Data center intelligent energy-saving control method, device, equipment and storage medium

Abstract

The present invention discloses a method, device, equipment and storage medium for intelligent energy-saving control of a data center. The method uses historical control data recorded by a data center to train an automatic machine learning Auto ML model based on a power usage efficiency PUE value to obtain a trained target Auto ML model; according to the target Auto ML model, a deep deterministic policy gradient DDPG is combined to obtain a minimum PUE value at each moment, and an optimal behavior strategy is generated; according to the optimal behavior strategy, the energy consumption of the data center is regulated, and the optimal parameter control combination at each moment can be provided in real time, effectively reducing the energy consumption of the data center, solving the defect that the existing data center PUE control method cannot realize fully automatic real-time control strategy output, saving the time consumed by the intelligent energy-saving control of the data center, and improving the speed and efficiency of the intelligent energy-saving control of the data center.

Description

Technical Field

[0001] The present invention relates to the field of cloud computing technology, and in particular to a data center intelligent energy-saving control method, device, equipment and storage medium. Background Art

[0002] With the rapid development of cloud computing and big data industries, data centers, as their important infrastructure platforms, are also growing rapidly in number and scale; at the same time, the huge energy consumption and environmental pressure brought by data centers are also increasingly attracting worldwide attention. Therefore, strengthening the monitoring and evaluation of data center energy efficiency indicators and optimizing data center energy use plans are of great significance to improving energy utilization efficiency and reducing environmental pressure. Among the many data center energy efficiency evaluation indicators, the most widely used one at this stage is Power Usage Effectiveness (PUE).

[0003] Existing data center energy-saving control methods rely heavily on manual experience for adjustment, fail to fully utilize the correlation between data, and the adjusted PUE is difficult to reach the optimal value, resulting in energy waste. Summary of the invention

[0004] The main purpose of the present invention is to provide a data center intelligent energy-saving control method, device, equipment and storage medium, aiming to solve the technical problems in the prior art that the data center energy-saving control method relies heavily on manual experience for adjustment, does not fully utilize the correlation between data, and the adjusted PUE is difficult to reach the optimal value, resulting in energy waste.

[0005] In a first aspect, the present invention provides a data center intelligent energy-saving control method, the data center intelligent energy-saving control method comprising the following steps:

[0006] Use the historical control data recorded by the data center to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value to obtain the trained target Auto ML model;

[0007] According to the target Auto ML model combined with deep deterministic policy gradient DDPG, the minimum PUE value at each moment is obtained to generate the optimal behavior strategy;

[0008] The energy consumption of the data center is regulated according to the optimal behavior strategy.

[0009] Optionally, the method of using the historical control data recorded by the data center to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value to obtain a trained target Auto ML model includes:

[0010] Eliminate null values ​​and refine historical data based on the power usage efficiency (PUE) value of the data center to obtain historical control data and preset training data for the automatic machine learning Auto ML model;

[0011] Constructing a parameter training model according to the historical control data and the preset training data;

[0012] The Auto ML model is trained and fine-tuned according to the parameter training model to obtain a trained target Auto ML model.

[0013] Optionally, constructing a parameter training model according to the historical control data and the preset training data includes:

[0014] Obtaining an algorithm set of an Auto ML model and adjustment parameters of the data center;

[0015] The historical control data is divided into K training sets by using K-fold cross validation, and the preset training data is divided into K test data by using K-fold cross validation;

[0016] A parameter training model is constructed according to the algorithm set, the adjustment parameters, the K training sets and the K test data by the following formula:

[0017]

[0018] Among them, A={A (1) ,…,A (R)} represents the algorithm set of the Auto ML model, each element of which represents a different classification algorithm, λ is the adjustment parameter of the data center, The historical control data are divided into K training data sets by K-fold cross validation, The preset training data is divided into K test data sets through K-fold cross validation, and L is The Loss value at this time.

[0019] Optionally, the training and fine-tuning the Auto ML model according to the parameter training model to obtain a trained target Auto ML model includes:

[0020] Collecting current input data of the data center, training the Auto ML model according to the parameter training model, obtaining a historical data Auto ML model trained with the historical control data, and obtaining a current data training Auto ML model corresponding to the current input data;

[0021] Obtaining a first weight corresponding to the historical data Auto ML model and a second weight corresponding to the current data training Auto ML model;

[0022] According to the historical data Auto ML model, the current data training Auto ML model, the first weight, and the second weight, the historical data Auto ML model and the current data training Auto ML model are fine-tuned by the following formula to obtain a fine-tuned Auto ML model, and the fine-tuned Auto ML model is used as the target Auto ML model after training:

[0023] M=w i M origin +w j M current

[0024] Among them, M is the fine-tuned Auto ML model, M origin is the Auto ML model for historical data, the Auto ML model for current data training, and w i is the first weight, w j is the second weight.

