CN115965160A - Data center energy consumption prediction method and device, storage medium and electronic equipment - Google Patents
Data center energy consumption prediction method and device, storage medium and electronic equipment Download PDFInfo
- Publication number
- CN115965160A CN115965160A CN202310090163.2A CN202310090163A CN115965160A CN 115965160 A CN115965160 A CN 115965160A CN 202310090163 A CN202310090163 A CN 202310090163A CN 115965160 A CN115965160 A CN 115965160A
- Authority
- CN
- China
- Prior art keywords
- energy consumption
- data center
- center energy
- data
- data set
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a data center energy consumption prediction method, a data center energy consumption prediction device, a storage medium and electronic equipment, wherein a data center energy consumption original data set is obtained, and a target data center energy consumption data set is obtained through a mutual information theory method and a data preprocessing method; constructing a graph convolution neural network based on a target data center energy consumption data set and a mutual information theoretical method; and establishing a data center energy consumption prediction model through a graph convolution neural network and a preset time convolution network based on the target data center energy consumption data set to obtain a data center energy consumption prediction result. The method utilizes the mutual information theory to respectively perform characteristic dimension reduction on the original energy consumption data sets of the data center, thereby effectively reducing the complexity of model construction; a combined data center energy consumption prediction model is established through a graph convolution network and a time convolution network, the former aggregates energy consumption characteristics of nodes in the neighborhood of each node, and the latter extracts relevant information of a time sequence, so that accurate prediction of data center energy consumption and energy consumption indexes is achieved.
Description
Technical Field
The invention relates to the technical field of energy conservation of data centers, in particular to a method and a device for predicting energy consumption of a data center, a storage medium and electronic equipment.
Background
With the development of information technology, the industries such as cloud computing, big data, artificial intelligence, 5G and the like are promoted to rise continuously, and the data center serves as specific equipment of global agreement and takes on the functions of transmitting, accelerating, displaying, calculating and storing data information on network infrastructure. With the continuous development and growth of the number and scale of data centers and the increasing of energy consumption level, how to effectively solve the outstanding energy consumption problem is a problem to be solved urgently at present in order to realize the energy-saving operation of the data centers on the premise of ensuring the safe and stable operation of the data centers. Therefore, the energy consumption of the data center is accurately predicted, and effective data support and guarantee can be provided for management personnel to manage the energy consumption.
At present, most of data center energy consumption predictions are made for a specific energy consumption object, including predicting the overall energy consumption of a data center, or predicting the energy consumption of a server of the data center, or researching the energy consumption of a cold source system of the data center, and the like. And the research on the energy consumption of a single object cannot calculate indexes such as PUE, PLF or CLF and the like, so that the energy consumption level of the data center cannot be comprehensively evaluated. Meanwhile, less researchers simultaneously research the overall energy consumption of the data center and the energy consumption of each system.
Disclosure of Invention
In view of this, embodiments of the present invention provide a method, an apparatus, a storage medium, and an electronic device for predicting energy consumption of a data center, so as to solve the technical problem in the prior art that an energy consumption level of the data center cannot be comprehensively evaluated.
The technical scheme provided by the invention is as follows:
in a first aspect, an embodiment of the present invention provides a data center energy consumption prediction method, where the data center energy consumption prediction method includes: acquiring an energy consumption original data set of a data center; based on the data center energy consumption original data set, a target data center energy consumption data set is obtained through a mutual information theory method and a data preprocessing method; constructing a graph convolution neural network based on the target data center energy consumption data set and the mutual information theoretical method; establishing a data center energy consumption prediction model through the graph convolution neural network and a preset time convolution network based on the target data center energy consumption data set; and obtaining a data center energy consumption prediction result through the data center energy consumption prediction model based on the data center energy consumption original data set.
With reference to the first aspect, in a possible implementation manner of the first aspect, obtaining a target data center energy consumption data set based on the data center energy consumption original data set through a mutual information theory method and a data preprocessing method includes: based on the data center energy consumption original data set, a first data center energy consumption data set is obtained through the mutual information theory method processing; and obtaining the target data center energy consumption data set through the data preprocessing method based on the first data center energy consumption data set.
With reference to the first aspect, in another possible implementation manner of the first aspect, constructing a graph convolution neural network based on the target data center energy consumption data set and the mutual information theory method includes: determining a correlation data set corresponding to the target data center energy consumption data set, wherein data in the correlation data set table represents the correlation degree of energy consumption nodes corresponding to every two data center energy consumption data in the target data center energy consumption data set; determining the connection relation of every two energy consumption nodes based on the correlation data set; and constructing the graph convolution neural network based on the connection relation of every two energy consumption nodes.
With reference to the first aspect, in a further possible implementation manner of the first aspect, before establishing a data center energy consumption prediction model based on the target data center energy consumption dataset through the graph convolution neural network and a preset time convolution network, the method further includes: determining a data center energy consumption data prediction sequence set through the preset time convolution network based on the data center energy consumption original data set; and determining a target loss function based on the data center energy consumption original data set and the data center energy consumption data prediction sequence set.
