A low-carbon economic dispatch method for virtual computing power center based on lyapunov optimization

Abstract

This invention discloses a low-carbon economic scheduling method for virtual computing power centers based on Lyapunov optimization, belonging to the field of collaborative optimization of power and computing resources. The method first integrates heterogeneous computing power nodes through a distributed computing power resource aggregation management platform, transforming them into equivalent CPU computing power to form a computing power cluster. It then combines energy storage systems with grid power purchase and sale to construct a comprehensive energy management system, establishing a net benefit maximization optimization objective and related constraints. Next, it introduces task backlog and virtual carbon deficit queues to handle coupled constraints. Finally, based on Lyapunov drift-reward theory, the long-term stochastic optimization problem is decoupled into two real-time solvable subproblems: computing power allocation and energy storage-carbon arbitrage, generating optimal scheduling instructions. This invention requires no predictive information, achieving low-carbon economic scheduling of computing power resources while meeting long-term carbon constraints and service quality requirements, thus improving the utilization efficiency and overall net benefit of computing power resources.

Description

Technical Field

[0001] This invention belongs to the field of collaborative optimization of power systems and computing resources, and specifically relates to a low-carbon economic dispatch method for virtual computing centers based on Lyapunov optimization. Background Technology

[0002] With the large-scale expansion of computing infrastructure, its high energy consumption and high carbon emissions have posed energy challenges. Currently, there is a structural contradiction in the allocation of computing resources: on the one hand, large data centers have strong energy efficiency rigidity, exacerbating the peak-valley difference in the power grid; on the other hand, massive heterogeneous computing resources at the edge and personal terminals are in a state of "computing islands," with high resource idle rates. Therefore, how to construct a standardized aggregation architecture to package massive heterogeneous and distributed computing power into schedulable virtual resources is the foundation for achieving efficient collaboration between computing power and electricity. Existing scheduling methods (such as model predictive control) face two major challenges:

[0003] 1) Time scale mismatch: Carbon emission assessment has a long-term cumulative characteristic, while scheduling decisions are mostly at the hour level, making it difficult to solve the long-short cycle matching problem when full-cycle information is lacking.

[0004] 2) Strong prediction dependence: Existing methods rely heavily on high-precision predictions of electricity prices, new energy sources and task arrival rates. Prediction bias can cause the strategy to deviate significantly from the optimal solution. Summary of the Invention

[0005] To address the above problems, this invention provides a low-carbon economic scheduling method for virtual computing power centers (VCCs) based on Lyapunov optimization. This method constructs a virtual computing power center aggregation architecture and designs an online optimization algorithm that does not require predictive information, transforming long-term carbon constraints into a real-time queue stability problem, thereby achieving economic scheduling and low-carbon operation of distributed computing resources.

[0006] The technical solution adopted in this invention is as follows: A low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization, comprising the following steps:

[0007] Step 1: The virtual computing center, through a distributed computing resource aggregation and management platform, completes the large-scale integration and unified management of heterogeneous distributed computing nodes. Using a normalized computing power method, it transforms the original computing power of different types of computing nodes into equivalent CPU computing power, forming... There are 1 computing power clusters, denoted as _ ... Simultaneously, an energy storage system is provided for the virtual computing center. Based on the virtual computing center's purchase and sale of electricity with the power grid, a comprehensive energy management system for the virtual computing center is constructed. Under the premise of meeting long-term carbon emission constraints and service quality requirements, the comprehensive energy system of the virtual computing center maximizes comprehensive net revenue through the coordinated operation of the electricity market and the carbon market. An optimization objective function is constructed with the goal of maximizing comprehensive net revenue under equal periodicity. The system is then uniformly scheduled in combination with the constraints considered by the comprehensive energy management system of the virtual computing center. The constraints include the energy balance constraint of the comprehensive energy management system of the virtual computing center, the operation constraint of the computing cluster, and the operation constraint of the energy storage system.

[0008] Step 2: Introduce two corresponding queue variables as the state of the virtual computing center's integrated energy system: a task backlog queue and a virtual carbon deficit queue. These are used to handle the coupled constraints of service quality stability and long-term carbon emission compliance. The task backlog queue represents the amount of unprocessed computing tasks currently accumulated in the virtual computing center, reflecting the system congestion level. The virtual carbon deficit queue is used to represent the satisfaction of long-term carbon emission constraints.

[0009] Step 3: Based on Lyapunov drift-reward theory, the long-term stochastic optimization problem is decoupled into two deterministic subproblems that can be solved in real time: the computing power resource allocation subproblem and the energy storage electricity-carbon arbitrage subproblem. The optimal scheduling instruction for the current time slot is generated to realize the low-carbon economic scheduling of the virtual computing power center.

[0010] Furthermore, in step one, the virtual computing center aggregates no fewer than 800 distributed computing nodes through the KubeEdge edge computing management platform. These nodes include data centers, edge nodes, and personal computers, and each node contains different raw computing power. Idle power consumption Maximum power consumption Network bandwidth and running status By using the normalized computing power method, the original computing power of distributed computing power nodes is converted into equivalent CPU computing power:

[0011]

[0012] In the formula: Represents a computing power cluster Maximum processing capacity; Represents a cluster polymerization Number of type nodes; for The normalization coefficient of computing power for type nodes;

[0013] The virtual computing center's integrated energy system, under the premise of meeting long-term carbon emission constraints and service quality requirements, maximizes comprehensive net revenue through the coordinated operation of the electricity market and the carbon market. An optimization objective function is constructed with the goal of maximizing comprehensive net revenue under isocyclic conditions, and its expression is as follows:

[0014]

[0015] In the formula: This represents the weekly net revenue of the integrated energy management system for virtual computing centers. Represents the mathematical expectation; Indicates the system in time slots Net income; This represents the system's weekly carbon settlement cost; Indicates the time slot number; This indicates the number of time periods optimized throughout the week;

[0016] The virtual computing center integrated energy management system in time slots The expression for net income is as follows:

