Method and system for optimizing energy consumption of data center

Abstract

The invention provides an energy consumption optimization method and system for a data center. The method comprises the following steps: acquiring an initial network topology, real-time traffic demand data and historical traffic demand data of the data center; predicting to obtain a predicted traffic demand matrix of a future time period based on the real-time and historical traffic demand data; performing feature extraction on the initial network topology to obtain an initial topology feature matrix; the predicted traffic demand matrix and the initial topological feature matrix are input into a neural network model, the model outputs an optimal topological feature matrix prediction result corresponding to the predicted traffic demand matrix, and the model is obtained through training by minimizing total loss including energy consumption loss in advance; enabling a difference value between an energy consumption optimization prediction result corresponding to the traffic demand matrix in the training data set and a set energy consumption optimal label to be smaller than a threshold value through minimizing the energy consumption loss; and reconstructing the initial network topology based on a prediction result of the optimal topology feature matrix to obtain an optimal network topology corresponding to the predicted traffic demand matrix.

Description

Technical Field

[0001] The present invention relates to the technical fields of data centers and energy consumption optimization, and particularly to an energy consumption optimization method and system for a data center. Background Art

[0002] With the rapid development of information technology and the wide popularization of applications such as big data and cloud computing, data centers are playing an increasingly important role in modern society. A data center is not only the core facility for storing and processing massive amounts of data, but also an important support for providing Internet and various information services. In the context of large model training and intelligent computing centers, data centers are facing severe challenges. With the rapid development of artificial intelligence (AI) technology, large model training has become an important direction in AI research and applications. Large model training usually requires processing massive amounts of data and complex computing tasks, with extremely high demands for network bandwidth and computing resources. Traditional data center architectures are difficult to efficiently support such demands. In this context, intelligent computing centers, as high-performance computing platforms dedicated to providing support for artificial intelligence and big data computing, have emerged. Intelligent computing centers not only need to provide powerful computing capabilities, but also need to have flexible network architectures and efficient energy consumption management to support large model training and other high-performance computing tasks.

[0003] Research shows that training large models requires consuming a large amount of energy and consumes more electricity than traditional data centers. A report by OpenAI pointed out that since 2012, the power demand for AI training applications has doubled every three to four months. The AI computing power has increased by at least four hundred thousand times in the past decade. AI large models have been called "power-consuming giants". AI servers and chips are the main devices generating energy consumption. A general-purpose server only needs two 800-watt server power supplies, while an AI server requires four 1800-watt high-power power supplies, greatly increasing the power demand. It is estimated that by 2027, the electricity consumption of newly manufactured artificial intelligence devices will be equivalent to that of countries such as the Netherlands, Sweden, and Argentina.

[0004] Thus, in the context of intelligent computing centers, how to efficiently utilize network resources and reduce energy consumption has become an urgent problem to be solved.

[0005] To address the energy consumption problem, researchers have proposed various energy consumption optimization technologies. These technologies are mainly divided into two major categories: hardware and software methods. Among them, hardware methods mainly focus on minimizing the physical parameters of the data center; software methods focus on designing efficient resource allocation technologies to achieve the best balance between energy consumption and data center performance. Although these optimization technologies can reduce the energy consumption of photoelectric hybrid data centers to a certain extent, there are still many deficiencies as follows:

[0006] 1) Limitations of fixed topology: The network architecture of traditional pure electric intelligent computing centers mainly adopts a fixed topology, which cannot flexibly respond to changing communication requirements and traffic distribution. This results in low resource utilization and high energy consumption. Moreover, it is difficult for a fixed topology to adapt to dynamically changing traffic. Especially during peak periods, it may lead to local resource overload, while during low-traffic periods, some resources may be idle, causing waste.

[0007] 2) Complexity of management and control: After introducing optical switches, the management and control of the optical-electric hybrid intelligent computing center become more complex. The differences in switching modes, resource allocation complexity, transmission delay, and power consumption management between optical circuit switches (OCS) and electric packet switches (EPS) will increase the management and control complexity of the entire optical-electric hybrid data center.

