CN115983445A

CN115983445A - PUE prediction method, and training method, device and equipment of PUE prediction model

Info

Publication number: CN115983445A
Application number: CN202211597374.7A
Authority: CN
Inventors: 赵波; 禹鑫; 刘强; 薛宁
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2022-12-12
Filing date: 2022-12-12
Publication date: 2023-04-18

Abstract

The disclosure provides a PUE prediction method, a PUE prediction model training device and PUE prediction model training equipment, and relates to the technical field of artificial intelligence. The specific implementation scheme is as follows: respectively extracting the characteristics of the reference PUE characteristic data in at least two reference periods to obtain PUE characteristic representation of the reference PUE characteristic data; the reference period is located before the target period; determining autocorrelation characteristics among reference PUE characteristic data in different reference periods; and predicting the PUE value of the data center in the target period according to the PUE characteristic representation and the autocorrelation characteristic. By the technical scheme, the accuracy of PUE prediction can be improved.

Description

PUE prediction method, and training method, device and equipment of PUE prediction model

Technical Field

The disclosure relates to the technical field of artificial intelligence, in particular to a PUE prediction method, a PUE prediction model training device and PUE prediction model training equipment.

Background

The energy efficiency index (PUE) is a key parameter for operation of the data center, is an index for evaluating the energy efficiency of the data center, and is a ratio of all energy consumed by the data center to energy consumed by the IT load.

At present, the prediction of the seasonal or annual PUE of the data center mainly depends on the statistical analysis of historical contemporaneous or recent data, and the average value, the extreme value or the median of a data segment is usually taken as a future seasonal or annual PUE target; in addition, considering the operation condition, the cabinet service condition, the key guarantee holiday and other related contents, the fine adjustment can be usually carried out based on the statistical values. However, due to the complexity of the operation conditions of the data center and the fusion of multiple systems, the method relates to multiple aspects of water, electricity and heat, and the statistical value estimation scheme is one-sided and has large errors. Thus, improvements are needed.

Disclosure of Invention

The disclosure provides a PUE prediction method, a PUE prediction model training device and equipment.

According to an aspect of the present disclosure, there is provided a PUE prediction method, including:

respectively extracting the characteristics of the reference PUE characteristic data in at least two reference periods to obtain PUE characteristic representation of the reference PUE characteristic data; the reference period is positioned before the target period;

determining autocorrelation characteristics among reference PUE characteristic data in different reference periods;

and predicting the PUE value of the data center in the target period according to the PUE characteristic representation and the autocorrelation characteristic.

According to another aspect of the present disclosure, there is provided a training method of a PUE prediction model, the method including:

respectively extracting the characteristics of the reference PUE characteristic data in at least two reference periods to obtain PUE characteristic representation of the reference PUE characteristic data; the reference period is located before the target period;

predicting a PUE value of the data center in a target period according to the PUE feature representation and the autocorrelation feature;

determining training loss according to the predicted PUE value of the data center in the target period and the PUE label value of the data center in the target period;

and training the PUE prediction model by adopting the training loss.

According to another aspect of the present disclosure, there is provided a PUE prediction apparatus, including:

the PUE feature extraction module is used for respectively extracting features of reference PUE feature data in at least two reference periods to obtain PUE feature representation of the reference PUE feature data; the reference period is positioned before the target period;

the autocorrelation characteristic determining module is used for determining autocorrelation characteristics among the reference PUE characteristic data in different reference periods;

and the PUE value prediction module is used for predicting the PUE value of the data center in the target period according to the PUE characteristic representation and the autocorrelation characteristic.

According to another aspect of the present disclosure, there is provided an apparatus for training a PUE prediction model, the apparatus including:

the PUE characteristic extraction module is used for respectively extracting the characteristics of reference PUE characteristic data in at least two reference periods to obtain PUE characteristic representation of the reference PUE characteristic data; the reference period is located before the target period;

the PUE value prediction module is used for predicting the PUE value of the data center in a target period according to the PUE characteristic representation and the autocorrelation characteristic;

the training loss determining module is used for determining the training loss according to the predicted PUE value of the data center in the target period and the PUE label value of the data center in the target period;

and the model training module is used for training the PUE prediction model by adopting the training loss.

According to another aspect of the present disclosure, there is provided an electronic device including:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a PUE prediction method, or a training method of a PUE prediction model, according to any of the embodiments of the present disclosure.

According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the PUE prediction method or the training method of the PUE prediction model according to any of the embodiments of the present disclosure.

According to another aspect of the present disclosure, a computer program product is provided, comprising a computer program, which when executed by a processor, implements the PUE prediction method, or the training method of the PUE prediction model, according to any of the embodiments of the present disclosure.