[0025] Optionally, obtaining a first weight corresponding to the Auto ML model for historical data and a second weight corresponding to the Auto ML model for current data training includes:

[0026] Obtaining an average value of the historical control data and obtaining an average value of the current input data;

[0027] According to the average value of the historical control data and the average value of the current input data, the first weight corresponding to the historical data Auto ML model and the second weight corresponding to the current data training Auto ML model are obtained by the following formula:

[0028]

[0029]

[0030] Among them, w i is the first weight, w j is the second weight, is the average value of the historical control data, is the average value of the current input data.

[0031] Optionally, obtaining the minimum PUE value at each moment according to the target Auto ML model in combination with deep deterministic policy gradient DDPG to generate an optimal behavior strategy includes:

[0032] According to the target Auto ML model combined with deep deterministic policy gradient DDPG, a policy measurement function for obtaining the minimum PUE value at each moment is obtained;

[0033] According to the strategy measurement function, the optimal behavior strategy is determined by the following formula:

[0034]

[0035] Among them, μ is the optimal behavior strategy and J is the strategy measurement function.

[0036] Optionally, the strategy measurement function for obtaining the minimum PUE value at each moment according to the target Auto ML model in combination with deep deterministic policy gradient DDPG includes:

[0037] According to the target Auto ML model combined with the deep deterministic policy gradient DDPG, the policy measurement function of the minimum PUE value at each moment is obtained by the following formula:

[0038]

[0039] Q u (s t , a t )=E[r(s t , a t )+γQ u (s t+1 ,μ(s t+1 ))]

[0040] Among them, J is the strategy measurement function, Q is the Q network in DDPG, and s t is the environmental state at time t, a t is the action taken at time t, r is the state of the environment in s t When executing a t The single-step reward value after the action, γ is the preset decay coefficient, and β is the behavior strategy used to introduce random noise.

[0041] In a second aspect, in order to achieve the above-mentioned purpose, the present invention further proposes an intelligent energy-saving control device for a data center, the intelligent energy-saving control device for a data center comprising:

[0042] A training module is used to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value by using the historical control data recorded by the data center to obtain a trained target Auto ML model;

[0043] A strategy generation module is used to obtain the minimum PUE value at each moment according to the target Auto ML model combined with deep deterministic policy gradient DDPG to generate an optimal behavior strategy;

[0044] A control module is used to control the energy consumption of the data center according to the optimal behavior strategy.

[0045] In the third aspect, to achieve the above-mentioned purpose, the present invention also proposes a data center intelligent energy-saving and control device, which includes: a memory, a processor, and a data center intelligent energy-saving and control program stored in the memory and executable on the processor, and the data center intelligent energy-saving and control program is configured to implement the steps of the data center intelligent energy-saving and control method as described above.

[0046] In a fourth aspect, in order to achieve the above-mentioned purpose, the present invention also proposes a storage medium, on which a data center intelligent energy-saving control program is stored, and when the data center intelligent energy-saving control program is executed by a processor, the steps of the data center intelligent energy-saving control method as described above are implemented.

[0047] The intelligent energy-saving control method for a data center proposed in the present invention uses historical control data recorded by the data center to train an automatic machine learning Auto ML model based on the power usage efficiency PUE value to obtain a trained target Auto ML model; according to the target Auto ML model combined with deep deterministic policy gradient DDPG, the minimum PUE value at each moment is obtained to generate an optimal behavior strategy; according to the optimal behavior strategy, the energy consumption of the data center is regulated, and the optimal parameter control combination at each moment can be provided in real time, effectively reducing the energy consumption of the data center, solving the defect that the existing data center PUE control method cannot realize fully automatic real-time control strategy output, saving the time consumed by the intelligent energy-saving control of the data center, and improving the speed and efficiency of the intelligent energy-saving control of the data center. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention;

[0049] Figure 2 This is a flow chart of the first embodiment of the intelligent energy-saving control method for a data center according to the present invention;

[0050] Figure 3 This is a flow chart of the second embodiment of the data center intelligent energy-saving control method of the present invention;

[0051] Figure 4 This is a functional module diagram of the first embodiment of the intelligent energy-saving control device for a data center of the present invention.

[0052] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0053] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0054] The solution of the embodiment of the present invention is mainly: by using the historical control data recorded by the data center to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value, a trained target Auto ML model is obtained; according to the target Auto ML model combined with the deep deterministic policy gradient DDPG, the minimum PUE value at each moment is obtained to generate the optimal behavior strategy; according to the optimal behavior strategy, the energy consumption of the data center is regulated, and the optimal parameter control combination at each moment can be provided in real time, effectively reducing the energy consumption of the data center, solving the defect that the existing data center PUE adjustment method cannot realize the full-automatic real-time adjustment strategy output, saving the time consumed by the intelligent energy-saving regulation of the data center, improving the speed and efficiency of the intelligent energy-saving regulation of the data center, and solving the technical problem that the energy-saving regulation method of the data center in the prior art relies heavily on manual experience for adjustment, does not fully utilize the correlation between data, and the adjusted PUE is difficult to reach the optimal value, resulting in energy waste.

[0055] Reference Figure 1 , Figure 1 The figure is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention.

[0056] like Figure 1 As shown, the device may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed RAM memory, or it may be a stable memory (Non-Volatile Memory), such as a disk memory. The memory 1005 may optionally be a storage device independent of the aforementioned processor 1001.