[0017]

[0018] In the formula: This indicates that the virtual computing center's integrated energy management system is in time slots. The overall benefits; This indicates that the virtual computing center's integrated energy management system is in time slots. Maintenance costs;

[0019] The virtual computing center integrated energy management system in time slots The total revenue consists of two parts: revenue from computing power services and revenue from electricity sales, expressed as follows:

[0020]

[0021] In the formula: and These represent the revenue from computing power services and electricity sales revenue of the integrated energy management system of the virtual computing center, respectively. and These are the weighted average service price and time slot for computing tasks accepted by the integrated energy management system of the virtual computing center. Electricity price; and The virtual computing center integrated energy management system is located in time slots. Assigned to computing power cluster The workload and the power purchased from the grid;

[0022] The virtual computing center integrated energy management system in time slots The expression for the operation and maintenance cost is as follows:

[0023]

[0024] In the formula: For computing power clusters Basic operational costs for handling a unit of task;

[0025] The expression for the weekly carbon settlement cost of the virtual computing center's integrated energy management system is as follows:

[0026]

[0027] In the formula: Penalties for violating carbon constraints; For time slots The average carbon emission factor of the power grid is affected by the dynamic changes in the proportion of regional renewable energy output; The upper limit set for the long-term average carbon emission rate; Indicates the duration of a unit of time period;

[0028] The constraints considered in the integrated energy management system of the virtual computing center include energy balance constraints, computing cluster operation constraints, and energy storage system operation constraints, as detailed below:

[0029] 1) Energy balance constraints of the virtual computing center integrated energy management system:

[0030]

[0031] In the formula: Represents a computing power cluster In the time slot Energy consumption; Indicates the energy storage system in time slots The charging power; Indicates the energy storage system in time slots The discharge power;

[0032] 2) Constraints on the operation of computing clusters:

[0033] The constraints on the operation of computing clusters include: capacity constraints, task balancing constraints, and energy consumption constraints, as detailed below:

[0034] The expression for the capacity constraint is as follows:

[0035]

[0036] The expression for the task balancing constraint is as follows:

[0037]

[0038] In the formula: Indicates time slot The number of tasks completed; This indicates the backlog of tasks in the previous time slot;

[0039] The energy consumption constraint is expressed as follows:

[0040]

[0041] In the formula: Represents a computing power cluster Total energy consumption for processing a unit of task; Represents a computing power cluster The computational energy consumption for processing a unit of task; This indicates that the task is being transmitted to the computing cluster. Transmission energy consumption; Represents a computing power cluster Static power consumption; Represents a computing power cluster Maximum energy consumption;

[0042] 3) Energy storage system operation constraints:

[0043] The operational constraints of energy storage systems include: charge / discharge power range constraints, state of charge / discharge constraints, state of charge constraints, energy storage capacity range constraints, and long-term cycling constraints, as detailed below:

[0044] The expression for the charge / discharge power range constraint is as follows:

[0045]

[0046] In the formula: This indicates the maximum charging power of the energy storage system; This indicates the maximum discharge power of the energy storage system;

[0047] The expression for the charge / discharge state constraint is as follows:

[0048]

[0049] The expression for the charge state constraint is as follows:

[0050]

[0051] In the formula: Indicates the energy storage system in time slots The amount of stored energy; Indicates the charging and discharging efficiency of the energy storage system;

[0052] The expression for the energy storage capacity range constraint is as follows:

[0053]

[0054] In the formula: This indicates the maximum energy storage capacity of the energy storage system;

[0055] The expression for a long-running cycle constraint is as follows:

[0056]

[0057] In the formula: This indicates the initial energy storage capacity of the energy storage system.

[0058] Furthermore, in step two, two corresponding queue variables are introduced as system states:

[0059] 1) Task backlog queue : Represents the current accumulated amount of unprocessed computing tasks in the virtual computing center, reflecting the degree of system congestion. Its dynamic evolution equation is:

[0060]

[0061] In the formula: Indicates the backlog queue for the next time slot, and keeps it in place. Its strong stability means that it guarantees limited delays in missions;

[0062] 2) Virtual carbon deficit queue : Used to characterize the satisfaction of long-term carbon emission constraints, its dynamic evolution equation is:

[0063]

[0064] In the formula: This represents the virtual carbon deficit queue for the next time slot; Indicates the instantaneous carbon emissions at the current moment; The degree of stockpiling directly reflects the magnitude of carbon default risk;

[0065] Furthermore, the integrated energy system of the virtual computing center is defined in time slots. The combined state vector is .

[0066] Furthermore, in step three, based on Lyapunov's drift-reward theory, the long-term stochastic optimization problem is decoupled into two sub-problems that can be solved in real time: the computing power resource allocation sub-problem and the energy storage-carbon arbitrage sub-problem. The optimal scheduling instruction for the current time slot is then generated to achieve low-carbon economic scheduling of the virtual computing center; specifically as follows:

[0067] To characterize the stability of the system, a quadratic Lyapunov scalar function is constructed. :

[0068]

[0069] Define one-step conditional Lyapunov drift The expected change in the system congestion level:

[0070]

[0071] According to stochastic network optimization theory, in order to maximize system benefits while ensuring queue stability, the original long-term optimization problem can be transformed into minimizing the upper bound of "drift minus weighted benefit" for each time slot, by introducing control parameters. , To balance queue stability and economic benefits, a single time slot objective function is constructed:

[0072]

[0073] In the formula: To control parameters, which physically represent the risk preferences of virtual computing center operators; using inequalities After scaling and expanding the constructed single-slot objective function and removing the constant term, it is equivalent to performing the same operation in each time slot. Minimize the following objective function :

[0074]

[0075] The complex multivariate coupled optimization problem is decomposed into two low-complexity linear programming subproblems, as follows:

[0076] 1) Sub-problem 1: Decision on distribution of computing power services

[0077] Extracting the objective function In terms of computing power allocation The relevant terms are then transformed into a form that maximizes returns:

[0078]

[0079] Define cluster The marginal net income index is:

[0080]

[0081] Decision-making logic: If Operating at full capacity; if The task will be temporarily stored in the queue to wait for the price to drop;

[0082] 2) Sub-problem two: Energy storage electricity-carbon dual arbitrage decision

[0083] Extracting the objective function The terms related to energy storage power are summarized to obtain the energy storage optimization objective:

[0084]

[0085] Define generalized nodal marginal electricity price Its physical meaning is the comprehensive cost per unit of electrical energy; the charging and discharging threshold is the exponentially weighted moving average of the generalized nodal marginal electricity price. As a reference benchmark, its update formula is:

[0086]

[0087] Where: attenuation coefficient This corresponds to an equivalent memory window of 20 time slots; It is the exponentially weighted moving average of the marginal electricity price of the previous time slot generalized node;

[0088] when At that time, the energy storage system is charging. , ;when or At that time, the energy storage system discharges. , Otherwise, the energy storage system remains in standby mode. .

[0089] Ultimately, this enables the energy storage system to automatically capture electricity price differences and carbon price differences, achieving low-carbon economic dispatch of the virtual computing center.

[0090] This invention also provides a low-carbon economic scheduling system for virtual computing centers based on Lyapunov optimization, used to implement the low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization as described above. The system includes a computing power aggregation management module, a comprehensive energy management module, a queue status management module, and a Lyapunov optimization scheduling module. These modules work together to achieve heterogeneous computing power integration, constraint control, and low-carbon economic scheduling of the virtual computing center.

[0091] The computing power aggregation management module is configured to integrate heterogeneous distributed computing power nodes on a large scale and manage them in a unified manner through a distributed computing power resource aggregation management platform. It adopts a normalized computing power method to convert the original computing power of different types of nodes into equivalent CPU computing power to form a schedulable computing power cluster.

[0092] The integrated energy management module is configured to carry an energy storage system and build a power purchase and sale interaction link between the virtual computing center and the power grid. Under the premise of meeting long-term carbon emission constraints and service quality requirements, and in combination with the needs of coordinated operation of the electricity market and carbon market, it constructs an optimization objective function with the goal of maximizing the comprehensive net income under equal-week conditions. At the same time, it integrates a constraint condition library of virtual computing center energy balance constraints, computing cluster operation constraints, and energy storage system operation constraints to provide a constraint basis for scheduling.

[0093] The queue status management module is configured to establish and update two system status queue variables in real time: a task backlog queue and a virtual carbon deficit queue. The task backlog queue represents the amount of unprocessed computing tasks currently accumulated in the virtual computing center and reflects the degree of system congestion. The virtual carbon deficit queue represents the satisfaction of long-term carbon emission constraints. The dynamic evolution of the two queues is used to monitor the coupling constraint status of service quality stability and long-term carbon emission compliance in real time.

[0094] The Lyapunov optimization scheduling module is configured based on the Lyapunov drift-reward theory, decoupling the long-term stochastic optimization problem into a computing resource allocation subproblem and an energy storage-carbon arbitrage subproblem. By constructing a Lyapunov scalar function, defining a one-step conditional Lyapunov drift, and introducing risk preference control parameters, it generates a single-slot optimization objective function and transforms it into a deterministic subproblem that can be solved in real time. It then outputs the optimal scheduling instruction for the current slot, realizing low-carbon economic scheduling of the virtual computing center.

[0095] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements a low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization as described above.

[0096] The present invention also provides a computer program product, including a computer program or instructions, which, when executed by a processor, implements a low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization as described above.

[0097] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements a low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization as described above.

[0098] This invention offers the following advantages and benefits: Compared with existing technologies, it achieves standardized encapsulation and unified scheduling of massive heterogeneous and distributed computing resources, transforming them into flexible resources that can be scheduled by the power grid. Addressing the challenge of online processing of long-term carbon constraints, a virtual carbon deficit queue mechanism is introduced, transforming the complex long-term carbon constraints into a queue stability problem. An online drift-reward algorithm based on Lyapunov optimization is designed, avoiding the need for precise prediction of uncertain future information such as electricity prices. This decouples the complex problem into a low-complexity real-time linear programming problem, achieving a balance between computational efficiency and optimization performance, and improving the real-time performance and feasibility of scheduling. Ultimately, this invention maximizes the long-term comprehensive net benefit of the virtual computing center while strictly meeting long-term low-carbon constraints, significantly improving the economic efficiency of computing resource utilization and the low-carbon nature of its operation. Attached Figure Description

[0099] Figure 1 Flowchart of a low-carbon economic scheduling method for virtual computing power centers;

[0100] Figure 2 This is a model diagram of the integrated energy management system for a virtual computing center.

[0101] Figure 3 A diagram showing the comparison of the system's cumulative net revenue under four scheduling methods;

[0102] Figure 4 This diagram illustrates the computing power load distribution under four scheduling methods.

[0103] Figure 5 This diagram illustrates the comparison between instantaneous and average carbon emissions of the system under four scheduling methods.

[0104] Figure 6 This is a schematic diagram illustrating the charging and discharging behavior of an energy storage system in response to generalized nodal electricity prices under the scheduling method of the present invention.

