Heterogeneous computing power network power supply scheduling method based on carbon intensity and related device

Abstract

The invention discloses a heterogeneous computing power network power supply scheduling method based on carbon intensity and a related device, and the method comprises the steps: obtaining the carbon intensity at a current moment in a preset time period; receiving power use state information from the heterogeneous computing power network nodes in real time; generating a newest power supply scheduling scheme of the heterogeneous computing power network according to the carbon intensity and the power use state information, so that the carbon emission generated by executing the newest power supply scheduling scheme is less than or equal to a carbon emission budget; the latest power supply scheduling scheme comprises the power supply quantity allocated to each heterogeneous computing power network node for deep learning operation; and according to the latest power supply scheduling scheme, distributing power supply quantity for the deep learning operation of each heterogeneous computing power network node. Through algorithm optimization and resource allocation, carbon emission minimization in the power supply process is realized, system efficiency and environmental protection are improved, low-carbon transformation of the society is promoted, and the method has important significance and value.

Description

Technical Field

[0001] The present invention relates to the field of intelligent power dispatching optimization of heterogeneous computing power networks, and in particular to a carbon intensity-based heterogeneous computing power network power supply dispatching method and related devices. Background Art

[0002] As the world pays more and more attention to climate change and environmental pollution, the use of clean energy has become one of the important ways to resolve the contradiction between energy and environment. In this context, the sustainable development and operation efficiency of the power system have become important issues. The traditional power system power supply dispatch is mainly based on the optimization of load demand and power generation cost, but often ignores the impact on the environment.

[0003] Carbon intensity is one of the important indicators to measure the impact of energy systems on the environment. It indicates the amount of carbon dioxide emissions generated per unit of energy. Reducing carbon intensity means reducing the negative impact on the environment, which is consistent with the development goal of clean energy. Therefore, taking carbon intensity into account in power supply dispatching has become an important issue.

[0004] Traditional power system dispatching mainly relies on centralized algorithms and systems. This approach can meet the needs of the system to a certain extent, but it has limitations when dealing with complex environments and uncertain factors. With the development of information technology, the concepts of distributed computing and heterogeneous computing have gradually been introduced into power system dispatching to cope with the dynamics and uncertainty of the system.

[0005] A heterogeneous computing network is a network that combines different types of computing resources. It can use both traditional centralized computing resources and distributed computing resources to achieve efficient dispatch of the power system. In a heterogeneous computing network, centralized computing resources can handle large-scale complex problems, while distributed computing resources can handle tasks with high real-time requirements, thereby improving the system's response speed and flexibility.

[0006] The present invention is proposed based on the above background and technical requirements, and aims to take carbon intensity into account in the power supply dispatch of the power system, and realize efficient dispatch and environmentally friendly operation of the power system by combining the advantages of heterogeneous computing. The proposal of the present invention will provide important support for the sustainable development of the power system and the utilization of clean energy, and has important theoretical and practical significance. Summary of the invention

[0007] In view of this, the present invention provides a carbon intensity-based heterogeneous computing power network power supply scheduling method and related devices, which fully utilize the advantages of heterogeneous computing power networks, realize dynamic scheduling of power systems and minimization of carbon emissions through carbon intensity optimization algorithms and heterogeneous computing resources, and improve the environmental friendliness and operation efficiency of power systems.

[0008] In order to achieve the above tasks, the present invention provides a method for power supply scheduling of a heterogeneous computing power network based on carbon intensity, comprising:

[0009] Get the carbon intensity at the current moment in the preset time period.

[0010] Receive power usage status information from heterogeneous computing network nodes in real time.

[0011] Generate a latest power supply scheduling plan for the heterogeneous computing network according to the carbon intensity and the power usage status information, so that the carbon emissions generated by executing the latest power supply scheduling plan are less than or equal to the carbon emission budget; the latest power supply scheduling plan includes the power supply allocated to each deep learning job of the heterogeneous computing network node;

[0012] According to the latest power supply scheduling scheme, power is allocated to the deep learning job of each of the heterogeneous computing network nodes.