[0057] Those skilled in the art will understand that Figure 1 The device structure shown in the figure does not constitute a limitation of the device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0058] like Figure 1As shown, the memory 1005 as a storage medium may include an operating device, a network communication module, a user interface module, and a data center intelligent energy-saving control program.

[0059] The device of the present invention calls the data center intelligent energy-saving control program stored in the memory 1005 through the processor 1001, and performs the following operations:

[0060] Use the historical control data recorded by the data center to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value to obtain the trained target Auto ML model;

[0061] According to the target Auto ML model combined with deep deterministic policy gradient DDPG, the minimum PUE value at each moment is obtained to generate the optimal behavior strategy;

[0062] The energy consumption of the data center is regulated according to the optimal behavior strategy.

[0063] The device of the present invention calls the data center intelligent energy-saving control program stored in the memory 1005 through the processor 1001, and also performs the following operations:

[0064] Eliminate null values ​​and refine historical data based on the power usage efficiency (PUE) value of the data center to obtain historical control data and preset training data for the automatic machine learning Auto ML model;

[0065] Constructing a parameter training model according to the historical control data and the preset training data;

[0066] The Auto ML model is trained and fine-tuned according to the parameter training model to obtain a trained target Auto ML model.

[0067] The device of the present invention calls the data center intelligent energy-saving control program stored in the memory 1005 through the processor 1001, and also performs the following operations:

[0068] Obtaining an algorithm set of an Auto ML model and adjustment parameters of the data center;

[0069] The historical control data is divided into K training sets by using K-fold cross validation, and the preset training data is divided into K test data by using K-fold cross validation;

[0070] A parameter training model is constructed according to the algorithm set, the adjustment parameters, the K training sets and the K test data by the following formula:

[0071]

[0072] Among them, A={A (1) ,…,A (R)} represents the algorithm set of the Auto ML model, each element of which represents a different classification algorithm, λ is the adjustment parameter of the data center, The historical control data are divided into K training data sets by K-fold cross validation, The preset training data is divided into K test data sets through K-fold cross validation, and L is The Loss value at this time.

[0073] The device of the present invention calls the data center intelligent energy-saving control program stored in the memory 1005 through the processor 1001, and also performs the following operations:

[0074] Collecting current input data of the data center, training the Auto ML model according to the parameter training model, obtaining a historical data Auto ML model trained with the historical control data, and obtaining a current data training Auto ML model corresponding to the current input data;

[0075] Obtaining a first weight corresponding to the historical data Auto ML model and a second weight corresponding to the current data training Auto ML model;

[0076] According to the historical data Auto ML model, the current data training Auto ML model, the first weight, and the second weight, the historical data Auto ML model and the current data training Auto ML model are fine-tuned by the following formula to obtain a fine-tuned Auto ML model, and the fine-tuned Auto ML model is used as the target Auto ML model after training:

[0077] M=w i M origin +w j M current

[0078] Among them, M is the fine-tuned Auto ML model, M origin is the Auto ML model for historical data, the Auto ML model for current data training, and w i is the first weight, w j is the second weight.

[0079] The device of the present invention calls the data center intelligent energy-saving control program stored in the memory 1005 through the processor 1001, and also performs the following operations:

[0080] Obtaining an average value of the historical control data and obtaining an average value of the current input data;

[0081] According to the average value of the historical control data and the average value of the current input data, the first weight corresponding to the historical data Auto ML model and the second weight corresponding to the current data training Auto ML model are obtained by the following formula:

[0082]

[0083]

[0084] Among them, w i is the first weight, w j is the second weight, is the average value of the historical control data, is the average value of the current input data.

[0085] The device of the present invention calls the data center intelligent energy-saving control program stored in the memory 1005 through the processor 1001, and also performs the following operations:

[0086] According to the target Auto ML model combined with deep deterministic policy gradient DDPG, a policy measurement function for obtaining the minimum PUE value at each moment is obtained;

[0087] According to the strategy measurement function, the optimal behavior strategy is determined by the following formula:

[0088]

[0089] Among them, μ is the optimal behavior strategy and J is the strategy measurement function.

[0090] The device of the present invention calls the data center intelligent energy-saving control program stored in the memory 1005 through the processor 1001, and also performs the following operations:

[0091] According to the target Auto ML model combined with the deep deterministic policy gradient DDPG, the policy measurement function of the minimum PUE value at each moment is obtained by the following formula:

[0092]

[0093] Q u (s t , a t )=E[r(s t , a t )+γQ u (s t+1 ,μ(s t+1 ))]

[0094] Among them, J is the strategy measurement function, Q is the Q network in DDPG, and st is the environmental state at time t, a t is the action taken at time t, r is the state of the environment in s t When executing a t The single-step reward value after the action, γ is the preset decay coefficient, and β is the behavior strategy used to introduce random noise.