[0105] Figure 7 This is a dynamic diagram illustrating the system task throughput and task backlog queue under the scheduling method of the present invention. Specific implementation methods

[0106] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0107] Example 1

[0108] like Figure 1 As shown, this embodiment provides a low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization, including the following steps:

[0109] Step 1: The virtual computing center aggregates 800 distributed computing nodes based on the KubeEdge edge computing management platform. Node types include cloud data centers, edge nodes, and personal computers. Different types of nodes... They have different computational characteristics and performance indicators, such as raw computing power. Idle power consumption Maximum power consumption Network bandwidth and running status This embodiment aggregates 800 distributed computing nodes into three typical heterogeneous computing clusters. The cloud data center possesses the largest processing capacity, suitable for handling large-scale computing tasks; edge nodes and green PC clusters, while having smaller capacities, exhibit lower energy consumption, thus demonstrating the complementarity of heterogeneous resources. The simulation uses real time-of-use electricity price data from a region in Northeast China. Then, using a normalized computing power method, the computing power of different types of nodes is converted into equivalent CPU computing power, forming... There are 1 computing power clusters, denoted as _ ... Simultaneously, it provides energy storage systems for the virtual computing center, enabling electricity purchase and sale between the virtual computing center and the power grid, such as... Figure 2 As shown, a comprehensive energy management system for a virtual computing center is constructed. Under the premise of meeting long-term carbon emission constraints and service quality requirements, this system maximizes comprehensive net revenue through the coordinated operation of the electricity market and the carbon market. An optimization objective function is constructed with the goal of maximizing comprehensive net revenue under equal-period conditions, and its expression is as follows:

[0110]

[0111] In the formula: This represents the weekly net revenue of the integrated energy management system for virtual computing centers. Represents the mathematical expectation; Indicates the system in time slots Net income; This represents the system's weekly carbon settlement cost; Indicates the time slot number; This indicates the number of time periods optimized throughout the week;

[0112] The virtual computing center integrated energy management system in time slots The expression for net income is as follows:

[0113]

[0114] In the formula: This indicates that the virtual computing center's integrated energy management system is in time slots. The overall benefits; This indicates that the virtual computing center's integrated energy management system is in time slots. Maintenance costs;

[0115] The virtual computing center integrated energy management system in time slots The total revenue consists of two parts: revenue from computing power services and revenue from electricity sales, expressed as follows:

[0116]

[0117] In the formula: and These represent the revenue from computing power services and electricity sales revenue of the integrated energy management system of the virtual computing center, respectively. and These are the weighted average service price and time slot for computing tasks accepted by the integrated energy management system of the virtual computing center. Electricity price; and The virtual computing center integrated energy management system is located in time slots. Assigned to computing power cluster The workload and the power purchased from the grid;

[0118] The virtual computing center integrated energy management system in time slots The expression for the operation and maintenance cost is as follows:

[0119]

[0120] In the formula: For computing power clusters Basic operational costs for handling a unit of task;

[0121] The expression for the weekly carbon settlement cost of the virtual computing center's integrated energy management system is as follows:

[0122]

[0123] In the formula: Penalties for violating carbon constraints; For time slots The average carbon emission factor of the power grid is affected by the dynamic changes in the proportion of regional renewable energy output; The upper limit set for the long-term average carbon emission rate; Indicates the duration of a unit of time period;

[0124] The constraints considered in the integrated energy management system of the virtual computing center include energy balance constraints, computing cluster operation constraints, and energy storage system operation constraints, as detailed below:

[0125] 1) Energy balance constraints of the virtual computing center integrated energy management system:

[0126]

[0127] In the formula: Represents a cluster In the time slot Energy consumption; Indicates the energy storage system in time slots The charging power; Indicates the energy storage system in time slots The discharge power;

[0128] 2) Constraints on the operation of computing clusters:

[0129] The constraints on the operation of computing clusters include: capacity constraints, task balancing constraints, and energy consumption constraints, as detailed below:

[0130] The expression for the capacity constraint is as follows:

[0131]

[0132] The expression for the task balancing constraint is as follows:

[0133]

[0134] In the formula: Indicates time slot The number of tasks completed; This indicates the backlog of tasks in the previous time slot;

[0135] The energy consumption constraint is expressed as follows:

[0136]

[0137] In the formula: Represents a computing power cluster Total energy consumption for processing a unit of task; Represents a computing power cluster The computational energy consumption for processing a unit of task; This indicates that the task is being transmitted to the computing cluster. Transmission energy consumption; Represents a computing power cluster Static power consumption; Represents a computing power cluster Maximum energy consumption;

[0138] 3) Energy storage system operation constraints:

[0139] The operational constraints of energy storage systems include: charge / discharge power range constraints, state of charge / discharge constraints, state of charge constraints, energy storage capacity range constraints, and long-term cycling constraints, as detailed below:

[0140] The expression for the charge / discharge power range constraint is as follows:

[0141]

[0142] In the formula: This indicates the maximum charging power of the energy storage system; This indicates the maximum discharge power of the energy storage system;

[0143] The expression for the charge / discharge state constraint is as follows:

[0144]

[0145] The expression for the charge state constraint is as follows:

[0146]

[0147] In the formula: Indicates the energy storage system in time slots The amount of stored energy; Indicates the charging and discharging efficiency of the energy storage system;

[0148] The expression for the energy storage capacity range constraint is as follows:

[0149]

[0150] In the formula: This indicates the maximum energy storage capacity of the energy storage system;

[0151] The expression for a long-running cycle constraint is as follows:

[0152]

[0153] In the formula: This indicates the initial energy storage capacity of the energy storage system.

[0154] Step 2: First, to address the coupled constraints of service quality stability and long-term carbon emission compliance, two corresponding queue variables are introduced as system states:

[0155] 1) Task backlog queue : Represents the current accumulated amount of unprocessed computing tasks in the virtual computing center, reflecting the degree of system congestion. Its dynamic evolution equation is:

[0156]

[0157] In the formula: Indicates the backlog queue for the next time slot, and keeps it in place. Its strong stability means that it guarantees limited delays in missions;

[0158] 2) Virtual carbon deficit queue : Used to characterize the satisfaction of long-term carbon emission constraints, its dynamic evolution equation is:

[0159]

[0160] In the formula: This represents the virtual carbon deficit queue for the next time slot; Indicates the instantaneous carbon emissions at the current moment; The degree of stockpiling directly reflects the magnitude of carbon default risk;

[0161] Furthermore, it defines the integrated energy system of the virtual computing center in time slots. The combined state vector is .

[0162] Step 3: Based on Lyapunov drift-reward theory, the long-running stochastic optimization problem is decoupled into two real-time solvable subproblems: the computing resource allocation subproblem and the energy storage-carbon arbitrage subproblem. The optimal scheduling instruction for the current time slot is generated to achieve low-carbon economic scheduling of the virtual computing center. To characterize the system's stability, a quadratic Lyapunov scalar function is constructed from the system's comprehensive state vector defined in Step 2. :

[0163]

[0164] Define one-step conditional Lyapunov drift The expected change in congestion level of the integrated energy management system for virtual computing centers:

[0165]

[0166] According to stochastic network optimization theory, to maximize system benefits while ensuring queue stability, the original long-term optimization problem can be transformed into minimizing the upper bound of "drift minus weighted benefit" for each time slot. This is achieved by introducing control parameters. ( To balance queue stability and economic benefits, a single-slot objective function is constructed:

[0167]

[0168] In the formula: To control parameters, which physically represent the risk appetite of virtual computing center operators: The larger the virtual computing center, the more inclined it is to tolerate temporary queue backlogs or carbon deficit fluctuations in pursuit of high returns. Utilizing inequalities... After scaling and expanding the above objectives and removing the constant term, it is equivalent to doing so in each time slot. Minimize the following objective function :

[0169]

[0170] Observing the objective function, we can see that the decision variables (Computing power allocation) and , The energy storage charging and discharging processes are physically independent. Therefore, the complex multivariate coupled optimization problem can be decomposed into two low-complexity linear programming subproblems, enabling efficient parallel solutions.

[0171] 1) Sub-problem 1: Decision on distribution of computing power services

[0172] extract In terms of computing power allocation The relevant terms are then transformed into a form that maximizes returns:

[0173]

[0174] Define computing power cluster The marginal net income index is:

[0175]

[0176] Decision-making logic: If If this indicates that handling the task brings positive overall value (including economic benefits and destocking value), then it will operate at full capacity; if This indicates that the current electricity price or carbon cost is too high, making the task unprofitable. Therefore, the load is proactively reduced, and the task is temporarily placed in the queue to wait for the price to fall.

[0177] 2) Sub-problem two: Energy storage electricity-carbon dual arbitrage decision

[0178] Extracting the objective function The terms related to energy storage power are summarized to obtain the energy storage optimization objective:

[0179]

[0180] Define generalized nodal marginal electricity price Its physical meaning is the comprehensive cost per unit of electrical energy (including electricity price cost and carbon cost).

[0181] The charge / discharge threshold is an exponentially weighted moving average of the generalized nodal marginal electricity price. As a reference benchmark, its update formula is:

[0182]

[0183] Where: attenuation coefficient This corresponds to an equivalent memory window of 20 time slots; It is the exponentially weighted moving average of the marginal electricity price of the previous time slot generalized node;

[0184] The energy storage charging and discharging decision logic is as follows:

[0185] when At that time, the energy storage system is charging. , ;

[0186] when or At that time, the energy storage system discharges. , ;

[0187] Otherwise, the energy storage system remains in standby mode. ;

[0188] Ultimately, this enables the energy storage system to automatically capture electricity price differences and carbon price differences, achieving low-carbon economic dispatch of the virtual computing center.

[0189] Example 2

[0190] This embodiment also provides a low-carbon economic scheduling system for virtual computing centers based on Lyapunov optimization, used to implement the low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization as described in Embodiment 1. The system includes a computing power aggregation management module, a comprehensive energy management module, a queue status management module, and a Lyapunov optimization scheduling module. These modules work together to achieve heterogeneous computing power integration, constraint control, and low-carbon economic scheduling of the virtual computing center.

[0191] The computing power aggregation management module is configured to integrate heterogeneous distributed computing power nodes on a large scale and manage them in a unified manner through a distributed computing power resource aggregation management platform. It adopts a normalized computing power method to convert the original computing power of different types of nodes into equivalent CPU computing power to form a schedulable computing power cluster.

[0192] The integrated energy management module is configured to carry an energy storage system and build a power purchase and sale interaction link between the virtual computing center and the power grid. Under the premise of meeting long-term carbon emission constraints and service quality requirements, and in combination with the needs of coordinated operation of the electricity market and carbon market, it constructs an optimization objective function with the goal of maximizing the comprehensive net income under equal-week conditions. At the same time, it integrates a constraint condition library of virtual computing center energy balance constraints, computing cluster operation constraints, and energy storage system operation constraints to provide a constraint basis for scheduling.

[0193] The queue status management module is configured to construct two system status queue variables: a task backlog queue and a virtual carbon deficit queue. The task backlog queue represents the amount of unprocessed computing tasks currently accumulated in the virtual computing center and reflects the degree of system congestion. The virtual carbon deficit queue represents the satisfaction of long-term carbon emission constraints. The dynamic evolution of the two queues is used to monitor the coupled constraint status of service quality stability and long-term carbon emission compliance in real time.

[0194] The Lyapunov optimization scheduling module is configured based on the Lyapunov drift-reward theory, decoupling the long-term stochastic optimization problem into a computing resource allocation subproblem and an energy storage-carbon arbitrage subproblem. By constructing a Lyapunov scalar function, defining a one-step conditional Lyapunov drift, and introducing risk preference control parameters, it generates a single-slot optimization objective function and transforms it into a deterministic subproblem that can be solved in real time. It then outputs the optimal scheduling instruction for the current slot, realizing low-carbon economic scheduling of the virtual computing center.

[0195] Example 3

[0196] To fully verify the engineering value of the virtual computing center architecture and the performance advantages of the proposed Lyapunov optimization algorithm, this embodiment sets up four operating schemes for comparative analysis. The evaluation is carried out from three dimensions: overall net benefit, spatiotemporal distribution characteristics of computing load, and carbon emission compliance.

[0197] Option 1: Non-aggregated mode. This mode simulates the current situation where computing resources are distributed and operate independently, without energy storage support, and adopts a traditional load balancing strategy.

[0198] Option 2: Baseline Strategy. This dispatching decision focuses solely on minimizing costs in the current time slot, lacking forward-looking consideration of future electricity price fluctuations and long-term carbon constraints.

[0199] Option 3: Model Predictive Control. The virtual computing center employs rolling time-domain optimization to make decisions based on predicted future information. To simulate a real-world scenario, Gaussian white noise with a standard deviation of 5% is superimposed on the actual electricity price versus carbon intensity curve as a prediction error.

[0200] Option 4: Lyapunov strategy. This uses the aforementioned online drift-weighted algorithm for scheduling without requiring any prior information about the future.

[0201] like Figure 3 As shown, the cumulative net profit trajectory of different schemes within 168 hours is illustrated in Table 1. The overall net profit and unit task profit of the four schemes are shown in Table 1. The non-aggregated mode has the worst economic performance, with an overall net profit of only 195,200 yuan, which is less than half of the optimal scheme under the virtual computing center architecture.

[0202] Table 1. Overall Net Income and Unit Task Revenue under Four Schemes

[0203]

[0204] The lack of unified scheduling and energy storage buffers for distributed computing nodes, coupled with the long-term standby status of many devices, leads to persistently high operating costs. In contrast, the three solutions under the virtual computing center architecture all show significant improved returns. Among them, the Lyapunov strategy proposed in this invention utilizes the buffering and adjustment effect of virtual queues to achieve coordinated arbitrage of electricity and carbon resources without the need for prediction, ultimately achieving a cumulative return of 432,800 yuan. This represents a significant improvement of 0.86% and 2.12% compared to the benchmark strategy and model predictive control, respectively, verifying the superiority of the proposed algorithm in highly stochastic environments.

[0205] like Figure 4 As shown, the results of computing load allocation for four strategies are presented, which intuitively reveal the differences in decision-making logic:

[0206] Non-aggregated mode: In this mode, task allocation is approximately constant, fluctuating passively only with the total load. This static load balancing strategy cannot take advantage of the energy efficiency differences between different computing power clusters, resulting in inefficient resource allocation.

[0207] Baseline Strategy: Exhibits a significant cutoff interval. Due to the use of fixed threshold control, the system immediately stops task allocation whenever the electricity price exceeds 0.8 yuan / kWh, resulting in uniform blank gaps in the load graph and a lack of flexibility.

[0208] Model predictive control exhibits more flexible relative adjustment characteristics. Its shutdown behavior depends on the price difference between the current and future. Therefore, during periods of high prices, if the predicted future electricity price is even higher, model predictive control will continue to operate; conversely, if the predicted future electricity price is lower, model predictive control will continue to operate. However, it can be observed that its load fill rate is significantly lower than the strategy proposed in this paper. This is because the prediction noise causes the algorithm to overestimate future risks and adopt a more conservative admission strategy, thus missing out on some service revenue.

[0209] Lyapunov strategy: demonstrates optimal flexible throughput capability. This is achieved thanks to virtual queues. The algorithm employs a backlog pressure mechanism. When the task backlog is high, the algorithm automatically increases its tolerance for high electricity prices to maintain queue stability. As shown in the figure, this strategy, while ensuring long-term carbon constraints, almost fills all available low-price and parity windows, achieving maximum task throughput and service revenue.

[0210] Figure 5 The instantaneous carbon emission curves for the four scenarios are shown. The average carbon emissions and carbon emissions per unit task under the four scenarios are shown in Table 2.

[0211] Table 2 Average carbon emissions and carbon emissions per unit task under the four scenarios

[0212]

[0213] The non-aggregated mode, lacking a multi-source coordination mechanism, passively fluctuates its carbon emission curve entirely based on grid carbon intensity and workload, frequently exceeding emission limits during peak carbon intensity and workload periods. While benchmark strategies and model predictive control incorporate energy storage regulation, they are often limited by short-sighted decision-making logic or accumulated prediction errors, prioritizing short-term low electricity prices at the expense of inherent high carbon emission risks, resulting in average carbon emissions even higher than the non-aggregated mode. The Lyapunov strategy demonstrates optimal carbon compliance by introducing a virtual carbon deficit queue. The algorithm transforms long-term carbon constraints into real-time virtual carbon price penalties. When cumulative emissions approach their limit, The rapid increase in emissions forced the system to proactively reduce electricity purchases or release low-carbon energy storage. Results showed that this strategy minimized carbon emissions per task (0.0375 kg / task).

[0214] Figure 6 The charging and discharging behavior of the energy storage system under the Lyapunov strategy was demonstrated. The virtual computing center effectively reshaped its net load curve by synergistically shifting computing tasks in time and space with charging and discharging energy storage. During peak electricity price and carbon intensity periods, the virtual computing center provided peak shaving support to the power grid by discharging energy storage and actively suppressing computing demand; during off-peak periods, it charged and concentrated on absorbing backlogged tasks, thus playing a role in valley filling.

[0215] Figure 7 This visually demonstrates the system's service capacity under the strategy outlined in this paper. The bar chart shows that the system maintained consistently high task throughput without any service interruptions. (Task backlog queue) The dynamic changes are as follows. Although there may be temporary backlogs (peaks) of tasks to facilitate electricity price arbitrage, the queue length is always bounded and drops rapidly during troughs, proving that the control algorithm meets the system's strong stability requirements, that is, it ensures limited latency of computing power services while achieving low-carbon economic scheduling.

[0216] In summary, this invention effectively solves the time scale mismatch problem between long-term carbon emission constraints and hourly real-time decision-making, without relying on precise predictions of future environmental information. Case studies demonstrate that this method has significant advantages in strongly stochastic environments, with overall net returns improved by 0.86% and 2.12% compared to the baseline strategy and model predictive control, respectively. Furthermore, leveraging the dynamic compensation mechanism of the virtual carbon deficit queue, the system can capture instantaneous electricity-carbon price arbitrage while strictly meeting long-term average carbon emission constraints. In addition, the virtual computing center exhibits peak-shaving and valley-filling characteristics similar to physical energy storage through the spatiotemporal shifting of computing tasks, effectively achieving the coordinated absorption of computing load and renewable energy.

Claims

1. A low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization, characterized in that, Includes the following steps: Step 1: The virtual computing center, through a distributed computing resource aggregation and management platform, completes the large-scale integration and unified management of heterogeneous distributed computing nodes. Using a normalized computing power method, it transforms the original computing power of different types of computing nodes into equivalent CPU computing power, forming... There are 1 computing power clusters, denoted as _ ... Simultaneously, an energy storage system is provided for the virtual computing center. Based on the virtual computing center's purchase and sale of electricity with the power grid, a comprehensive energy management system for the virtual computing center is constructed. Under the premise of meeting long-term carbon emission constraints and service quality requirements, the comprehensive energy system of the virtual computing center maximizes comprehensive net revenue through the coordinated operation of the electricity market and the carbon market. An optimization objective function is constructed with the goal of maximizing comprehensive net revenue under equal periodicity. The system is then uniformly scheduled in combination with the constraints considered by the comprehensive energy management system of the virtual computing center. The constraints include the energy balance constraint of the comprehensive energy management system of the virtual computing center, the operation constraint of the computing cluster, and the operation constraint of the energy storage system. Step 2: Introduce two corresponding queue variables as the state of the virtual computing center's integrated energy system: a task backlog queue and a virtual carbon deficit queue. These are used to handle the coupled constraints of service quality stability and long-term carbon emission compliance. The task backlog queue represents the amount of unprocessed computing tasks currently accumulated in the virtual computing center, reflecting the system congestion level. The virtual carbon deficit queue is used to represent the satisfaction of long-term carbon emission constraints. Step 3: Based on Lyapunov drift-reward theory, the long-term stochastic optimization problem is decoupled into two deterministic subproblems that can be solved in real time: the computing power resource allocation subproblem and the energy storage electricity-carbon arbitrage subproblem. The optimal scheduling instruction for the current time slot is generated to realize the low-carbon economic scheduling of the virtual computing power center.

2. The low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization according to claim 1, characterized in that: In step one, the virtual computing center aggregates no fewer than 800 distributed computing nodes through the KubeEdge edge computing management platform. The node types include data centers, edge nodes, and personal computers, and the nodes contain different raw computing power. Idle power consumption Maximum power consumption Network bandwidth and running status By using the normalized computing power method, the original computing power of distributed computing power nodes is converted into equivalent CPU computing power: In the formula: Represents a computing power cluster Maximum processing capacity; Represents a computing power cluster polymerization Number of type nodes; for The normalization coefficient of computing power for type nodes; The virtual computing center's integrated energy system, under the premise of meeting long-term carbon emission constraints and service quality requirements, maximizes comprehensive net revenue through the coordinated operation of the electricity market and the carbon market. An optimization objective function is constructed with the goal of maximizing comprehensive net revenue under isocyclic conditions, and its expression is as follows: In the formula: This represents the weekly net revenue of the integrated energy management system for virtual computing centers. Represents the mathematical expectation; Indicates the system in time slots Net income; This represents the system's weekly carbon settlement cost; Indicates the time slot number; This indicates the number of time periods optimized throughout the week; The virtual computing center integrated energy management system in time slots The expression for net income is as follows: In the formula: This indicates that the virtual computing center's integrated energy management system is in time slots. The overall benefits; This indicates that the virtual computing center's integrated energy management system is in time slots. Maintenance costs; The virtual computing center integrated energy management system in time slots The total revenue consists of two parts: revenue from computing power services and revenue from electricity sales, expressed as follows: In the formula: and These represent the revenue from computing power services and electricity sales revenue of the integrated energy management system of the virtual computing center, respectively. and These are the weighted average service price and time slot for computing tasks accepted by the integrated energy management system of the virtual computing center. Electricity price; and The virtual computing center integrated energy management system is located in time slots. Assigned to computing power cluster The workload and the power purchased from the grid; Based on the purchase and sale of virtual computing power centers and power grids, the integrated energy management system of virtual computing power centers operates in time slots. The expression for the operation and maintenance cost is as follows: In the formula: For computing power clusters Basic operational costs for handling a unit of task; The expression for the weekly carbon settlement cost of the virtual computing center's integrated energy management system is as follows: In the formula: Penalties for violating carbon constraints; For time slots The average carbon emission factor of the power grid is affected by the dynamic changes in the proportion of regional renewable energy output; The upper limit set for the long-term average carbon emission rate; Indicates the duration of a unit of time period; The constraints considered in the integrated energy management system of the virtual computing center include energy balance constraints, computing cluster operation constraints, and energy storage system operation constraints, as detailed below: 1) Energy balance constraints of the virtual computing center integrated energy management system: In the formula: Represents a computing power cluster In the time slot Energy consumption; Indicates the energy storage system in time slots The charging power; Indicates the energy storage system in time slots The discharge power; 2) Constraints on the operation of computing clusters: The constraints on the operation of computing clusters include: capacity constraints, task balancing constraints, and energy consumption constraints, as detailed below: The expression for the capacity constraint is as follows: The expression for the task balancing constraint is as follows: In the formula: Indicates time slot The number of tasks completed; This indicates the backlog of tasks in the previous time slot; The energy consumption constraint is expressed as follows: In the formula: Represents a computing power cluster Total energy consumption for processing a unit of task; Represents a computing power cluster The computational energy consumption for processing a unit of task; This indicates that the task is being transmitted to the computing cluster. Transmission energy consumption; Represents a computing power cluster Static power consumption; Represents a computing power cluster Maximum energy consumption; 3) Energy storage system operation constraints: The operational constraints of energy storage systems include: charge / discharge power range constraints, state of charge / discharge constraints, state of charge constraints, energy storage capacity range constraints, and long-term cycling constraints, as detailed below: The expression for the charge / discharge power range constraint is as follows: In the formula: This indicates the maximum charging power of the energy storage system; This indicates the maximum discharge power of the energy storage system; The expression for the charge / discharge state constraint is as follows: The expression for the charge state constraint is as follows: In the formula: Indicates the energy storage system in time slots The amount of stored energy; Indicates the charging and discharging efficiency of the energy storage system; The expression for the energy storage capacity range constraint is as follows: In the formula: This indicates the maximum energy storage capacity of the energy storage system; The expression for a long-running cycle constraint is as follows: In the formula: This indicates the initial energy storage capacity of the energy storage system.

