Artificial intelligence-based coordinated scheduling method for load and energy consumption of distributed computing power center

Abstract

The present invention relates to the technical field of coordinated scheduling of load and energy consumption, and provides an artificial intelligence-based coordinated scheduling method and system for load and energy consumption of a distributed computing power center, a storage medium, and an electronic device. In the present invention, a GAT and LSTM are used to construct a computing power center node workload prediction model, a computing power center node workload energy consumption prediction model, and an energy consumption prediction model for a cooling system; the three models work cooperatively by means of a cascaded architecture; and the output of a preceding model is used as the input of a subsequent model, so as to enhance the data interaction capability between the models, thereby improving overall prediction performance. In addition, powerful language generation and knowledge reasoning capabilities of generative artificial intelligence are used to generate a flexible and efficient scheduling strategy, thereby significantly improving the flexibility and intelligence level of a system, and better adapting to complex and dynamic computing power requirement scenarios.

Description

A Distributed Computing Center Load and Energy Coordination Scheduling Method Based on Artificial Intelligence

[0001] Cross-references to related applications

[0002] This application claims priority to Chinese patent application 202510367674.3, filed on March 26, 2025, entitled “A Coordinated Scheduling Method for Load and Energy Consumption of Distributed Computing Centers Based on Artificial Intelligence”, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This invention relates to the field of load and energy consumption collaborative scheduling technology, specifically to a method, system, storage medium, and electronic device for load and energy consumption collaborative scheduling of a distributed computing center based on artificial intelligence. Background Technology

[0004] In the wave of digital transformation, distributed computing centers have become a key infrastructure driving the development of various industries. However, distributed computing centers face serious challenges during operation, including uneven workloads and excessive energy consumption.

[0005] On the one hand, the workload distribution among computing nodes is extremely uneven due to the vastly different computing power requirements of various applications and the randomness of task arrival times and durations. Some nodes may become overloaded due to excessive workloads, leading to performance degradation or even crashes. Other nodes may remain idle, resulting in wasted resources. On the other hand, the energy consumption of computing centers cannot be ignored. Computing nodes consume a large amount of electricity during operation, and the cooling system also requires significant energy to maintain its normal operating temperature. Excessive energy consumption not only increases operating costs but also puts considerable pressure on the environment. Therefore, achieving coordinated scheduling of workload and energy consumption in distributed computing centers, improving resource utilization, and reducing energy consumption have become critical issues that urgently need to be addressed.

[0006] Taking edge computing centers as an example, as a typical application scenario of distributed computing centers, their nodes are widely distributed and resources are limited. The dynamic nature of task workloads and the complexity of the network environment further exacerbate the problems of uneven workload and excessive energy consumption. For instance, in smart cities or the Industrial Internet, edge nodes need to process large amounts of localized data in real time. However, due to uneven task distribution, some nodes may be overloaded and unable to meet real-time requirements, while other nodes operate inefficiently. At the same time, the energy supply for edge devices is usually limited, and excessive energy consumption will significantly shorten equipment lifespan and increase maintenance costs.

[0007] Currently, most methods only focus on the study of computing center workload or energy consumption, while ignoring the strong correlation between computing center workload, computing center workload energy consumption and cooling system energy consumption. This makes it difficult to accurately identify energy hotspots and flow paths during energy management and optimization, thus making it impossible to make targeted improvements. Summary of the Invention

[0008] (a) Technical problems to be solved

[0009] To address the shortcomings of existing technologies, this invention provides a method, system, storage medium, and electronic device for coordinated scheduling of load and energy consumption in distributed computing centers based on artificial intelligence, which solves the technical problem of unreasonable resource allocation caused by individually scheduling the workload or energy consumption of computing centers.

[0010] (II) Technical Solution

[0011] To achieve the above objectives, the present invention provides the following technical solution:

[0012] A method for coordinated scheduling of load and energy consumption in a distributed computing center based on artificial intelligence, comprising:

[0013] Collect and preprocess historical data of node workload, historical data of workload requirements for new computing tasks, historical data of node workload energy consumption, historical data of node environment, and historical data of cooling system energy consumption.

[0014] Based on the preprocessed historical data of the node workload and the historical data of the workload requirements of the new computing power task, and dividing the first training set and the first test set, a computing power center node workload prediction model is constructed using GAT and LSTM to obtain node workload prediction data.

[0015] Based on the preprocessed historical energy consumption data of the node workload and the historical environmental data of the node, combined with the predicted data of the node workload, and divided into a second training set and a second test set, a computing center node workload energy consumption prediction model is constructed using GAT and LSTM to obtain the predicted energy consumption data of the node workload.

[0016] Based on the preprocessed historical energy consumption data of the cooling system and the historical environmental data of the node, combined with the energy consumption prediction data of the node workload, and divided into a third training set and a third test set, an energy consumption prediction model of the cooling system is constructed using LSTM.

[0017] Collect workload demand data for new computing tasks, as well as the latest historical data on node workload, workload energy consumption, node environment, and cooling system energy consumption. Then, using the computing center node workload prediction model, the computing center node workload energy consumption prediction model, and the cooling system energy consumption prediction model, predict the node's workload, workload energy consumption, and cooling system energy consumption data within a future time window.

[0018] The workload, workload energy consumption, and cooling system energy consumption data within the future time window are filled into a preset prompt information template and used as input for generative artificial intelligence to obtain the final distributed computing center workload-energy consumption collaborative scheduling scheme.

[0019] Preferably, the workload prediction model for the computing center node includes two LSTM layers and two GAT layers; wherein:

[0020] The first LSTM layer is used to receive historical node workload data at multiple historical time steps from multiple computing nodes, and pass its output features to the first GAT layer to obtain the first feature through multiple attention heads.

[0021] The second LSTM layer is used to receive historical data on the workload requirements of multiple new computing tasks, and after reshaping the data structure of its output features, the second feature is obtained.

[0022] The second GAT layer is used to receive the merging result of the first feature and the second feature, and obtain node workload prediction data for multiple historical time steps of multiple computing nodes.

[0023] Preferably, the computing center node workload energy consumption prediction model includes three LSTM layers and two GAT layers; wherein:

[0024] The third LSTM layer is used to receive node workload prediction data from multiple computing nodes at multiple historical time steps to obtain a third feature;

[0025] The fourth LSTM layer is used to receive historical energy consumption data of node workloads at multiple historical time steps from multiple computing nodes to obtain a fourth feature.

[0026] The fifth LSTM layer is used to receive historical environmental data from nodes at multiple historical time steps, and after reshaping the data structure of the output features, the fifth feature is obtained.