[0013] Furthermore, generating the latest power supply scheduling plan for the heterogeneous computing network according to the carbon intensity and the power usage status information includes:

[0014] Calculate the carbon emissions of each node of the heterogeneous computing network according to the carbon intensity and the power usage status information;

[0015] Generating a power supply priority queue of the heterogeneous computing power network according to the total carbon emissions of the nodes of the heterogeneous computing power network;

[0016] The heterogeneous computing network node with the highest priority in the priority queue increases the power allocated by Δe on the basis of the original power;

[0017] Repeat the above steps until the termination condition is reached to obtain the latest power supply scheduling plan, wherein the termination condition is: the total power in the heterogeneous computing power network has been allocated, and / or the carbon emissions of the heterogeneous computing power network exceed the carbon emission budget.

[0018] Furthermore, the carbon emissions of each of the heterogeneous computing network nodes are calculated according to the carbon intensity and the power usage status information, including:

[0019] According to the power usage status information and the amount of electricity that has been allocated to the deep learning job of each of the heterogeneous computing network nodes, the first job completion energy consumption and the second job completion energy consumption of the deep learning job of each of the heterogeneous computing network nodes are calculated; wherein the first job completion energy consumption represents the energy consumption required for power supply and transmission of the heterogeneous computing network node according to the allocated amount of electricity; the second job completion energy consumption represents the energy consumption required for power supply and transmission of the heterogeneous computing network node after the power supply is increased by Δe;

[0020] According to the energy consumption of completing the first operation and the energy consumption of completing the second operation, the first operation carbon emissions and the second operation carbon emissions of the deep learning operation of the heterogeneous computing network node are calculated; wherein the first operation carbon emissions are the carbon emissions generated during the power supply transmission process according to the allocated power; and the second operation emissions are the carbon emissions generated during the power supply transmission process after the power supply is increased by Δe;

[0021] Compare the carbon emissions of the first operation and the second operation of the deep learning operation of the heterogeneous computing power network node, and take the lower carbon emission value of the two as the total carbon emissions of the heterogeneous computing power network node.

[0022] A carbon intensity-based heterogeneous computing network power supply scheduling device, comprising:

[0023] A carbon intensity collection module, used to obtain the carbon intensity at the current moment every preset time period and send the carbon intensity to the coordination module;

[0024] Coordination module, including network status real-time monitoring and feedback module, power resource allocation module and job placement module;

[0025] The network status real-time monitoring feedback module is used to receive in real time the power usage status information from the heterogeneous computing power network nodes and the carbon intensity sent by the carbon intensity collection module;

[0026] The power resource allocation module is used to generate the latest power supply scheduling plan for the heterogeneous computing network according to the carbon intensity and the power usage status information, so as to determine the power supply of each node of the heterogeneous computing network, so that the carbon emissions generated by executing the latest power supply scheduling plan are less than or equal to the carbon emission budget, and the latest power supply scheduling plan includes the power supply allocated to each node of the heterogeneous computing network, which is used to train the node deep learning model to meet the computing requirements of the deep learning job;

[0027] The job placement module is used to allocate power supply to the deep learning job of each of the heterogeneous computing power network nodes according to the latest power supply scheduling plan;

[0028] The heterogeneous computing network node is used to execute the deep learning job according to the power supply allocated by the latest power supply scheduling method.

[0029] Furthermore, the power resource allocation module generates the latest power supply scheduling plan for the heterogeneous computing network according to the carbon intensity and the power usage status information, including:

[0030] Calculate the carbon emissions of each node of the heterogeneous computing network according to the carbon intensity and the power usage status information;

[0031] Generating a power supply priority queue of the heterogeneous computing power network according to the carbon emissions of the nodes of the heterogeneous computing power network;

[0032] For the job with the highest priority in the priority queue, Δe is added to the originally allocated power; Δe represents a preset power value;

[0033] Repeat the above steps until the termination condition is reached to obtain the latest power supply scheduling plan, wherein the termination condition is: the total power in the heterogeneous computing power network has been allocated, and / or the carbon emissions of the heterogeneous computing power network exceed the carbon emission budget.

