Data center carbon emission optimization method based on dynamic load balancing, storage medium and equipment
By optimizing data center load balancing using time series models and genetic algorithms, the problems of lag and resource mismatch in carbon emission optimization in traditional methods are solved, and real-time carbon emission optimization and load balancing of data centers are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-18
- Publication Date
- 2026-04-10
AI Technical Summary
Traditional data center carbon emission optimization methods cannot respond to dynamic changes in load and carbon emission factors in real time, leading to task backlog and resource mismatch, and ignoring global carbon emission targets, resulting in increased carbon emissions.
A dynamic load balancing approach is adopted, which uses a time series model to predict the load range of the data center and combines it with a genetic algorithm to determine the optimal scheduling strategy, optimize the allocation of data tasks, and dynamically adjust the scheduling strategy with the goal of global carbon emission optimization.
It enables real-time optimization of carbon emissions in data centers, avoids overload and performance degradation, ensures load balance, and improves the carbon emission optimization effect of data centers.
Smart Images
Figure CN121833170A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of cloud computing resource scheduling, and particularly relates to a data center carbon emission optimization method based on dynamic load balancing, a storage medium and equipment. BACKGROUND
[0002] With the popularization of cloud computing, AI and 5G technologies, data centers consume more and more energy, leading to gradually increasing carbon emissions. In order to cope with environmental pressure, data centers, as a key emission reduction field, need to reduce carbon footprints through optimization to reduce carbon emissions. For example, when an enterprise uses data center resources of multiple cloud service providers at the same time, dynamic task allocation is needed to reduce the overall carbon emissions of the data center.
[0003] However, traditional data center carbon emission optimization methods rely on static scheduling algorithms such as round robin and weighted minimum connection number, which cannot respond to the dynamic changes of data center load and carbon emission factors in real time, resulting in task accumulation during high carbon periods. At the same time, existing data center carbon emission optimization methods only consider the energy consumption of the data center itself, ignoring the global carbon emissions of the data center, thereby causing deviations in the target amount of carbon emissions to be optimized, which may cause resource mismatch and increase in carbon emissions. In addition, the load prediction part and the scheduling decision part in the traditional data center carbon emission optimization method run independently, and the predicted load is not embedded in the scheduling decision, resulting in scheduling lag. SUMMARY
[0004] In view of the problems in the prior art, the application provides a data center carbon emission optimization method based on dynamic load balancing, a storage medium and equipment, which optimizes the scheduling strategy of data tasks based on the predicted load interval of the data center, and further improves the effect of data center carbon emission optimization.
[0005] To achieve the above technical purposes, the application adopts the following technical solutions:
[0006] A data center carbon emission optimization method based on dynamic load balancing comprises the following steps: Step S1: Number all data centers, and obtain the historical load, real-time carbon emission coefficient and real-time resource state of each data center; Step S2: respectively predict the load interval of each data center by using a time series model for the historical load and real-time resource state of each data center; Step S3: according to the real-time carbon emission coefficient of each data center and the load interval of the corresponding data center, set several scheduling strategies for data tasks, take global carbon emission optimization as the target, determine the optimal scheduling strategy based on a genetic algorithm, and distribute data tasks to the corresponding data center for processing according to the optimal scheduling strategy.
[0007] Further, step S2 comprises the following sub-steps: Step S2.1: extracting periodic features and sliding window statistical features from the historical load of each data center; Step S2.2: splicing the extracted periodic features and sliding window statistical features and real-time resource status to obtain a spliced feature sequence, inputting the input layer of the Transformer model, and adding position encoding to the spliced feature sequence; Step S2.3: inputting the spliced feature sequence with position encoding into the Encoder layer of the Transformer model, capturing long-range dependencies between different time steps and features from the spliced feature sequence with position encoding through self-attention mechanism, and outputting a hidden state sequence; Step S2.4: taking the hidden state of the last time step of the hidden state sequence as input to the output layer of the Transformer model, and outputting the load interval of the corresponding data center.