[0095] Through the above scheme, this embodiment uses the historical control data recorded by the data center to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value, so as to obtain a trained target Auto ML model; according to the target Auto ML model, the minimum PUE value at each moment is obtained in combination with the deep deterministic policy gradient DDPG, and the optimal behavior strategy is generated; according to the optimal behavior strategy, the energy consumption of the data center is regulated, and the optimal parameter control combination at each moment can be provided in real time, effectively reducing the energy consumption of the data center, solving the defect that the existing data center PUE adjustment method cannot realize fully automatic real-time adjustment strategy output, saving the time consumed by the intelligent energy-saving regulation of the data center, and improving the speed and efficiency of the intelligent energy-saving regulation of the data center.

[0096] Based on the above hardware structure, an embodiment of the data center intelligent energy-saving control method of the present invention is proposed.

[0097] Reference Figure 2 , Figure 2 It is a flow chart of the first embodiment of the intelligent energy-saving control method for data center of the present invention.

[0098] In a first embodiment, the data center intelligent energy-saving control method includes the following steps:

[0099] Step S10: Use the historical control data recorded by the data center to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value to obtain a trained target Auto ML model.

[0100] It should be noted that the historical control data recorded by the data center can be used to train an automatic machine learning (Auto Machine Learning, Auto ML) model based on the prediction of power usage efficiency (Power Usage Effectiveness, PUE) value to obtain an Auto ML model with higher prediction accuracy, that is, to obtain a trained target Auto ML model.

[0101] Step S20: According to the target Auto ML model, the minimum PUE value at each moment is obtained in combination with the deep deterministic policy gradient DDPG to generate the optimal behavior strategy.

[0102] It can be understood that according to the target Auto ML model combined with Deep Deterministic Policy Gradient (DDPG), the optimal behavior strategy corresponding to the parameter combination with the minimum PUE result at each moment can be obtained. The optimal behavior strategy is the optimal parameter combination of the data center, which is generally the combination of setting parameters corresponding to the minimum PUE result corresponding to different scenarios.

[0103] Furthermore, the step S20 specifically includes the following steps:

[0104] According to the target Auto ML model combined with deep deterministic policy gradient DDPG, a policy measurement function for obtaining the minimum PUE value at each moment is obtained;

[0105] According to the strategy measurement function, the optimal behavior strategy is determined by the following formula:

[0106]

[0107] Among them, μ is the optimal behavior strategy and J is the strategy measurement function.

[0108] It should be understood that the parameter output method based on DDPG is to use DDPG to construct the PUE minimization parameter combination output modeling based on the obtained PUE value Auto ML model. This part is based on the setting parameters of the PUE intelligent minimization parameters of the data center. Through the setting parameters, the optimal behavior strategy is the optimal setting parameter combination of the data center.

[0109] Furthermore, the step obtains a policy measurement function of the minimum PUE value at each moment according to the target Auto ML model combined with the deep deterministic policy gradient DDPG, specifically including the following steps:

[0110] According to the target Auto ML model combined with the deep deterministic policy gradient DDPG, the policy measurement function of the minimum PUE value at each moment is obtained by the following formula:

[0111]

[0112] Q u (s t , a t )=E[r(s t , a t )+γQ u (s t+1 ,μ(s t+1 ))]

[0113] Among them, J is the strategy measurement function, Q is the Q network in DDPG, and s tis the environmental state at time t, a t is the action taken at time t, r is the state of the environment in s t When executing a t The single-step reward value after the action, γ is the preset decay coefficient, and β is the behavior strategy used to introduce random noise.

[0114] It should be noted that, according to the fine-tuned optimal data center PUE prediction model combined with DDPG to build the optimal behavior strategy for parameter optimization, the parameter setting with the minimum PUE result at each moment can be obtained, and the intelligent fully automatic real-time regulation of data center energy saving can be realized. γ is the preset decay coefficient, which can generally be set to a value between 0 and 1. β is the behavior strategy, which can be used to introduce random noise.

[0115] Step S30: regulating the energy consumption of the data center according to the optimal behavior strategy.

[0116] It should be understood that after obtaining the optimal behavior strategy, the energy consumption of the data center can be regulated according to the corresponding parameters based on the setting parameters corresponding to the optimal behavior strategy, thereby realizing intelligent regulation of data center energy saving and effectively reducing the energy consumption of the data center.

[0117] Through the above scheme, this embodiment uses the historical control data recorded by the data center to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value, so as to obtain a trained target Auto ML model; according to the target Auto ML model, the minimum PUE value at each moment is obtained in combination with the deep deterministic policy gradient DDPG, and the optimal behavior strategy is generated; according to the optimal behavior strategy, the energy consumption of the data center is regulated, and the optimal parameter control combination at each moment can be provided in real time, effectively reducing the energy consumption of the data center, solving the defect that the existing data center PUE adjustment method cannot realize fully automatic real-time adjustment strategy output, saving the time consumed by the intelligent energy-saving regulation of the data center, and improving the speed and efficiency of the intelligent energy-saving regulation of the data center.

[0118] Further, Figure 3 FIG. 1 is a flow chart of a second embodiment of the intelligent energy-saving control method for a data center according to the present invention. Figure 3 As shown, based on the first embodiment, a second embodiment of the data center intelligent energy-saving control method of the present invention is proposed. In this embodiment, step S10 specifically includes the following steps:

[0119] Step S11: Eliminate null value data and over-refine historical data of the data center based on the power usage efficiency PUE value to obtain historical control data and obtain preset training data for the automatic machine learning Auto ML model.