[0008] 3) Insufficient real-time performance and scalability: Existing network optimization algorithms and energy consumption management strategies have deficiencies in both real-time performance and scalability. Moreover, as the network scale expands and application requirements increase, it is difficult for traditional methods to balance system scalability while ensuring real-time response. This poses higher requirements for the efficient operation and energy consumption optimization of large-scale data centers. Summary of the Invention

[0009] In view of this, embodiments of the present invention provide an energy consumption optimization method and system for a data center to eliminate or improve one or more defects existing in the prior art.

[0010] One aspect of the present invention provides an energy consumption optimization method for a data center, the method comprising the following steps:

[0011] Obtain the initial network topology, real-time traffic demand data, and historical traffic demand data of the data center;

[0012] Predict the predicted traffic demand matrix for a future time period based on the real-time traffic demand data and historical traffic demand data;

[0013] Extract features from the initial network topology to obtain an initial topology feature matrix;

[0014] Input the predicted traffic demand matrix and the initial topology feature matrix into a neural network model, so that the neural network model outputs a predicted result of the optimal topology feature matrix corresponding to the predicted traffic demand matrix. Among them, the neural network model is pre-trained by minimizing the total loss, and the total loss includes energy consumption loss. By minimizing the energy consumption loss, the difference between the energy consumption optimization prediction result corresponding to the traffic demand matrix in the dataset used to train the neural network model and the set optimal energy consumption label is less than the first threshold. The energy consumption optimization prediction result is based on the energy consumption of the optimized topology feature matrix prediction result corresponding to the traffic demand matrix output by the neural network model and the corresponding initial topology feature matrix after inputting the traffic demand matrix and the corresponding initial topology feature matrix in the dataset into the neural network model;

[0015] Reconstruct the initial network topology based on the predicted result of the optimal topology feature matrix to obtain the optimal network topology corresponding to the predicted traffic demand matrix.

[0016] In some embodiments of the present invention, the neural network model is pre-trained through the following steps:

[0017] Train a preset neural network model based on the dataset, so that the trained neural network model outputs a predicted result of the optimal topology feature matrix corresponding to the input traffic demand matrix based on the input traffic demand matrix and the corresponding initial topology feature matrix. The dataset includes multiple traffic demand matrices with their respective optimal energy consumption labels, as well as corresponding multiple initial topology feature matrices and multiple optimal topology feature matrices.

[0018] In some embodiments of the present invention, the total loss further includes a topology loss. By minimizing the topology loss, the difference between the optimized topology feature matrix prediction result and the corresponding optimal topology feature matrix is less than the second threshold. At this time, the optimized topology feature matrix prediction result output by the neural network model is the predicted result of the optimal topology feature matrix.

[0019] In some embodiments of the present invention, the network topology includes multiple nodes and multiple links connecting the nodes, and the nodes are used to represent the devices in the data center; the topology feature matrix includes the zero-load energy consumption, load rate, available bandwidth, and delay sensitivity of each node in the corresponding network topology, as well as the transmission distance, available bandwidth, and delay sensitivity of each link.

[0020] In some embodiments of the present invention, the energy consumption of the network topologies corresponding to the optimized topology feature matrix prediction result and the initial topology feature matrix is obtained based on an energy consumption model and their respective topology feature matrices. The energy consumption model includes a node energy consumption model and a link energy consumption model. The node energy consumption model is obtained based on the zero-load energy consumption and the load rate of each node, and the link energy consumption model is obtained based on the transmission distance of each link.

[0021] In some embodiments of the present invention, the traffic demand includes the traffic patterns, bandwidth requirements, and delay constraints of the data streams of each node and the links between each node in the network topology.