With reference to the first aspect, in yet another possible implementation manner of the first aspect, establishing a data center energy consumption prediction model based on the target data center energy consumption dataset through the graph convolution neural network and a preset time convolution network includes: obtaining a data center energy consumption coupling characteristic data set through the graph convolution neural network based on the target data center energy consumption data set; based on the data center energy consumption coupling characteristic data set, obtaining a time characteristic data set through the preset time convolution network; and establishing the data center energy consumption prediction model through the target loss function based on the data center energy consumption coupling characteristic data set and the time characteristic data set.
With reference to the first aspect, in yet another possible implementation manner of the first aspect, after establishing a data center energy consumption prediction model based on the target data center energy consumption dataset through the graph convolution neural network and a preset time convolution network, the method further includes: acquiring a hyper-parameter set of the data center energy consumption prediction model; and initializing the hyper-parameter set.
With reference to the first aspect, in a further possible implementation manner of the first aspect, after the data center energy consumption prediction model is built based on the target data center energy consumption dataset through the graph convolution neural network and a preset time convolution network, the method further includes: acquiring a preset evaluation index system set; and evaluating the data center energy consumption prediction model based on each evaluation index in the preset evaluation index system set to obtain a prediction performance evaluation result of the data center energy consumption prediction model.
In a second aspect, an embodiment of the present invention provides a data center energy consumption prediction apparatus, where the data center energy consumption prediction apparatus includes: the acquisition module is used for acquiring an energy consumption original data set of the data center; the processing module is used for obtaining a target data center energy consumption data set through a mutual information theory method and a data preprocessing method based on the data center energy consumption original data set; the construction module is used for constructing a graph convolution neural network based on the target data center energy consumption data set and the mutual information theoretical method; the establishing module is used for establishing a data center energy consumption prediction model through the graph convolution neural network and a preset time convolution network based on the target data center energy consumption data set; and the prediction module is used for obtaining a data center energy consumption prediction result through the data center energy consumption prediction model based on the data center energy consumption original data set.
In a third aspect, an embodiment of the present invention provides a computer-readable storage medium, where computer instructions are stored, and the computer instructions are configured to cause the computer to execute the data center energy consumption prediction method according to any one of the first aspect and the first aspect of the embodiments of the present invention.
In a fourth aspect, an embodiment of the present invention provides an electronic device, including: the data center energy consumption prediction method comprises a memory and a processor, wherein the memory and the processor are connected in communication with each other, the memory stores computer instructions, and the processor executes the computer instructions to execute the data center energy consumption prediction method according to the first aspect and any one of the first aspect of the embodiments of the invention.
The technical scheme provided by the invention has the following effects:
in the feature selection, a mutual information theory is provided to respectively perform feature dimension reduction on the original data set of the energy consumption of the data center, and the data center operation state parameter and the environment parameter which are most relevant to each energy consumption data are selected, so that the method is not only suitable for real-value random variables, but also suitable for discrete random variables; a combined data center energy consumption prediction model is established through a graph convolution network and a time convolution network, the former aggregates energy consumption characteristics of nodes in the neighborhood of each node, and the latter extracts relevant information of a time sequence, so that accurate prediction of data center energy consumption and energy consumption indexes is achieved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a data center energy consumption prediction method according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of a graph convolutional neural network provided according to an embodiment of the present invention;
FIG. 3 is a diagram of a causal dilation convolution provided in accordance with an embodiment of the present invention;
FIG. 4 is a schematic flow chart diagram of a residual module provided according to an embodiment of the present invention;
FIG. 5 is a block diagram of a data center energy consumption prediction apparatus according to an embodiment of the present invention;
FIG. 6 is a schematic structural diagram of a computer-readable storage medium provided according to an embodiment of the present invention;
fig. 7 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above 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.
The embodiment of the invention provides a data center energy consumption prediction method, as shown in fig. 1, the method comprises the following steps:
step 101: and acquiring an energy consumption original data set of the data center.
The data center energy consumption original data in the data center energy consumption original data set are used for representing characteristic historical data such as energy consumption data and operation state parameters based on operation of a data center machine room, and mainly comprise IT energy consumption, refrigeration energy consumption, other energy consumption, total energy consumption, other operation state parameters and monitored environment parameter data.
Step 102: and obtaining a target data center energy consumption data set through a mutual information theory method and a data preprocessing method based on the data center energy consumption original data set.
Specifically, the sample of the raw data of the data center includes energy consumption data and characteristic data related to energy consumption, the related characteristic attributes are not related to the energy consumption to the same extent, and some attributes are closely related to a certain energy consumption but are weakly related to other energy consumption. In order to effectively reduce the complexity of the model and improve the prediction accuracy, the Mutual Information (MI) theory of the correlation analysis method is adopted to perform feature screening on the original energy consumption data set of the data center in the embodiment of the invention.