[0120] It should be noted that by eliminating null value data and over-refining the historical data of the power usage efficiency PUE value of the data center, historical control data can be obtained. That is, after preprocessing the input historical data of the data center PUE value to eliminate null value data and over-refined data, the corresponding historical control data can be obtained. The preset training data is the test data pre-set by the Auto ML model for training the model.

[0121] Step S12: constructing a parameter training model according to the historical control data and the preset training data.

[0122] It can be understood that a parameter training model for high-precision progress prediction of the Auto ML model can be constructed through the historical control data and the preset training data.

[0123] Furthermore, the step S12 specifically includes the following steps:

[0124] Obtaining an algorithm set of an Auto ML model and adjustment parameters of the data center;

[0125] The historical control data is divided into K training sets by using K-fold cross validation, and the preset training data is divided into K test data by using K-fold cross validation;

[0126] A parameter training model is constructed according to the algorithm set, the adjustment parameters, the K training sets and the K test data by the following formula:

[0127]

[0128] Among them, A={A (1) ,…,A (R)} represents the algorithm set of the Auto ML model, each element of which represents a different classification algorithm, λ is the adjustment parameter of the data center, The historical control data are divided into K training data sets by K-fold cross validation, The preset training data is divided into K test data sets through K-fold cross validation, and L is The Loss value at this time.

[0129] It should be noted that in order to find the algorithm A and data center parameters corresponding to the minimum Loss case, the Auto-Sklearn framework can be used on the basis of the historical control data of the given data center to obtain the optimal solution to the problem in an iterative manner, and the corresponding algorithm A and parameters at this time can be obtained, and finally the optimal Auto ML model can be output. The Auto ML model based on PUE value prediction can be trained using the historical control data and training data recorded by the data center through the above formula, and an Auto ML model with higher prediction accuracy can be obtained. The historical control data can be divided into K training data sets through K-fold cross validation, and the preset training data can be divided into K test data sets through K-fold cross validation.

[0130] Step S13: Train and fine-tune the Auto ML model according to the parameter training model to obtain a trained target Auto ML model.

[0131] It should be understood that constructing the parameter training model can train and fine-tune the Auto ML model to obtain a fine-tuned model, that is, the model training can be deemed complete and the fine-tuned model can be used as the target Auto ML model.

[0132] Furthermore, the step S13 specifically includes the following steps:

[0133] Collecting current input data of the data center, training the Auto ML model according to the parameter training model, obtaining a historical data Auto ML model trained with the historical control data, and obtaining a current data training Auto ML model corresponding to the current input data;

[0134] Obtaining a first weight corresponding to the historical data Auto ML model and a second weight corresponding to the current data training Auto ML model;

[0135] According to the historical data Auto ML model, the current data training Auto ML model, the first weight, and the second weight, the historical data Auto ML model and the current data training Auto ML model are fine-tuned by the following formula to obtain a fine-tuned Auto ML model, and the fine-tuned Auto ML model is used as the target Auto ML model after training:

[0136] M=w i M origin +w j M current

[0137] Among them, M is the fine-tuned Auto ML model, Morigin is the Auto ML model for historical data, the Auto ML model for current data training, and w i is the first weight, w j is the second weight.

[0138] It should be noted that the specific fine-tuning method of the Auto ML model is: the fine-tuned Auto ML model is obtained by calculating the above formula according to the current control data of the data center.

[0139] Furthermore, the step of obtaining a first weight corresponding to the historical data Auto ML model and a second weight corresponding to the current data training Auto ML model includes:

[0140] Obtaining an average value of the historical control data and obtaining an average value of the current input data;

[0141] According to the average value of the historical control data and the average value of the current input data, the first weight corresponding to the historical data Auto ML model and the second weight corresponding to the current data training AutoML model are obtained by the following formula:

[0142]

[0143]

[0144] Among them, w i is the first weight, w j is the second weight, is the average value of the historical control data, is the average value of the current input data.

[0145] It should be understood that the weights corresponding to the model obtained according to the historical control data and the model obtained by the current data input can be determined by the above formula through the average value of the historical control data and the average value of the current input data, and then the fine-tuned Auto ML model is obtained based on the historical data Auto ML model, the current data trained Auto ML model, the first weight and the second weight calculation.

[0146] Through the above scheme, this embodiment obtains historical control data by eliminating null value data and over-refining the historical data of the data center based on the power usage efficiency PUE value, and obtains preset training data of the automatic machine learning Auto ML model; builds a parameter training model according to the historical control data and the preset training data; trains and fine-tunes the Auto ML model according to the parameter training model to obtain a trained target Auto ML model, and can obtain an Auto ML model with high prediction accuracy, thereby ensuring the accuracy of intelligent energy-saving control of the data center and improving the speed and efficiency of intelligent energy-saving control of the data center.

[0147] Accordingly, the present invention further provides an intelligent energy-saving control device for a data center.