[0022] In some embodiments of the present invention, the method further includes:

[0023] Using a sandbox module to verify whether the optimal network topology and the corresponding energy consumption can simultaneously meet the traffic demand of the predicted traffic demand matrix and the preset energy consumption optimization goal. If so, storing the optimal network topology in a preset knowledge base; if not, recalling from the preset knowledge base a network topology that is similar to the optimal network topology and simultaneously meets the traffic demand of the predicted traffic demand matrix and the preset energy consumption optimization goal as the optimal network topology corresponding to the predicted traffic demand matrix, where the knowledge base includes multiple predefined network topologies that meet their respective traffic demands and preset energy consumption optimization goals.

[0024] Another aspect of the present invention provides an energy consumption optimization system for a data center. The system includes: a computer device, the computer device includes a processor and a memory, and computer instructions are stored in the memory. The processor is configured to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the system implements the steps of the foregoing method.

[0025] Another aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the foregoing method are implemented.

[0026] Another aspect of the present invention provides a computer program product, including computer instructions, which implement the steps of the foregoing method when executed by a processor.

[0027] The energy consumption optimization method and system for the data center of the present invention can meet the higher requirements of large-scale data centers, especially optoelectronic hybrid data centers, for efficient operation and energy consumption optimization, and achieve efficient energy consumption management and resource optimization allocation for data centers of types such as optoelectronic hybrid. By dynamically adjusting according to different traffic demands to obtain the corresponding optimal network topology that meets the goal of minimizing energy consumption, the energy consumption of the data center can be effectively reduced, the resource utilization rate of the data center can be improved, and the operation and maintenance costs can be reduced. At the same time, this method has good scalability and flexibility and can adapt to the requirements of data centers of different scales and types.

[0028] Additional advantages, objects, and features of the present invention will be partly described below, and will partly become apparent to those of ordinary skill in the art after studying the following, or can be learned from the practice of the present invention. The objects and other advantages of the present invention can be realized and obtained by the structure specifically pointed out in the specification and the drawings.

[0029] Those skilled in the art will understand that the objects and advantages that can be achieved by the present invention are not limited to the above specifically described, and the above and other objects that the present invention can achieve will be more clearly understood according to the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings described herein are used to provide a further understanding of the present invention, form a part of this application, and do not limit the present invention.

[0031] Figure 1 It is a schematic flowchart of the energy consumption optimization method for the data center in an embodiment of the present invention;

[0032] Figure 2 It is a schematic diagram of the principle of the sandbox module recalling the network topology in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] To make the objects, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in combination with the embodiments and the drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but do not limit the present invention.

[0034] Here, it should also be noted that in order to avoid obscuring the present invention due to unnecessary details, only the structures and / or processing steps closely related to the solution of the present invention are shown in the drawings, and other details less related to the present invention are omitted.

[0035] It should be emphasized that the term "comprising / including" when used herein refers to the presence of features, elements, steps, or components, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0036] Here, it should also be noted that, unless otherwise specified, the term "connection" in this text can not only refer to a direct connection, but also an indirect connection with intermediaries.

[0037] In the following, embodiments of the present invention will be described with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0038] In order to overcome the limitations of the fixed network topology in the existing energy consumption optimization methods for data centers, the complexity of data center network management and control after introducing optical switches, as well as the deficiencies in real-time performance and scalability, embodiments of the present invention propose an energy consumption optimization method and system for data centers, which can meet the higher requirements for efficient operation and energy consumption optimization in large-scale data centers, especially optical and electrical hybrid data centers, and achieve efficient energy consumption management and resource optimization allocation for data centers of types such as optical and electrical hybrid.

[0039] Figure 1 It is a schematic flowchart of the energy consumption optimization method for a data center in an embodiment of the present invention. As Figure 1 shown, the method includes the following steps:

[0040] Step S110, obtain the initial network topology, real-time traffic demand data, and historical traffic demand data of the data center.