And further, preprocessing the screened data set to obtain a target data center energy consumption data set.
Step 103: and constructing a graph convolution neural network based on the target data center energy consumption data set and the mutual information theory method.
Where a convolutional neural network is used to generalize the convolution operation from traditional data (images or meshes) to graph data, this convolutional neural network is shown in figure 2.
The input layer is a graph G = (V, E) with node features and structural features, the node set V is a node feature vector, and the association existing between nodes is represented by E; the hidden layer is a graph convolution layer, and the node characteristics are updated in a spectrum convolution mode or a space convolution mode; the ReLU layer is similar to an activation function of a convolutional neural network, and nonlinear transformation is carried out on the nodes; the output layer shows different forms according to different tasks, and can be used for node classification tasks, graph representation classification tasks and the like.
By constructing the graph convolution network, the implicit internal coupling characteristics among the energy consumption of each system can be effectively extracted, the energy consumption of each system in the data center can be predicted, the prediction precision can be improved, and the prediction of the multidimensional evaluation index of the data center is realized.
Step 104: and establishing a data center energy consumption prediction model through the graph convolution neural network and a preset time convolution network based on the target data center energy consumption data set.
Most of general energy consumption prediction models only study the energy consumption change rule of a single object or directly study the PUE historical data, but cannot systematically and comprehensively analyze and study the energy consumption of each system of a data center.
In the embodiment of the invention, the target data center energy consumption data set comprises the overall energy consumption of the data center and the energy consumption of each system, so that the trained data center energy consumption prediction model based on the data set can directly predict the IT energy consumption, the refrigeration energy consumption, other energy consumption and the total energy consumption of the data center, and can simultaneously predict the multidimensional energy efficiency index of the data center, and the definitions of the multidimensional energy efficiency index are respectively shown in the following relational expressions (1), (2) and (3):
in the formula:
expressing a PUE (Power Usage efficiency, index for evaluating energy efficiency of a data center) predicted value; />
Representing a predicted value of the refrigeration load coefficient; />
Representing a power supply load coefficient predicted value; />
Representing a total energy consumption predicted value; />
Representing an IT energy consumption predicted value; />
Representing a predicted value of refrigeration energy consumption; />
Representing other energy consumption prediction values.
Further, the Time Convolutional Network (TCN) has the advantages of stable gradient and low training memory. The TCN can complete the prediction in a parallelization way and can solve the sequence modeling task with causal constraint, namely, for the value of the time t of the upper layer, only the value of the time t of the next layer and the value before the time t of the next layer are relied on, which means that no information leakage occurs from the past to the future. By adopting a Time Convolution Network (TCN) to replace a traditional Recurrent Neural Network (RNN) and a variant long-short term memory network (LSTM) thereof and a gated cyclic unit (GRU), parallelization processing can be realized, and the problem of gradient explosion/disappearance is avoided.
Specifically, a Time Convolution Network (TCN) is employed in embodiments of the present invention to extract timing characteristics. As shown in fig. 3, which is a diagram of a causal dilation convolution structure, the dilation factor D =1,2,4, and the convolution kernel size is 3.
Therefore, the data center energy consumption prediction model is established based on the graph convolution neural network and the preset time convolution network, and the prediction accuracy of the data center energy consumption and the energy consumption index can be improved.
Step 105: and obtaining a data center energy consumption prediction result through the data center energy consumption prediction model based on the data center energy consumption original data set.
And predicting the IT energy consumption, the refrigeration energy consumption, other energy consumption and total energy consumption of the data center based on the trained neural network model, and obtaining the energy consumption prediction result of each system of the data center through inverse normalization processing.
Specifically, the obtained energy consumption original data of each system in the data center are input into the constructed data center energy consumption prediction model, and then the corresponding data center energy consumption prediction result can be obtained.
In the method for predicting the energy consumption of the data center, provided by the embodiment of the invention, in the feature selection, a mutual information theory is provided to respectively perform feature dimension reduction on the original data set of the energy consumption of the data center, and the data center operation state parameter and the environment parameter which are most relevant to each energy consumption data are selected, so that the method is not only suitable for real-value random variables, but also suitable for discrete random variables; a combined data center energy consumption prediction model is established through a graph convolution network and a time convolution network, the former aggregates energy consumption characteristics of nodes in the neighborhood of each node, and the latter extracts relevant information of a time sequence, so that accurate prediction of data center energy consumption and energy consumption indexes is achieved.
As an optional implementation manner of the embodiment of the present invention, step 102 includes: based on the data center energy consumption original data set, a first data center energy consumption data set is obtained through the mutual information theory method processing; and obtaining the target data center energy consumption data set through the data preprocessing method based on the first data center energy consumption data set.
Since the factors influencing the energy consumption of the data center are complex and large in quantity, however, for such a high-dimensional data set, which contains part of useless features, how to obtain the key features thereof, and reducing the feature dimension of the model is a key issue of research.