3. The low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization according to claim 2, characterized in that: In step two, two corresponding queue variables are introduced as system states: 1) Task backlog queue : Represents the current accumulated amount of unprocessed computing tasks in the virtual computing center, reflecting the degree of system congestion. Its dynamic evolution equation is: In the formula: Indicates the backlog queue for the next time slot, and keeps it in place. Its strong stability means that it guarantees limited delays in missions; 2) Virtual carbon deficit queue : Used to characterize the satisfaction of long-term carbon emission constraints, its dynamic evolution equation is: In the formula: This represents the virtual carbon deficit queue for the next time slot; Indicates the instantaneous carbon emissions at the current moment; The degree of stockpiling directly reflects the magnitude of carbon default risk; Furthermore, it defines the integrated energy system of the virtual computing center in time slots. The combined state vector is .

4. The low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization according to claim 3, characterized in that: In step three, based on Lyapunov's drift-reward theory, the long-running stochastic optimization problem is decoupled into two subproblems that can be solved in real time: the computing power resource allocation subproblem and the energy storage-carbon arbitrage subproblem. The optimal scheduling instruction for the current time slot is then generated to achieve low-carbon economic scheduling of the virtual computing center; specifically as follows: Construct a quadratic Lyapunov scalar function based on the aforementioned comprehensive state vector. : Define one-step conditional Lyapunov drift The expected change in congestion level of the integrated energy management system for virtual computing centers: To balance queue stability and economic benefits, a single time slot objective function is constructed: In the formula: For control parameters, Physically, it represents the risk appetite of virtual computing center operators; Scaling and expanding the constructed single-slot objective function, and removing the constant term, is equivalent to performing the same operation in each time slot. Minimize the following objective function : The complex multivariate coupled optimization problem is decomposed into two low-complexity linear programming subproblems, as follows: 1) Sub-problem 1: Decision on distribution of computing power services Extracting the objective function In terms of computing power allocation The relevant terms are then transformed into a form that maximizes returns: Define computing power cluster The marginal net income index is: Decision-making logic: If Operating at full capacity; if The task will be temporarily stored in the queue to wait for the price to drop; 2) Sub-problem two: Energy storage electricity-carbon dual arbitrage decision Extracting the objective function The terms related to energy storage power are summarized to obtain the energy storage optimization objective: Define generalized nodal marginal electricity price Its physical meaning is the comprehensive cost per unit of electrical energy; The charge / discharge threshold is an exponentially weighted moving average of the generalized nodal marginal electricity price. As a reference benchmark, its update formula is: Where: attenuation coefficient This corresponds to an equivalent memory window of 20 time slots; It is the exponentially weighted moving average of the marginal electricity price of the previous time slot generalized node; The energy storage charging and discharging decision logic is as follows: when At that time, the energy storage system is charging. , ; when or At that time, the energy storage system discharges. , ; Otherwise, the energy storage system remains in standby mode. ; Ultimately, this enables the energy storage system to automatically capture electricity price differences and carbon price differences, achieving low-carbon economic dispatch of the virtual computing center.

5. A low-carbon economic scheduling system for virtual computing centers based on Lyapunov optimization, used to implement the low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization as described in any one of claims 1-4, characterized in that, The system includes a computing power aggregation management module, a comprehensive energy management module, a queue status management module, and a Lyapunov optimization scheduling module. These modules work together to achieve heterogeneous computing power integration, constraint control, and low-carbon economic scheduling in the virtual computing power center. The computing power aggregation management module is configured to integrate heterogeneous distributed computing power nodes on a large scale and manage them in a unified manner through a distributed computing power resource aggregation management platform. It adopts a normalized computing power method to convert the original computing power of different types of nodes into equivalent CPU computing power to form a schedulable computing power cluster. The integrated energy management module is configured to carry an energy storage system and build a power purchase and sale interaction link between the virtual computing center and the power grid. Under the premise of meeting long-term carbon emission constraints and service quality requirements, and in combination with the needs of coordinated operation of the electricity market and carbon market, it constructs an optimization objective function with the goal of maximizing the comprehensive net income under equal-week conditions. At the same time, it integrates a constraint condition library of virtual computing center energy balance constraints, computing cluster operation constraints, and energy storage system operation constraints to provide a constraint basis for scheduling. The queue status management module is configured to establish and update two system status queue variables in real time: a task backlog queue and a virtual carbon deficit queue. The task backlog queue represents the amount of unprocessed computing tasks currently accumulated in the virtual computing center and reflects the degree of system congestion. The virtual carbon deficit queue represents the satisfaction of long-term carbon emission constraints. The dynamic evolution of the two queues is used to monitor the coupling constraint status of service quality stability and long-term carbon emission compliance in real time. The Lyapunov optimization scheduling module is configured based on the Lyapunov drift-reward theory, decoupling the long-term stochastic optimization problem into a computing resource allocation subproblem and an energy storage-carbon arbitrage subproblem. By constructing a Lyapunov scalar function, defining a one-step conditional Lyapunov drift, and introducing risk preference control parameters, it generates a single-slot optimization objective function and transforms it into a deterministic subproblem that can be solved in real time. It then outputs the optimal scheduling instruction for the current slot, realizing low-carbon economic scheduling of the virtual computing center.

6. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and when the processor executes the computer program, it implements a low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization as described in any one of claims 1-4.

7. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by the processor, they implement the low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization as described in any one of claims 1-4.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a low-carbon economic scheduling method for virtual computing centers based on Lyapunov optimization as described in any one of claims 1-4.

Images