[0148] Reference Figure 4 , Figure 4 This is a functional module diagram of the first embodiment of the intelligent energy-saving control device for a data center of the present invention.

[0149] In the first embodiment of the intelligent energy-saving control device for a data center of the present invention, the intelligent energy-saving control device for a data center comprises:

[0150] The training module 10 is used to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value by using the historical control data recorded by the data center to obtain a trained target Auto ML model.

[0151] The strategy generation module 20 is used to obtain the minimum PUE value at each moment according to the target Auto ML model in combination with the deep deterministic policy gradient DDPG, and generate the optimal behavior strategy.

[0152] The control module 30 is used to control the energy consumption of the data center according to the optimal behavior strategy.

[0153] The training module 10 is also used to eliminate null value data and over-refine the historical data of the data center based on the power usage efficiency PUE value, obtain historical control data, and obtain preset training data for the automatic machine learning Auto ML model; build a parameter training model according to the historical control data and the preset training data; train and fine-tune the Auto ML model according to the parameter training model to obtain a trained target Auto ML model.

[0154] The training module 10 is further used to obtain the algorithm set of the Auto ML model and the adjustment parameters of the data center; divide the historical control data into K training sets using K-fold cross validation, and divide the preset training data into K test data using K-fold cross validation; and construct a parameter training model according to the algorithm set, the adjustment parameters, the K training sets and the K test data by the following formula:

[0155]

[0156] Among them, A={A (1) ,…,A (R)} represents the algorithm set of the Auto ML model, each element of which represents a different classification algorithm, λ is the adjustment parameter of the data center, The historical control data are divided into K training data sets by K-fold cross validation, The preset training data is divided into K test data sets through K-fold cross validation, and L is The Loss value at this time.

[0157] The training module 10 is further used to collect current input data of the data center, train the Auto ML model according to the parameter training model, obtain a historical data AutoML model obtained by training corresponding to the historical control data, and obtain a current data training Auto ML model corresponding to the current input data; obtain a first weight corresponding to the historical data Auto ML model and a second weight corresponding to the current data training Auto ML model; fine-tune the historical data Auto ML model and the current data training Auto ML model according to the historical data Auto ML model, the current data training Auto ML model, the first weight and the second weight through the following formula to obtain a fine-tuned Auto ML model, and use the fine-tuned Auto ML model as the target Auto ML model after training:

[0158] M=w i M origin +w j M current

[0159] Among them, M is the fine-tuned Auto ML model, M origin is the Auto ML model for historical data, the Auto ML model for current data training, and w i is the first weight, w j is the second weight.

[0160] The training module 10 is further used to obtain the average value of the historical control data and the average value of the current input data; and obtain the first weight corresponding to the historical data Auto ML model and the second weight corresponding to the current data training Auto ML model according to the average value of the historical control data and the average value of the current input data by the following formula:

[0161]

[0162]

[0163] Among them, w i is the first weight, w j is the second weight, is the average value of the historical control data, is the average value of the current input data.

[0164] The strategy generation module 20 is also used to obtain the minimum PUE value at each moment according to the target Auto ML model combined with the deep deterministic policy gradient DDPG, and generate the optimal behavior strategy, including: obtaining a strategy measurement function of the minimum PUE value at each moment according to the target Auto ML model combined with the deep deterministic policy gradient DDPG;

[0165] According to the strategy measurement function, the optimal behavior strategy is determined by the following formula:

[0166]

[0167] Among them, μ is the optimal behavior strategy and J is the strategy measurement function.

[0168] The strategy generation module 20 is also used to obtain a strategy measurement function of the minimum PUE value at each moment according to the target Auto ML model combined with the deep deterministic policy gradient DDPG through the following formula:

[0169]

[0170] Q u (s t , a t )=E[r(s t , a t )+γQ u (s t+1 ,μ(s t+1 ))]

[0171] Among them, J is the strategy measurement function, Q is the Q network in DDPG, and s t is the environmental state at time t, a tis the action taken at time t, r is the state of the environment in s t When executing a t The single-step reward value after the action, γ is the preset decay coefficient, and β is the behavior strategy used to introduce random noise.

[0172] Among them, the steps implemented by each functional module of the data center intelligent energy-saving control device can refer to the various embodiments of the data center intelligent energy-saving control method of the present invention, and will not be repeated here.

[0173] In addition, an embodiment of the present invention further provides a storage medium, on which a data center intelligent energy-saving control program is stored. When the data center intelligent energy-saving control program is executed by a processor, the following operations are implemented:

[0174] Use the historical control data recorded by the data center to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value to obtain the trained target Auto ML model;

[0175] According to the target Auto ML model combined with deep deterministic policy gradient DDPG, the minimum PUE value at each moment is obtained to generate the optimal behavior strategy;

[0176] The energy consumption of the data center is regulated according to the optimal behavior strategy.

