the final energy consumption data for the target date. Here, the fitting formula refers to the fitting formula corresponding to the inspection method calculated above.

[0066] Furthermore, the process by which the optimization module uses different long short-term memory network models to obtain energy consumption data for the estimated date is the same as the process described above of using different long short-term memory network models corresponding to the inspection method to estimate the final energy consumption data for a target date, so it will not be repeated here.

[0067] According to another aspect of the embodiments of this application, reference is made to... Figure 2 As shown, this application also provides an artificial intelligence-based data center energy consumption optimization system, including a preparation module, an analysis module, and an optimization module, to implement the artificial intelligence-based data center energy consumption optimization method described above.

[0068] The functions of each module are as follows:

[0069] The preparation module is used to determine several consecutive monitoring dates in the past, and to obtain the energy consumption data generated by the data center on each monitoring date in chronological order, and to set the later monitoring dates among all the monitoring dates as target dates;

[0070] The analysis module is used to organize all monitoring dates, except for a number of target dates, into monitoring date intervals in chronological order. It then uses different date checking methods to identify several monitoring dates from these intervals as key dates. For each checking method, it divides the monitoring date interval into several monitoring date sub-intervals based on the corresponding key dates. It sets important values ​​for all energy consumption data within each sub-interval. Based on the energy consumption data and corresponding important values ​​for each date within the monitoring date interval, it trains and generates different Long Short-Term Memory (LSTM) network models. Furthermore, it uses these different LTM network models corresponding to various checking methods to predict the final energy consumption data for a number of target dates, recording the LTM network model corresponding to the checking method with the highest prediction accuracy.

[0071] The optimization module is used to obtain energy consumption data for the estimated date using different recorded long short-term memory network models in order to perform energy consumption optimization processing.

[0072] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0073] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0074] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A data center energy consumption optimization method based on artificial intelligence, characterized in that, The method includes the following steps: The preparation module determines several consecutive monitoring dates in the past, obtains the energy consumption data generated by the data center on each monitoring date in chronological order, and sets the later monitoring dates among all the monitoring dates as target dates. The analysis module organizes all monitoring dates, except for a number of target dates, into monitoring date intervals according to date order. Then, it uses different date checking methods to check out a number of monitoring dates from the monitoring date intervals and treats them as a number of key dates. For each inspection method, the analysis module divides the monitoring date interval into several monitoring date sub-intervals based on several key dates corresponding to the inspection method, sets important values ​​for all energy consumption data in each monitoring date sub-interval, and trains and generates different long short-term memory network models based on the energy consumption data of each date in the monitoring date interval and the important values ​​corresponding to the energy consumption data. The analysis module uses different long short-term memory network models corresponding to each inspection method to predict the final energy consumption data for several target dates, records the different long short-term memory network models corresponding to the inspection method with the highest prediction accuracy, and the optimization module uses the recorded different long short-term memory network models to obtain the energy consumption data for the predicted dates to perform energy consumption optimization processing. The date checking method includes: for each monitoring date in the monitoring date interval except for the first and last monitoring dates, calculate the average of all energy consumption data before the monitoring date and the average of all energy consumption data after the monitoring date. When the difference between the two is greater than a preset threshold, the monitoring date is determined to be a critical date. The date checking method includes: for each monitoring date in the monitoring date interval except for the first monitoring date, calculating the fluctuation value of the energy consumption data for the monitoring date; and for each monitoring date in the monitoring date interval except for the first and last monitoring dates, calculating the average of all fluctuation values ​​before the monitoring date and the average of all fluctuation values ​​after the monitoring date; when the difference between the two is greater than a preset threshold, the monitoring date is determined to be a critical date. The date checking method includes: identifying key factors that can affect the energy consumption data generated by the data center, including the number of servers and ambient temperature, and taking the monitoring date on which the key factors change significantly as the key date; The analysis module sets important values ​​for all energy consumption data in each monitoring date sub-interval, including setting the same important value for all energy consumption data in each monitoring date sub-interval, and setting the important value to be larger for later monitoring date sub-intervals.

2. The method according to claim 1, characterized in that, Before generating different Long Short-Term Memory (LSTM) network models based on energy consumption data and corresponding key values ​​for each date within the monitoring date range, the analysis module includes the following steps: The analysis module uses the least squares method to perform data fitting on the energy consumption data for all dates within the monitoring date range to obtain a fitting formula; For each date within the monitoring date range, the analysis module uses the date's energy consumption data to subtract the fitted data output by the fitted formula corresponding to the date to correct the date's energy consumption data.

3. The method according to claim 2, characterized in that, The analysis module generates different long short-term memory network models based on energy consumption data for each date within the monitoring date range and the corresponding important values ​​of the energy consumption data, including the following steps: For each date within the monitoring date range, the analysis module uses the date, the energy consumption data for that date, and the important values ​​corresponding to the energy consumption data to form a data record. The analysis module then classifies all the data records according to the preset date classification rules. For each category, the analysis module sorts all data records in the category according to date, and based on all sorted data records in each category, the analysis module trains and generates a long short-term memory network model.

4. The method according to claim 3, characterized in that, Based on all sorted data records in each category, the analysis module trains to generate a Long Short-Term Memory (LSTM) network model, including the following steps: For each category, the analysis module normalizes the energy consumption data and the corresponding important values ​​of all sorted data records in the category. For each category, the analysis module sets a time step and generates different practice data using all sorted data records in the category according to the set time step. Each practice data includes energy consumption data for the number of time steps and a target energy consumption data. The significance value of each practice data is set to the mean of the significance values ​​corresponding to all the energy consumption data contained therein. For each category, the analysis module sequentially inputs the energy consumption data of the number of time steps in all the training data into the Long Short-Term Memory network model, calculates the result value of the weighted loss function, updates the parameters of the Long Short-Term Memory network model through backpropagation, and repeats this step until the stopping condition is met.

5. The method according to claim 4, characterized in that, The analysis module uses different long short-term memory network models corresponding to various inspection methods to predict the final energy consumption data for several target dates, including the following steps: The analysis module classifies several target dates according to the date classification rules. For each target date in each category, the analysis module uses a long short-term memory network model with the same category as each inspection method to estimate the energy consumption data of each target date. For each inspection method, the analysis module sorts the estimated energy consumption data for all target dates in chronological order. For each target date, the estimated energy consumption data is added to the fitted data output by the fitting formula corresponding to the target date to obtain the final energy consumption data for the target date.

6. An artificial intelligence-based data center energy consumption optimization system, used to implement the method as described in any one of claims 1 to 5, characterized in that, Includes the following modules: The preparation module is used to determine several consecutive monitoring dates in the past, and to obtain the energy consumption data generated by the data center on each monitoring date in chronological order, and to set the later monitoring dates among all the monitoring dates as target dates; The analysis module is used to organize all monitoring dates, except for a number of target dates, into monitoring date intervals in chronological order. It then uses different date checking methods to identify several monitoring dates from these intervals as key dates. For each checking method, it divides the monitoring date interval into several monitoring date sub-intervals based on the corresponding key dates. It sets important values ​​for all energy consumption data within each sub-interval. Based on the energy consumption data and corresponding important values ​​for each date within the monitoring date interval, it trains and generates different Long Short-Term Memory (LSTM) network models. Furthermore, it uses these different LTM network models corresponding to various checking methods to predict the final energy consumption data for a number of target dates, recording the LTM network model corresponding to the checking method with the highest prediction accuracy. The optimization module is used to obtain energy consumption data for the estimated date using different recorded long short-term memory network models in order to perform energy consumption optimization processing.

Images