[0034] Furthermore, the carbon emissions of each of the heterogeneous computing network nodes are calculated according to the carbon intensity and the power usage status information, including:

[0035] According to the power usage status information and the amount of electricity that has been allocated to the deep learning job of each of the heterogeneous computing network nodes, the first job completion energy consumption and the second job completion energy consumption of the deep learning job of each of the heterogeneous computing network nodes are calculated; wherein the first job completion energy consumption represents the energy consumption required for power supply and transmission of the heterogeneous computing network node according to the allocated amount of electricity; the second job completion energy consumption represents the energy consumption required for power supply and transmission of the heterogeneous computing network node after the power supply is increased by Δe;

[0036] According to the energy consumption of completing the first operation and the energy consumption of completing the second operation, the first operation carbon emissions and the second operation carbon emissions of the deep learning operation of the heterogeneous computing network node are calculated; wherein the first operation carbon emissions are the carbon emissions generated during the power supply transmission process according to the allocated power; and the second operation emissions are the carbon emissions generated during the power supply transmission process after the power supply is increased by Δe;

[0037] Compare the carbon emissions of the first operation and the second operation of the deep learning operation of the heterogeneous computing power network node, and take the lower carbon emission value of the two as the total carbon emissions of the heterogeneous computing power network node.

[0038] A carbon intensity-based heterogeneous computing power network power supply scheduling system, comprising: a computer-readable storage medium and a processor;

[0039] The computer-readable storage medium is used to store executable instructions;

[0040] The processor is used to read the executable instructions stored in the computer-readable storage medium to execute the carbon intensity-based heterogeneous computing network power supply scheduling method.

[0041] A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the carbon intensity-based heterogeneous computing network power supply scheduling method.

[0042] Compared with the prior art, the present invention has the following characteristics:

[0043] 1) The present invention comprehensively considers the changes in carbon intensity of heterogeneous computing network nodes in different environments and at different time periods, obtains carbon intensity information through real-time monitoring and regular sampling, combines the power usage status of the nodes, and uses these factors as the basis for generating the optimal power supply scheduling plan, which is more flexible and reasonable, and improves system efficiency and environmental protection.

[0044] 2) The present invention adjusts the latest power supply scheduling plan according to the real-time carbon intensity, thereby ensuring that the nodes of the heterogeneous computing power network do not exceed the given carbon emission budget when executing deep learning jobs, and realizes the transfer of workload from nodes with high carbon intensity to nodes with low carbon intensity, making the entire heterogeneous computing power network carbon-efficient.

[0045] 3) The deep learning training model used in the present invention is not very complex and can minimize additional costs while optimizing carbon emissions in power supply scheduling, and has higher practicality and scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a structural schematic diagram of a power supply scheduling device for a heterogeneous computing power network based on carbon intensity according to an embodiment of the present invention;

[0047] Figure 2 A schematic diagram of the scheduling process of the deep learning job provided by the present invention;

[0048] Figure 3 The present invention is a flowchart of a method for power supply scheduling of a heterogeneous computing power network based on carbon intensity according to an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] See also Figure 1 The embodiment of the present invention provides a heterogeneous computing power network power supply scheduling device based on carbon intensity, including:

[0051] The carbon intensity acquisition module is used to acquire the carbon intensity at the current moment every preset time period and send the carbon intensity to the coordination module.

[0052] In this embodiment, the time interval is set to collect carbon intensity once every hour. Carbon intensity represents the amount of carbon dioxide emissions generated per unit of energy consumption. The same deep learning job is executed in different time periods, resulting in the same amount of energy consumption, but different carbon emissions. In this embodiment, the carbon intensity acquisition module acquires the amount of carbon dioxide generated by the transmission of electricity in the heterogeneous computing network in real time, and sends this data to the coordination module, so that the coordination module can adjust the power supply scheduling plan in real time according to the change of carbon intensity.

[0053] The coordination module includes: a network status real-time monitoring and feedback module, a power resource allocation module and a job placement module.

[0054] The network status real-time monitoring feedback module is used to receive the network status information from the heterogeneous computing power in real time. In this embodiment, the network status real-time monitoring feedback module obtains the network status information of the heterogeneous computing power network at the current moment, and the network status information includes the power usage status information and carbon intensity of each network node in the network at the current moment, thereby helping the power resource allocation module to make accurate scheduling decisions based on the real-time network status.

[0055] The power resource allocation module is used to generate an updated power supply scheduling plan for the heterogeneous computing network based on the carbon intensity and the power usage status information, so that the carbon emissions generated by executing the updated power supply scheduling plan are less than or equal to the carbon emission budget; the updated power supply scheduling plan includes the power supply allocated to each node of the heterogeneous computing network; the power supply allocated to each node of the heterogeneous computing network is used to train the node deep learning model to meet the computing requirements of the deep learning job.

