CN-116033717-B - Transformer-based data center energy-saving control method

CN116033717BCN 116033717 BCN116033717 BCN 116033717BCN-116033717-B

Abstract

The invention provides a data center energy-saving control method based on a transducer, which belongs to the technical field of data center control and comprises the following steps of S10, obtaining operation data of a data center including cold station indexes, terminal equipment indexes and environment indexes, S20, preprocessing the obtained operation data, S30, extracting features of the preprocessed operation data to construct a data set, S40, creating a PUE prediction model based on the transducer, training the PUE prediction model by using the data set, S50, publishing the trained PUE prediction model, performing online tuning on index parameters of the published PUE prediction model, and controlling the operation of the data center based on the tuned index parameters. The invention has the advantage of greatly improving the energy-saving effect of the data center.

Inventors

ZHOU JIANMING
LIN JUNDE
CHEN LIFENG
LIN CHENGHAN

Assignees

福建新大陆软件工程有限公司

Dates

Publication Date: 20260505
Application Date: 20230109

Claims (7)

1. A data center energy-saving control method based on a transducer is characterized by comprising the following steps: Step S10, acquiring operation data of a data center including cold station indexes, terminal equipment indexes and environment indexes; step S20, preprocessing the acquired operation data; Step S30, extracting characteristics of the preprocessed operation data to construct a data set; step S40, a PUE prediction model is established based on a transducer, and the PUE prediction model is trained by utilizing the data set; And S50, issuing the PUE prediction model after training, performing online tuning on index parameters of the issued PUE prediction model, and controlling the operation of a data center based on the tuned index parameters.
2. The method is characterized in that in the step S10, the cold station indexes at least comprise the number of operating cooling towers, the water outlet temperature of cooling water of the cooling towers, the number of operating chilled water pumps, the water supply pressure of a chilled water main pipe, the water supply pressure of the chilled water main pipe, the water return temperature of the chilled water main pipe, the wet bulb temperature, the number of operating water sets, the water return temperature of the chilled water sets, the temperature difference of an evaporator of the chilled water sets, the chilled water flow rate of the chilled water sets, the total active power input by a chilled pump, the total active power input by the chilled pump, the water return pressure of the chilled water main pipe, the water return temperature of the chilled water main pipe, the current percentage of the chilled water sets, the water outlet temperature of the chilled water sets, the temperature difference of a condenser of the chilled water sets, the output frequency of the chilled water sets and variable frequency feedback of a fan of the cooling tower; The end equipment indexes at least comprise the number of water pumps, the number of CT, the number of CH, the frequency and the outlet water temperature; the environmental indicators at least comprise IT load rate, outdoor temperature, outdoor humidity and outdoor wind speed.
3. The method for energy-saving control of a data center based on a transducer according to claim 1, wherein the step S20 is specifically: Firstly, unified dimension standardization is carried out on the acquired operation data through a z-score method, then repeated operation data are removed through a deleting method, abnormal operation data are deleted through a quartile method, missing data in the operation data are complemented through an adjacent n-bit mean method, and finally parameters of the operation data are combined through an index normalization method, so that preprocessing of the operation data is completed.
4. The method for energy-saving control of a data center based on a transducer according to claim 1, wherein the step S30 is specifically: extracting the characteristic index related to the PUE from the preprocessed operation data by a variance method; Extracting the collinearity relation between the characteristic indexes by a pearson correlation coefficient method, and screening the characteristic indexes with strong collinearity based on the collinearity relation; then, reducing the dimension of each characteristic index of the strong collinearity by a PCA method so as to remove the characteristic index of the strong collinearity; And finally, constructing a data set based on the residual characteristic indexes.
5. The method for energy saving control of a data center based on a transducer according to claim 1, wherein in the step S40, the PUE prediction model comprises a DSSM multi-tower network, a transducer network and a DNN output network; The DSSM multi-tower network includes a cold station tower model created based on Embedding and DNN, an end device tower model created based on Embedding and DNN; And the input end of the converter network is connected with the output end of the cold station tower model and the output end of the terminal equipment tower model, and the output end of the converter network is connected with the input end of the DNN output network.
6. The method for energy-saving control of a data center based on a transducer according to claim 1, wherein in the step S40, the loss function of the PUE prediction model is a mean square error (RMSE) to evaluate the error between the predicted value and the actual value; In the training process of the PUE prediction model, index parameters of the PUE prediction model are optimized through a AdamW optimizer fused with Warmup strategies, so that the loss value of the loss function is minimized.
7. The method for energy-saving control of a data center based on a transducer according to claim 1, wherein the step S50 is specifically: and releasing the PUE prediction model after training to a server, performing online optimization on index parameters of the released PUE prediction model by utilizing an ant colony algorithm to enable the PUE to be at an optimal value, performing multi-layer filtration on the optimized index parameters based on an SLA protocol, and controlling the operation of a data center based on the filtered index parameters.