According to the technology disclosed by the invention, the accuracy of PUE prediction can be improved.

It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present disclosure, nor do they limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.

Drawings

The drawings are included to provide a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

FIG. 1 is a flow chart of a PUE prediction method provided in accordance with an embodiment of the present disclosure;

FIG. 2 is a flow chart of another PUE prediction method provided in accordance with an embodiment of the present disclosure;

FIG. 3 is a flow chart of another PUE prediction method provided in accordance with an embodiment of the present disclosure;

FIG. 4A is a flowchart of a training method of a PUE prediction model according to an embodiment of the disclosure;

FIG. 4B is a schematic diagram of a training process of a PUE prediction model provided according to an embodiment of the present disclosure;

FIG. 5 is a schematic structural diagram of a PUE prediction apparatus according to an embodiment of the present disclosure;

FIG. 6 is a schematic structural diagram of a training apparatus for a PUE prediction model according to an embodiment of the present disclosure;

fig. 7 is a block diagram of an electronic device for implementing a PUE prediction method or a training method of a PUE prediction model according to an embodiment of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of embodiments of the present disclosure are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

It is to be understood that the terms "reference" and "target" and the like in the description and claims of the invention and the above-described drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

In addition, in the technical scheme of the invention, the collection, storage, use, processing, transmission, provision, disclosure and other processing of the reference PUE characteristic data and the like all meet the regulations of relevant laws and regulations and do not violate the good custom of the public order.

Fig. 1 is a flowchart of a PUE prediction method according to an embodiment of the present disclosure. The method is suitable for the situation of predicting the PUE index of the data center. The method may be executed by a PUE prediction apparatus, which may be implemented in software and/or hardware, and may be integrated in an electronic device, such as a server, that carries the PUE prediction function. As shown in fig. 1, the PUE prediction method of this embodiment may include:

s101, respectively carrying out feature extraction on the reference PUE feature data in at least two reference periods to obtain PUE feature representation of the reference PUE feature data.

In this embodiment, the cycle refers to a time period in which the PUE value of the data center is calculated once, and may be, for example, a week, a day, several hours, or the like. The target period is a period in which the PUE value of the data center needs to be predicted, and may be a certain period in the history or a certain period in the future. The reference period refers to a period referred to when the PUE value of the data center is predicted; alternatively, the reference period may be a certain period of the history; it should be noted that, the reference period is located before the target period; further, the different reference periods may be consecutive to each other. For example, assuming that the target period is day N, the at least two reference periods may be days 1 to N-1, where N is a natural number equal to or greater than 3.

The reference PUE characteristic data is data used for predicting the PUE value of the data center; alternatively, the reference PUE feature data may be PUE values of the data center at a reference period. Further, data cleaning may be performed on the reference PUE characteristic data, for example, data of which the reference PUE value is less than or equal to 1 and the reference PUE value is greater than or equal to 1 in the reference PUE characteristic data may be cleaned and readjusted according to actual working conditions and empirical data, so as to avoid an influence of an extreme value on prediction of a subsequent PUE value. Furthermore, data standardization processing such as chemotaxis processing and dimensionless processing can be carried out on the reference PUE characteristic data so as to enhance the identifiability of the PUE characteristic data and the comparability of the data; for example, a Min-Max normalization process may be employed, assuming that the reference PUE profile for at least two reference periods is { x } ₁ ，x ₂ ，x ₃ ，...x _i ...，x _n In which x _n Reference PUE characteristic data, x, representing the nth reference period _i The reference PUE feature data representing the ith reference period may then be normalized by the following equation:

wherein x is _{i_s} Feature data, x, of the normalized reference PUE feature data representing the ith reference period _i Reference PUE characteristic data, x, representing the ith reference period _max Maximum reference PUE characteristic data, x, of the reference PUE characteristic data representing at least two reference periods _min The smallest reference PUE signature data of the reference PUE signature data representing at least two reference periods.

The PUE feature representation refers to data obtained by feature extraction of reference PUE data, and may include, but is not limited to, sequences, matrices, and vectors. It should be noted that the reference period corresponds to the reference PUE feature representation one to one, and the reference PUE feature data corresponds to the PUE feature representation one to one.

Specifically, a deep neural network may be adopted to perform feature extraction on the reference PUE feature data in at least two reference periods, respectively, to obtain the PUE feature representation of each reference PUE feature data. The deep neural network may be a long-term memory (LSTM) network or the like.

And S102, determining autocorrelation characteristics among the reference PUE characteristic data in different reference periods.

In this embodiment, the autocorrelation feature refers to a feature used for characterizing the correlation between the reference PUE feature data in different reference periods, and may be represented in a vector or matrix form.