[0027] The third GAT layer is used to receive the combined results of the third feature, the fourth feature, and the fifth feature, and pass them to the fourth GAT layer to obtain energy consumption prediction data of node workloads at multiple historical time steps of multiple computing nodes.

[0028] Preferably, the LSTM-based energy consumption prediction model for the cooling system comprises two LSTMs; wherein:

[0029] After reshaping the data structure of the energy consumption prediction data of the node workload at multiple historical time steps of multiple computing nodes, the sixth feature is obtained; the energy consumption history data of the cooling system is used as the seventh feature, and the preprocessed node environment history data is used as the eighth feature.

[0030] The sixth LSTM layer is used to receive the combined results of the sixth feature, the seventh feature, and the eighth feature, and pass them to the seventh LSTM layer to obtain energy consumption prediction data for multiple historical time steps of the cooling system.

[0031] Preferably, the generative artificial intelligence adopts GPT-4.

[0032] Preferably, the node workload historical data includes any one or a combination of several of the following: CPU utilization, average load, memory usage, free memory, memory utilization rate, disk read / write speed, disk read / write count, disk I / O wait time, number of running processes, task queue length, CPU time occupied by each process, number of transactions processed per second, and amount of data transferred per second.

[0033] Preferably, the historical data of the workload requirements of the new computing power task includes any one or a combination of several of the following: task type, task priority, data volume, model complexity, number of users, frequency of user operations, number of instructions to be executed per second, number of floating-point operations per second, frequency of data reading and writing, and parallelism of the task.

[0034] Preferably, the historical energy consumption data of the node workload includes any one or a combination of the following: the overall energy consumption of the node within the input step, the average energy consumption of the node within the input step, the energy consumption of the CPU, the energy consumption of the memory, and the energy consumption of the storage device.

[0035] Preferably, the node environment historical data includes any one or a combination of the following: the node's ambient temperature, the computer room temperature of the computing center node, the difference between the weather temperature and the node's computer room temperature, ambient humidity, and computer room humidity.

[0036] Preferably, the energy consumption history data of the cooling system includes any one or a combination of the overall energy consumption of the node cooling system within the input step and the average energy consumption of the node cooling system within the input step.

[0037] A distributed computing center load and energy consumption collaborative scheduling system based on artificial intelligence, comprising:

[0038] The data acquisition and preprocessing module is used to collect and preprocess historical data of node workload, historical data of workload requirements of new computing tasks, historical data of node workload energy consumption, historical data of node environment, and historical data of cooling system energy consumption.

[0039] The model training module is used to construct a computing center node workload prediction model based on the preprocessed historical data of the node workload and the historical data of the workload requirements of the new computing power task, and to divide the first training set and the first test set, using GAT and LSTM to obtain node workload prediction data.

[0040] Based on the preprocessed historical energy consumption data of the node workload and the historical environmental data of the node, combined with the predicted data of the node workload, and divided into a second training set and a second test set, a computing center node workload energy consumption prediction model is constructed using GAT and LSTM to obtain the predicted energy consumption data of the node workload.

[0041] And based on the preprocessed historical energy consumption data of the cooling system and the historical environmental data of the node, combined with the energy consumption prediction data of the node workload, and divided into a third training set and a third test set, an energy consumption prediction model of the cooling system is constructed using LSTM.

[0042] The data prediction module is used to collect workload demand data for new computing power tasks, as well as the latest historical data on node workload, workload energy consumption, node environment, and cooling system energy consumption. Through the computing power center node workload prediction model, the computing power center node workload energy consumption prediction model, and the cooling system energy consumption prediction model, the module predicts the node's workload, workload energy consumption, and cooling system energy consumption data within a future time window.

[0043] The scheme generation module is used to fill the workload, workload energy consumption, and cooling system energy consumption data within the future time window into a preset prompt information template, and use it as input for generative artificial intelligence to obtain the final distributed computing center workload-energy consumption collaborative scheduling scheme.

[0044] A storage medium storing a computer program for the coordinated scheduling of load and energy consumption in a distributed computing center based on artificial intelligence, wherein the computer program causes a computer to execute the coordinated scheduling method for load and energy consumption in a distributed computing center as described above.

[0045] An electronic device, comprising:

[0046] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing the distributed computing center load and energy consumption collaborative scheduling method as described above.

[0047] (III) Beneficial Effects

[0048] This invention provides a method, system, storage medium, and electronic device for collaborative scheduling of load and energy consumption in a distributed computing center based on artificial intelligence. Compared with existing technologies, it has the following advantages:

[0049] In this invention, GAT and LSTM are used to construct a workload prediction model for the computing center node, a workload energy consumption prediction model for the computing center node, and an energy consumption prediction model for the cooling system, respectively. These three models work collaboratively to establish a quantitative relationship between workload and energy consumption, enabling data sharing and interaction. Furthermore, leveraging the powerful language generation and knowledge reasoning capabilities of generative artificial intelligence, flexible and efficient scheduling strategies are generated, significantly improving the system's flexibility and intelligence, and better adapting to complex and ever-changing computing power demands. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 is a block diagram of a distributed computing center load and energy consumption collaborative scheduling method based on artificial intelligence provided by an embodiment of the present invention;

[0052] Figure 2 is a flowchart of a distributed computing center load and energy consumption collaborative scheduling method based on artificial intelligence according to an embodiment of the present invention;

[0053] Figure 3 is a flowchart of the construction process of a computing center node workload prediction model according to an embodiment of the present invention;

[0054] Figure 4 is a flowchart of the construction of a computing center node workload energy consumption prediction model according to an embodiment of the present invention;

[0055] Figure 5 is a flowchart of the construction of an energy consumption prediction model for a cooling system according to an embodiment of the present invention;

[0056] Figure 6 is a structural block diagram of a distributed computing center load and energy consumption collaborative scheduling system based on artificial intelligence provided in an embodiment of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] This application provides a method, system, storage medium, and electronic device for coordinated scheduling of load and energy consumption in a distributed computing center based on artificial intelligence, which solves the technical problem of unreasonable resource allocation caused by scheduling the workload or energy consumption of the computing center separately.

[0059] The technical solution in this application embodiment is to solve the above-mentioned technical problems, and the key points are as follows:

[0060] 1. Graph structure modeling: GAT can model the nodes of the distributed computing center and their interrelationships as a graph structure, and capture the complex relationships between nodes through the graph attention mechanism.

[0061] 2. Node Feature Fusion: In the energy consumption prediction module, GAT can effectively fuse features such as node workload and node environment, and can also comprehensively consider the workload of adjacent nodes and the communication link status between nodes, thereby more accurately predicting the node workload and energy consumption trend.