Furthermore, the sample of the raw data of the data center comprises energy consumption data and characteristic data related to energy consumption, the correlation degree of the related characteristic attributes and the energy consumption is different, and part of the attributes are closely related to a certain energy consumption but have weak correlation with other energy consumption. In order to effectively reduce the complexity of the model and improve the prediction precision, the embodiment of the invention utilizes a Mutual Information (Mutual Information) theory based on a correlation analysis method to perform feature screening on an original data set, selects the features with strong correlation characteristics with energy consumption, and reduces the feature dimension of the model and the risk of overfitting the model.
Specifically, characteristic screening is respectively carried out on IT energy consumption, refrigeration energy consumption, other energy consumption and total energy consumption of the data center, and operation state parameters and environment parameters of the data center according to Mutual Information (MI), and characteristic parameters with high correlation degree with each energy consumption of the data center are selected. The MI is an information measurement mode of the information theory, is used for characterizing the degree of interdependence between variables, and is defined by the following relation (4):
in the formula: x and Y represent two random variables; p (X) represents the edge density function of X; p (Y) represents the edge density function of Y; p (X, Y) represents the combined density function of X and Y.
Through an MI theoretical algorithm, model input of data center energy consumption prediction is finally determined, and therefore model training time can be effectively reduced and prediction accuracy can be improved. After determining the input parameters of the model, performing data preprocessing on the feature-screened data set, wherein the data preprocessing mainly comprises data normalization according to the following steps of 7:2:1 the data set is divided into a training set, a verification set and a test set.
The normalization method is shown in the following relation (5):
in the formula: x represents original sample data; x is the number of min Representing the minimum value of the sample data set; x is the number of max Representing the maximum value of the sample data set.
As an optional implementation manner of the embodiment of the present invention, step 103 includes: determining a correlation data set corresponding to the energy consumption data set of the target data center; determining the connection relation of every two energy consumption nodes based on the correlation data set; and constructing the graph convolution neural network based on the connection relation of every two energy consumption nodes.
And the data in the correlation data set represents the correlation degree of the energy consumption nodes corresponding to every two data center energy consumption data in the target data center energy consumption data set.
In order to fully extract the coupling characteristics among the energy consumptions of the systems, based on the characteristics of the energy consumption data of the data center, a proper graph structure is constructed by utilizing mutual information, and the method comprises the following steps: and measuring the degree of correlation between the two nodes, determining the connection relation of the two nodes, and constructing a graph convolutional neural network.
Specifically, the adjacency matrix is generally defined by a node distance or a graph feature similarity, as shown in the following relation (6):
in the formula: a (i, j) represents an adjacency matrix of a graph structure, which reflects a connected state between two nodes, 1 represents connection, and 0 represents disconnection; mutual information MI (h) i ,h j ) Representing energy consumption data mutual information values corresponding to the node i and the node j; and t represents the average value of the mutual information values of the energy consumption data of all the nodes.
Further, according to the description in step 103, the graph structure of the energy consumption coupling characteristics of the data center system is modeled as G = (V, E), where V represents a set of energy consumption nodes of the data center system, and in this embodiment of the present invention, IT energy consumption, cooling energy consumption, other energy consumption, and total energy consumption, that is, the graph structure has four nodes; e represents the set of edges in the graph structure, determined in this application by the relation (6).
As an optional implementation manner of the embodiment of the present invention, before step 104, the method further includes: determining a data center energy consumption data prediction sequence set through the preset time convolution network based on the data center energy consumption original data set; and determining a target loss function based on the data center energy consumption original data set and the data center energy consumption data prediction sequence set.
Specifically, as described in step 104, the time series modeling network is a non-linear mapping function f X → Y that must satisfy the causal constraint, i.e., is determined only from past inputs and not from any future inputs. I.e. training the network at a given input x by supervised learning 0 ,x 1 ,…,x T Time-wise computing a prediction sequence
Such that the predicted sequence and a given real output sequence y 0 ,y 1 ,…,y T The loss function between takes a minimum.
Namely, the objective loss function is expressed by the following relation (7):
in the formula: n represents the number of nodes, namely the number of energy consumption objects of the data center system;
a predicted value representing energy consumption; y is i Representing the true value of energy consumption.
As an optional implementation manner of the embodiment of the present invention, step 104 includes: based on the target data center energy consumption data set, obtaining a data center energy consumption coupling characteristic data set through the graph convolution neural network; based on the data center energy consumption coupling characteristic data set, obtaining a time characteristic data set through the preset time convolution network; and establishing the data center energy consumption prediction model through the target loss function based on the data center energy consumption coupling characteristic data set and the time characteristic data set.
Firstly, after determining the structure of the graph convolution neural network, extracting the coupling features, wherein the GCN is used as a neural network layer, and the information transmission mode between layers is shown as the following relation (8):
in the formula:
i represents an identity matrix; />
Represents->
A degree matrix of (c); h represents the characteristic of each layer, and the input layer H is an input characteristic X; w represents a weight matrix; σ denotes a nonlinear activation function.