Specifically, the autocorrelation operation may be performed on the reference PUE characteristic data in different reference periods to obtain autocorrelation characteristics between the reference PUE characteristic data in different reference periods.

S103, predicting the PUE value of the data center in the target period according to the PUE characteristic representation and the autocorrelation characteristic.

Specifically, based on a preset prediction rule, the PUE value of the data center in the target period may be predicted according to the PUE feature representation of each reference PUE feature data and the autocorrelation feature between different reference PUE feature data. For example, the PUE feature representation of each reference PUE feature data and the autocorrelation features between different reference PUE feature data may be fused to obtain a fused feature, a dimension reduction operation may be performed on the fused feature to obtain a one-dimensional feature, and then an activation operation (such as a Sigmoid operation) may be used to predict the one-dimensional feature to obtain a PUE value of the data center in a target period.

According to the technical scheme provided by the embodiment of the disclosure, the PUE characteristic representation of the reference PUE characteristic data is obtained by respectively carrying out characteristic extraction on the reference PUE characteristic data in at least two reference periods, then the autocorrelation characteristics among the reference PUE characteristic data in different reference periods are determined, and further the PUE value of the data center in the target period is predicted according to the PUE characteristic representation and the autocorrelation characteristics. According to the technical scheme, the characteristics of the PUE characteristic data of different reference periods can be fully mined by extracting the characteristics of the PUE characteristic data of a plurality of reference periods; meanwhile, the autocorrelation characteristics are introduced, the dependency relationship and the mutual influence degree among the PUE characteristic data of different reference periods can be fully mined, and therefore a good foundation is laid for improving the prediction accuracy of the PUE value.

On the basis of the above embodiment, as an optional manner of the present disclosure, the reference PUE feature data of the data center in the reference period includes the reference PUE value of the data center in the reference period and a time feature corresponding to the reference PUE value.

The reference PUE value refers to a real PUE value of the data center in a reference period. The time characteristic corresponding to the reference PUE value is used for representing the characteristic of the time period in which the reference PUE value is located, and can be represented in a vector or matrix form; alternatively, the temporal features may be vectors formed by time stamps corresponding to the reference PUE values.

For example, on the basis of the foregoing embodiment, as an optional manner of the present disclosure, determining the time characteristic corresponding to the reference PUE value may be that, for each reference period, a time stamp corresponding to the reference PUE value in the reference period is obtained by the data center; determining the time sequence of the reference period according to the time stamp corresponding to the reference period; and converting the time sequence of the reference period to obtain the time characteristic corresponding to the reference PUE value in the reference period.

Specifically, for each reference period, a timestamp corresponding to a reference PUE value of the data center in the reference period is obtained, then the timestamp corresponding to the reference period is split, for example, feature items such as month, quarter and period of the timestamp are split, the split data is processed to obtain a time sequence of the reference period, and then the time sequence of the reference period can be converted in a preset manner, for example, the time sequence can be subjected to sine and cosine conversion to obtain a time feature corresponding to the reference PUE value in the reference period. The trend, the seasonality, the later-period type and the fluctuation characteristics in the PUE characteristic data are fully mined to comprehensively consider the state of the PUE characteristic data of the target period, so that a foundation is laid for obtaining reasonable prediction data.

For example, a timestamp corresponding to a reference PUE value of a certain reference period is 12 months and 1 day 2022 years, the timestamp corresponds to 12 months, 4 quarters and fourths after being split, the corresponding time sequence is {12, 4}, and further, after sine and cosine transformation is performed on the time sequence {12, 4}, time characteristics {0.21,0.98,0.07,0.99 } are obtained.

It can be understood that, by introducing the time characteristic corresponding to the reference PUE value, the reference PUE characteristic data can be further enriched, and the coupling between the reference PUE value and the time characteristic is increased, thereby further guaranteeing the accuracy of the prediction of the PUE value.

Fig. 2 is a flowchart of another PUE prediction method provided according to an embodiment of the present disclosure. Based on the above embodiment, this embodiment provides an alternative implementation scheme for further optimizing "predicting PUE values of a data center in a target period according to PUE feature representation and autocorrelation features". As shown in fig. 2, the PUE prediction method of this embodiment may include:

s201, respectively carrying out feature extraction on the reference PUE feature data in at least two reference periods to obtain PUE feature representation of the reference PUE feature data.

Wherein the reference period is located before the target period.

S202, determining autocorrelation characteristics among the reference PUE characteristic data in different reference periods.

S203, processing the PUE characteristic representation of the reference PUE characteristic data and the autocorrelation characteristic of the reference PUE characteristic data by adopting an attention network in the PUE prediction model to obtain the attention characteristic of the reference PUE characteristic data.