[0041] Specifically, the data center includes optical and electrical hybrid data centers, etc. The optical and electrical hybrid data center includes various devices such as optical circuit switches (OCS) and electrical packet switches (EPS), as well as each link between the various devices. The current network topology of the data center can be obtained and analyzed through the network controller in the data center, and the current network topology is regarded as the initial network topology. The network topology of the data center includes multiple nodes and multiple edges connecting the nodes. Among them, each node is each device in the data center, and each edge is each link for connecting devices.

[0042] Step S120, predict the predicted traffic demand matrix for the future time period based on the real-time traffic demand data and the historical traffic demand data.

[0043] Specifically, the traffic demand (real-time traffic demand data, historical traffic demand data, or predicted traffic demand matrix) includes the traffic patterns, bandwidth requirements, and latency constraints of the data streams of each node and each link between nodes in the network topology. Among them, the traffic pattern includes the daily peak value and periodic fluctuation law of the data stream with spatio-temporal distribution characteristics represented by a time series, as well as the traffic volume, etc.; the bandwidth requirement includes the instantaneous bandwidth usage allocated on demand, etc. Specifically, the real-time traffic demand data and historical traffic demand data can be input into a short-term traffic prediction model, such as a Long Short-Term Memory (LSTM) model, etc., so that the model outputs the predicted traffic demand data represented in matrix form for a corresponding future period of time, that is, the predicted traffic demand matrix, to represent the expected traffic of each node and each link in the future time period. For example, it is predicted that node A transmits 1TB of traffic data to node B, etc., which can realize the prediction of the change trend of business demand in the future period of time, and the predicted traffic demand matrix can be used to guide the corresponding network topology optimization to ensure that the reconstructed network topology can timely respond to, carry, and meet the traffic demand in the changed future time period.

[0044] Step S130: Extract features from the initial network topology to obtain an initial topology feature matrix.

[0045] Specifically, the network controller can use network representation learning technology to extract the features in the initial network topology. For example, the Graph Attention Network (GAT) is used to analyze each node, each link (edge) and their relationships in the initial network topology to capture the global dependencies of each node and each link, so as to generate the feature representation of the initial network topology, that is, the initial topology feature matrix. The extracted features include the features of each node and each link in the initial network topology. Among them, the node features include the zero-load energy consumption, load rate, available bandwidth, and latency sensitivity of the node, etc., and the link features include the transmission distance, available bandwidth, and latency sensitivity of the link, etc.

[0046] Step S140: Input the predicted traffic demand matrix and the initial topology feature matrix into a neural network model, so that the neural network model outputs a predicted result of the optimal topology feature matrix corresponding to the predicted traffic demand matrix. The neural network model is pre-trained by minimizing a total loss, where the total loss includes an energy consumption loss. By minimizing the energy consumption loss, the difference between the energy consumption optimization prediction result corresponding to the traffic demand matrix in the dataset used for training the neural network model and the set energy consumption optimal label is less than a first threshold. The energy consumption optimization prediction result is obtained based on the energy consumption of the optimized topology feature matrix prediction result corresponding to the traffic demand matrix output by the neural network model and the corresponding initial topology feature matrix after inputting the traffic demand matrix and the corresponding initial topology feature matrix in the dataset into the neural network model.

[0047] In some embodiments, the neural network model is pre-trained through the following steps:

[0048] Train a preset neural network model based on the dataset, so that the trained neural network model outputs a predicted result of the optimal topology feature matrix corresponding to the input traffic demand matrix based on the input traffic demand matrix and the corresponding initial topology feature matrix. The dataset includes multiple traffic demand matrices each with its own energy consumption optimal label, as well as the corresponding multiple initial topology feature matrices and multiple optimal topology feature matrices.