Through the GCN layer, the coupling characteristics among the energy consumption nodes and other variable characteristics are aggregated from the neighborhood nodes to the target node, so that the extraction of the coupling characteristics among the energy consumption nodes of each system of the data center is realized, and a foundation is provided for a downstream time sequence prediction task.
General energy consumption prediction only considers the energy consumption of a single object system, neglects that in the same data center, the energy consumption of different systems has a complex coupling relation, for example, the energy consumption between a cold source system and an IT system has strong correlation, the energy consumption of the cold source system is not only greatly influenced by external environmental factors, but also strongly influenced by the energy consumption of the IT system, and when the data center is in a high-working-condition task running state, that is, the energy consumption of the IT system is high, the cold source system is often at a high energy consumption point. Therefore, in the embodiment of the invention, the graph neural convolution network GCN can be used for effectively extracting the implicit intrinsic coupling characteristic, and the prediction precision of the overall energy consumption of the data center and the energy consumption of each system can be effectively improved.
Further, the timing characteristics are extracted using a Time Convolutional Network (TCN) as described in step 104.
In order to solve causal constraint and increase the receptive field of the model, the TCN introduces causal dilation convolution to ensure that future information is not leaked. Given a one-dimensional timing input x ∈ R n And a convolution kernel F: {0, \8230:, k-1, the expansion convolution operation F of the element s is defined as shown in the following relation (9):
in the formula: d represents a swelling factor; * Representing a convolution operation; k represents the convolution kernel size; x is the number of s-d·i Indicating the path direction.
Further, the TCN network is composed of a one-dimensional full convolution network and a plurality of residual modules, which are shown in fig. 4.
When the input of the residual block is x, the output o is shown in the following relation (10):
o＝Activation(x+F(x)) (10)
in the formula: x represents the input of the residual module; o represents an output; f () represents a mapping function in a convolutional network; activation () represents an Activation function.
In the embodiment of the present invention, the input x of the residual error module is the output of the above relation (8), that is, the node of the graph volume network layer outputs the data center energy consumption coupling characteristic data set, that is; and o is the energy consumption prediction result at the current and future time.
And finally, obtaining a corresponding data center energy consumption prediction model by utilizing the data center energy consumption coupling characteristic data set extracted by the graph convolution neural network and the time characteristic data set extracted by the time convolution network through target loss function training.
The TCN is adopted to extract the time sequence characteristics in the energy consumption data of each system of the data center, and the causal constraint is solved through the expansion convolution structure, so that the information leakage cannot occur in the past and the future; compared with LSTM and GRU, it can adopt parallel operation and solve the problem of gradient explosion/disappearance.
As an optional implementation manner of the embodiment of the present invention, after step 104, the method further includes: acquiring a hyper-parameter set of the data center energy consumption prediction model; and initializing the hyper-parameter set.
The hyper-parameter set can comprise an input step length, an output step length, an initial weight, an initial bias, a batch size, iteration times, a regular method, coefficients and a learning rate of the data center energy consumption prediction model, and an empirical method is adopted to set the hyper-parameter set of the data center energy consumption prediction model in the embodiment of the invention.
Further, each hyper-parameter in the hyper-parameter set is initialized.
As an optional implementation manner of the embodiment of the present invention, after step 104, the method further includes: acquiring a preset evaluation index system set; and evaluating the data center energy consumption prediction model based on each evaluation index in the preset evaluation index system set to obtain a prediction performance evaluation result of the data center energy consumption prediction model.
Specifically, the model prediction performance in the present application can be evaluated by using various evaluation indexes. In the embodiment of the present invention, the centralized evaluation indexes of the preset evaluation index system may include a Mean Absolute Error (MAE), a Root Mean Square Error (RMSE), and a Mean Absolute Percentage Error (MAPE), which are defined as the following relations (11), (12), and (13), respectively:
in the formula: k represents the size of the output sample; y is t Representing an actual energy consumption value;
representing the predicted energy consumption value.
Further, the evaluation performance of the data center energy consumption prediction model is evaluated by using the multiple evaluation indexes, so that a prediction performance evaluation result of the data center energy consumption prediction model can be obtained.
An embodiment of the present invention further provides a data center energy consumption prediction apparatus, as shown in fig. 5, the apparatus includes:
an obtaining module 501, configured to obtain an energy consumption original data set of a data center; for details, reference is made to the description relating to step 101 in the above-described method embodiment.
A processing module 502, configured to obtain a target data center energy consumption data set through a mutual information theory method and a data preprocessing method based on the data center energy consumption original data set; see the above description of step 102 in the method embodiment for details.
A building module 503, configured to build a graph convolution neural network based on the target data center energy consumption data set and the mutual information theoretical method; see the above description of step 103 in the method embodiments for details.
The establishing module 504 is configured to establish a data center energy consumption prediction model through the graph convolution neural network and a preset time convolution network based on the target data center energy consumption data set; see the above description of step 104 in the method embodiment for details.