[0177] Furthermore, when the data center intelligent energy-saving control program is executed by the processor, the following operations are also implemented:

[0178] Eliminate null values ​​and refine historical data based on the power usage efficiency (PUE) value of the data center to obtain historical control data and preset training data for the automatic machine learning Auto ML model;

[0179] Constructing a parameter training model according to the historical control data and the preset training data;

[0180] The Auto ML model is trained and fine-tuned according to the parameter training model to obtain a trained target Auto ML model.

[0181] Furthermore, when the data center intelligent energy-saving control program is executed by the processor, the following operations are also implemented:

[0182] Obtaining an algorithm set of an Auto ML model and adjustment parameters of the data center;

[0183] The historical control data is divided into K training sets by using K-fold cross validation, and the preset training data is divided into K test data by using K-fold cross validation;

[0184] A parameter training model is constructed according to the algorithm set, the adjustment parameters, the K training sets and the K test data by the following formula:

[0185]

[0186] Among them, A={A (1) ,…,A (R)} represents the algorithm set of the Auto ML model, each element of which represents a different classification algorithm, λ is the adjustment parameter of the data center, The historical control data are divided into K training data sets by K-fold cross validation, The preset training data is divided into K test data sets through K-fold cross validation, and L is The Loss value at this time.

[0187] Furthermore, when the data center intelligent energy-saving control program is executed by the processor, the following operations are also implemented:

[0188] Collecting current input data of the data center, training the Auto ML model according to the parameter training model, obtaining a historical data Auto ML model trained with the historical control data, and obtaining a current data training Auto ML model corresponding to the current input data;

[0189] Obtaining a first weight corresponding to the historical data Auto ML model and a second weight corresponding to the current data training Auto ML model;

[0190] According to the historical data Auto ML model, the current data training Auto ML model, the first weight, and the second weight, the historical data Auto ML model and the current data training Auto ML model are fine-tuned by the following formula to obtain a fine-tuned Auto ML model, and the fine-tuned Auto ML model is used as the target Auto ML model after training:

[0191] M=w i M origin +w j M current

[0192] Among them, M is the fine-tuned Auto ML model, M origin is the Auto ML model for historical data, the Auto ML model for current data training, and w i is the first weight, w j is the second weight.

[0193] Furthermore, when the data center intelligent energy-saving control program is executed by the processor, the following operations are also implemented:

[0194] Obtaining an average value of the historical control data and obtaining an average value of the current input data;

[0195] According to the average value of the historical control data and the average value of the current input data, the first weight corresponding to the historical data Auto ML model and the second weight corresponding to the current data training Auto ML model are obtained by the following formula:

[0196]

[0197]

[0198] Among them, w i is the first weight, w j is the second weight, is the average value of the historical control data, is the average value of the current input data.

[0199] Furthermore, when the data center intelligent energy-saving control program is executed by the processor, the following operations are also implemented:

[0200] According to the target Auto ML model combined with deep deterministic policy gradient DDPG, a policy measurement function for obtaining the minimum PUE value at each moment is obtained;

[0201] According to the strategy measurement function, the optimal behavior strategy is determined by the following formula:

[0202]

[0203] Among them, μ is the optimal behavior strategy and J is the strategy measurement function.

[0204] Furthermore, when the data center intelligent energy-saving control program is executed by the processor, the following operations are also implemented:

[0205] According to the target Auto ML model combined with the deep deterministic policy gradient DDPG, the policy measurement function of the minimum PUE value at each moment is obtained by the following formula:

[0206]

[0207] Q u (s t , a t )=E[r(s t , a t )+γQ u (s t+1 ,μ(st+1 ))]

[0208] Among them, J is the strategy measurement function, Q is the Q network in DDPG, and s t is the environmental state at time t, a t is the action taken at time t, r is the state of the environment in s t When executing a t The single-step reward value after the action, γ is the preset decay coefficient, and β is the behavior strategy used to introduce random noise.

[0209] Through the above scheme, this embodiment uses the historical control data recorded by the data center to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value, so as to obtain a trained target Auto ML model; according to the target Auto ML model, the minimum PUE value at each moment is obtained in combination with the deep deterministic policy gradient DDPG, and the optimal behavior strategy is generated; according to the optimal behavior strategy, the energy consumption of the data center is regulated, and the optimal parameter control combination at each moment can be provided in real time, effectively reducing the energy consumption of the data center, solving the defect that the existing data center PUE adjustment method cannot realize fully automatic real-time adjustment strategy output, saving the time consumed by the intelligent energy-saving regulation of the data center, and improving the speed and efficiency of the intelligent energy-saving regulation of the data center.

[0210] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0211] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0212] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A data center intelligent energy-saving control method, characterized in that: The data center intelligent energy-saving control method comprises: Use the historical control data recorded by the data center to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value to obtain the trained target Auto ML model; According to the target Auto ML model combined with deep deterministic policy gradient DDPG, the minimum PUE value at each moment is obtained to generate the optimal behavior strategy; regulating the energy consumption of the data center according to the optimal behavior strategy; The method of obtaining the minimum PUE value at each moment according to the target Auto ML model combined with the deep deterministic policy gradient DDPG and generating the optimal behavior strategy includes: According to the target Auto ML model combined with deep deterministic policy gradient DDPG, a policy measurement function for obtaining the minimum PUE value at each moment is obtained; According to the strategy measurement function, the optimal behavior strategy is determined by the following formula: Among them, μ is the optimal behavior strategy and J is the strategy measurement function.