In this embodiment, the attention feature refers to feature data obtained through attention mechanism operation, and may be represented in a vector or matrix form. It should be noted that, in this embodiment, the dimension of the attention feature and the dimension of the reference PUE feature data may be the same.

Specifically, the PUE feature representation of the reference PUE feature data and the autocorrelation feature of the reference PUE feature data may be input to an attention network in the PUE prediction model, and the attention feature of the reference PUE feature data may be obtained through network learning. For example, the PUE feature representation of the reference PUE feature data is learned by LSTM, and the PUE feature representation may include network weights and output features of LSTM network outputs, and further may include Q matrices with network weights as attention network inputs, K matrices with output features as attention network inputs, and V matrices with autocorrelation features as attention network inputs, and further the attention network may obtain attention features by the following formula:

wherein Context represents attention feature, Q represents network weight, K represents output feature, V represents autocorrelation feature, dimension (K) represents solving dimension of K matrix, namely output feature, and K ^T Representing the result of transposing the K matrix, i.e., the output feature, and Softmax () representing the normalized exponential function.

And S204, predicting the PUE value of the data center in the target period according to the attention characteristics.

Specifically, a flattening Flatten operation can be used for reducing the dimension of the attention feature to obtain a one-dimensional feature, and then an activation operation such as a Sigmoid operation can be performed on the one-dimensional feature to predict the PUE value of the data center in the target period.

According to the technical scheme provided by the embodiment of the disclosure, the PUE characteristic representation of the reference PUE characteristic data is obtained by respectively carrying out characteristic extraction on the reference PUE characteristic data in at least two reference periods, then the autocorrelation characteristics among the reference PUE characteristic data in different reference periods are determined, the attention network in the PUE prediction model is adopted to process the PUE characteristic representation of the reference PUE characteristic data and the autocorrelation characteristics of the reference PUE characteristic data to obtain the attention characteristics of the reference PUE characteristic data, and finally the PUE value of the data center in the target period is predicted according to the attention characteristics. According to the technical scheme, attention features are introduced, the dependency degree between the PUE feature representation and the autocorrelation features is further extracted, so that the depth and the breadth of PUE feature data are fully mined, and the finally predicted PUE value is more accurate and reasonable.

Fig. 3 is a flowchart of another PUE prediction method provided in accordance with an embodiment of the present disclosure. This example provides an alternative embodiment to further optimize the "predicting PUE value of data center in target period according to attention characteristics" based on the above example. As shown in fig. 3, the PUE prediction method of this embodiment may include:

s301, respectively carrying out feature extraction on the reference PUE feature data in at least two reference periods to obtain PUE feature representation of the reference PUE feature data.

Wherein the reference period is located before the target period.

And S302, determining autocorrelation characteristics among the reference PUE characteristic data in different reference periods.

S303, processing the PUE characteristic representation of the reference PUE characteristic data and the autocorrelation characteristic of the reference PUE characteristic data by adopting an attention network in the PUE prediction model to obtain the attention characteristic of the reference PUE characteristic data.

S304, weighting the reference PUE characteristic data in at least two reference periods to obtain weighted PUE characteristic data.

In this embodiment, the weighted PUE feature data refers to feature data obtained by weighting the reference PUE feature data.

Optionally, a preset weighting manner may be adopted to respectively weight the reference PUE feature data in at least two reference periods, so as to obtain weighted PUE feature data corresponding to each reference PUE feature data. For example, for each reference period, determining a reference weight for the reference period; and giving corresponding reference weight to the reference PUE characteristic data in the reference period to obtain weighted PUE characteristic data in the reference period.

The reference weight refers to the importance of the reference PUE feature data in the reference period to the final predicted PUE value, and the greater the reference weight is, the greater the importance is.

Specifically, for each reference period, a reference weight of the reference period is determined, and a result of multiplying the reference PUE feature data in the reference period by the reference weight of the reference period is used as weighted PUE feature data in the reference period. Further, the result of multiplying the reference PUE feature data in the reference period by the reference weight of the reference period may be added to the reference PUE feature data in the reference period, and the added result may be used as the weighted PUE feature data in the reference period.

It can be understood that a certain weight is given to the reference PUE feature data in the reference period, and the influence degree of each reference period on the PUE prediction of the subsequent target period can be fully mined, so that the PUE value prediction of the target period is more reasonable.

S305, fusing the weighted PUE characteristic data and the attention characteristic to obtain a fused characteristic.

In this embodiment, the fusion feature refers to a feature obtained by fusing the attention feature and the weighted PUE feature data corresponding to each reference PUE feature data, and may be represented in a vector or matrix form.