[0049] Specifically, each traffic demand matrix with a corresponding energy consumption optimal label in the dataset, as well as the corresponding initial topology feature matrix and optimal topology feature matrix, can actually be historical energy consumption optimization records of the data center. For example, "Closing link L1 under a certain traffic pattern can reduce energy consumption by 20%". Among them, the traffic demand matrix, as well as the initial topology feature matrix and the optimal topology feature matrix before and after closing link L1, are all known, and "reducing energy consumption by 20%" is the corresponding energy consumption optimal label. During the training process, it is necessary to satisfy the bandwidth constraint condition that the bandwidth capacity of the network topology corresponding to the optimized topology feature matrix prediction result output by the neural network (NN) model is not lower than the bandwidth demand of the corresponding traffic demand matrix, which can ensure reasonable bandwidth allocation.

[0050] In some embodiments, the total loss further includes a topology loss. By minimizing the topology loss, the difference between the optimized topology feature matrix prediction result and the corresponding optimal topology feature matrix is less than a second threshold. At this time, the optimized topology feature matrix prediction result output by the neural network model is the predicted result of the optimal topology feature matrix.

[0051] In this embodiment, during the training process, by jointly minimizing the energy consumption loss and the topology loss, the energy consumption of the network topologies (topology feature matrices) before and after optimization is compared to obtain the energy consumption optimization prediction result, and it is continuously made to approach the preset optimal energy consumption label. This not only enables the energy consumption of the optimal topology feature matrix prediction result corresponding to the predicted traffic demand matrix output by the finally trained neural network model to meet the energy consumption minimization objective, but also makes the optimal topology feature matrix prediction result closer to the true optimal topology feature matrix corresponding to the predicted traffic demand matrix, thereby enabling the trained neural network model to have the best prediction performance for the optimal topology feature and improving the energy consumption optimization ability of the model. Among them, both the energy consumption loss and the topology loss can be calculated using the cross-entropy loss function.

[0052] In some embodiments, the energy consumption of the network topologies corresponding to the optimized topology feature matrix prediction result and the initial topology feature matrix is obtained based on the energy consumption model and their respective topology feature matrices. The energy consumption model includes a node energy consumption model and a link energy consumption model. The node energy consumption model is obtained based on the zero-load energy consumption and the load rate of each node, and the link energy consumption model is obtained based on the transmission distance of each link.

[0053] Before optimizing the energy consumption of the data center, this method pre-establishes an energy consumption model for quantifying the energy consumption situation in the entire data center network, so as to provide effective energy consumption data support and reference during the topology optimization process. Specifically, the energy consumption model of the data center is obtained through the node energy consumption model and the link energy consumption model of the data center. Among them, the node energy consumption model simplifies the calculation of node energy consumption by using the linear relationship between the average energy consumption value of the node and the load situation, and the link energy consumption model simplifies the calculation of network transmission energy consumption by using the linear relationship between the average energy consumption value of the link and the transmission distance. The expression of the energy consumption model of the data center can be as follows:

[0054]

[0055] Where E total represents the total energy consumption of the entire data center, N represents the number of nodes in the data center, represents the energy consumption of the i-th node, represents the energy consumption of the link between the i-th node and its adjacent node j, K represents the number of adjacent nodes j of the i-th node; α represents the energy consumption coefficient of the node, that is, the average energy consumption value of the node; load i represents the load rate of the i-th node; represents the zero-load energy consumption of the i-th node, which is a constant; β represents the energy consumption coefficient of the link, that is, the average energy consumption value of the link; dist i,j represents the transmission distance between the i-th node and its adjacent node j.

[0056] According to the expression of the above energy consumption model, it can be seen that the energy consumption values calculated based on different topological feature matrices generated from different traffic demand matrices are also different. This is mainly because the network topological structures corresponding to different topological feature matrices are different, that is, the nodes and links therein are also different. In the process of network topology optimization, this method combines the training of the energy consumption model and the neural network model and uses it in the training process of the neural network model, which can enable the trained neural network model to directly output the prediction result of the optimal topological feature matrix corresponding to the predicted traffic demand matrix and meet the optimal energy consumption target, thus effectively realizing the energy consumption optimization of the data center.

[0057] Step S150, reconstruct the initial network topology based on the prediction result of the optimal topological feature matrix to obtain the optimal network topology corresponding to the predicted traffic demand matrix.