2. The data center intelligent energy-saving control method according to claim 1, characterized in that: The method uses the historical control data recorded by the data center to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value to obtain a trained target Auto ML model, including: Eliminate null values ​​and refine historical data based on the power usage efficiency (PUE) value of the data center to obtain historical control data and preset training data for the automatic machine learning Auto ML model; Constructing a parameter training model according to the historical control data and the preset training data; The Auto ML model is trained and fine-tuned according to the parameter training model to obtain a trained target Auto ML model.

3. The data center intelligent energy-saving control method according to claim 2, characterized in that: The step of constructing a parameter training model according to the historical control data and the preset training data includes: Obtaining an algorithm set of an Auto ML model and adjustment parameters of the data center; The historical control data is divided into K training sets by using K-fold cross validation, and the preset training data is divided into K test data by using K-fold cross validation; A parameter training model is constructed according to the algorithm set, the adjustment parameters, the K training sets and the K test data by the following formula: Among them, A={A (1) ,…,A (R) } represents the algorithm set of the Auto ML model, each element of which represents a different classification algorithm, λ is the adjustment parameter of the data center, The historical control data are divided into K training data sets by K-fold cross validation, The preset training data is divided into K test data sets through K-fold cross validation, and L is The Loss value at this time.

4. The data center intelligent energy-saving control method according to claim 2, characterized in that: The step of training and fine-tuning the Auto ML model according to the parameter training model to obtain a trained target Auto ML model includes: Collecting current input data of the data center, training the Auto ML model according to the parameter training model, obtaining a historical data Auto ML model trained with the historical control data, and obtaining a current data training Auto ML model corresponding to the current input data; Obtaining a first weight corresponding to the historical data Auto ML model and a second weight corresponding to the current data training Auto ML model; According to the historical data Auto ML model, the current data training Auto ML model, the first weight, and the second weight, the historical data Auto ML model and the current data training Auto ML model are fine-tuned by the following formula to obtain a fine-tuned Auto ML model, and the fine-tuned Auto ML model is used as the target Auto ML model after training: M=w i M origin +w j M current Among them, M is the fine-tuned Auto ML model, M origin is the Auto ML model for historical data, the Auto ML model for current data training, and w i is the first weight, w j is the second weight.

5. The data center intelligent energy-saving control method according to claim 4, characterized in that: The obtaining of the first weight corresponding to the historical data Auto ML model and the second weight corresponding to the current data training Auto ML model includes: Obtaining an average value of the historical control data and obtaining an average value of the current input data; According to the average value of the historical control data and the average value of the current input data, the first weight corresponding to the historical data Auto ML model and the second weight corresponding to the current data training Auto ML model are obtained by the following formula: Among them, w i is the first weight, w j is the second weight, is the average value of the historical control data, is the average value of the current input data.

6. The data center intelligent energy-saving control method according to claim 1, characterized in that: The strategy measurement function for obtaining the minimum PUE value at each moment according to the target Auto ML model combined with the deep deterministic policy gradient DDPG includes: According to the target Auto ML model combined with the deep deterministic policy gradient DDPG, the policy measurement function of the minimum PUE value at each moment is obtained by the following formula: Q u (s t ,a t )=E[r(s t ,a t )+γQ u (s t+1 ,μ(s t+1 ))] Among them, J is the strategy measurement function, Q is the Q network in DDPG, and s t is the environmental state at time t, a t is the action taken at time t, r is the state of the environment in s t When executing a t The single-step reward value after the action, γ is the preset decay coefficient, and β is the behavior strategy used to introduce random noise.

7. An intelligent energy-saving control device for a data center, characterized in that: The data center intelligent energy-saving control device comprises: A training module is used to train the automatic machine learning Auto ML model based on the power usage efficiency PUE value by using the historical control data recorded by the data center to obtain a trained target Auto ML model; A strategy generation module is used to obtain the minimum PUE value at each moment according to the target Auto ML model combined with deep deterministic policy gradient DDPG to generate an optimal behavior strategy; A control module, used to control the energy consumption of the data center according to the optimal behavior strategy; The strategy generation module is also used to obtain a strategy measurement function of the minimum PUE value at each moment according to the target Auto ML model combined with the deep deterministic policy gradient DDPG; According to the strategy measurement function, the optimal behavior strategy is determined by the following formula: Among them, μ is the optimal behavior strategy and J is the strategy measurement function.

8. A data center intelligent energy-saving control device, characterized in that: The data center intelligent energy-saving and control device includes: a memory, a processor, and a data center intelligent energy-saving and control program stored in the memory and executable on the processor, wherein the data center intelligent energy-saving and control program is configured to implement the steps of the data center intelligent energy-saving and control method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium stores a data center intelligent energy-saving control program, and when the data center intelligent energy-saving control program is executed by the processor, the steps of the data center intelligent energy-saving control method according to any one of claims 1 to 6 are implemented.

Images