Specifically, a preset fusion mode may be adopted to fuse the attention feature and the weighted PUE feature data corresponding to each reference PUE feature data to obtain a fusion feature. For example, the attention feature and the weighted PUE feature data corresponding to each reference PUE feature data may be spliced, and the spliced feature data may be used as the fusion feature. For another example, the attention feature and the weighted PUE feature data corresponding to each reference PUE feature data may be added, and the added feature data may be used as the fusion feature.

S306, predicting the PUE value of the data center in the target period according to the fusion characteristics.

Specifically, a flattening Flatten operation can be used for reducing the dimension of the fusion feature to obtain a one-dimensional feature, and then an activation operation such as a Sigmoid operation can be performed on the one-dimensional feature to predict and obtain a PUE value of the data center in a target period.

According to the technical scheme provided by the embodiment of the disclosure, the PUE characteristic representation of the reference PUE characteristic data is obtained by respectively carrying out characteristic extraction on the reference PUE characteristic data in at least two reference periods, the autocorrelation characteristics among the reference PUE characteristic data in different reference periods are determined, then the attention network in the PUE prediction model is adopted to process the PUE characteristic representation of the reference PUE characteristic data and the autocorrelation characteristics of the reference PUE characteristic data to obtain the attention characteristics of the reference PUE characteristic data, further the reference PUE characteristic data in at least two reference periods are weighted to obtain weighted PUE characteristic data, the weighted PUE characteristic data and the attention characteristics are fused to obtain the fusion characteristics, and finally the PUE value of the data center in the target period is predicted according to the fusion characteristics. According to the technical scheme, the weighted PUE characteristic data are introduced, and the deviation degree of the PUE value in the predicted target period can be adjusted, so that the predicted PUE value is more fit with an actual value.

FIG. 4A is a flowchart of a training method of a PUE prediction model provided according to an embodiment of the disclosure; fig. 4B is a schematic diagram of a training process of a PUE prediction model provided according to an embodiment of the present disclosure. The method is suitable for the situation of predicting the PUE index of the data center. The method may be performed by a training apparatus for the PUE prediction model, which may be implemented in software and/or hardware, and may be integrated into an electronic device, such as a server, that carries a training function of the PUE prediction model. As shown in fig. 4A and 4B, the training method of the PUE prediction model of this embodiment may include:

s401, respectively carrying out feature extraction on the reference PUE feature data in at least two reference periods to obtain PUE feature representation of the reference PUE feature data.

Here, the period refers to a time period during which the PUE value of the data center is calculated once, and may be, for example, a week, a day, or several hours. The target period is a period in which the PUE value of the data center needs to be predicted, and may be a certain period of the history. The reference period refers to a period referred to when the PUE value of the data center is predicted; alternatively, the reference period may be a certain period of the history; it should be noted that, the reference period is located before the target period; further, the different reference periods may be consecutive to each other.

The reference PUE characteristic data is data used for predicting PUE values of the data center; alternatively, the reference PUE feature data may be PUE values of the data center at a reference period. Furthermore, data cleaning and data standardization processing can be carried out on the reference PUE characteristic data.

Specifically, a deep neural network may be adopted to perform feature extraction on the reference PUE feature data in at least two reference periods, so as to obtain the PUE feature representation of the reference PUE feature data. The deep neural network may be a long-term memory network (LSTM) or the like.

S402, determining autocorrelation characteristics among the reference PUE characteristic data in different reference periods.

In this embodiment, the autocorrelation features refer to features used for characterizing the correlation between reference PUE feature data in different reference periods, and may be represented in a vector or matrix form.

Specifically, the autocorrelation operation may be performed on the reference PUE characteristic data in different reference periods to obtain autocorrelation characteristics between the reference PUE characteristic data in different reference periods. For example, the autocorrelation processing layer of the PUE prediction model may be used to perform autocorrelation operation on the reference PUE feature data in different reference periods, so as to obtain autocorrelation features between the reference PUE feature data in different reference periods.

And S403, predicting the PUE value of the data center in the target period according to the PUE characteristic representation and the autocorrelation characteristic.

Specifically, based on a preset prediction rule, the PUE value of the data center in the target period may be predicted according to the PUE feature representation of each reference PUE feature data and the autocorrelation feature between different reference PUE feature data. For example, the PUE feature representation of each reference PUE feature data and the autocorrelation feature between different reference PUE feature data may be fused to obtain a fused feature, the dimension reduction operation is performed on the fused feature to obtain a one-dimensional feature, and then the one-dimensional feature is predicted by using an activation operation (e.g., a Sigmoid operation) to obtain a PUE value of the data center in a target period.