[0058] In this step, the network controller reconstructs and optimizes the network topology according to the prediction result of the optimal topological feature matrix. After obtaining the optimal network topology through reconstruction, the data center operates according to the relevant configuration of the optimal network topology, and adjusts the configuration of devices and links, including optical switch status, circuit switch status, and transmission path selection, etc. For example, selectively turning off network devices such as optical switches can minimize the energy consumption of the data center while meeting the predicted traffic demand. At the same time, this method can also dynamically adjust the optimal network topology according to different demands, optimize the network resource utilization rate of the data center, and minimize energy consumption as much as possible.

[0059] In some embodiments, the method further includes the following steps:

[0060] Use the sandbox module to verify whether the optimal network topology and the corresponding energy consumption can simultaneously meet the traffic demand of the predicted traffic demand matrix and the preset energy consumption optimization target. If so, store the optimal network topology in a preset knowledge base; if not, recall from the preset knowledge base a network topology similar to the optimal network topology and simultaneously meeting the traffic demand of the predicted traffic demand matrix and the preset energy consumption optimization target as the optimal network topology corresponding to the predicted traffic demand matrix, where the knowledge base includes multiple predefined network topologies that meet their respective corresponding traffic demands and preset energy consumption optimization targets.

[0061] The energy consumption of the optimal network topology is obtained by inputting the prediction result of the optimal network topology features into the energy consumption model. To avoid the risk of unpredictable error estimation caused by the interpretability of the neural network, avoid economic losses caused by malicious attacks, and provide highly reliable network topology reconstruction, this method also designs a sandbox module. If the optimal network topology obtained in step S150 causes the delay of a certain critical link to exceed the standard, the sandbox module recalls a congestion-free network topology from the knowledge base according to the above predicted optimal topology features and configures the network to ensure that the network topology can meet the bandwidth required by the service. The steps and processes of the sandbox module recalling the network topology according to the estimated features (i.e., the predicted optimal topology features) are as Figure 2 shown. First, the sandbox module searches the knowledge base for the top k network topologies with high similarity to the estimated features as the candidate network topology set. The sandbox model performs post-verification on each network topology in the above candidate network topology set in descending order of similarity to detect whether there are congested links in the network topology. Moreover, the sandbox module verifies whether the bandwidth, delay, and energy consumption of each network topology meet the standards (i.e., whether they can simultaneously meet the traffic requirements of the predicted traffic demand matrix and the preset energy consumption optimization goal). If all can meet the standards, it stops the availability detection and recalls the network topology. At this time, the obtained network topology can ensure the requirements of the service traffic. Through the post-verification method of the sandbox module, the satisfaction of the service traffic requirements can be ensured, and highly reliable network operation and maintenance can be achieved.

[0062] Correspondingly, the present invention also provides an energy consumption optimization system for a data center. The system includes a computer device, the computer device includes a processor and a memory, the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps of the foregoing method.

[0063] The embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the foregoing method. The computer-readable storage medium may be a tangible storage medium, such as a random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium known in the technical field.

[0064] The embodiment of the present invention also provides a computer program product, including computer instructions, which implement the steps of the foregoing method when executed by a processor.

[0065] Those of ordinary skill in the art should understand that the various exemplary components, systems, and methods described in connection with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Specifically, whether to implement it in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a functional card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium or transmitted through a data signal carried in a carrier wave over a transmission medium or a communication link.

[0066] It should be clear that the present invention is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and illustrated as examples. However, the method process of the present invention is not limited to the specific steps described and illustrated. Those skilled in the art can make various changes, modifications, and additions, or change the order between steps after understanding the spirit of the present invention.

[0067] In the present invention, the features described and / or illustrated for one embodiment can be used in the same or a similar manner in one or more other embodiments, and / or combined with the features of other embodiments or replace the features of other embodiments.