The prediction module 505 is configured to obtain a data center energy consumption prediction result through the data center energy consumption prediction model based on the data center energy consumption original data set; see the above description of step 105 in the method embodiment for details.
In the feature selection, the data center energy consumption prediction device provided by the embodiment of the invention provides a mutual information theory to respectively perform feature dimension reduction on the data center energy consumption original data set, and selects the data center operation state parameter and the environment parameter which are most relevant to each energy consumption data, so that the device is not only suitable for real-value random variables, but also suitable for discrete random variables; a combined data center energy consumption prediction model is established through a graph convolution network and a time convolution network, the former aggregates energy consumption characteristics of nodes in the neighborhood of each node, and the latter extracts relevant information of a time sequence, so that accurate prediction of data center energy consumption and energy consumption indexes is achieved.
As an optional implementation manner of the embodiment of the present invention, the processing module includes: the first processing submodule is used for obtaining a first data center energy consumption data set through the mutual information theory method processing based on the data center energy consumption original data set; and the second processing submodule is used for obtaining the target data center energy consumption data set through the data preprocessing method based on the first data center energy consumption data set.
As an optional implementation manner of the embodiment of the present invention, the building module includes: the first determining submodule is used for determining a correlation data set corresponding to the target data center energy consumption data set, and data in the correlation data set table represents the correlation degree of energy consumption nodes corresponding to every two data center energy consumption data in the target data center energy consumption data set; a second determining submodule, configured to determine, based on the correlation data set, a connection relationship between every two energy consumption nodes; and the construction submodule is used for constructing the graph convolution neural network based on the connection relation of every two energy consumption nodes.
As an optional implementation manner of the embodiment of the present invention, the apparatus further includes: the first determining module is used for determining a data center energy consumption data prediction sequence set through the preset time convolution network based on the data center energy consumption original data set; and the second determining module is used for determining a target loss function based on the data center energy consumption original data set and the data center energy consumption data prediction sequence set.
As an optional implementation manner of the embodiment of the present invention, the establishing module includes: the first extraction submodule is used for obtaining a data center energy consumption coupling characteristic data set through the graph convolution neural network based on the target data center energy consumption data set; the second extraction submodule is used for obtaining a time characteristic data set through the preset time convolution network based on the data center energy consumption coupling characteristic data set; and the establishing submodule is used for establishing the data center energy consumption prediction model through the target loss function based on the data center energy consumption coupling characteristic data set and the time characteristic data set.
As an optional implementation manner of the embodiment of the present invention, the apparatus further includes: the first acquisition module is used for acquiring a hyper-parameter set of the data center energy consumption prediction model; and the first processing module is used for carrying out initialization processing on the hyper-parameter set.
As an optional implementation manner of the embodiment of the present invention, the apparatus further includes: the second acquisition module is used for acquiring a preset evaluation index system set; and the evaluation module is used for evaluating the data center energy consumption prediction model based on each evaluation index in the preset evaluation index system set to obtain a prediction performance evaluation result of the data center energy consumption prediction model.
The functional description of the data center energy consumption prediction device provided by the embodiment of the invention refers to the description of the data center energy consumption prediction method in the above embodiment in detail.
An embodiment of the present invention further provides a storage medium, as shown in fig. 6, on which a computer program 601 is stored, where the instructions, when executed by a processor, implement the steps of the data center energy consumption prediction method in the foregoing embodiments. The storage medium may be a magnetic Disk, an optical Disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a Flash Memory (Flash Memory), a Hard Disk Drive (HDD), or a Solid State Drive (SSD); the storage medium may also comprise a combination of memories of the kind described above.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic Disk, an optical Disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a Flash Memory (Flash Memory), a Hard Disk (Hard Disk Drive, HDD), a Solid-State Drive (SSD), or the like; the storage medium may also comprise a combination of memories of the kind described above.
An embodiment of the present invention further provides an electronic device, as shown in fig. 7, the electronic device may include a processor 71 and a memory 72, where the processor 71 and the memory 72 may be connected through a bus or in another manner, and fig. 7 takes the connection through the bus as an example.
The processor 71 may be a Central Processing Unit (CPU). The Processor 71 may also be other general purpose processors, digital Signal Processors (DSPs), application Specific Integrated Circuits (ASICs), field Programmable Gate Arrays (FPGAs) or other Programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components, or combinations thereof.
The memory 72, which is a non-transitory computer-readable storage medium, may be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the corresponding program instructions/modules in the embodiments of the present invention. The processor 71 executes various functional applications and data processing of the processor by executing non-transitory software programs, instructions and modules stored in the memory 72, namely, the data center energy consumption prediction method in the above method embodiment is implemented.
The memory 72 may include a storage program area and a storage data area, wherein the storage program area may store an operating device, an application program required for at least one function; the storage data area may store data created by the processor 71, and the like. Further, the memory 72 may include high speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory 72 may optionally include memory located remotely from the processor 71, which may be connected to the processor 71 via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The one or more modules are stored in the memory 72 and, when executed by the processor 71, perform the data center energy consumption prediction method of the embodiment shown in fig. 1-4.