[0056] The step of generating the latest power supply scheduling scheme for the heterogeneous computing network according to the carbon intensity and the power usage status information includes:

[0057] Calculate the carbon emissions of each node of the heterogeneous computing network according to the carbon intensity and the power usage status information;

[0058] Generating a power supply priority queue of the heterogeneous computing power network according to the carbon emissions of the nodes of the heterogeneous computing power network;

[0059] For the job with the highest priority in the priority queue, Δe is added to the originally allocated power; Δe represents a preset power value;

[0060] Repeat the above steps until the termination condition is reached to obtain the latest power supply scheduling plan, wherein the termination condition is: the total power in the heterogeneous computing power network has been allocated, and / or the carbon emissions of the heterogeneous computing power network exceed the carbon emission budget.

[0061] The step of calculating the carbon emissions of each of the heterogeneous computing network nodes according to the carbon intensity and the power usage status information includes:

[0062] According to the power usage status information and the amount of electricity that has been allocated to the deep learning job of each of the heterogeneous computing network nodes, the first job completion energy consumption and the second job completion energy consumption of the deep learning job of each of the heterogeneous computing network nodes are calculated; wherein the first job completion energy consumption represents the energy consumption required for power supply and transmission of the heterogeneous computing network node according to the allocated amount of electricity; the second job completion energy consumption represents the energy consumption required for power supply and transmission of the heterogeneous computing network node after the power supply is increased by Δe;

[0063] According to the energy consumption of completing the first operation and the energy consumption of completing the second operation, the first operation carbon emissions and the second operation carbon emissions of the deep learning operation of the heterogeneous computing network node are calculated; wherein the first operation carbon emissions are the carbon emissions generated during the power supply transmission process according to the allocated power; and the second operation emissions are the carbon emissions generated during the power supply transmission process after the power supply is increased by Δe;

[0064] Compare the carbon emissions of the first operation and the second operation of the deep learning operation of the heterogeneous computing power network node, and take the lower carbon emission value of the two as the total carbon emissions of the heterogeneous computing power network node.

[0065] The job placement module is used to allocate power supply to each deep learning job of the heterogeneous computing power network node according to the latest power supply scheduling plan;

[0066] The heterogeneous computing network executes the deep learning operation according to the power supply allocated by the latest power supply scheduling method.

[0067] Figure 2 This is a schematic diagram of the scheduling process of a deep learning job. In the job submission stage, it is necessary to submit the data set, implementation code, termination conditions, optimizer and other information required for the deep learning job. It is worth noting that different deep learning jobs have different parameter quantities and data set sizes, and the energy consumption required to execute the job is different, and the final carbon emissions generated are also different.

[0068] In the power resource allocation stage, reference Figure 2The coordination module in this embodiment receives the network status information from the heterogeneous computing network in real time through the network status real-time monitoring feedback module, and calculates the power supply allocated to the deep learning job submitted by each heterogeneous computing network node through the power resource allocation module, that is, according to the carbon emission budget of the heterogeneous computing network and the current carbon intensity of the network, allocates power to the deep learning job of each heterogeneous computing network node, and finally generates the latest power scheduling plan.

[0069] like Figure 1 As shown, the job placement stage is implemented by the job placement module of the coordination module, and power is allocated to each deep learning job of the heterogeneous computing network node according to the latest power supply scheduling plan. In this embodiment, the process needs to record the real-time carbon intensity of the time at fixed intervals, and use the coordination module to generate the latest power supply scheduling plan according to the current carbon intensity to ensure that the network carbon emissions do not exceed the carbon emission budget. If the carbon intensity at the current moment is high, the scheduling of jobs with high carbon intensity in the latest power supply scheduling plan can be appropriately reduced; similarly, if the carbon intensity at the current moment is low, the scheduling of jobs with high carbon intensity in the latest power supply scheduling plan can be appropriately increased, so as to adjust the latest power supply plan in real time and transfer the workload from the time of high carbon intensity to the time of low carbon intensity to improve the carbon efficiency of the system.

[0070] See also Figure 3 The embodiment of the present invention further provides a method for power supply scheduling of a heterogeneous computing power network based on carbon intensity, which is applied to the power supply scheduling device described in the embodiment of the present invention, and the method includes:

[0071] Step S1, obtaining the carbon intensity at the current moment in a preset time period.

[0072] Step S2, receiving power usage status information from heterogeneous computing network nodes in real time.