S404, determining the training loss according to the predicted PUE value of the data center in the target period and the PUE label value of the data center in the target period.

In this embodiment, the PUE value refers to a predicted PUE value of the data center in the target period. The PUE tag value is the real PUE value of the data center in the target period.

Specifically, the training loss may be determined according to the predicted PUE value of the data center in the target period and the PUE label value of the data center in the target period based on a preset loss function, such as a cross entropy loss function.

S405, training the PUE prediction model by adopting the training loss.

Specifically, the PUE prediction model may be trained using the training loss until the training stopping condition is satisfied, and the training of the PUE prediction model is stopped. The training stop condition may be that the training loss is stable within a set range, or the number of iterations reaches a set value, or the like. Wherein, the setting range and the setting value can be set by the technicians in the field according to the actual situation.

As a specific example, as shown in FIG. 4B, the PUE prediction model may include at least one feature extraction sub-model and a prediction network, each feature extraction sub-model including an LSTM network, an autocorrelation processing layer and an attention network. Specifically, reference PUE feature data (marked as original data) of a data center in at least two reference periods are input into an LSTM network in a first feature extraction sub-model to obtain network weight (Q) and output feature (K) output by the LSTM network, wherein the LSTM comprises a plurality of LSTM neural cell layers, and H marks hidden state transmission among the plurality of layers of neural networks; meanwhile, inputting reference PUE characteristic data (marked as original data) of the data center in at least two reference periods into an autocorrelation processing network in a first characteristic extraction submodel, performing autocorrelation operation to obtain autocorrelation characteristics (V) among the reference PUE characteristic data in different reference periods, and then taking network weight (Q), output characteristics (K) and the autocorrelation characteristics (V) as the input of an attention network in the first characteristic extraction submodel to perform attention operation to obtain attention characteristics output by the first characteristic extraction submodel; then, inputting the attention feature output by the first feature extraction submodel into a next feature extraction submodel of the first feature extraction submodel; and so on. And then, fusing the attention feature output by the last feature extraction submodule in the PUE prediction model and the weighted PUE feature data obtained after the weighted residual is carried out on the reference PUE feature data in at least two reference periods to obtain a fusion feature, inputting the fusion feature into a prediction network of the PUE prediction model for prediction, for example, the prediction network adopts a Sigmoid activation function to calculate the fusion feature, and predicting to obtain a PUE value of the data center in a target period and recording as Yi. And finally, determining the training loss according to the predicted PUE value of the data center in the target period and the PUE label value of the data center in the target period, and training the PUE prediction model by adopting the training loss.

It should be noted that, when the PUE prediction model is trained, a 10-fold cross validation training mode may be adopted, that is, the training set is randomly divided into 10 parts, wherein 9 parts of data are used for model training, 1 part of data are used for model testing, one training cycle is completed, and after every 1 part of data completes the model testing, the training is completed.

According to the technical scheme provided by the embodiment of the disclosure, the PUE characteristic representation of the reference PUE characteristic data is obtained by respectively carrying out characteristic extraction on the reference PUE characteristic data in at least two reference periods, then the autocorrelation characteristics among the reference PUE characteristic data in different reference periods are determined, further, the PUE value of the data center in the target period is predicted according to the PUE characteristic representation and the autocorrelation characteristics, finally, the training loss is determined according to the predicted PUE value of the data center in the target period and the PUE label value of the data center in the target period, and the PUE prediction model is trained by adopting the training loss. According to the technical scheme, the characteristics of the PUE characteristic data of different reference periods can be fully mined by extracting the characteristics of the PUE characteristic data of a plurality of reference periods; meanwhile, the autocorrelation characteristics are introduced, the dependency relationship and the mutual influence degree among the PUE characteristic data of different reference periods can be fully mined, and therefore a good foundation is laid for improving the prediction accuracy of the PUE value.

Fig. 5 is a schematic structural diagram of a PUE prediction apparatus according to an embodiment of the present disclosure. The embodiment is suitable for the situation of how to predict the PUE index of the data center. The apparatus can be implemented in software and/or hardware, and can be integrated into an electronic device, such as a server, that carries the PUE prediction function. As shown in fig. 5, the PUE prediction apparatus 500 of the present embodiment may include:

a PUE feature extraction module 501, configured to perform feature extraction on reference PUE feature data in at least two reference periods, respectively, to obtain PUE feature representation of the reference PUE feature data; the reference period is located before the target period;

an autocorrelation feature determining module 502, configured to determine autocorrelation features between reference PUE feature data in different reference periods;

and a PUE value prediction module 503, configured to predict a PUE value of the data center in the target period according to the PUE feature representation and the autocorrelation feature.