Description

Transformer-based data center energy-saving control method Technical Field The invention relates to the technical field of data center control, in particular to a data center energy-saving control method based on a transducer. Background The popularity of cloud computing and the rapid increase of cloud computing in scale promote the development of large-scale data centers, and the operation and maintenance of the mass data centers cannot leave excessive electricity consumption. For operation enterprises of the data center, the energy consumption of the data center is huge, the occupation of electric charge in the operation cost is higher, the reduction of the energy consumption to save the electric charge cost is equivalent to the increase of the profit of the enterprise, and how to promote effective energy-saving practice by means of artificial intelligence becomes an important target of continuous attention of the enterprise. For energy-saving control of a data center, there are two methods conventionally including 1, product-level energy-saving technology, such as PID control mode of an air-conditioning cooling system, optimizing by calculating control quantity by utilizing proportion, integral and differential, although the design is simple and the cost is low, the method is difficult to obtain ideal operation effect by simple proportion control, in the practical application process, cold station condition, end equipment, service load and environment change of each machine room are huge, the method cannot consider interaction and comprehensive influence between cold end and environment, 2, artificial intelligent optimizing technology, namely, a neural network model used in industry represented by Google corporation, establishes an AI prediction model for historical data such as environment, air-conditioning and energy consumption, and adjusts operation of the data center by the AI prediction model, however, most AI prediction models face complex environment change, professional cooling system and numerous control parameters at present, only input all acquired indexes, analysis of coupling characteristics between indexes is lacking, interference analysis is lacking in consideration of irrelevant indexes, and the best effect of approaching long-term PUE (PowerUsageEffectiveness) to power supply use efficiency 1 is difficult to be achieved. Therefore, how to provide a data center energy-saving control method based on a transducer, so as to achieve the effect of improving the energy saving of the data center, is a technical problem to be solved urgently. Disclosure of Invention The invention aims to solve the technical problem of providing a data center energy-saving control method based on a transducer, so as to achieve the effect of improving the energy saving of the data center. The invention discloses a data center energy-saving control method based on a transducer, which comprises the following steps: Step S10, acquiring operation data of a data center including cold station indexes, terminal equipment indexes and environment indexes; step S20, preprocessing the acquired operation data; Step S30, extracting characteristics of the preprocessed operation data to construct a data set; step S40, a PUE prediction model is established based on a transducer, and the PUE prediction model is trained by utilizing the data set; And S50, issuing the PUE prediction model after training, performing online tuning on index parameters of the issued PUE prediction model, and controlling the operation of a data center based on the tuned index parameters. Further, in the step S10, the cold station index at least includes a number of operating cooling towers, a water outlet temperature of cooling water of the cooling towers, a number of operating chilled water pumps, a number of operating cooling water pumps, a water supply pressure of a chilled water main pipe, a water return temperature of the chilled water main pipe, a wet bulb temperature, a number of operating water sets, a water return temperature of the chilled water sets, a temperature difference of an evaporator of the chilled water sets, a chilled water flow rate of the chilled water sets, an input total active power of the chilled pump, an input total active power of the chilled water pump, a water return pressure of the chilled water main pipe, a water return temperature of the chilled water main pipe, a flow rate of the chilled water sets, a current percentage of the chilled water sets, a water outlet temperature of the chilled water sets, a temperature difference of the condenser of the chilled water sets, an output frequency of the chilled water sets, and a variable frequency feedback of a fan of the cooling towers; The end equipment indexes at least comprise the number of water pumps, the number of CT, the number of CH, the frequency and the outlet water temperature; the environmental indicators at least comprise IT load rate, outdoor temperature, outdoor h