[0073] Step S3: Generate a latest power supply scheduling plan for the heterogeneous computing network based on the carbon intensity and the power usage status information, so that the carbon emissions generated by executing the latest power supply scheduling plan are less than or equal to the carbon emission budget; the latest power supply scheduling plan includes the power supply allocated to each deep learning job of the heterogeneous computing network node.

[0074] Step S4: Allocate power supply to the deep learning job of each of the heterogeneous computing network nodes according to the latest power supply scheduling plan.

[0075] Step S5: Execute the deep learning operation according to the power supply allocated by the latest power supply scheduling method.

[0076] In step S3, generating the latest power supply scheduling plan for the heterogeneous computing network according to the carbon intensity and the power usage status information includes:

[0077] Step S31, calculating the carbon emissions of each of the heterogeneous computing network nodes according to the carbon intensity and the power usage status information. Step S31 specifically includes:

[0078] Step S31-1, according to the power usage status information and the amount of power that has been allocated to the deep learning job of each of the heterogeneous computing network nodes, calculate the first job completion energy consumption and the second job completion energy consumption of the deep learning job of each of the heterogeneous computing network nodes; wherein the first job completion energy consumption represents the energy consumption required for power supply and transmission of the heterogeneous computing network node according to the allocated power; the second job completion energy consumption is the energy consumption required for power supply and transmission of the heterogeneous computing network node after the power supply is increased by Δe;

[0079] For example, for a deep learning job Ω that has been allocated power e, the energy consumption of the first job to complete represents the energy consumption required to transmit the power e, that is, E 1 =Ω*b+α*d*e; the second operation energy consumption represents the energy consumption required to transmit the amount of electricity e+Δe, which is E 2 =Ω*b+α*d*(e+Δe). Δe represents a preset power value, which can be adjusted according to the total power generation of the heterogeneous computing network; b represents the energy consumed by executing a unit computing task of deep learning. This coefficient depends on the hardware efficiency and energy management strategy of the heterogeneous computing network nodes, d represents the power transmission distance, and α is the weighting factor between the energy consumption of calculating deep learning and the energy consumption of power transmission.

[0080] In this embodiment, the greater the power supply of the deep learning operation of the heterogeneous computing network node, the greater its transmission energy consumption, and the greater the final total carbon emissions. The historical deep learning operation data can be used as a sample, and the parameter value of the deep learning operation and the power supply obtained by the power supply scheduling plan can be input to establish a neural network model to predict the current operation data.

[0081] Step S31-2, calculate the first operation carbon emissions and the second operation carbon emissions of the deep learning operation of the heterogeneous computing network node according to the energy consumption of completing the first operation and the energy consumption of completing the second operation; wherein, the first operation carbon emissions are the carbon emissions generated by the power supply transmission process according to the allocated power; the second operation emissions are the carbon emissions generated by the power supply transmission process after the power supply is increased by Δe.

[0082] In this embodiment, the carbon emissions of the deep learning operations of the heterogeneous computing network nodes are the product of carbon intensity and power transmission energy consumption.

[0083] For example, for a deep learning job Ω that has been allocated with an amount of electricity e, the carbon emissions of the first job represent the carbon emissions generated by the transmission of the amount of electricity e, that is, C 1=c*E 1 The carbon emission of the second operation represents the carbon emission generated by the transmission of electricity e+Δe, which is C 2 =c*E 2 . c represents the carbon intensity of the heterogeneous computing network node at the current sampling moment.

[0084] Step S31-3, compare the carbon emissions of the first operation and the second operation of the deep learning operation of the heterogeneous computing network node, and take the lower carbon emission value of the two as the total carbon emissions of the heterogeneous computing network node. The total carbon emissions of the heterogeneous computing network node are expressed as Ctotal=min(C1, C2).

[0085] Step S32: generating a power supply priority queue of the heterogeneous computing power network according to the total carbon emissions of the nodes of the heterogeneous computing power network.

[0086] For example, a node i in a heterogeneous computing network executes a deep learning job Ω i The total carbon emissions generated are less than the heterogeneous computing network node j executing the deep learning job Ω j The total carbon emissions generated, then the priority of heterogeneous computing network node i is higher than that of heterogeneous computing network node j.

[0087] Step S33: The heterogeneous computing network node with the highest priority in the priority queue increases Δe based on the originally allocated power.