Further, the PUE value prediction module 503 includes:

an attention feature determination unit, configured to process the PUE feature representation of the reference PUE feature data and the autocorrelation feature of the reference PUE feature data by using an attention network in the PUE prediction model, to obtain an attention feature of the reference PUE feature data;

and the PUE value prediction unit is used for predicting the PUE value of the data center in the target period according to the attention characteristics.

Further, the PUE value prediction unit is specifically configured to:

a weighted PUE characteristic determining subunit, configured to weight the reference PUE characteristic data in the at least two reference periods to obtain weighted PUE characteristic data;

a fusion characteristic determining subunit, configured to fuse the weighted PUE characteristic data and the attention characteristic to obtain a fusion characteristic;

and the PUE value prediction subunit is used for predicting the PUE value of the data center in the target period according to the fusion characteristics.

Further, the weighted PUE feature determination subunit is specifically configured to:

for each reference period, determining a reference weight of the reference period;

and giving corresponding reference weight to the reference PUE characteristic data in the reference period to obtain weighted PUE characteristic data in the reference period.

Further, the reference PUE feature data includes a reference PUE value of the data center in a reference period and a time feature corresponding to the reference PUE value.

Further, the apparatus further comprises:

the time stamp determining module is used for acquiring a time stamp corresponding to a reference PUE value of the data center in each reference period;

a time sequence determining module, configured to determine a time sequence of the reference period according to the timestamp corresponding to the reference period;

and the time characteristic determining module is used for converting the time sequence of the reference period to obtain the time characteristic corresponding to the reference PUE value in the reference period.

Fig. 6 is a schematic structural diagram of a training apparatus for a PUE prediction model according to an embodiment of the present disclosure. The embodiment is suitable for the situation of how to predict the PUE index of the data center. The device can be implemented in software and/or hardware, and can be integrated into an electronic device, such as a server, that carries the training function of the PUE prediction model. As shown in fig. 6, the training apparatus 600 for the PUE prediction model of the present embodiment may include:

a PUE feature extraction module 601, configured to perform feature extraction on reference PUE feature data in at least two reference periods, respectively, to obtain PUE feature representation of the reference PUE feature data; the reference period is located before the target period;

an autocorrelation feature determining module 602, configured to determine autocorrelation features between reference PUE feature data in different reference periods;

a PUE value prediction module 603, configured to predict a PUE value of the data center in the target period according to the PUE feature representation and the autocorrelation feature;

a training loss determining module 604, configured to determine a training loss according to the predicted PUE value of the data center in the target period and the PUE tag value of the data center in the target period;

and a model training module 605, configured to train the PUE prediction model by using the training loss.

The present disclosure also provides an electronic device, a readable storage medium, and a computer program product according to embodiments of the present disclosure.

FIG. 7 is a block diagram of an electronic device for implementing a PUE prediction method or a training method of a PUE prediction model according to an embodiment of the present disclosure. FIG. 7 illustrates a schematic block diagram of an example electronic device 700 that can be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular telephones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 7, the electronic device 700 includes a computing unit 701, which may perform various appropriate actions and processes according to a computer program stored in a Read Only Memory (ROM) 702 or a computer program loaded from a storage unit 708 into a Random Access Memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the electronic device 700 can also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other by a bus 704. An input/output (I/O) interface 705 is also connected to bus 704.

A number of components in the electronic device 700 are connected to the I/O interface 705, including: an input unit 706 such as a keyboard, a mouse, or the like; an output unit 707 such as various types of displays, speakers, and the like; a storage unit 708 such as a magnetic disk, optical disk, or the like; and a communication unit 709 such as a network card, modem, wireless communication transceiver, etc. The communication unit 709 allows the electronic device 700 to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.

Computing unit 701 may be a variety of general purpose and/or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various dedicated Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and so forth. The calculation unit 701 performs the respective methods and processes described above, such as the PUE prediction method or the training method of the PUE prediction model. For example, in some embodiments, the PUE prediction method or the training method of the PUE prediction model may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program may be loaded and/or installed onto the electronic device 700 via the ROM 702 and/or the communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the PUE prediction method or the training method of the PUE prediction model described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the PUE prediction method or the training method of the PUE prediction model in any other suitable manner (e.g., by means of firmware).

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, field Programmable Gate Arrays (FPGAs), application Specific Integrated Circuits (ASICs), application Specific Standard Products (ASSPs), system on a chip (SOCs), complex Programmable Logic Devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions/operations specified in the flowchart and/or block diagram to be performed. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user may provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), wide Area Networks (WANs), and the Internet.

The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server may be a cloud server, a server of a distributed system, or a server with a combined blockchain.