[0068] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, various changes and variations can be made to the embodiments of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An energy consumption optimization method for a data center, characterized in that The method includes: Obtaining the initial network topology, real-time traffic demand data, and historical traffic demand data of the data center; Predicting the predicted traffic demand matrix for a future time period based on the real-time traffic demand data and the historical traffic demand data; Performing feature extraction on the initial network topology to obtain an initial topology feature matrix; Inputting the predicted traffic demand matrix and the initial topology feature matrix into a neural network model, so that the neural network model outputs a prediction result of the optimal topology feature matrix corresponding to the predicted traffic demand matrix, where the neural network model is pre-trained by minimizing the total loss, the total loss includes energy consumption loss, and by minimizing the energy consumption loss, the difference between the energy consumption optimization prediction result corresponding to the traffic demand matrix in the dataset used for training the neural network model and the set energy consumption optimal label is less than a first threshold, and the energy consumption optimization prediction result is based on the energy consumption of the optimized topology feature matrix prediction result corresponding to the traffic demand matrix output by the neural network model and the corresponding initial topology feature matrix after inputting the traffic demand matrix and the corresponding initial topology feature matrix in the dataset into the neural network model; Reconstructing the initial network topology based on the prediction result of the optimal topology feature matrix to obtain the optimal network topology corresponding to the predicted traffic demand matrix.

2. The method according to claim 1, wherein The neural network model is pre-trained through the following steps: Training a preset neural network model based on the dataset, so that the trained neural network model outputs a prediction result of the optimal topology feature matrix corresponding to the input traffic demand matrix based on the input traffic demand matrix and the corresponding initial topology feature matrix, and the dataset includes multiple traffic demand matrices each provided with an energy consumption optimal label, as well as corresponding multiple initial topology feature matrices and multiple optimal topology feature matrices.

3. The method according to claim 2, wherein The total loss further includes a topology loss, and by minimizing the topology loss, the difference between the optimized topology feature matrix prediction result and the corresponding optimal topology feature matrix is less than a second threshold, and at this time, the optimized topology feature matrix prediction result output by the neural network model is the prediction result of the optimal topology feature matrix.

4. The method according to claim 1, wherein The network topology includes multiple nodes and multiple links connecting the nodes, and the nodes are used to represent the devices in the data center; the topology feature matrix includes the zero-load energy consumption, load rate, available bandwidth, and delay sensitivity of each node in the corresponding network topology, as well as the transmission distance, available bandwidth, and delay sensitivity of each link.

5. The method according to claim 4, wherein The energy consumption of the network topologies corresponding to the optimized topology feature matrix prediction result and the initial topology feature matrix is obtained based on an energy consumption model and their respective topology feature matrices. The energy consumption model includes a node energy consumption model and a link energy consumption model. The node energy consumption model is based on the zero-load energy consumption and load rate of each node, and the link energy consumption model is based on the transmission distance of each link.

6. The method according to claim 1, characterized in that, The traffic demand includes the traffic pattern, bandwidth demand, and delay constraint of the data flow of each node and each link in the network topology.

7. The method according to any one of claims 1 to 6, characterized in that The method further includes: Use a sandbox module to verify whether the optimal network topology and the corresponding energy consumption can simultaneously meet the traffic demands of the predicted traffic demand matrix and the preset energy consumption optimization goal. If so, store the optimal network topology in a preset knowledge base; if not, recall from the preset knowledge base a network topology that is similar to the optimal network topology and simultaneously meets the traffic demands of the predicted traffic demand matrix and the preset energy consumption optimization goal as the optimal network topology corresponding to the predicted traffic demand matrix. Among them, the knowledge base includes multiple predefined network topologies that meet their respective corresponding traffic demands and preset energy consumption optimization goals.

8. An energy consumption optimization system for a data center, comprising a processor, a memory, and computer instructions stored on the memory, characterized in that, The processor is used to execute the computer instructions, and when the computer instructions are executed, the system implements the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer program product, comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, it implements the steps of the method according to any one of claims 1 to 7.

Images