[0062] 3. Workload-Energy Consumption Correlation Analysis: The node workload prediction module, the node workload energy consumption prediction module, and the cooling system energy consumption prediction module work together to establish a quantitative relationship between workload and energy consumption through in-depth analysis of workload and energy consumption data.

[0063] 4. Intelligent scheduling scheme generation: Based on the powerful language generation and knowledge reasoning capabilities of generative artificial intelligence, the optimal workload-energy consumption collaborative scheduling scheme is generated by constructing prompts based on workload and energy consumption prediction results.

[0064] 5. Adaptive Adjustment: Generative artificial intelligence can dynamically adjust the scheduling scheme according to the real-time workload and energy consumption of the computing center to adapt to changing needs and ensure that the computing center is always in a highly efficient and energy-saving operating state.

[0065] In addition, the following explanations of terms used in the embodiments of this invention are provided:

[0066] 1) Graph Attention Network (GAT) is a neural network model based on graph structure, designed to address some limitations of Graph Convolutional Network (GCN).

[0067] 2) Long short-term memory (LSTM) is a special type of recurrent neural network (RNN) mainly designed to solve the gradient vanishing and gradient explosion problems during long sequence training.

[0068] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0069] Example 1:

[0070] As shown in Figure 1, this embodiment of the invention provides a method for coordinated scheduling of load and energy consumption in a distributed computing center based on artificial intelligence, including:

[0071] S1. Collect and preprocess historical data of node workload, historical data of workload requirements for new computing tasks, historical data of node workload energy consumption, historical data of node environment, and historical data of cooling system energy consumption.

[0072] S2. Based on the preprocessed historical data of the node workload and the historical data of the workload requirements of the new computing power task, and dividing the first training set and the first test set, a computing power center node workload prediction model is constructed using GAT and LSTM to obtain node workload prediction data.

[0073] S3. Based on the preprocessed historical energy consumption data of the node workload and the historical environmental data of the node, combined with the predicted data of the node workload, and dividing into a second training set and a second test set, a computing center node workload energy consumption prediction model is constructed using GAT and LSTM to obtain the predicted energy consumption data of the node workload.

[0074] S4. Based on the preprocessed historical energy consumption data of the cooling system and the historical environmental data of the node, combined with the energy consumption prediction data of the node workload, and dividing the system into a third training set and a third test set, an energy consumption prediction model for the cooling system is constructed using LSTM.

[0075] S5. Collect the workload requirement data of new computing power tasks, as well as the latest historical data of node workload, workload energy consumption, node environment, and cooling system energy consumption. Through the computing power center node workload prediction model, the computing power center node workload energy consumption prediction model, and the cooling system energy consumption prediction model, predict the node's workload, workload energy consumption, and cooling system energy consumption data within the future time window.

[0076] S6. Fill the workload, workload energy consumption, and cooling system energy consumption data within the future time window into the preset prompt information template, and use them as input for generative artificial intelligence to obtain the final distributed computing center workload-energy consumption collaborative scheduling scheme.

[0077] In this embodiment of the invention, GAT and LSTM are used to construct a workload prediction model for the computing center node, a workload energy consumption prediction model for the computing center node, and an energy consumption prediction model for the cooling system, respectively. These three models work collaboratively to establish a quantitative relationship between workload and energy consumption, enabling data sharing and interaction. Furthermore, leveraging the powerful language generation and knowledge reasoning capabilities of generative artificial intelligence, flexible and efficient scheduling strategies are generated, significantly improving the system's flexibility and intelligence, and better adapting to complex and ever-changing computing power demand scenarios.

[0078] Figure 2 shows a flowchart of a distributed computing center load and energy consumption collaborative scheduling method based on artificial intelligence.

[0079] The following section will detail each step of the above scheme with reference to Figure 2:

[0080] In step S1, historical data of node workload, historical data of workload requirements of new computing tasks, historical data of node workload energy consumption, historical data of node environment, and historical data of cooling system energy consumption are collected and preprocessed.

[0081] This step involves collecting and preprocessing historical data on node workloads, historical data on workload requirements for new computing tasks, historical data on node workload energy consumption, historical data on node environment, and historical data on cooling system energy consumption.

[0082] For example:

[0083] The node workload history data includes CPU utilization, average load, memory usage, free memory, memory utilization rate, disk read / write speed, disk read / write count, disk I / O wait time, number of running processes, task queue length, CPU time used by each process, number of transactions processed per second, and amount of data transferred per second.

[0084] The historical data on the workload requirements of the new computing power tasks include task type, task priority, data volume, model complexity, number of users, frequency of user operations, number of instructions to be executed per second, number of floating-point operations per second, frequency of data reading and writing, and task parallelism.

[0085] The historical energy consumption data of the node workload includes the overall energy consumption of the node within the input step size, the average energy consumption of the node within the input step size, the energy consumption of the CPU, the energy consumption of the memory, and the energy consumption of the storage device.

[0086] The historical environmental data of the nodes includes the ambient temperature of the nodes, the temperature of the computing center node's computer room, the difference between the weather temperature and the node's computer room temperature, the ambient humidity, and the computer room humidity.

[0087] The energy consumption history data of the cooling system includes the overall energy consumption of the node cooling system within the input step size, and the average energy consumption of the node cooling system within the input step size.

[0088] Furthermore, the preprocessing process includes data cleaning of outlier data, numericalization of non-numerical data, and normalization of data to remove noise and outliers, thereby improving data consistency and comparability.

[0089] In step S2, based on the preprocessed historical data of the node workload and the historical data of the workload requirements of the new computing power task, and dividing the first training set and the first test set, a computing power center node workload prediction model is constructed using GAT and LSTM to obtain node workload prediction data.

[0090] This step includes: dividing the dataset based on the preprocessed historical workload data of nodes and the historical workload demand data of new computing tasks; constructing a computing center node workload prediction model; and training and optimizing the model.

[0091] For example, in this step, the dataset is randomly divided into a first training set and a first test set in an 8:2 ratio.

[0092] As shown in Figure 3, the workload prediction model for the computing center node includes two LSTM layers and two GAT layers; wherein:

[0093] The first LSTM layer is used to receive historical node workload data at multiple historical time steps from multiple computing nodes, and pass its output features to the first GAT layer to obtain the first feature through multiple attention heads.

[0094] The second LSTM layer is used to receive historical data on the workload requirements of multiple new computing tasks, and after reshaping the data structure of its output features, the second feature is obtained.

[0095] The second GAT layer is used to receive the merging result of the first feature and the second feature, and obtain node workload prediction data for multiple historical time steps of multiple computing nodes.