[0088] Step S34, repeat the above steps until the termination condition is reached, the latest power supply scheduling plan, the termination condition is: the total power in the heterogeneous computing power network has been allocated, and / or the carbon emissions of the heterogeneous computing power network exceed the carbon emission budget.

[0089] The embodiment of the present invention uses a dynamic programming algorithm to allocate power for the deep learning jobs of each network node in the heterogeneous computing network. The power allocation problem of the deep learning job of each network node is solved in sequence, and the power allocated to the deep learning job of the previous network node is sorted according to the priority. When solving the power allocation problem of the deep learning job of the next network node, the power allocation should be subtracted from the total power supply. When solving the power allocation problem of the deep learning job of each network node, various possible power allocation decisions are listed, and decisions that exceed the amount of power required for the network node to execute the deep learning job are discarded, and the local solution that may reach the optimal value is retained. Re-execute steps S31-S34 to generate a new priority queue until the termination condition is reached.

[0090] In this embodiment, each time a cycle is executed, the power supply of the node with higher priority in the heterogeneous computing network will increase by Δe, and the carbon intensity of the node during this period is lower than that of other nodes in the heterogeneous computing network. This can reduce the overall carbon emissions of the network. After several cycles, the optimal power supply scheduling strategy for the heterogeneous computing network can be obtained.

[0091] In this embodiment, based on the carbon intensity and the electricity usage status information, electricity can be preferentially allocated to heterogeneous computing network nodes with low carbon emissions, and the latest power supply scheduling plan can be dynamically adjusted to reduce the total carbon emissions of the heterogeneous computing network so that it does not exceed the budget.

[0092] Another embodiment of the present invention provides a carbon intensity-based heterogeneous computing network power supply scheduling system, including: a computer-readable storage medium and a processor;

[0093] The computer-readable storage medium is used to store executable instructions;

[0094] The processor is used to read the executable instructions stored in the computer-readable storage medium to execute the carbon intensity-based heterogeneous computing network power supply scheduling method.

[0095] Another embodiment of the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the carbon intensity-based heterogeneous computing network power supply scheduling method.

[0096] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0097] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0098] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0099] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for power supply scheduling of heterogeneous computing power network based on carbon intensity, characterized in that: include: Get the carbon intensity at the current moment in the preset time period; Receive power usage status information from heterogeneous computing network nodes in real time; Generate a latest power supply scheduling plan for the heterogeneous computing network according to the carbon intensity and the power usage status information, so that the carbon emissions generated by executing the latest power supply scheduling plan are less than or equal to the carbon emission budget; The latest power supply scheduling scheme includes the power supply amount allocated to each of the heterogeneous computing power network nodes for deep learning operations; According to the latest power supply scheduling scheme, power is allocated to the deep learning job of each of the heterogeneous computing network nodes.

2. The method for power supply scheduling of heterogeneous computing power network based on carbon intensity according to claim 1, characterized in that: According to the carbon intensity and the power usage status information, the latest power supply scheduling plan for the heterogeneous computing network is generated, including: Calculate the carbon emissions of each node of the heterogeneous computing network according to the carbon intensity and the power usage status information; Generating a power supply priority queue of the heterogeneous computing power network according to the total carbon emissions of the nodes of the heterogeneous computing power network; The heterogeneous computing network node with the highest priority in the priority queue increases the power allocated by Δe on the basis of the original power; Repeat the above steps until the termination condition is reached to obtain the latest power supply scheduling plan, wherein the termination condition is: the total power in the heterogeneous computing power network has been allocated, and / or the carbon emissions of the heterogeneous computing power network exceed the carbon emission budget.

3. The device for dispatching power supply to heterogeneous computing power network based on carbon intensity according to claim 2, characterized in that: The step of calculating the carbon emissions of each of the heterogeneous computing network nodes according to the carbon intensity and the power usage status information includes: According to the power usage status information and the amount of electricity that has been allocated to the deep learning job of each of the heterogeneous computing network nodes, the first job completion energy consumption and the second job completion energy consumption of the deep learning job of each of the heterogeneous computing network nodes are calculated; wherein the first job completion energy consumption represents the energy consumption required for power supply and transmission of the heterogeneous computing network node according to the allocated amount of electricity; the second job completion energy consumption represents the energy consumption required for power supply and transmission of the heterogeneous computing network node after the power supply is increased by Δe; According to the energy consumption of completing the first operation and the energy consumption of completing the second operation, the first operation carbon emissions and the second operation carbon emissions of the deep learning operation of the heterogeneous computing network node are calculated; wherein the first operation carbon emissions are the carbon emissions generated during the power supply transmission process according to the allocated power; and the second operation emissions are the carbon emissions generated during the power supply transmission process after the power supply is increased by Δe; Compare the carbon emissions of the first operation and the second operation of the deep learning operation of the heterogeneous computing power network node, and take the lower carbon emission value of the two as the total carbon emissions of the heterogeneous computing power network node.

4. A power supply scheduling device for heterogeneous computing power network based on carbon intensity, characterized in that: include: A carbon intensity collection module, used to obtain the carbon intensity at the current moment every preset time period and send the carbon intensity to the coordination module; Coordination module, including network status real-time monitoring and feedback module, power resource allocation module and job placement module; The network status real-time monitoring feedback module is used to receive in real time the power usage status information from the heterogeneous computing power network nodes and the carbon intensity sent by the carbon intensity collection module; The power resource allocation module is used to generate the latest power supply scheduling plan for the heterogeneous computing network according to the carbon intensity and the power usage status information, so as to determine the power supply of each node of the heterogeneous computing network, so that the carbon emissions generated by executing the latest power supply scheduling plan are less than or equal to the carbon emission budget, and the latest power supply scheduling plan includes the power supply allocated to each node of the heterogeneous computing network, which is used to train the node deep learning model to meet the computing requirements of the deep learning job; The job placement module is used to allocate power supply to the deep learning job of each of the heterogeneous computing power network nodes according to the latest power supply scheduling plan; The heterogeneous computing network node is used to execute the deep learning job according to the power supply allocated by the latest power supply scheduling method.

5. The device for dispatching power supply to heterogeneous computing power network based on carbon intensity according to claim 4, characterized in that: The power resource allocation module generates a latest power supply scheduling plan for the heterogeneous computing network according to the carbon intensity and the power usage status information, including: Calculate the carbon emissions of each node of the heterogeneous computing network according to the carbon intensity and the power usage status information; Generating a power supply priority queue of the heterogeneous computing power network according to the carbon emissions of the nodes of the heterogeneous computing power network; For the job with the highest priority in the priority queue, Δe is added to the originally allocated power; Δe represents a preset power value; Repeat the above steps until the termination condition is reached to obtain the latest power supply scheduling plan, wherein the termination condition is: the total power in the heterogeneous computing power network has been allocated, and / or the carbon emissions of the heterogeneous computing power network exceed the carbon emission budget.

6. The device for dispatching power supply to heterogeneous computing power network based on carbon intensity according to claim 5, characterized in that: The step of calculating the carbon emissions of each of the heterogeneous computing network nodes according to the carbon intensity and the power usage status information includes: According to the power usage status information and the amount of electricity that has been allocated to the deep learning job of each of the heterogeneous computing network nodes, the first job completion energy consumption and the second job completion energy consumption of the deep learning job of each of the heterogeneous computing network nodes are calculated; wherein the first job completion energy consumption represents the energy consumption required for power supply and transmission of the heterogeneous computing network node according to the allocated amount of electricity; the second job completion energy consumption represents the energy consumption required for power supply and transmission of the heterogeneous computing network node after the power supply is increased by Δe; According to the energy consumption of completing the first operation and the energy consumption of completing the second operation, the first operation carbon emissions and the second operation carbon emissions of the deep learning operation of the heterogeneous computing network node are calculated; wherein the first operation carbon emissions are the carbon emissions generated during the power supply transmission process according to the allocated power; and the second operation emissions are the carbon emissions generated during the power supply transmission process after the power supply is increased by Δe; Compare the carbon emissions of the first operation and the second operation of the deep learning operation of the heterogeneous computing power network node, and take the lower carbon emission value of the two as the total carbon emissions of the heterogeneous computing power network node.

7. A carbon intensity-based heterogeneous computing network power supply scheduling system, comprising: A computer readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is used to read the executable instructions stored in the computer-readable storage medium to execute the carbon intensity-based heterogeneous computing network power supply scheduling method described in any one of claims 1 to 3.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the carbon intensity-based heterogeneous computing network power supply scheduling method described in any one of claims 1 to 3.

Images