The details of the electronic device may be understood by referring to the corresponding descriptions and effects in the embodiments shown in fig. 1 to fig. 4, and are not described herein again.
Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations fall within the scope defined by the appended claims.
Claims (10)
1. A data center energy consumption prediction method is characterized by comprising the following steps:
acquiring an energy consumption original data set of a data center;
based on the data center energy consumption original data set, a target data center energy consumption data set is obtained through a mutual information theory method and a data preprocessing method;
constructing a graph convolution neural network based on the target data center energy consumption data set and the mutual information theoretical method;
establishing a data center energy consumption prediction model through the graph convolution neural network and a preset time convolution network based on the target data center energy consumption data set;
and obtaining a data center energy consumption prediction result through the data center energy consumption prediction model based on the data center energy consumption original data set.
2. The method of claim 1, wherein obtaining a target data center energy consumption data set based on the data center energy consumption raw data set through a mutual information theory method and a data preprocessing method comprises:
based on the data center energy consumption original data set, a first data center energy consumption data set is obtained through the mutual information theory method processing;
and obtaining the target data center energy consumption data set through the data preprocessing method based on the first data center energy consumption data set.
3. The method of claim 1, wherein constructing a convolutional neural network based on the target data center energy consumption dataset and the mutual information theory method comprises:
determining a correlation data set corresponding to the target data center energy consumption data set, wherein data in the correlation data set table represents the correlation degree of energy consumption nodes corresponding to every two data center energy consumption data in the target data center energy consumption data set;
determining the connection relation of every two energy consumption nodes based on the correlation data set;
and constructing the graph convolution neural network based on the connection relation of every two energy consumption nodes.
4. The method of claim 1, wherein before establishing a data center energy consumption prediction model based on the target data center energy consumption dataset through the graph convolution neural network and a preset time convolution network, the method further comprises:
determining a data center energy consumption data prediction sequence set through the preset time convolution network based on the data center energy consumption original data set;
and determining a target loss function based on the data center energy consumption original data set and the data center energy consumption data prediction sequence set.
5. The method of claim 4, wherein establishing a data center energy consumption prediction model based on the target data center energy consumption dataset through the graph convolution neural network and a preset time convolution network comprises:
based on the target data center energy consumption data set, obtaining a data center energy consumption coupling characteristic data set through the graph convolution neural network;
based on the data center energy consumption coupling characteristic data set, obtaining a time characteristic data set through the preset time convolution network;
and establishing the data center energy consumption prediction model through the target loss function based on the data center energy consumption coupling characteristic data set and the time characteristic data set.
6. The method of claim 1, wherein after establishing a data center energy consumption prediction model based on the target data center energy consumption dataset through the graph convolution neural network and a preset time convolution network, the method further comprises:
acquiring a hyper-parameter set of the data center energy consumption prediction model;
and initializing the hyper-parameter set.
7. The method of claim 6, wherein after establishing a data center energy consumption prediction model based on the target data center energy consumption dataset through the graph convolution neural network and a preset time convolution network, the method further comprises:
acquiring a preset evaluation index system set;
and evaluating the data center energy consumption prediction model based on each evaluation index in the preset evaluation index system set to obtain a prediction performance evaluation result of the data center energy consumption prediction model.
8. An apparatus for predicting energy consumption of a data center, the apparatus comprising:
the acquisition module is used for acquiring an energy consumption original data set of the data center;
the processing module is used for obtaining a target data center energy consumption data set through a mutual information theory method and a data preprocessing method based on the data center energy consumption original data set;
the construction module is used for constructing a graph convolution neural network based on the target data center energy consumption data set and the mutual information theoretical method;
the establishing module is used for establishing a data center energy consumption prediction model through the graph convolution neural network and a preset time convolution network based on the target data center energy consumption data set;
and the prediction module is used for obtaining a data center energy consumption prediction result through the data center energy consumption prediction model based on the data center energy consumption original data set.
9. A computer-readable storage medium storing computer instructions for causing a computer to perform the method of predicting energy consumption of a data center of any one of claims 1 to 7.