[0096] Specifically:

[0097] The parameters involved in the model construction process are: the number of computing nodes (Nodes_Num), Time_Step (the time step of the output), Nodes_Features_1 (the number of features of the load data of the computing nodes), and Batch_Size (the batch size of the samples).

[0098] The first LSTM layer takes as input the historical time-step workload data of the original task from Nodes_Num computing nodes. Leveraging the powerful time-series processing capabilities of the LSTM network, this layer can deeply mine and extract the time-series features from the input historical workload data. The number of hidden layer neurons in this layer is set to Nodes_Features_1, the input shape is [Nodes_Num × Time_Step, Nodes_Features_1], and the output structure is the features of the first LSTM layer with the structure [Nodes_Num × Time_Step, Nodes_Features_1]. This processing method fully utilizes the temporal correlation in the historical workload data, providing strong support for subsequent analysis and prediction.

[0099] The first GAT layer: This layer primarily handles automatically capturing the historical load dependencies between computational nodes. The first GAT layer has 1 attention head and 1 hidden channel. Its input data is in the shape of [Nodes_Num × Time_Step, Nodes_Features_1]. By setting up an attention mechanism with 1 attention head, this layer can perform deep feature mining on the input data and transform it into an output of the shape [Nodes_Num × Time_Step, Hidden_Channels_1 × Attention_Head_1]. This multi-attention head design allows the model to capture data features from different perspectives, significantly improving the comprehensiveness and accuracy of feature extraction.

[0100] The second LSTM layer focuses on processing the load data of new computing tasks, assuming a maximum limit of Task_Limit for the load requirements of new computing tasks. This layer has Nodes_Features_1 hidden neurons. It receives load data from Task_Limit new tasks, with the task load features represented by Nodes_Features_1. If the number of computing tasks is less than Task_Limit, it is padded with 0s. The input and output shapes of this layer are both [Task_Limit, Nodes_Features_1]. This process effectively extracts the features of the new task load data, laying the foundation for subsequent fusion with historical load data and further analysis.

[0101] Data Structure Restructuring and Feature Merging: To fuse the historical workload data processed by the first GAT layer and the new task workload data processed by the second LSTM layer, the output of the second LSTM layer needs to be restructured into [Nodes_Num×Time_Step, Nodes_Features_1]. The restructured data is then merged with the output of the first GAT layer (structure [Nodes_Num×Time_Step, Hidden_Channels_1×Attention_Head_1]) to achieve the following sample dimension merging: the fused structure is [Nodes_Num×Time_Step, Hidden_Channels_1×Attention_Head_1+Nodes_Features_1].

[0102] The second GAT layer: The main function of this layer is to deeply fuse the historical workload data processed by the first GAT layer and the new computing workload data processed by the second LSTM layer, and to complement the dynamic workload feature information. The second GAT layer has one attention head and the number of hidden layer neurons is set to Nodes_Features_1. The input data structure of this layer is [Nodes_Num×Time_Step,Hidden_Channels_1×Attention_Head_1+Nodes_Features_1], and the output structure is [Nodes_Num×Time_Step,Nodes_Features_1], which is the data structure of the predicted workload data.

[0103] Furthermore, this step also optimizes the hyperparameters of the computing center node workload prediction model through optimization algorithms such as grid search. After obtaining the optimal computing center node workload prediction model, its output is saved for subsequent prediction of node workload.

[0104] Training process:

[0105] Before starting formal training, determine the hyperparameters that need to be optimized by grid search, such as the learning rate and the number of attention mechanism heads, and define the range of values ​​for these hyperparameters.

[0106] In each training epoch, set the model to training mode `model_1.train()`. For different hyperparameter combinations, perform the following operations:

[0107] Iterate through the training data loader train_loader_1. For each batch of sample data, first clear the optimizer gradient to zero using optimizer.zero_grad().

[0108] Input the input data data_1.x and the corresponding computing node data_1.edge into the computing center node workload prediction model to obtain the output out_1. Calculate the loss using the measured load data data_1.y corresponding to the input data and out_1.

[0109] Backpropagation calculates the gradient loss.backward() and updates the model parameters optimizer.step(). The loss for each batch is accumulated, and the average loss for each training epoch under this hyperparameter combination is recorded.

[0110] Evaluation process:

[0111] After training all hyperparameter combinations is complete, set the model to evaluation mode using model_1.eval().

[0112] For each combination of hyperparameters tested during training, the test data loader test_loader_1 is traversed. For each batch of data test_data_1, the data is input into the model to obtain the output test_out_1, and the loss test_loss is calculated.

[0113] Accumulate the losses from all test batches and record the average loss on the test set for each hyperparameter combination. By comparing the average loss on the test set under different hyperparameter combinations, identify the hyperparameter combination that optimizes the model's performance, thereby improving the workload prediction model. Finally, print the average loss on the test set under the optimal hyperparameter combination.

[0114] In step S3, based on the preprocessed historical energy consumption data of the node workload and the historical data of the node environment, combined with the predicted data of the node workload, and dividing into a second training set and a second test set, a computing center node workload energy consumption prediction model is constructed using GAT and LSTM to obtain the predicted energy consumption data of the node workload.

[0115] This step includes: dividing the dataset based on historical energy consumption data of node workload, historical node environment data, and predicted node workload data; constructing a computing center node workload energy consumption prediction model; and training and optimizing the model.

[0116] It is important to note that this step combines the node load prediction values ​​from the computing center node workload prediction model obtained in the previous step with the node's historical load energy consumption and historical environmental data, and then inputs them into the computing center node load energy consumption prediction model. This is primarily to improve the accuracy of load energy consumption prediction through multi-dimensional data fusion. The node load prediction values ​​provide the dynamic trend of future load, while historical load energy consumption data reflects the energy consumption patterns of nodes under different loads, and environmental data captures the impact of external conditions on energy consumption. Combining these data allows for a more comprehensive analysis of the complex relationship between workload and load energy consumption, especially the direct impact of environmental factors on equipment heat dissipation efficiency and energy consumption.

[0117] For example, this step randomly divides the dataset into a second training set and a second test set in an 8:2 ratio. Furthermore, a model structure is constructed based on the input data format. The data structures of the second training set and the second test set are transformed according to the input format of the workload energy consumption prediction model, and the energy consumption prediction model for the computing center node workload is fitted and optimized until the mean absolute error of the loss function no longer decreases significantly.

[0118] As shown in Figure 4, the computing center node workload energy consumption prediction model includes three LSTM layers and two GAT layers; wherein:

[0119] The third LSTM layer is used to receive node workload prediction data from multiple computing nodes at multiple historical time steps to obtain a third feature;

[0120] The fourth LSTM layer is used to receive historical energy consumption data of node workloads at multiple historical time steps from multiple computing nodes to obtain a fourth feature.