[0008] Further, step S3 comprises the following sub-steps: Step S3.1: determining the computing amount and delay tolerance of each sub-task in the data task; Step S3.2: setting constraint conditions of the data task scheduling strategy according to the delay tolerance of each sub-task and the predicted load interval of each data center; Step S3.3: initializing a plurality of scheduling strategies according to the set constraint conditions, mapping each scheduling strategy to a chromosome of a genetic algorithm, encoding the gene position on the chromosome to represent the data center number allocated to each sub-data task in the scheduling strategy, and setting the maximum number of iterations of the genetic algorithm; Step S3.4: combining each chromosome with the real-time carbon emission coefficient of each data center, the load interval of the data center, and the computing amount of each sub-task to calculate the fitness function of global carbon emission optimization; Step S3.5: retaining the chromosome with the smallest fitness function, and updating the remaining chromosomes through low-carbon preference crossover and carbon-sensitive mutation, wherein the updated chromosomes all need to meet the constraint conditions of the data task scheduling strategy; Step S3.6: repeating steps S3.4-S3.5 until the maximum number of iterations is reached, taking the chromosome with the smallest fitness function as the optimal scheduling strategy, and distributing the data task to the corresponding data center for processing according to the optimal scheduling strategy.
[0009] Further, the constraint conditions of the data task scheduling strategy comprise: determining the load interval constraint of the data center according to the predicted load interval of the data center:
[0010] wherein, denotes the real-time load of the data center numbered denotes the number of sub-tasks in the data task, denotes the index of denotes the load required for the thsub-task to be allocated to the data center numbered denotes the scheduling strategy of the thsub-task, if the thsub-task is allocated to the data center numbered , otherwise, ; denotes the minimum value of the predicted load interval of the data center numbered denotes the maximum value of the predicted load interval of the data center numbered ; determines the delay constraint of each sub-task according to the delay tolerance of the sub-task:
[0011] wherein, denotes the completion time of the thsub-task in the data task, denotes the delay tolerance of the thsub-task in the data task.
[0012] Further, the calculation process of the fitness function of the global carbon emission optimization is:
[0013] wherein, denotes the fitness function of the global carbon emission optimization, denotes the number of all data centers, denotes the index of denotes the number of sub-tasks in the data task denotes the index of denotes the load required for the thsub-task to be allocated to the data center numbered denotes the scheduling strategy of the thsub-task, if the thsub-task is allocated to the data center numbered , otherwise, ; Indicates the first The computational cost of each subtask Indicates the number is The real-time carbon emission coefficient of the data center. Indicates the number is The predictive computing power of the data center , Indicates the number is Minimum of the predicted load range for the data center Indicates the number is The maximum value of the predicted load range for the data center. Indicates the number is The real-time load of the data center Indicates the number is The data center's full-load computing power.
[0014] Furthermore, the process of updating the remaining chromosomes in step S3.5 through low-carbon preference crossover and carbon-sensitive mutation is as follows: The carbon emissions for each gene locus are calculated based on the computational cost of the subtasks at each gene locus on the chromosome and the real-time carbon emission coefficient of the corresponding data center. The carbon emissions of each gene locus are normalized according to the minimum-maximum carbon emissions of all chromosome gene loci to obtain the carbon emission ratio of each gene locus. The lower the carbon emission ratio of a gene locus on a chromosome, the more likely the corresponding gene locus is to undergo crossover. The higher the carbon emission ratio of a gene locus on a chromosome, the more likely the corresponding gene locus is to mutate.
[0015] Furthermore, the calculation process for the carbon emissions of each gene locus is as follows:
[0016] in, In a certain scheduling strategy, the first... Carbon emissions of each sub-task Indicates the first The computational cost of each subtask Indicates the number is The real-time carbon emission coefficient of the data center. Indicates the number is The predictive computing power of the data center , Indicates the number is Minimum of the predicted load range for the data center Indicates the number is The maximum value of the predicted load range for the data center. Indicates the number is real-time load of the data center, representing the full-load computing power of the data center numbered
[0017] Further, the method further comprises: obtaining the load of each data center after the data task is dispatched to the corresponding data center for processing according to the optimal scheduling strategy, and triggering parameter updating of the time series model if the load of the data center exceeds the predicted load interval of the corresponding data center.
[0018] Further, the application further provides a computer readable storage medium storing a computer program, wherein the computer program enables a computer to execute the data center carbon emission optimization method based on dynamic load balancing.
[0019] Further, the application further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the data center carbon emission optimization method based on dynamic load balancing when executing the computer program.
[0020] Compared with the prior art, the application has the following beneficial effects: The data center carbon emission optimization method based on dynamic load balancing analyzes the historical load mode of each data center through a time series model, and predicts the future load of the corresponding data center in combination with the real-time resource state, so that the predicted load of the data center is considered in the scheduling strategy of the data task, and the scheduling strategy can be adjusted in real time according to the change of the load of the data center, thereby migrating part of the sub-tasks to a data center with low carbon emission and low load in advance, avoiding performance degradation or excessive carbon emission caused by overload; at the same time, the application takes global carbon emission optimization as the goal, and ensures that the data task allocation meets the low-carbon requirement and avoids local overload by coordinating the resources and carbon emission of all data centers, so as to ensure load balancing and make the scheduling strategy of the data task globally optimal, thereby further improving the effect of data center carbon emission optimization. BRIEF DESCRIPTION OF DRAWINGS
[0021] Figure 1 is a flowchart of the data center carbon emission optimization method based on dynamic load balancing of the application; Figure 2 is a flowchart of the optimal scheduling strategy determination based on the genetic algorithm of the application. DETAILED DESCRIPTION
[0022] The technical solutions of the application will be further explained in combination with the drawings.
[0023] As Figure 1 A flowchart of the data center carbon emission optimization method based on dynamic load balancing of the application, the data center carbon emission optimization method comprising the following steps: Step S1: Number all data centers, obtain the historical load, real-time carbon emission coefficient and real-time resource state of each data center, wherein the historical load of each data center includes CPU usage, memory occupancy, disk I / O and network bandwidth, and the real-time resource state of each data center includes current running virtual machine data, container quantity and application load condition.
[0024] Step S2: The historical load and real-time resource state of each data center are respectively predicted by using a time series model to predict the load interval of the corresponding data center, the long-term use rule of the data center is reflected through the historical load, the instantaneous capacity of the data center is reflected through the real-time resource state, and a more accurate load interval prediction is generated by combining the long-term use rule and the instantaneous capacity of the data center, thereby significantly improving the accuracy of the data center load interval prediction; comprising the following sub-steps: Step S2.1: Extract periodic features and sliding window statistical features from the historical load of each data center, so that the Transformer model can learn the long-term rule of the data center load and avoid misjudgment caused by short-term noise; the sliding window statistical features include mean, variance, maximum and minimum to reflect the short-term volatility of the load, which can enhance the sensitivity of the Transformer model to abnormal load; Step S2.2: The real-time resource state provides instantaneous capacity information of the data center, which can avoid the Transformer model from over-relying on historical information, the extracted periodic features and sliding window statistical features and the real-time resource state are spliced to obtain a spliced feature sequence, and the input layer of the Transformer model is inputted, and position encoding is added to the spliced feature sequence; Step S2.3: The spliced feature sequence with position encoding is inputted into the Encoder layer of the Transformer model, the long-range dependency relationship between different time steps and features is captured from the spliced feature sequence with position encoding through the self-attention mechanism, the hidden state sequence is outputted, the dependency relationship between different time steps is directly calculated through the self-attention mechanism without relying on the recursive structure, the long-range dependency can be effectively learned, at the same time, the position encoding adds a unique code to each time step, the time sequence information is preserved, and the prediction error caused by sequence disorder is avoided; Step S2.4: The hidden state of the last time step of the hidden state sequence is inputted into the output layer of the Transformer model, and the load interval of the corresponding data center is outputted.
[0025] Step S3: According to the real-time carbon emission coefficient of each data center and the load interval of the corresponding data center, a plurality of scheduling strategies of the data task are set, the global carbon emission optimization is taken as the target, the optimal scheduling strategy is determined based on the genetic algorithm, the data task is distributed to the corresponding data center for processing according to the optimal scheduling strategy, the predicted load of the data center is considered in the scheduling strategy of the data task, so that the scheduling strategy can be adjusted in real time with the change of the load of the data center, thereby migrating part of the subtasks to the data center with low carbon and low load in advance to avoid performance degradation or carbon emission surge caused by overload; at the same time, the present application takes the global carbon emission optimization as the target, and through the overall planning of the resources and carbon emission of all data centers, it ensures that the data task allocation meets the low carbon requirement and avoids local overload, ensures load balancing, so that the scheduling strategy of the data task is globally optimal, and the effect of carbon emission optimization of the data center is further improved.
[0026] As Figure 2 The optimal scheduling strategy determination process of the present application based on the genetic algorithm includes the following sub-steps: Step S3.1: Determine the calculation amount and delay tolerance of each subtask in the data task; Step S3.2: Set the constraint condition of the data task scheduling strategy according to the delay tolerance of each subtask and the predicted load interval of each data center; Step S3.3: Initialize a plurality of scheduling strategies according to the set constraint condition, map each scheduling strategy to a chromosome of the genetic algorithm, the gene site coded on the chromosome represents the data center number allocated to each sub-data task in the scheduling strategy, and the maximum iteration number of the genetic algorithm is set; Step S3.4: Calculate the fitness function of global carbon emission optimization by combining each chromosome with the real-time carbon emission coefficient of each data center, the load interval of the data center and the calculation amount of each subtask; Step S3.5: Keep the chromosome with the minimum fitness function, and update the remaining chromosomes through low-carbon preference crossover and carbon-sensitive mutation, and strengthen the update of the chromosome gene site in the low-carbon direction, and the updated chromosomes all need to meet the constraint condition of the data task scheduling strategy; Step S3.6: Repeat step S3.4-Step S3.5 until the maximum iteration number is reached, take the chromosome with the minimum fitness function as the optimal scheduling strategy, and distribute the data task to the corresponding data center for processing according to the optimal scheduling strategy.
[0027] In one technical solution of the present application, the constraint condition of the data task scheduling strategy includes: To avoid the increase of data task delay and the decrease of throughput caused by data center overload, the load interval constraint of the data center is determined according to the predicted load interval of the data center, the data center with high load risk is excluded in advance, the data task is ensured to run in the environment with sufficient resources, and meanwhile, the resource idling of the data center with low load can be avoided. The determination process of the load interval constraint of the data center in the application is as follows:
[0028] wherein, represents the real-time load of the data center numbered as represents the number of subtasks in the data task, represents the index of represents the load required for the subtask numbered as to be allocated to the data center numbered as represents the scheduling strategy of the subtask numbered as If the subtask numbered as is allocated to the data center numbered as , , otherwise, ; represents the minimum value of the predicted load interval of the data center numbered as represents the maximum value of the predicted load interval of the data center numbered as ; The delay constraint of the subtask is determined according to the delay tolerance of each subtask, so that each subtask is allocated to the data center with sufficient resources and low delay, and the delay exceeding the standard caused by resource competition is avoided. The determination process of the delay constraint of the subtask in the application is as follows:
[0029] wherein, represents the completion time of the subtask numbered as in the data task, represents the delay tolerance of the subtask numbered as in the data task.
[0030] In one technical solution of the application, the fitness function of global carbon emission optimization can accurately reflect the carbon emission intensity of different data centers at different times by introducing the real-time carbon emission coefficient of the data center and the predicted load interval, and the calculation process of the fitness function of global carbon emission optimization is as follows:
[0031] wherein, represents the fitness function of global carbon emission optimization, represents the number of all data centers, represents an index of, represents the number of subtasks in a data task represents an index of, represents the load required for the subtask to be allocated to the data center numbered represents the scheduling strategy of the subtask, if the subtask is allocated to the data center numbered , otherwise, ; represents the computation amount of the subtask, represents the real-time carbon emission coefficient of the data center numbered represents the predicted computing power of the data center numbered , represents the minimum value of the predicted load interval of the data center numbered represents the maximum value of the predicted load interval of the data center numbered represents the real-time load of the data center numbered represents the full-load computing power of the data center numbered
[0032] In one technical solution of the present application, the process of updating the remaining chromosomes through low-carbon preference crossover and carbon-sensitive mutation is as follows: According to the computation amount of each subtask on each gene site of the chromosome and the real-time carbon emission coefficient of the corresponding data center, the carbon emission amount of each gene site is calculated:
[0033] wherein, represents the carbon emission amount of the subtask in a certain scheduling strategy, represents the computation amount of the subtask, represents the real-time carbon emission coefficient of the data center numbered represents the predicted computing power of the data center numbered , represents the minimum value of the predicted load interval of the data center numbered a minimum of a load interval predicted by a data center of the data task a maximum of a load interval predicted by a data center of the data task numbered a real-time load of a data center numbered a full-load computing power of a data center numbered
[0034] The carbon emission of each gene locus is normalized according to the minimum-maximum value of the carbon emission of all chromosome loci, to obtain the carbon emission proportion of each gene locus, and to realize accurate carbon emission of each subtask under the corresponding scheduling strategy; The lower the carbon emission proportion of the gene locus on the chromosome, the more likely the corresponding gene locus is to undergo crossing, which can avoid destroying the discovered low-carbon sub-scheduling strategy and accelerate convergence to the global low-carbon optimal solution; The higher the carbon emission proportion of the gene locus on the chromosome, the more likely the corresponding gene locus is to undergo mutation, which can jump out of the local optimum and enhance the global search ability of the genetic algorithm.
[0035] In one technical solution of the present application, the load of each data center is obtained after the data task is dispatched to the corresponding data center for processing under the optimal scheduling strategy, and if the load of the data center exceeds the corresponding data center predicted load interval, the parameter update of the time series model is triggered, the time series model is optimized through closed-loop feedback, the error of the predicted load interval is reduced, the correct data center load interval is provided for the scheduling strategy optimization of the data task, and the precision of the data center carbon emission optimization is improved.
[0036] In one technical solution of the present application, a computer readable storage medium is provided, which stores a computer program, and the computer program enables a computer to execute the data center carbon emission optimization method based on dynamic load balancing.
[0037] In one technical solution of the present application, an electronic device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the data center carbon emission optimization method based on dynamic load balancing is realized.
[0038] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium which can contain or store programs for use by or in connection with an instruction execution system, apparatus or device. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any suitable combination of the above. More specific examples of the computer storage medium can include one or more wires, portable computer disks, hard drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.
[0039] Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed in the present application can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.
[0040] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiment. Any technical solution falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled in the art, some improvements and refinements without departing from the principles of the present application shall be considered within the protection scope of the present application.
Claims
1. A method for data center carbon emission optimization based on dynamic load balancing, characterized in that, The method comprises the following steps: Step S1: numbering all data centers, obtaining historical load, real-time carbon emission coefficient and real-time resource state of each data center; Step S2: respectively using time series model to predict the load interval of each data center according to the historical load and real-time resource state of each data center; Step S3: setting several scheduling strategies of data task according to the real-time carbon emission coefficient of each data center and the load interval of the corresponding data center, taking global carbon emission optimization as the goal, determining the optimal scheduling strategy based on genetic algorithm, and distributing the data task to the corresponding data center for processing according to the optimal scheduling strategy.
2. The method of claim 1, wherein, Step S2 comprises the following sub-steps: Step S2.1: extracting periodic features and sliding window statistical features from the historical load of each data center; Step S2.2: splicing the extracted periodic features and sliding window statistical features and real-time resource state to obtain a spliced feature sequence, inputting the input layer of the Transformer model, and adding position encoding to the spliced feature sequence; Step S2.3: inputting the spliced feature sequence with position encoding into the Encoder layer of the Transformer model, capturing the long-range dependence between different time steps and features from the spliced feature sequence with position encoding through the self-attention mechanism, and outputting a hidden state sequence; Step S2.4: taking the hidden state of the last time step of the hidden state sequence as input to the output layer of the Transformer model, and outputting the load interval of the corresponding data center.
3. The method of claim 1, wherein, Step S3 comprises the following sub-steps: Step S3.1: determining the computing amount and delay tolerance of each sub-task in the data task; Step S3.2: setting constraint conditions of the data task scheduling strategy according to the delay tolerance of each sub-task and the predicted load interval of each data center; Step S3.3: initializing several scheduling strategies according to the set constraint conditions, mapping each scheduling strategy to a chromosome of genetic algorithm, encoding the gene bits on the chromosome to represent the data center number allocated to each sub-data task in the scheduling strategy, and setting the maximum iteration number of the genetic algorithm; Step S3.4: combining each chromosome with the real-time carbon emission coefficient of each data center, the load interval of the data center, and the computing amount of each sub-task to calculate the fitness function of global carbon emission optimization; Step S3.5: retaining the chromosome with the smallest fitness function, and updating the remaining chromosomes through low-carbon preference crossover and carbon-sensitive mutation, wherein the updated chromosomes all need to meet the constraint conditions of the data task scheduling strategy; Step S3.6: repeating steps S3.4-S3.5 until the maximum iteration number is reached, taking the chromosome with the smallest fitness function as the optimal scheduling strategy, and distributing the data task to the corresponding data center for processing according to the optimal scheduling strategy.
4. The method of claim 3, wherein, The constraint conditions of the data task scheduling strategy comprise: determining the load interval constraint of the data center according to the predicted load interval of the data center; wherein, denotes the real-time load of the data center numbered , denotes the number of sub-tasks in the data task, denotes the index of , denotes the load required for the th sub-task to be assigned to the data center numbered , denotes the scheduling strategy of the th sub-task, if the th sub-task is assigned to the data center numbered , , otherwise, ; denotes the minimum value of the predicted load interval of the data center numbered , denotes the maximum value of the predicted load interval of the data center numbered ; determining the delay constraint of the sub-task according to the delay tolerance of each sub-task: wherein, denotes the completion time of a subtask in a data task, denotes the delay tolerance of a subtask in a data task, denotes the delay tolerance of a subtask in a data task, 5. The method of claim 3, wherein, The calculation process of the global carbon emission optimization fitness function is as follows: wherein, represents a fitness function of global carbon emission optimization, represents the number of all data centers, represents an index of represents the number of subtasks in a data task represents an index of represents the load required for the th subtask to be allocated to the data center numbered represents the scheduling strategy of the th subtask, if the th subtask is allocated to the data center numbered , otherwise, ; represents the computation amount of the th subtask, represents the real-time carbon emission coefficient of the data center numbered represents the predicted computing power of the data center numbered , represents the minimum value of the predicted load interval of the data center numbered represents the maximum value of the predicted load interval of the data center numbered represents the real-time load of the data center numbered represents the full-load computing power of the data center numbered . 6. The method of claim 3, wherein, The process of updating the rest of the chromosomes in step S3.5 by low-carbon bias crossover and carbon-sensitive mutation is as follows: According to the calculation amount of each gene site of the chromosome and the real-time carbon emission coefficient of the corresponding data center, the carbon emission of each gene site is calculated; The carbon emission of each gene site is normalized according to the minimum-maximum value of the carbon emission of all gene sites of the chromosome to obtain the carbon emission proportion of each gene site; The lower the carbon emission proportion of the gene site on the chromosome, the more likely the corresponding gene site is to occur crossover; The higher the carbon emission proportion of the gene site on the chromosome, the more likely the corresponding gene site is to occur mutation.
7. The method of claim 6, wherein, The calculation process of the carbon emission of each gene site is as follows: in, In a certain scheduling strategy, the first... Carbon emissions of each sub-task Indicates the first The computational cost of each subtask Indicates the number is The real-time carbon emission coefficient of the data center. Indicates the number is The predictive computing power of the data center , Indicates the number is Minimum of the predicted load range for the data center Indicates the number is The maximum value of the predicted load range for the data center. Indicates the number is The real-time load of the data center Indicates the number is The data center's full-load computing power.
8. The method of claim 1, wherein, Further comprising: After the data task is sent to the corresponding data center for processing according to the optimal scheduling strategy, the load of each data center is obtained, and if the load of the data center exceeds the predicted load interval of the corresponding data center, the parameter update of the time series model is triggered.
9. A computer readable storage medium storing a computer program, characterized in that, The computer program enables the computer to execute the data center carbon emission optimization method based on dynamic load balancing according to any one of claims 1-8.
10. An electronic device, comprising: Including: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the data center carbon emission optimization method based on dynamic load balancing according to any one of claims 1-8.