Artificial intelligence is the subject of research that makes computers simulate some human mental processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), both at the hardware level and at the software level. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing, and the like; the artificial intelligence software technology mainly comprises a computer vision technology, a voice recognition technology, a natural language processing technology, a machine learning/deep learning technology, a big data processing technology, a knowledge map technology and the like.

Cloud computing (cloud computing) refers to accessing an elastically extensible shared physical or virtual resource pool through a network, where resources may include servers, operating systems, networks, software, applications, storage devices, and the like, and may be a technical system that deploys and manages resources in a self-service manner as needed. Through the cloud computing technology, high-efficiency and strong data processing capacity can be provided for technical application and model training of artificial intelligence, block chains and the like.

It should be understood that various forms of the flows shown above, reordering, adding or deleting steps, may be used. For example, the steps described in the present disclosure may be executed in parallel, sequentially, or in different orders, and are not limited herein as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved.

The above detailed description should not be construed as limiting the scope of the disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present disclosure should be included in the protection scope of the present disclosure.

Claims

1. A PUE (Power efficiency index) prediction method comprises the following steps:

2. The method of claim 1, wherein predicting the PUE value of the data center in the target period based on the PUE feature representation and the autocorrelation feature comprises:

processing the PUE characteristic representation of the reference PUE characteristic data and the autocorrelation characteristic of the reference PUE characteristic data by adopting an attention network in a PUE prediction model to obtain the attention characteristic of the reference PUE characteristic data;

and predicting the PUE value of the data center in the target period according to the attention feature.

3. The method of claim 2, wherein the predicting, from the attention feature, a PUE value for a data center in a target period comprises:

weighting the reference PUE characteristic data in at least two reference periods to obtain weighted PUE characteristic data;

fusing the weighted PUE characteristic data and the attention characteristic to obtain a fused characteristic;

and predicting the PUE value of the data center in the target period according to the fusion characteristics.

4. The method of claim 3, wherein the weighting the reference PUE feature data in the at least two reference periods to obtain weighted PUE feature data comprises:

5. The method of any of claims 1-4, wherein the reference PUE characteristic data comprises a reference PUE value of the data center in a reference period and a time characteristic corresponding to the reference PUE value.

6. The method of claim 5, further comprising:

acquiring a time stamp corresponding to a reference PUE value of the data center in each reference period;

determining the time sequence of the reference period according to the time stamp corresponding to the reference period;

and converting the time sequence of the reference period to obtain the time characteristic corresponding to the reference PUE value in the reference period.

7. A training method of a PUE prediction model comprises the following steps:

and training the PUE prediction model by adopting the training loss.

8. A PUE prediction apparatus comprising:

9. The apparatus of claim 8, wherein the PUE value prediction module comprises:

an attention feature determining unit, configured to process the PUE feature representation of the reference PUE feature data and the autocorrelation feature of the reference PUE feature data by using an attention network in a PUE prediction model, to obtain an attention feature of the reference PUE feature data;

and the PUE value prediction unit is used for predicting the PUE value of the data center in the target period according to the attention feature.

10. The apparatus according to claim 9, wherein the PUE value prediction unit is specifically configured to:

a fusion feature determining subunit, configured to fuse the weighted PUE feature data and the attention feature to obtain a fusion feature;

11. The apparatus of claim 10, wherein the weighted PUE feature determination subunit is specifically configured to:

12. The apparatus of any of claims 8-11, wherein the reference PUE feature data comprises a reference PUE value of a data center in a reference period and a time feature corresponding to the reference PUE value.

13. The apparatus of claim 12, the apparatus further comprising:

14. An apparatus for training a PUE prediction model, comprising:

the PUE characteristic extraction module is used for respectively extracting the characteristics of reference PUE characteristic data in at least two reference periods to obtain PUE characteristic representation of the reference PUE characteristic data; the reference period is positioned before the target period;

the system comprises an autocorrelation characteristic determining module, a correlation analysis module and a correlation analysis module, wherein the autocorrelation characteristic determining module is used for determining autocorrelation characteristics among reference PUE characteristic data in different reference periods;

the PUE value prediction module is used for predicting the PUE value of the data center in the target period according to the PUE characteristic representation and the autocorrelation characteristic;

15. An electronic device, comprising:

at least one processor; and

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the PUE prediction method of any one of claims 1-6 or the training method of the PUE prediction model of claim 7.

16. A non-transitory computer readable storage medium storing computer instructions for causing a computer to perform the PUE prediction method of any one of claims 1-6 or the training method of the PUE prediction model of claim 7.

17. A computer program product comprising a computer program which, when executed by a processor, implements the PUE prediction method of any one of claims 1-6, or the training method of the PUE prediction model of claim 7.