[0121] The fifth LSTM layer is used to receive historical environmental data from nodes at multiple historical time steps, and after reshaping the data structure of the output features, the fifth feature is obtained.

[0122] The third GAT layer is used to receive the combined results of the third feature, the fourth feature, and the fifth feature, and pass them to the fourth GAT layer to obtain energy consumption prediction data of node workloads at multiple historical time steps of multiple computing nodes.

[0123] Specifically:

[0124] The parameters involved in the model construction process are: the number of computing nodes (Nodes_Num), Time_Step (representing the output time step), Nodes_Features_1 (representing the number of features of the predicted load data of the computing center node workload prediction model), Nodes_Features_2 (representing the number of features of the historical load energy consumption data of the computing nodes), Environment_Features_3 (representing the number of features of the historical environment data of the computing nodes), and the batch size of the samples is Batch_Size.

[0125] The third LSTM layer: The number of hidden layer neurons in this layer is Nodes_Features_1. Its input is the Time_Step predicted load data of Nodes_Num computing nodes output from the computing power center node workload prediction model. The input shape is [Nodes_Num×Time_Step,Nodes_Features_1]. This layer can perform in-depth mining and extraction of time series features in the input predicted load data, and the output structure is features of [Nodes_Num×Time_Step,Nodes_Features_1].

[0126] The fourth LSTM layer has 2 hidden neurons. Its input consists of Time_Step historical load energy consumption data from Nodes_Num computing nodes. The input shape is [Nodes_Num×Time_Step,Nodes_Features_2]. This layer can perform in-depth mining and extraction of time series features from the input historical load energy consumption data, and the output structure is features of [Nodes_Num×Time_Step,Nodes_Features_2].

[0127] The fifth LSTM layer: The number of hidden neurons in this layer is Environment_Features_3×Nodes_Num. Its input is Time_Step historical environment data, and the input shape is [Time_Step, Environment_Features_3]. The output structure is the features of [Time_Step, Environment_Features_3×Nodes_Num].

[0128] Data structure reshaping: The output features of the fifth LSTM layer are restructured into [Nodes_Num×Time_Step,Environment_Features_3].

[0129] Feature merging: The output features of the restructured fifth LSTM layer are merged with the output features of the third LSTM layer (structure [Nodes_Num×Time_Step,Nodes_Features_1]) and the output features of the fourth LSTM layer (structure [Nodes_Num×Time_Step,Nodes_Features_2]) in terms of sample dimensions. The fused structure is [Nodes_Num×Time_Step,Nodes_Features_1+Nodes_Features_2+Environment_Features_3].

[0130] The third GAT layer: The main function of this layer is to deeply fuse and dynamically complement the node prediction load data processed by the third LSTM layer, the node historical load energy consumption data processed by the fourth LSTM layer, and the historical environment data processed by the fifth LSTM layer. The third GAT layer has 3 attention heads and 3 hidden channel neurons. Its input data is in the shape of [Nodes_Num×Time_Step, Nodes_Features_1+Nodes_Features_2+Environment_Features_3]. By setting the attention mechanism with 3 attention heads, this layer can perform deep feature mining on the input data and transform it into an output of the shape [Nodes_Num×Time_Step, Hidden_Channels_3×Attention_Head_3].

[0131] Fourth GAT Layer: The fourth GAT layer has 1 attention head and 2 hidden layer neurons. The input data structure of this layer is [Nodes_Num×Time_Step,Hidden_Channels_3×Attention_Head_3], and the output structure is [Nodes_Num×Time_Step,Nodes_Features_2]. This is the predicted load energy consumption data output by the computing center node workload energy consumption prediction model.

[0132] Furthermore, this step also optimizes the hyperparameters of the computing center node workload energy consumption prediction model through optimization algorithms such as grid search. After obtaining the optimal computing center node workload energy consumption prediction model, its output is saved for subsequent prediction of node workload energy consumption.

[0133] Training process:

[0134] Before starting formal training, determine the hyperparameters that need to be optimized by grid search, such as the learning rate and the number of attention mechanism heads, and define the range of values ​​for these hyperparameters.

[0135] In each training epoch, set the model to training mode `model_2.train()`. For different hyperparameter combinations, perform the following operations:

[0136] Iterate through the training data loader train_loader_2. For each batch of sample data, first clear the optimizer gradient to zero using optimizer.zero_grad().

[0137] Input the input data data_2.x and the corresponding computing node data_2.edge into the computing center node workload energy consumption prediction model to obtain the output out_2. Calculate the loss using the measured load data data_2.y corresponding to the input data and out_2.

[0138] Backpropagation calculates the gradient loss.backward() and updates the model parameters optimizer.step(). The loss for each batch is accumulated, and the average loss for each training epoch under this hyperparameter combination is recorded.

[0139] Evaluation process:

[0140] After training all hyperparameter combinations is complete, set the model to evaluation mode using model_2.eval().

[0141] For each combination of hyperparameters tested during training, the test data loader test_loader_2 is traversed. For each batch of data test_data_2, the data is input into the model to obtain the output test_out_2, and the loss test_loss is calculated.

[0142] Accumulate the losses from all test batches and record the average loss of each hyperparameter combination on the test set. By comparing the average loss of the test set under different hyperparameter combinations, identify the hyperparameter combination that optimizes the model's performance, thereby optimizing the workload energy consumption prediction model for the computing center nodes. Finally, print the average loss of the test set under the optimal hyperparameter combination.

[0143] In step S4, based on the preprocessed historical energy consumption data of the cooling system and the historical environmental data of the node, combined with the energy consumption prediction data of the node workload, and dividing the system into a third training set and a third test set, an energy consumption prediction model for the cooling system is constructed using LSTM.

[0144] This step includes: dividing the dataset based on the preprocessed historical energy consumption data of the cooling system, historical data of the node environment, and energy consumption prediction data of the node workload; constructing an energy consumption prediction model for the computing center cooling system; and training and optimizing the model.

[0145] It's important to note that there's often a correlation between load energy consumption and cooling system energy consumption. Increased load energy consumption often leads to increased equipment heat generation, requiring the cooling system to increase its cooling capacity, thus raising cooling system energy consumption. Environmental data can help correct load energy consumption and cooling energy consumption predictions, avoiding biases caused by relying solely on historical cooling energy consumption and predicted load energy consumption. A two-layer LSTM can learn from historical and predicted data to uncover this causal relationship, providing a basis for subsequent system adjustments.

[0146] For example, the dataset is randomly divided into a third training set and a third test set in an 8:2 ratio. Furthermore, the data structures of the third training set and the third test set are transformed according to the format of the input data for model construction, and the time step of the input data is set. The energy consumption prediction model for the computing center cooling system is then fitted based on the transformed training set until the mean absolute error of the loss function no longer decreases significantly.

[0147] As shown in Figure 5, the LSTM-based energy consumption prediction model for the cooling system comprises two LSTMs; wherein:

[0148] After reshaping the data structure of the energy consumption prediction data of the node workload at multiple historical time steps of multiple computing nodes, the sixth feature is obtained; the energy consumption history data of the cooling system is used as the seventh feature, and the preprocessed node environment history data is used as the eighth feature.

[0149] The sixth LSTM layer is used to receive the combined results of the sixth feature, the seventh feature, and the eighth feature, and pass them to the seventh LSTM layer to obtain energy consumption prediction data for multiple historical time steps of the cooling system.

[0150] Specifically:

[0151] Data structure reshaping: The predicted load energy consumption data of the Nodes_Num computing nodes output by the computing center node workload energy consumption prediction model is transformed into [Time_Step, Nodes_Num×Time_Step, Nodes_Features_2] with the input shape [Time_Step, Nodes_Num×Nodes_Features_2].

[0152] Feature merging: The predicted load energy consumption data after reshaping, the historical cooling system energy consumption data with the structure [Time_Step, Cooling_Features_3], and the historical environmental data with the structure [Time_Step, Environment_Features_3] are merged in terms of sample dimensions. The merged data structure is [Time_Step, Nodes_Num×Nodes_Features_2+Cooling_Features_3+Environment_Features_3].

[0153] The sixth LSTM layer: This layer extracts the coupled features of predicted load energy consumption data, historical cooling system energy consumption data, and historical environmental data. The number of hidden layer neurons in this layer is Nodes_Features_2 + Cooling_Features_3 + Environment_Features_3. The merged data from the previous step is input into the sixth LSTM layer. The input structure of this layer is [Time_Step, Nodes_Num × Nodes_Features_2 + Cooling_Features_3 + Environment_Features_3], and the output structure is [Time_Step, Nodes_Features_2 + Cooling_Features_3 + Environment_Features_3].

[0154] The seventh LSTM layer: This layer further mines deeper and more complex feature patterns and long-term dependencies in the data. The number of hidden layer neurons in this layer is Cooling_Features_3. Its input structure is [Time_Step, Nodes_Features_2+Cooling_Features_3+Environment_Features_3], and its output structure is [Time_Step, Cooling_Features_3], which is the data structure for the predicted cooling system energy consumption.

[0155] Furthermore, this step also optimizes the hyperparameters of the computing center cooling system energy consumption prediction model through optimization algorithms such as grid search. After obtaining the optimal computing center cooling system energy consumption prediction model, its output is saved for subsequent prediction of the cooling system energy consumption of nodes.

[0156] Training process:

[0157] Before starting formal training, determine the hyperparameters that need to be optimized by grid search, such as learning rate and number of neurons in hidden layer, and define the range of values ​​for these hyperparameters.

[0158] In each training epoch, set the model to training mode `model_3.train()`. For different hyperparameter combinations, perform the following operations:

[0159] Iterate through the training data loader train_loader_3. For each batch of sample data, first clear the optimizer gradient to zero using optimizer.zero_grad().

[0160] Input the input data data_3.x and the correspondence between computing nodes data_3.edge into the cooling system energy consumption prediction model of the computing center to obtain the output out_3. Calculate the loss using the historical measured cooling system energy consumption data data_3.y and out_3 corresponding to the input data.

[0161] Backpropagation calculates the gradient loss.backward() and updates the model parameters optimizer.step(). The loss for each batch is accumulated, and the average loss for each training epoch under this hyperparameter combination is recorded.

[0162] Evaluation process:

[0163] After training all hyperparameter combinations is complete, set the model to evaluation mode using model_3.eval().

[0164] For each combination of hyperparameters tested during training, the test data loader test_loader_3 is traversed. For each batch of data test_data_3, the data is input into the model to obtain the output test_out_3, and the loss test_loss is calculated.

[0165] Accumulate the losses from all test batches and record the average loss of each hyperparameter combination on the test set. By comparing the average loss of the test set under different hyperparameter combinations, identify the hyperparameter combination that optimizes the model's performance, thereby improving the energy consumption prediction model for the computing center's cooling system. Finally, print the average loss of the test set under the optimal hyperparameter combination.

[0166] In step S5, the workload requirement data of new computing power tasks, as well as the latest historical data of node workload, workload energy consumption, node environment, and cooling system energy consumption are collected. Through the computing power center node workload prediction model, the computing power center node workload energy consumption prediction model, and the cooling system energy consumption prediction model, the workload, workload energy consumption, and cooling system energy consumption data of the node in the future time window are predicted.

[0167] In step S6, the workload, workload energy consumption, and cooling system energy consumption data within the future time window are filled into a preset prompt information template and used as input for generative artificial intelligence to obtain the final distributed computing center workload-energy consumption collaborative scheduling scheme.

[0168] This step integrates the prediction results from the three aforementioned prediction models—the latest predicted data for node workload, the latest predicted data for node workload energy consumption, and the latest predicted data for node cooling system energy consumption—to construct a prompt message. This prompt message is then input into generative artificial intelligence (GPT-4, for example) to generate a distributed computing center workload-energy consumption collaborative scheduling scheme. Based on this scheme, the workload and energy consumption of the distributed computing center are then collaboratively scheduled.

[0169] For example, a feasible prompt message template can be constructed as follows:

[0170] Based on the task requirements, construct a prompt message template for inputting GPT-4, providing a data interface for the information to be filled in. An example of the prompt message template is as follows:

[0171] "The following is relevant forecast data for the distributed computing center. Please generate a workload-energy consumption collaborative scheduling scheme:"

[0172] Latest predicted node workload data: "{nodes_load}"

[0173] Latest predicted energy consumption data for node workloads: "{nodes_energy_consumption}"

[0174] Latest predicted energy consumption data for cooling systems:

[0175] ″′{cooling_system_energy_consumption}″′

[0176] The current computing center has a new task: "{nodes_calculation_task}"

[0177] Based on the information above, please provide a specific workload-energy consumption coordinated scheduling scheme, including which nodes the tasks will be assigned to, whether the node operating parameters need to be adjusted, and whether the cooling system's operating strategy needs to be adjusted.

[0178] Therefore, the process of constructing the prompt information in this step is as follows:

[0179] Integrate the computing power requirement data of new computing tasks, the latest predicted data of node workload, the latest predicted data of node workload energy consumption, and the latest predicted data of cooling system energy consumption to construct a prompt message. Populate the prompt message template with the integrated prediction data, and the corresponding relationship is as follows:

[0180] Computing power requirement data for new computing tasks: nodes_calculation_task

[0181] Latest predicted node workload data: nodes_load

[0182] Latest predicted energy consumption data for node workloads: nodes_energy_consumption

[0183] Latest predicted energy consumption data for cooling systems:

[0184] cooling_system_energy_consumption.

[0185] After determining the aforementioned prompt information, it is used as input to the generative artificial intelligence to obtain the final distributed computing center workload-energy consumption collaborative scheduling scheme. Specifically, the generative artificial intelligence response includes: encapsulating a function of the OpenAI interface, inputting the prompt information (after constructing a template based on the prompts) into the generative artificial intelligence model to obtain the distributed computing center workload-energy consumption collaborative scheduling scheme.

[0186] Thus, this embodiment of the invention completes the entire process of the distributed computing center load and energy consumption collaborative scheduling method based on artificial intelligence.

[0187] Example 2:

[0188] As shown in Figure 6, this embodiment of the invention provides a distributed computing center load and energy consumption collaborative scheduling system based on artificial intelligence, comprising:

[0189] The data acquisition and preprocessing module is used to collect and preprocess historical data of node workload, historical data of workload requirements of new computing tasks, historical data of node workload energy consumption, historical data of node environment, and historical data of cooling system energy consumption.

[0190] The model training module is used to construct a computing center node workload prediction model based on the preprocessed historical data of the node workload and the historical data of the workload requirements of the new computing power task, and to divide the first training set and the first test set, using GAT and LSTM to obtain node workload prediction data.

[0191] Based on the preprocessed historical energy consumption data of the node workload and the historical environmental data of the node, combined with the predicted data of the node workload, and divided into a second training set and a second test set, a computing center node workload energy consumption prediction model is constructed using GAT and LSTM to obtain the predicted energy consumption data of the node workload.

[0192] And based on the preprocessed historical energy consumption data of the cooling system and the historical environmental data of the node, combined with the energy consumption prediction data of the node workload, and divided into a third training set and a third test set, an energy consumption prediction model of the cooling system is constructed using LSTM.

[0193] The data prediction module is used to collect workload demand data for new computing power tasks, as well as the latest historical data on node workload, workload energy consumption, node environment, and cooling system energy consumption. Through the computing power center node workload prediction model, the computing power center node workload energy consumption prediction model, and the cooling system energy consumption prediction model, the module predicts the node's workload, workload energy consumption, and cooling system energy consumption data within a future time window.

[0194] The scheme generation module is used to fill the workload, workload energy consumption, and cooling system energy consumption data within the future time window into a preset prompt information template, and use it as input for generative artificial intelligence to obtain the final distributed computing center workload-energy consumption collaborative scheduling scheme.

[0195] Example 3:

[0196] This invention provides a storage medium storing a computer program for the coordinated scheduling of load and energy consumption in a distributed computing center based on artificial intelligence. The computer program enables a computer to execute the coordinated scheduling method for load and energy consumption in a distributed computing center as described in Embodiment 1.

[0197] Example 4:

[0198] This invention provides an electronic device, comprising:

[0199] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing the distributed computing center load and energy consumption collaborative scheduling method as described in Example 1.

[0200] It is understood that the AI-based distributed computing center load and energy consumption collaborative scheduling system, storage medium and electronic device provided in the embodiments of the present invention correspond to the AI-based distributed computing center load and energy consumption collaborative scheduling method provided in the embodiments of the present invention. The explanation, examples and beneficial effects of the relevant contents can be referred to the corresponding parts of the distributed computing center load and energy consumption collaborative scheduling method, and will not be repeated here.

[0201] In summary, compared with existing technologies, it has the following beneficial effects:

[0202] 1. Improve the accuracy and adaptability of workload scheduling: By leveraging GAT to model the nodes and their interrelationships in the distributed computing center as a graph structure, and using a graph attention mechanism to capture complex relationships between nodes and the diversity of tasks, combined with the intelligent decision-making capabilities of generative artificial intelligence, scheduling strategies can be dynamically adjusted based on real-time workload conditions and task characteristics, responding promptly and effectively to dynamic changes in workload and achieving precise workload balancing.

[0203] 2. Reduce energy management costs and overcome technical bottlenecks: Through software innovation, an energy consumption prediction model is built using GAT to accurately predict the energy consumption of node computing tasks and cooling systems. Based on these accurate energy consumption predictions, combined with a workload-energy consumption collaborative scheduling scheme generated by generative artificial intelligence, the operating status and task allocation of nodes are optimized without relying on a large number of hardware upgrades, and energy consumption is reasonably controlled. This overcomes the technical bottlenecks at the hardware level and achieves a significant reduction in energy consumption at a lower cost.

[0204] 3. Achieve effective coordination between workload and energy consumption: Through the collaborative work of the node workload prediction module, the node workload energy consumption prediction module, and the cooling system energy consumption prediction module, a quantitative relationship between workload and energy consumption is established, enabling data sharing and interaction.

[0205] 4. Enhance flexibility and intelligence: Taking into account various complex factors such as the network structure of the distributed computing center, the computing power of node devices, and the impact of environmental factors on workload, the system leverages the powerful language generation and knowledge reasoning capabilities of generative artificial intelligence to generate flexible and efficient scheduling strategies, significantly improving the system's flexibility and intelligence, and better adapting to complex and ever-changing computing power demand scenarios.

[0206] 5. Eliminate energy management lag and achieve optimal overall energy consumption control: It can proactively analyze workload changes, predict energy consumption trends in advance, and avoid prolonged task response times due to delayed adjustments. The introduced generative artificial intelligence can quickly adapt to dynamic changes in the workload and energy consumption of the computing center. When faced with sudden large-scale computing tasks, it can rapidly generate reasonable scheduling schemes to ensure timely task processing while controlling energy consumption growth.

[0207] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0208] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for coordinated scheduling of load and energy consumption in a distributed computing center based on artificial intelligence, characterized in that, include: Collect and preprocess historical data of node workload, historical data of workload requirements for new computing tasks, historical data of node workload energy consumption, historical data of node environment, and historical data of cooling system energy consumption. Based on the preprocessed historical data of the node workload and the historical data of the workload requirements of the new computing power task, and dividing the first training set and the first test set, a computing power center node workload prediction model is constructed using GAT and LSTM to obtain node workload prediction data. Based on the preprocessed historical energy consumption data of the node workload and the historical environmental data of the node, combined with the predicted data of the node workload, and divided into a second training set and a second test set, a computing center node workload energy consumption prediction model is constructed using GAT and LSTM to obtain the predicted energy consumption data of the node workload. Based on the preprocessed historical energy consumption data of the cooling system and the historical environmental data of the node, combined with the energy consumption prediction data of the node workload, and divided into a third training set and a third test set, an energy consumption prediction model of the cooling system is constructed using LSTM. Collect workload demand data for new computing tasks, as well as the latest historical data on node workload, workload energy consumption, node environment, and cooling system energy consumption. Then, using the computing center node workload prediction model, the computing center node workload energy consumption prediction model, and the cooling system energy consumption prediction model, predict the node's workload, workload energy consumption, and cooling system energy consumption data within a future time window. The workload, workload energy consumption, and cooling system energy consumption data within the future time window are filled into a preset prompt information template and used as input for generative artificial intelligence to obtain the final distributed computing center workload-energy consumption collaborative scheduling scheme.

2. The distributed computing center load and energy consumption collaborative scheduling method as described in claim 1, characterized in that, The workload prediction model for the computing center node includes two LSTM layers and two GAT layers; wherein: The first LSTM layer is used to receive historical node workload data at multiple historical time steps from multiple computing nodes, and pass its output features to the first GAT layer to obtain the first feature through multiple attention heads. The second LSTM layer is used to receive historical data on the workload requirements of multiple new computing tasks, and after reshaping the data structure of its output features, the second feature is obtained. The second GAT layer is used to receive the merging result of the first feature and the second feature, and obtain node workload prediction data for multiple historical time steps of multiple computing nodes.

3. The distributed computing center load and energy consumption collaborative scheduling method as described in claim 2, characterized in that, The computing center node workload energy consumption prediction model includes three LSTM layers and two GAT layers; wherein: The third LSTM layer is used to receive node workload prediction data from multiple computing nodes at multiple historical time steps to obtain a third feature; The fourth LSTM layer is used to receive historical energy consumption data of node workloads at multiple historical time steps from multiple computing nodes to obtain a fourth feature. The fifth LSTM layer is used to receive historical environmental data from nodes at multiple historical time steps, and after reshaping the data structure of the output features, the fifth feature is obtained. The third GAT layer is used to receive the combined results of the third feature, the fourth feature, and the fifth feature, and pass them to the fourth GAT layer to obtain energy consumption prediction data of node workloads at multiple historical time steps of multiple computing nodes.

4. The distributed computing center load and energy consumption collaborative scheduling method as described in claim 3, characterized in that, The LSTM-based energy consumption prediction model for the cooling system comprises two LSTMs; wherein: After reshaping the data structure of the energy consumption prediction data of the node workload at multiple historical time steps of multiple computing nodes, the sixth feature is obtained; the energy consumption history data of the cooling system is used as the seventh feature, and the preprocessed node environment history data is used as the eighth feature. The sixth LSTM layer is used to receive the combined results of the sixth feature, the seventh feature, and the eighth feature, and pass them to the seventh LSTM layer to obtain energy consumption prediction data for multiple historical time steps of the cooling system.

5. The distributed computing center load and energy consumption collaborative scheduling method as described in claim 1, characterized in that, The generative artificial intelligence used is GPT-4.

6. The distributed computing center load and energy consumption collaborative scheduling method as described in any one of claims 1 to 5, characterized in that, The node workload history data includes any one or a combination of the following: CPU utilization, average load, memory usage, free memory, memory utilization, disk read / write speed, disk read / write count, disk I / O wait time, number of running processes, task queue length, CPU time used by each process, number of transactions processed per second, and amount of data transferred per second. and / or The historical data on the workload requirements of the new computing power task includes any one or a combination of the following: task type, task priority, data volume, model complexity, number of users, frequency of user operations, number of instructions to be executed per second, number of floating-point operations per second, frequency of data reading and writing, and parallelism of the task. and / or The historical energy consumption data of the node workload includes any one or a combination of the following: the overall energy consumption of the node within the input step, the average energy consumption of the node within the input step, the energy consumption of the CPU, the energy consumption of the memory, and the energy consumption of the storage device. and / or The node environment historical data includes any one or a combination of the following: node ambient temperature, computing center node computer room temperature, the difference between weather temperature and node computer room temperature, ambient humidity, and computer room humidity. and / or The energy consumption history data of the cooling system includes any one or a combination of several of the following: the overall energy consumption of the node cooling system within the input step length, and the average energy consumption of the node cooling system within the input step length.

7. A distributed computing center load and energy consumption collaborative scheduling system based on artificial intelligence, characterized in that, include: The data acquisition and preprocessing module is used to collect and preprocess historical data of node workload, historical data of workload requirements of new computing tasks, historical data of node workload energy consumption, historical data of node environment, and historical data of cooling system energy consumption. The model training module is used to construct a computing center node workload prediction model based on the preprocessed historical data of the node workload and the historical data of the workload requirements of the new computing power task, and to divide the first training set and the first test set, using GAT and LSTM to obtain node workload prediction data. Based on the preprocessed historical energy consumption data of the node workload and the historical environmental data of the node, combined with the predicted data of the node workload, and divided into a second training set and a second test set, a computing center node workload energy consumption prediction model is constructed using GAT and LSTM to obtain the predicted energy consumption data of the node workload. And based on the preprocessed historical energy consumption data of the cooling system and the historical environmental data of the node, combined with the energy consumption prediction data of the node workload, and divided into a third training set and a third test set, an energy consumption prediction model of the cooling system is constructed using LSTM. The data prediction module is used to collect workload demand data for new computing power tasks, as well as the latest historical data on node workload, workload energy consumption, node environment, and cooling system energy consumption. Through the computing power center node workload prediction model, the computing power center node workload energy consumption prediction model, and the cooling system energy consumption prediction model, the module predicts the node's workload, workload energy consumption, and cooling system energy consumption data within a future time window. The scheme generation module is used to fill the workload, workload energy consumption, and cooling system energy consumption data within the future time window into a preset prompt information template, and use it as input for generative artificial intelligence to obtain the final distributed computing center workload-energy consumption collaborative scheduling scheme.

8. A storage medium, characterized in that, It stores a computer program for the coordinated scheduling of load and energy consumption of a distributed computing center based on artificial intelligence, wherein the computer program causes the computer to execute the coordinated scheduling method for load and energy consumption of a distributed computing center as described in any one of claims 1 to 6.

9. An electronic device, characterized in that, include: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing the distributed computing center load and energy consumption collaborative scheduling method as described in any one of claims 1 to 6.

Images