10. An electronic device, comprising: a memory and a processor, the memory and the processor being communicatively coupled to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method of predicting energy consumption of a data center according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310090163.2A CN115965160B (en) | 2023-01-18 | 2023-01-18 | Data center energy consumption prediction method and device, storage medium and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310090163.2A CN115965160B (en) | 2023-01-18 | 2023-01-18 | Data center energy consumption prediction method and device, storage medium and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115965160A true CN115965160A (en) | 2023-04-14 |
| CN115965160B CN115965160B (en) | 2023-08-08 |
Family
ID=87361528
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310090163.2A Active CN115965160B (en) | 2023-01-18 | 2023-01-18 | Data center energy consumption prediction method and device, storage medium and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115965160B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116663747A (en) * | 2023-07-19 | 2023-08-29 | 广东云下汇金科技有限公司 | Intelligent early warning method and system based on data center infrastructure |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10417565B1 (en) * | 2018-06-30 | 2019-09-17 | Carbon Lighthouse, Inc. | System, method, and computer program for modeling and predicting energy consumption in a building |
| CN114422381A (en) * | 2021-12-14 | 2022-04-29 | 西安电子科技大学 | Communication network flow prediction method, system, storage medium and computer equipment |
| CN114721835A (en) * | 2022-06-10 | 2022-07-08 | 湖南工商大学 | Method, system, device and medium for predicting energy consumption of edge data center server |
| CN114792158A (en) * | 2022-04-01 | 2022-07-26 | 三峡大学 | Multi-wind-farm short-term power prediction method based on space-time fusion graph neural network |
| CN115496286A (en) * | 2022-09-26 | 2022-12-20 | 重庆德宜高大数据科技有限公司 | Neural network carbon emission prediction method based on big data environment and application |
-
2023
- 2023-01-18 CN CN202310090163.2A patent/CN115965160B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10417565B1 (en) * | 2018-06-30 | 2019-09-17 | Carbon Lighthouse, Inc. | System, method, and computer program for modeling and predicting energy consumption in a building |
| CN114422381A (en) * | 2021-12-14 | 2022-04-29 | 西安电子科技大学 | Communication network flow prediction method, system, storage medium and computer equipment |
| CN114792158A (en) * | 2022-04-01 | 2022-07-26 | 三峡大学 | Multi-wind-farm short-term power prediction method based on space-time fusion graph neural network |
| CN114721835A (en) * | 2022-06-10 | 2022-07-08 | 湖南工商大学 | Method, system, device and medium for predicting energy consumption of edge data center server |
| CN115496286A (en) * | 2022-09-26 | 2022-12-20 | 重庆德宜高大数据科技有限公司 | Neural network carbon emission prediction method based on big data environment and application |
Non-Patent Citations (1)
| Title |
|---|
| 廖恩红 等: "基于时序卷积网络的云服务器性能预测模型", 《华南师范大学学报(自然科学版)》, vol. 52, no. 04, pages 107 - 113 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116663747A (en) * | 2023-07-19 | 2023-08-29 | 广东云下汇金科技有限公司 | Intelligent early warning method and system based on data center infrastructure |
| CN116663747B (en) * | 2023-07-19 | 2024-04-12 | 广东云下汇金科技有限公司 | Intelligent early warning method and system based on data center infrastructure |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115965160B (en) | 2023-08-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xuan et al. | Multi-model fusion short-term load forecasting based on random forest feature selection and hybrid neural network | |
| Wang et al. | A compound framework for wind speed forecasting based on comprehensive feature selection, quantile regression incorporated into convolutional simplified long short-term memory network and residual error correction | |
| Bendre et al. | Time series decomposition and predictive analytics using MapReduce framework | |
| Li et al. | Dynamic structure embedded online multiple-output regression for streaming data | |
| Jichang et al. | Water quality prediction model based on GRU hybrid network | |
| CN115204502A (en) | Training and predicting method, system, equipment and storage medium of pressure prediction model | |
| CN116562514B (en) | Method and system for immediately analyzing production conditions of enterprises based on neural network | |
| Stergiou et al. | Application of deep learning and chaos theory for load forecasting in Greece | |
| CN115965160B (en) | Data center energy consumption prediction method and device, storage medium and electronic equipment | |
| CN116187835A (en) | Data-driven-based method and system for estimating theoretical line loss interval of transformer area | |
| Shi et al. | Data recovery algorithm based on generative adversarial networks in crowd sensing Internet of Things | |
| Manoj et al. | FWS-DL: forecasting wind speed based on deep learning algorithms | |
| CN117439053A (en) | Method, device and storage medium for predicting electric quantity of Stacking integrated model | |
| Naoui et al. | Integrating iot devices and deep learning for renewable energy in big data system | |
| Xu et al. | Residual autoencoder-LSTM for city region vehicle emission pollution prediction | |
| Nandal et al. | Healthcare based financial decision making system using artificial intelligence | |
| CN116721327A (en) | Neural network architecture searching method based on generalization boundary | |
| Yan et al. | Transferability and robustness of a data-driven model built on a large number of buildings | |
| Tong et al. | A prediction model for complex equipment remaining useful life using gated recurrent unit complex networks | |
| Chang et al. | Enhanced road information representation in graph recurrent network for traffic speed prediction | |
| CN115080795A (en) | Multi-charging-station cooperative load prediction method and device | |
| Wang et al. | Ultra-short-term wind speed prediction based on empirical wavelet transform and combined model | |
| Jin et al. | Short-term load forecasting method based on adaptive graph convolutional recurrent network | |
| Xu et al. | Water Level Prediction Based on SSA-LSTM Model | |
| Sun et al. | Production quality early warning method for firearms components based on cloud edge collaboration |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |