Area-oriented computing and power collaborative scheduling method and system for computing power center cluster

Abstract

The application provides a kind of algorithm electricity coordination scheduling method for regional computing power center cluster, comprising: based on the operation ability of each computing power center and various transaction behaviors respectively constructing local algorithm electricity coupling feasible region and each transaction model, and coupling local algorithm electricity coupling feasible region with each transaction model to construct generalized Nash game model with each computing power center as game subject;Based on the generalized Nash game model, introduce Lagrange multiplier to construct augmented Lagrange function, and based on KKT optimality condition and auxiliary variable to establish distributed equation system, use Paillier encryption combined with ADMM algorithm to solve the equation system iteratively, get the optimal coordination scheduling plan within a day.The application integrates modeling of local algorithm electricity coupling feasible region and multi-model coupling, deeply integrates computing power-power operation mechanism and describes the strong correlation characteristics of multi-agent decision-making, and solves through Paillier encryption combined with ADMM algorithm, avoids the problem that sensitive information is easily leaked in plaintext distributed solution.

Description

Technical Field

[0001] This invention relates to the field of computing power collaborative scheduling, specifically to a computing power collaborative scheduling method and system for regional computing power center clusters. Background Technology

[0002] With the explosive growth of artificial intelligence, large-scale model training and inference, and cloud computing, computing infrastructure has become the core foundation of the digital economy, and the coupling depth between computing load and the power system continues to increase. At the national strategic level, "computing-power synergy" has been incorporated into the core deployment of new infrastructure construction. The Government Work Report clearly proposed to implement new infrastructure projects such as ultra-large-scale intelligent computing clusters and computing-power synergy, and to strengthen the national integrated monitoring and scheduling of computing power. Policies such as the "Special Action Plan for Green and Low-Carbon Development of Data Centers" and the "Action Plan for Accelerating the Construction of a New Power System (2024-2027)" have been successively introduced, clearly requiring that by the end of 2025, a two-way synergy mechanism between computing power and power should be initially formed, and the green electricity ratio of newly built data centers in national hub nodes should exceed 80%. Pilot projects for computing-power synergy have been deployed in computing hubs such as Beijing-Tianjin-Hebei and the Yangtze River Delta, as well as in clean energy-rich areas such as Qinghai and Xinjiang.

[0003] From the perspective of industry operation characteristics, the cluster of computing centers in the region has evolved from traditional passive high-energy-consuming loads into "computing-power coupling nodes" with capabilities in computing processing, energy consumption regulation, energy storage coordination, green electricity consumption, and grid-friendly interaction. Different computing centers exhibit significant heterogeneity in task arrival characteristics, offline task transferability, server dynamic voltage and frequency adjustment capabilities, renewable energy configuration, private energy storage capacity, and electricity price response characteristics. The region naturally possesses the potential for spatiotemporal complementarity in load, wind and solar power output, and energy storage resources, and has the engineering foundation to achieve collaborative optimization through peer-to-peer (P2P) electricity trading and shared energy storage.

[0004] In existing academic research and engineering practice, the relevant technologies can be divided into two main branches: The first category is energy optimization and scheduling technology for data center clusters. Existing research mostly focuses on energy consumption optimization, green electricity substitution, and flexible load scheduling within a single data center. It explores the flexible computing potential of data centers through methods such as online / offline task classification modeling, offline task time-shift constraints, server scheduling based on Dynamic Voltage and Frequency Scaling (DVFS), and Power Usage Effectiveness (PUE) load aggregation. Some studies have further explored scheduling methods for data centers participating in P2P electricity trading and shared energy storage leasing, but three core shortcomings remain: First, the coupled competitive relationship among multiple computing centers within a regional cluster is insufficiently characterized, failing to fully consider the coupling between shared energy storage leasing prices and total system demand, and the strong correlation characteristics of multi-agent decision-making resulting from the bilateral power conservation of P2P transactions. Second, the unique characteristics of IT-physical coupling of computing centers are not deeply integrated with energy trading and energy storage leasing in modeling, making it difficult to adapt to the actual operational needs of computing-electricity collaboration. Third, most studies employ centralized or plaintext distributed solutions, failing to consider data privacy protection issues during multi-agent collaboration.

[0005] The second category is privacy protection technologies for shared energy storage and P2P transactions. To address the privacy leakage risks in multi-party collaboration, existing research has proposed a game theory modeling method based on Generalized Nash Equilibrium (GNE). This method uses variational equilibrium and Karush-Kuhn-Tucker (KKT) conditions to equivalently reconstruct the game problem into a convex optimization problem, and employs the Alternating Direction Method of Multipliers (ADMM) for distributed solution, reducing the direct transmission of original sensitive parameters. However, existing technologies still have significant shortcomings: First, privacy protection methods are mostly designed for residents and ordinary industrial and commercial producers and consumers, failing to consider the unique load mechanisms of computing centers such as task scheduling and server operation, and thus cannot be directly adapted to computing-electricity collaboration scenarios. Second, the simple plaintext ADMM distributed solution still carries the risk of inferring the cost structure, marginal preferences, and operational characteristics of participating entities through intermediate information such as iterative residuals, leased capacity iterations, and transaction power iterations.

[0006] In summary, there is currently a lack of a unified scheduling method for regional computing power center clusters that integrates computing load mechanism modeling, multi-entity coupled game theory, shared energy storage and P2P transaction collaboration, and full-process privacy protection, making it difficult to support the large-scale implementation of the computing power collaboration strategy. Summary of the Invention

[0007] To address the shortcomings of existing technologies in collaborative scheduling of regional computing power center clusters, such as insufficient characterization of multi-entity coupling relationships, disconnect between computing power characteristics and energy trading modeling, and easy leakage of sensitive information during collaborative solution, this invention proposes a computing-electricity collaborative scheduling method for regional computing power center clusters. This method is applied to a regional computing-electricity collaborative interaction architecture primarily composed of P2P trading centers, shared energy storage power stations, and multiple computing power centers, and includes: Based on the operational capabilities of each computing center, the P2P transaction behavior between each computing center, the leasing behavior of each computing center to shared energy storage power stations, and the electricity purchase and sale behavior of each computing center to the main grid, a local computing-electricity coupling feasible domain and a transaction model of multiple transaction types are constructed respectively. The local computing-electricity coupling feasible domain is coupled with each transaction model to construct a generalized Nash game model with each computing center as the main player. Based on the generalized Nash game model, an augmented Lagrange function is constructed by introducing Lagrange multipliers, and a distributed solution system of equations is established based on the KKT optimality condition and auxiliary variables. The system of equations is solved iteratively with distributed privacy protection using Paillier encryption combined with the ADMM algorithm to obtain the optimal intraday cooperative scheduling plan. The regional computing center cluster is scheduled based on the optimal collaborative scheduling plan.

[0008] Preferably, the transaction model includes a private energy storage model, a shared energy storage capacity leasing model, a P2P directed transaction model, and a main grid power purchase and sale model.

[0009] Preferably, the transaction model, which constructs a local computing-power coupling feasible domain and multiple transaction types based on the operational capabilities of each computing center, P2P transaction behavior between computing centers, leasing behavior of each computing center from shared energy storage power stations, and electricity purchase and sale behavior of each computing center from the main grid, includes: Based on the computing power task scheduling capability and server scheduling capability of each computing power center, calculate the task arrival constraint, offline task time shift constraint, server processing capability constraint and DVFS level power consumption of each computing power center. Based on the task arrival constraint, offline task time shift constraint, server processing capability constraint and DVFS level power consumption, form the local computing power coupling feasible domain of each computing power center. The private energy storage usage cost of each computing center is calculated based on the charging and discharging power of the private energy storage configured in each computing center during the scheduling period, and a private energy storage model is constructed with the private energy storage usage cost as the objective function. Based on the leasing behavior of each computing power center to the shared energy storage power station, and according to the leasing capacity of the computing power center and the competing computing power center to the shared energy storage power station and the charging and discharging power of the shared energy storage power station during the scheduling period, the shared energy storage leasing cost and shared energy storage usage cost of each computing power center are calculated respectively. Taking the shared energy storage leasing cost and shared energy storage usage cost as the objective function, and the shared energy storage charging and discharging power constraint, the shared energy storage charged energy constraint and the shared energy storage total capacity constraint as the constraint conditions, a shared energy storage capacity leasing model is constructed. Based on the P2P transaction behavior between various computing power centers, the full-cycle cost of P2P transactions is calculated according to the line usage cost and transaction electricity price between different computing power centers. The full-cycle cost of P2P transactions is used as the objective function and the P2P bilateral power conservation constraint is used as the constraint condition to construct a P2P directed transaction model. Based on the electricity purchase and sale behavior of each computing center to the main grid, the electricity purchase and sale cost of the main grid is calculated according to the electricity purchased and sold by each computing center to the main grid. The main grid electricity purchase and sale model is constructed with the electricity purchase and sale cost as the objective function and the electricity purchase and sale power constraint as the constraint condition.

[0010] Preferably, the step of coupling the local computing power coupling feasible domain with each transaction model to construct a generalized Nash game model with each computing power center as the game subject includes: Based on the local computing power coupling feasible domain and the coupling relationship between each model, local decision variables and coupled decision variables of each computing power center are extracted. The full-cycle operating cost of each computing power center is obtained by summing the cost determined by the local decision variables and the cost determined by the coupled decision variables. Each computing power center is taken as a game subject, and the optimization objective function is to minimize the full-cycle operating cost of each game subject. A generalized Nash game model is constructed that couples the game subjects with the local computing power coupled feasible domain.

[0011] Preferably, the step of constructing an augmented Lagrange function by introducing Lagrange multipliers based on the generalized Nash game model, and establishing a distributed solution system of equations based on the KKT optimality conditions and auxiliary variables, includes: Based on the generalized Nash game model, an augmented Lagrange function is constructed by introducing Lagrange multipliers, and a distributed solution system of equations is established based on the KKT optimality condition. By introducing auxiliary variables, the P2P bilateral power conservation constraints and the shared total energy storage capacity constraints are rewritten as consistency constraints with the original coupling variables, and the equation set is reconstructed into a distributed solution equation set with a two-block structure adapted to the ADMM algorithm.

[0012] Preferably, the step of using Paillier encryption combined with the ADMM algorithm to perform distributed privacy-preserving iterative solution of the equation system to obtain the intraday optimal cooperative scheduling plan includes: Initialize the iteration parameters and generate a random mask factor in each iteration; Based on the random masking factor, the original coupling variable and the auxiliary variable, a consistent intermediate quantity is constructed, and the consistent intermediate quantity is then co-computed with Paillier encryption using ciphertext. Based on the encrypted state, the original coupled variables, auxiliary variables and dual variables are iteratively updated sequentially, and the original residuals and dual residuals are calculated. Based on the original residual and the dual residual, the iterative convergence is determined, and the parameters corresponding to the iterative convergence are taken as the intraday optimal scheduling parameters. The optimal intraday collaborative scheduling plan is obtained based on the intraday optimal scheduling parameters.

[0013] Preferably, the collaborative scheduling plan includes: a local offline task time-shifting plan, a server DVFS allocation plan, a private energy storage charging and discharging plan, a shared energy storage charging and discharging plan, a P2P transaction plan, a main grid power purchase and sales plan, and a shared energy storage total capacity allocation and execution plan.

[0014] Based on the same inventive concept, this invention also provides a computing power collaborative scheduling system for regional computing power center clusters, comprising: The model building module is used to construct local computing and electricity coupling feasible domains and transaction models of multiple transaction types based on the operational capabilities of each computing power center, P2P transaction behavior between each computing power center, leasing behavior of each computing power center to shared energy storage power stations, and electricity purchase and sale behavior of each computing power center to the main grid. The local computing and electricity coupling feasible domains are coupled with each transaction model to construct a generalized Nash game model with each computing power center as the main player. The reconstructed solution module is used to construct an augmented Lagrange function based on the generalized Nash game model by introducing Lagrange multipliers, and to establish a distributed solution system of equations based on the KKT optimality conditions and auxiliary variables. The system of equations is solved iteratively with distributed privacy protection using Paillier encryption combined with the ADMM algorithm to obtain the optimal collaborative scheduling plan for the day. The regional computing center cluster is then scheduled based on the optimal collaborative scheduling plan.

[0015] Preferably, the model building module includes: The feasible domain construction unit is used to calculate the task arrival constraints, offline task time shift constraints, server processing capacity constraints, and DVFS tier power consumption of each computing power center based on the computing power task scheduling capacity and server scheduling capacity of each computing power center, and to form the local computing power coupling feasible domain of each computing power center based on the task arrival constraints, offline task time shift constraints, server processing capacity constraints, and DVFS tier power consumption. The private energy storage model construction unit is used to calculate the private energy storage usage cost of each computing center based on the charging and discharging power of the private energy storage configured in each computing center during the scheduling period, and to construct a private energy storage model with the private energy storage usage cost as the objective function. The shared energy storage capacity leasing model construction unit is used to calculate the shared energy storage leasing cost and shared energy storage usage cost of each computing power center based on the leasing behavior of each computing power center to the shared energy storage power station, according to the leasing capacity of the computing power center and the competing computing power center to the shared energy storage power station and the charging and discharging power of the shared energy storage power station during the scheduling period. The shared energy storage leasing cost and shared energy storage usage cost are used as objective functions, and the shared energy storage charging and discharging power constraint, the shared energy storage charged energy constraint and the shared energy storage total capacity constraint are used as constraints to construct the shared energy storage capacity leasing model. The P2P directed transaction model construction unit is used to calculate the full-cycle cost of P2P transactions based on the P2P transaction behavior between various computing power centers, according to the line usage cost and transaction electricity price between different computing power centers, and to construct a P2P directed transaction model with the full-cycle cost of P2P transactions as the objective function and the P2P bilateral power conservation constraint as the constraint condition. The main grid power purchase and sale model construction unit is used to calculate the main grid power purchase and sale cost based on the power purchase and sale behavior of each computing power center to the main grid, and to construct the main grid power purchase and sale model with the main grid power purchase and sale cost as the objective function and the power purchase and sale constraint as the constraint condition. The generalized Nash game model construction unit is used to extract local decision variables and coupled decision variables for each computing power center based on the coupling relationship between the local computing power coupled feasible domain and each model. The unit calculates the full-cycle operating cost of each computing power center by summing the costs determined by the local decision variables and the costs determined by the coupled decision variables. Each computing power center is taken as a game subject, and the optimization objective function is to minimize the full-cycle operating cost of each game subject. The unit constructs a generalized Nash game model that is coupled and related to each game subject and the local computing power coupled feasible domain.

[0016] Preferably, the reconstruction solution module includes: The reconstruction unit is used to construct an augmented Lagrange function by introducing Lagrange multipliers based on the generalized Nash game model, and to establish a distributed solution system based on the KKT optimality conditions; it introduces auxiliary variables to rewrite the P2P bilateral power conservation constraints and the shared energy storage total capacity constraints into consistency constraints with the original coupling variables, and reconstructs the equation system into a distributed solution system with a two-block structure adapted to the ADMM algorithm; The solution unit initializes the iteration parameters and generates a random mask factor in each iteration. Based on the random mask factor and the original coupled variables and auxiliary variables, a consistent intermediate quantity is constructed. Paillier encryption is used to perform encrypted collaborative computation on the consistent intermediate quantity. The original coupled variables, auxiliary variables, and dual variables are iteratively updated sequentially under encrypted conditions, and the original residual and dual residual are calculated. Iteration convergence is determined based on the original residual and dual residual, and the parameters corresponding to the iteration convergence are used as the intraday optimal scheduling parameters. The intraday optimal collaborative scheduling plan is obtained based on the intraday optimal scheduling parameters.

[0017] In another aspect, the present invention also provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, a computing and power collaborative scheduling method for regional computing center clusters as described above is implemented.

[0018] In another aspect, the present invention also provides a readable storage medium on which an executable program is stored, which, when executed, implements a computing and power collaborative scheduling method for a regional computing center cluster as described above.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a computing-power collaborative scheduling method for regional computing center clusters, comprising: constructing local computing-power coupled feasible domains and transaction models for multiple transaction types based on the operational capabilities of each computing center, P2P transaction behavior between computing centers, leasing behavior of each computing center from shared energy storage power stations, and electricity purchase and sale behavior of each computing center from the main grid; coupling the local computing-power coupled feasible domains with each transaction model to construct a generalized Nash game model with each computing center as the game subject; based on the generalized Nash game model, introducing Lagrange multipliers to construct augmented Lagrange functions, and establishing a distributed solution system of equations based on KKT optimality conditions and auxiliary variables; using Paillier encryption combined with the ADMM algorithm to perform distributed privacy-preserving iterative solution of the equations to obtain the intraday optimal collaborative scheduling plan; and scheduling the regional computing center cluster based on the optimal collaborative scheduling plan. This invention integrates local computation-power coupling feasible domain and multi-model coupling into an integrated modeling approach, deeply fusing the computing power-power operation mechanism and characterizing the strong correlation characteristics of multi-agent decision-making. At the same time, it avoids the problem of easy leakage of sensitive information in plaintext distributed solving by using Paillier encryption combined with ADMM algorithm. Attached Figure Description

[0020] Figure 1 This is a flowchart of a computing and power collaborative scheduling method for regional computing center clusters according to the present invention; Figure 2 This is a schematic diagram of a regional computing and power collaborative interaction architecture according to the present invention; Figure 3 This is a schematic diagram illustrating the specific process of a computing and power collaborative scheduling method for a regional computing center cluster according to the present invention. Figure 4 This is a schematic diagram of an electronic device for a computing and power collaborative scheduling method for a regional computing center cluster according to the present invention. Detailed Implementation

[0021] This invention proposes a computing-electricity collaborative scheduling method, system, equipment, and medium for regional computing power center clusters, specifically a collaborative scheduling method for shared energy storage and point-to-point power trading privacy protection for regional computing power center clusters. This method focuses on a computing-power-coordinated system comprised of multiple computing centers, shared energy storage operators, and a P2P transaction coordination center within a region. First, it constructs a local computing-power-coordinated feasible region for each computing center, considering task arrival volume, offline task time shift, server processing capacity, and DVFS tier power consumption. Then, it couples this local feasible region with shared energy storage capacity leasing, P2P directed transactions, main grid power purchase and sales, and private energy storage models to construct a generalized Nash game model where both the objective function and feasible region are influenced by other entities. Further, based on variational equilibrium, it is equivalently reconstructed into a convex quadratic programming problem, and an auxiliary variable is introduced to form a dual-block coupling structure adapted for ADMM solution. Finally, a selective Paillier homomorphic encryption and random mask security computing mechanism is constructed for P2P transaction power, shared energy storage capacity leasing variables, auxiliary variables, and consistency residuals, enabling each computing center to complete distributed collaborative scheduling without disclosing task parameters, server operating parameters, PUE parameters, and cost preferences. The specific embodiments of this invention are further described in detail below with reference to the accompanying drawings.

[0022] Example 1: A computing and power collaborative scheduling method for regional computing center clusters, such as Figure 1 As shown, it includes: Step 1: Based on the operational capabilities of each computing power center, the P2P transaction behavior between each computing power center, the leasing behavior of each computing power center to the shared energy storage power station, and the electricity purchase and sale behavior of each computing power center to the main grid, construct a local computing power coupling feasible domain and a transaction model for multiple transaction types, and couple the local computing power coupling feasible domain with each transaction model to construct a generalized Nash game model with each computing power center as the main player. Step 2: Based on the generalized Nash game model, an augmented Lagrange function is constructed by introducing Lagrange multipliers, and a distributed solution system of equations is established based on the KKT optimality condition and auxiliary variables. The system of equations is solved iteratively with distributed privacy protection using Paillier encryption combined with the ADMM algorithm to obtain the optimal collaborative scheduling plan for the day. The regional computing center cluster is then scheduled based on the optimal collaborative scheduling plan.

[0023] In this embodiment, the computing power collaborative scheduling method is applied to a regional computing power collaborative interaction architecture mainly composed of P2P trading centers, shared energy storage power stations, and multiple computing power centers, such as... Figure 2As shown, in this architecture, the local scheduling terminal of the computing power center retains the online task arrival volume, offline task queue, DVFS parameters, PUE parameters, task time-shift penalty coefficient, and local cost function; the shared energy storage operation terminal only maintains the total power capacity, total energy capacity, leasing price parameters, and shared energy storage auxiliary variables; the P2P transaction coordination terminal only maintains the bilateral power conservation constraints and P2P auxiliary variables. The three types of entities only exchange coupled intermediate quantities processed by encryption or random masking, and do not exchange sensitive plaintext data locally at the computing power center. The transaction models include: a private energy storage model, a shared energy storage capacity leasing model, a P2P directed transaction model, and a mains grid power purchase and sale model. Figure 3 The specific steps of the computer-computer collaborative scheduling method are further explained as shown.

[0024] Specifically, in step 1, the construction of a local computing-power coupling feasible domain and a transaction model for multiple transaction types based on the operational capabilities of each computing center, P2P transaction behavior between computing centers, leasing behavior of each computing center to shared energy storage power stations, and electricity purchase and sale behavior of each computing center to the main grid includes: Based on the computing power task scheduling capability and server scheduling capability of each computing power center, calculate the task arrival constraint, offline task time shift constraint, server processing capability constraint and DVFS level power consumption of each computing power center. Based on the task arrival constraint, offline task time shift constraint, server processing capability constraint and DVFS level power consumption, form the local computing power coupling feasible domain of each computing power center. The private energy storage usage cost of each computing center is calculated based on the charging and discharging power of the private energy storage configured in each computing center during the scheduling period, and a private energy storage model is constructed with the private energy storage usage cost as the objective function. Based on the leasing behavior of each computing power center to the shared energy storage power station, and according to the leasing capacity of the computing power center and the competing computing power center to the shared energy storage power station and the charging and discharging power of the shared energy storage power station during the scheduling period, the shared energy storage leasing cost and shared energy storage usage cost of each computing power center are calculated respectively. Taking the shared energy storage leasing cost and shared energy storage usage cost as the objective function, and the shared energy storage charging and discharging power constraint, the shared energy storage charged energy constraint and the shared energy storage total capacity constraint as the constraint conditions, a shared energy storage capacity leasing model is constructed. Based on the P2P transaction behavior between various computing power centers, the full-cycle cost of P2P transactions is calculated according to the line usage cost and transaction electricity price between different computing power centers. The full-cycle cost of P2P transactions is used as the objective function and the P2P bilateral power conservation constraint is used as the constraint condition to construct a P2P directed transaction model. Based on the electricity purchase and sale behavior of each computing center to the main grid, the electricity purchase and sale cost of the main grid is calculated according to the electricity purchased and sold by each computing center to the main grid. The main grid electricity purchase and sale model is constructed with the electricity purchase and sale cost as the objective function and the electricity purchase and sale power constraint as the constraint condition.

[0025] In this embodiment, unlike ordinary producer-consumer load, the computing center load is not a given rigid power consumption curve, but an adjustable power load determined by offline task time shift, task arrival volume, server processing capacity, and DVFS tier power consumption. Therefore, this step transforms the computing task processing process into a local computing-power coupling feasible domain that can be invoked for subsequent energy trading and shared energy storage leasing decisions. The computing load processed by the computing center is generally divided into two categories: online tasks and offline tasks. Online tasks have strong real-time constraints and must be completed within the arrival time period; offline tasks are allowed to be shifted across time periods within the optimization cycle, but must satisfy the upper and lower bounds of the shift and the period conservation constraints. To characterize the adjustability of offline tasks in the time dimension, a net offline task shift is introduced, and a penalty term is used to reflect the scheduling cost caused by task delay or reordering.

[0026] The time-shift constraint for offline tasks is:

[0027]

[0028]

[0029] in, Indicates computing power center i During the period t The net amount of offline tasks moved; Indicates the time-shift ratio of offline tasks; Indicates computing power center i During the period t The number of offline tasks arriving; Indicates computing power center i During the period t Offline task arrival count after time-shift adjustment; T This represents the set of scheduling periods.

[0030] The time-shift scheduling cost for offline tasks is expressed by the following formula:

[0031] in, Indicates computing power center i Offline task time-shift scheduling cost; This represents the unit translation penalty coefficient for offline tasks.

[0032] Based on the task type classification, a server scheduling constraint based on DVFS is further constructed. The server operating frequency is discretized into five tiers, with different tiers corresponding to different single server power and service capabilities. To avoid introducing integer variables and control the problem scale, a continuous representation of the equivalent number of servers is used to describe the number of servers deployed in each tier.

[0033] The task arrival constraint is:

[0034]

[0035] in, Indicates computing power center i During the period t The number of online tasks received; This indicates the length of a single scheduling period; in this invention, it is taken as 1 hour. Indicates computing power center i The server is in the slot k Processing capacity; and They represent computing centers respectively. i During the period t Assign online and offline tasks to running gears k The equivalent number of servers; Indicates computing power center i A set of server tiers.

[0036] The server processing capacity constraint is:

[0037]

[0038] in, Indicates computing power center i The total number of servers.

[0039] Based on the server scheduling results above, the data center computing load for each time period can be obtained. The Power Usage Effectiveness (PUE) is used to uniformly calculate the cooling and auxiliary system loads onto the IT load.

[0040] DVFS power consumption is expressed by the following formula:

[0041] in, Indicates computing power center i During the period t Calculated power; Indicates computing power center i The standby power of a single server; Indicates computing power centeri The server is in the slot k Power parameters.

[0042] The total power load of a data center is expressed by the following formula:

[0043] in, Indicates computing power center i During the period t The total power load of the data center, including IT, cooling and auxiliary system loads; Indicates computing power center i During the period t The efficiency of electricity use.

[0044] Shared energy storage is centrally configured by an independent operator and open to multiple data centers within the region. Each computing center within the region can lease its power and energy capacity separately. The energy state of the shared energy storage leased by the computing centers must meet the following constraints: The power constraints for shared energy storage charging and discharging are:

[0045] in, and They represent computing centers respectively. i During the period t The charging and discharging power of leased shared energy storage; Indicates computing power center i Power capacity leased from shared energy storage power stations.

[0046] The energy constraint for shared energy storage charge is:

[0047]

[0048]

[0049] in, Indicates computing power center i Leased shared energy storage during the period t The final charged energy; Indicates computing power center i Leased shared energy storage during the period t -1 is the energy of the charged body at the end; and These represent the charging efficiency and discharging efficiency of shared energy storage, respectively. and These represent the lower and upper limits of the shared energy storage state of charge, respectively. Indicates computing power centeri Energy capacity leased from shared energy storage power stations; Indicates computing power center i Leasing shared energy storage during dispatch periods T The final charged energy; Indicates computing power center i The leased shared energy storage is the charged energy in the initial period.

[0050] The total capacity constraint for shared energy storage is:

[0051]

[0052] in, and These represent the upper limit of the total available power capacity and the upper limit of the total available energy capacity of shared energy storage, respectively. This represents the set of computing centers within a region.

[0053] The sum of the shared energy storage power capacity and energy capacity leased by all computing centers shall not exceed the total available capacity of shared energy storage.

[0054] To reflect the changing rental prices of shared energy storage capacity and energy capacity as aggregate demand increases, a linear pricing model is adopted. The rental cost of shared energy storage is expressed by the following formula:

[0055] in, Indicates computing power center i The cost of shared energy storage leasing; , , and These represent the shared energy storage price coefficients; and They represent computing centers respectively. j The power and energy capacity leased from shared energy storage power stations, computing center j With computing center i They are competitors.

[0056] The cost of using shared energy storage is expressed by the following formula:

[0057] in, Indicates computing power center i The cost of using shared energy storage; This indicates the charging and discharging cost of a shared energy storage unit.

[0058] The computing center is equipped with user-side energy storage for intraday energy movement and peak shaving.

[0059] The cost of using private energy storage is expressed by the following formula:

[0060] in, Indicates computing power center i The cost of using private energy storage; Indicates computing power center i The cost of charging and discharging private energy storage units; and They represent computing centers respectively. i Private energy storage during the period t The charging power and discharging power.

[0061] P2P transactions are directed transactions, defined as... For time period t Computing Center i To the computing center j Directed transaction power, For time period t Computing Center j To the computing center i Directed transaction power, agreed upon Indicates computing power center i To the computing center j Electricity sales Indicates computing power center i To the computing center j Purchase electricity.

[0062] The P2P two-sided power conservation constraint is:

[0063] in, and They represent computing centers respectively. i With computing center j The lower and upper bounds of the capacity of the connecting lines between them.

[0064] P2P transactions incur both energy costs corresponding to the transaction electricity price and potential line usage costs. To maintain the linear structure of the model, this paper introduces a line loss auxiliary variable to linearize the relevant absolute value terms.

[0065] The total cost of a P2P transaction cycle is expressed by the following formula:

[0066] in, Indicates computing power center i The total cost of a P2P transaction throughout its entire lifecycle; Indicates computing power center iWith computing center j The cost of using the network for P2P transactions between them; Indicates computing power center i With computing center j Between periods t P2P transaction electricity price; Indicates computing power center i With computing center j Between periods t The auxiliary variable for linearizing the line loss cost of P2P transactions.

[0067] The power purchase and sale between the computing center and the main power grid meets the power purchase and sale constraints:

[0068] in, and They represent computing centers respectively. i During the period t The amount of electricity purchased from and sold to the main power grid.

[0069] The cost of purchasing and selling electricity from the main grid is expressed by the following formula:

[0070] in, Indicates computing power center i The cost of purchasing and selling electricity from the main power grid; and They represent time periods respectively. t The purchase price and sale price of electricity on the main power grid.

[0071] In each time period t Computing Center i The power balance relationship is as follows:

[0072] in, Indicates computing power center i During the period t Local renewable energy output.

[0073] Specifically, in step 1, the coupling of the local computing power coupling feasible region with each transaction model to construct a generalized Nash game model with each computing power center as the game subject includes: Based on the local computing power coupling feasible domain and the coupling relationship between each model, local decision variables and coupled decision variables of each computing power center are extracted. The full-cycle operating cost of each computing power center is obtained by summing the cost determined by the local decision variables and the cost determined by the coupled decision variables. Each computing power center is taken as a game subject, and the optimization objective function is to minimize the full-cycle operating cost of each game subject. A generalized Nash game model is constructed that couples the game subjects with the local computing power coupled feasible domain.

[0074] In this embodiment, each computing center is considered a non-cooperative decision-making entity with an independent cost function and local privacy constraints. The focus is on characterizing the dual coupling effects of shared energy storage capacity leasing price, total shared energy storage capacity constraint, and P2P bilateral power conservation constraint on the objective function and local computing-power coupling feasible region of each computing center. Taking each computing center as an independent game entity and minimizing its own total operating cost over the entire lifecycle as the optimization objective, the decision variables are decomposed into two categories: 1) Local decision variables. It only affects its own operating costs and does not directly couple with the objective functions of other entities, including task scheduling variables, server state variables, private energy storage charging and discharging variables, main grid electricity purchase and sales variables, and P2P transaction variables. 2) Coupled decision variables This directly affects the cost function of all stakeholders, including shared energy storage capacity leasing and energy leasing capacity. Let... For computing power center i Costs determined solely by local decision variables For computing power center i Costs determined by coupled decision variables.

[0075] The cost determined by local decision variables is expressed as follows:

[0076] The cost determined by the coupled decision variables is expressed as follows:

[0077] So, computing center i The cost function under the combined influence of local and coupled decision variables is expressed as follows:

[0078] in, Indicates computing power center i The cost function under the combined influence of local decision variables and coupled decision variables; Indicates computing power center i The sum of all decision variables, i.e., local decision variables and coupled decision variables; This indicates that the region includes computing centers. i The set of all decision variables for all computing centers other than those mentioned above.

[0079] Based on this, and taking into account the coupling relationship between shared energy storage leasing prices and total system leasing demand, as well as the bilateral power conservation constraints of P2P transactions, a coupling relationship between the optimization objectives of each entity and the feasible domain is constructed, forming a generalized Nash game model for the collaborative scheduling of regional computing center clusters. :

[0080] in, Represents the set of constraints for all computing power centers; This represents the set of cost functions for all computing power centers. It fully characterizes the competitive and cooperative interactions among multiple entities.

[0081] At the same time, the individual problems of each participant can be further written into a unified and compact form:

[0082]

[0083] in, Indicates computing power center i The set of constraints; Indicates computing power center i The set of local constraints; Indicates computing power center i A set of coupling constraints affected by other entities.

[0084] Specifically, in step 2, the step of constructing an augmented Lagrange function by introducing Lagrange multipliers based on the generalized Nash game model, and establishing a distributed solution system of equations based on the KKT optimality conditions and auxiliary variables, includes: Based on the generalized Nash game model, an augmented Lagrange function is constructed by introducing Lagrange multipliers, and a distributed solution system of equations is established based on the KKT optimality condition. By introducing auxiliary variables, the P2P bilateral power conservation constraints and the shared total energy storage capacity constraints are rewritten as consistency constraints with the original coupling variables, and the equation set is reconstructed into a distributed solution equation set with a two-block structure adapted to the ADMM algorithm.

[0085] In this embodiment, for the constructed generalized Nash game model, since the objective function of each computing center is a convex function with respect to its own decision variables, all constraints are affine constraints, and the feasible region satisfies the Slater condition, the variational equilibrium of the game can be characterized by a KKT system with shared multipliers. Furthermore, the variational equilibrium of the original game problem is equivalent to the optimal solution of a centralized convex quadratic programming problem. Through this equivalent reconstruction, the generalized Nash game, where the objective functions and feasible regions of each computing center are coupled, can be transformed into a convex quadratic programming problem with unified optimality conditions. This provides a standardized problem structure for subsequently introducing auxiliary variables, constructing ADMM distributed updates, and implementing encrypted consistency computation.

[0086]

[0087] To enable distributed solution using ADMM, this step introduces auxiliary variables for both the P2P trading center and the shared energy storage operator. The P2P bilateral power conservation constraint and the total capacity constraint of shared energy storage are rewritten as consistency constraints between local variables and auxiliary variables, thus forming a bi-block structure adapted to ADMM distributed solution.

[0088] Introduce an auxiliary variable, P2P directed transaction power, into the P2P trading center. The original P2P bilateral power conservation constraints are replaced with the following two sets of constraints:

[0089] in, Indicates time period t Computing Center i To the computing center j Auxiliary variables for the power of directed transactions; Indicates time period t Computing Center j To the computing center i Auxiliary variables for the power of directed transactions.

[0090] Introduce an auxiliary variable regarding the power capacity of leased shared energy storage to the shared energy storage operator. Auxiliary variables for energy capacity of shared energy storage and leasing The original shared total energy storage capacity constraint is replaced with the following constraint:

[0091]

[0092] By introducing auxiliary variables, the above problem can be rewritten as:

[0093] Specifically, in step 2, the distributed privacy-preserving iterative solution of the equation system using Paillier encryption combined with the ADMM algorithm to obtain the intraday optimal cooperative scheduling plan includes: Initialize the iteration parameters and generate a random mask factor in each iteration; Based on the random masking factor, the original coupling variable and the auxiliary variable, a consistent intermediate quantity is constructed, and the consistent intermediate quantity is then co-computed with Paillier encryption using ciphertext. Based on the encrypted state, the original coupled variables, auxiliary variables and dual variables are iteratively updated sequentially, and the original residuals and dual residuals are calculated. Based on the original residual and the dual residual, the iterative convergence is determined, and the parameters corresponding to the iterative convergence are taken as the intraday optimal scheduling parameters. The optimal intraday collaborative scheduling plan is obtained based on the intraday optimal scheduling parameters.

[0094] In this embodiment, the standard ADMM transmits intermediate variables such as leased capacity, transaction power, and consistency residuals during iterations. Third parties can infer sensitive information such as the computing power center's task parameters, load flexibility, cost function, and transaction intent through the iteration sequence. To address the issue of sensitive information leakage, this invention uses Paillier homomorphic encryption and random masking to process the most sensitive consistency intermediate quantities. This ensures that the computing power center, shared energy storage operator, and P2P transaction coordination center only exchange ciphertext or masked intermediate quantities throughout the entire iteration process, without exchanging local sensitive plaintext data. The specific iteration steps are as follows: Step 201: Initialize iteration parameters and key.

[0095] First, initialize the iteration counter. Set penalty parameters Proximal term parameters Maximum number of iterations Original residual threshold and dual residual threshold For the original coupling variable (i.e., P2P directed transaction power) Power capacity of leased shared energy storage and shared energy storage capacity Updates and auxiliary variables (i.e., auxiliary variables for P2P directed transaction power) Auxiliary variables for the power capacity of leased shared energy storage Auxiliary variables for energy capacity of shared energy storage and leasing Update the definition of proximal coefficients separately. Each computing center, shared energy storage operator, and P2P transaction coordination center generates a Paillier public-private key pair, and only publishes the public key for subsequent encrypted communication; the private key is kept locally by its respective node and is not sent externally. Simultaneously, each participating entity independently generates a random masking factor in each iteration. and It is used to randomize the consistency residuals.

[0096] Step 202: Construct a cryptographically consistent intermediate quantity.

[0097] Let the first i The computing center in the first r The original coupling variables of the round iteration are denoted as The corresponding auxiliary variable is denoted as To avoid directly transmitting plaintext residuals, this invention defines a consistent intermediate quantity with a random mask. :

[0098] Accordingly, when updating auxiliary variables, intermediate values ​​are defined in mirror form. :

[0099] The above and Essentially, it is a random mask representation of the ADMM consistency residual. Because it contains private quantities held by both the computing center and the coordination center, neither party can recover the original residual without knowing the other party's random factor, thus blocking the back-reasoning path based on the iterative intermediate quantities.

[0100] Step 203: Paillier-based ciphertext collaborative computation For the aforementioned masked consistency intermediate quantities, this invention does not directly exchange them in the plaintext field, but instead uses Paillier homomorphic encryption for collaborative computation. The main process is as follows: 1) Computing Center i Using local public key Encrypt its own intermediate variables to form ciphertext. .in, This represents a public-key encryption function.

[0101] 2) Shared energy storage operators or P2P transaction coordination centers, exploiting Paillier's homomorphic property, transfer their own auxiliary variables without knowing the plaintext. and random mask factor Integrating ciphertext computation, constructing residual expressions in the ciphertext domain. .

[0102] 3) The encrypted message is sent back to the computing center. i Then, by the computing center i Use local private key Decrypt, and then multiply by the local random factor. This yields the final masked consistency intermediate. .in, This represents the private key decryption function.

[0103] Through the above process, the participants ultimately only obtain the intermediate product with random masking, without directly obtaining the plaintext variables or original residuals of the other party, thus ensuring the confidentiality of locally sensitive parameters throughout the iteration process.

[0104] Step 204: Update local variables of the computing center in encrypted state.

[0105] After completing the secure computation of the consistent intermediate quantity, each computing center solves the local subproblem in parallel. For the first... i For each computing center, the update of all decision variables is written as follows:

[0106] in, Indicates computing power center i In the r +1 round of all decision variables; Indicates computing power center i In the r All decision variables in each iteration; Indicates computing power center i In the r The dual variable of the round iteration; Indicates computing power center i The coefficients of the proximal terms.

[0107] Compared to plaintext ADMM, this invention does not directly use it as a consistency term, but instead uses it obtained through collaborative computation using random masking and homomorphic encryption. This allows for privacy-preserving solutions to local subproblems without altering the fundamental structure of distributed updates.

[0108] Step 205: Update auxiliary variables in encrypted state.

[0109] After each computing center completes its local update, the shared energy storage operator and the P2P trading center update their respective auxiliary variables. For the P2P trading coordination center, the auxiliary variable update is written as follows:

[0110] in, Indicates the first rAuxiliary variables for P2P directed transaction power in +1 iteration; Indicates the first r Auxiliary variables for the power of P2P directed transactions in round-by-round iteration; This is an auxiliary variable representing the P2P directed transaction power maintained by the P2P trading center. The auxiliary variable represents the feasible region that satisfies the P2P bilateral power conservation constraint. The proximal term coefficients of the auxiliary variable representing the power of P2P directed transactions; Indicates the first r The mirror intermediate quantity of the auxiliary variable of the power of P2P directed transactions; Indicates the first r The dual variable of the auxiliary variable of the power of P2P directed transactions.

[0111] For shared energy storage operators, the auxiliary variables of power capacity and energy capacity are uniformly denoted as... Its updates can be summarized as follows:

[0112] in, Indicates the first r Auxiliary variables for the power capacity or energy capacity of leased shared energy storage in +1 iteration; Indicates the first r Auxiliary variables for the power capacity or energy capacity of leased and shared energy storage in round-of-cycle iteration; An auxiliary variable representing the power capacity or energy capacity of the leased shared energy storage maintained by the shared energy storage operator; The feasible region of auxiliary variables that satisfy the total capacity constraint of shared energy storage; The proximal term coefficients of the auxiliary variable representing the power capacity or energy capacity of leased shared energy storage; Indicates the first r The mirror intermediate quantity of the auxiliary variable of power capacity or energy capacity of the leased shared energy storage in the round of iteration; Indicates the first r The dual variable of the auxiliary variable of the power capacity or energy capacity of the leased shared energy storage in the round of iteration.

[0113] The aforementioned auxiliary variable updates still maintain the binary structure of ADMM, but the consistency term no longer depends on the plaintext residual, but on the mirror intermediate quantity obtained from the ciphertext interaction. Replacement.

[0114] Step 206: Update dual variables.

[0115] After updating the local and auxiliary variables, update the dual variables according to the deviation of the consistency constraints. Its general form is written as:

[0116] in, Indicates the first r +1 rounds of iteration dual variables; Indicates the first r The dual variable of the round iteration; Indicates the first r +1 round of original coupling variables; Indicates the first r +1 auxiliary variable for iteration; Indicates the first r The penalty parameter for each iteration, which is different in each iteration, can also be further written as a time-varying form related to the random mask. ,in, and The two participating parties are indicated in the [number]th [month]. r The random mask factor generated in rounds of iteration.

[0117] Step 207: Convergence Judgment and Iteration Termination To determine whether the algorithm has converged, this invention uses a joint stopping test with the original residual and the dual residual. The original residual is defined as... Define the dual residual as .

[0118] When satisfied and Or the number of iterations reaches the maximum value. If the iteration stops, then stop; otherwise, repeat the above steps of calculating the encrypted residual, updating local variables, updating auxiliary variables, and updating dual variables.

[0119] After iterative convergence, each computing center only obtains its local intraday collaborative scheduling plan, including the local offline task time-shift plan, server DVFS allocation plan, private energy storage charging and discharging plan, shared energy storage charging and discharging plan, P2P transaction plan, and main grid power purchase and sales plan. Shared energy storage operators only obtain the total shared energy storage capacity allocation and execution plan, as well as auxiliary variables for power capacity and energy capacity that satisfy the total shared energy storage capacity constraint. The P2P transaction coordination center only obtains auxiliary variables for P2P transaction power that satisfy the bilateral power conservation constraint. No coordinating party can obtain the complete task arrival sequence, server operating parameters, PUE parameters, and local cost function from the computing center.

[0120] This invention, through the construction of a generalized Nash game framework, fully characterizes the coupling between shared energy storage leasing prices and total system demand, as well as the strong correlation characteristics of multi-agent decision-making brought about by the bilateral power conservation of P2P transactions. It achieves global optimization of system operating costs and solves the problem of insufficient characterization of the collaborative behavior of multiple computing centers within regional clusters in existing technologies.

[0121] This invention integrates the unique IT-side characteristics of computing centers, such as online / offline task time-shifting, DVFS server scheduling, and PUE load aggregation, with the power-side characteristics of energy trading and energy storage charging and discharging. It deeply merges the computing power-power operation mechanism, fully exploring the multi-dimensional flexible adjustment potential of computing centers. This solves the core problem that existing privacy protection methods are designed for ordinary producers and consumers and cannot adapt to computing-power collaboration scenarios.

[0122] The ADMM algorithm designed in this invention, featuring a "random mask + Paillier homomorphic encryption" dual privacy protection, solves the problems of easily reverse-engineered iterative intermediate quantities and easily leaked sensitive information in existing plaintext distributed solutions. It also avoids the drawbacks of traditional homomorphic encryption algorithms, such as slow convergence speed and high computational cost. At the privacy protection level, core sensitive information such as task arrival data, load flexibility parameters, cost preferences, and transaction intentions at each computing center are stored and computed locally throughout the entire process, without needing to be uploaded to any third party. All iterative intermediate quantities transmitted across entities are first superimposed with a local private random mask and then encrypted using the Paillier public key. Third parties can only access the encrypted data and cannot reconstruct the original iterative plaintext, nor can they reverse-engineer any sensitive information from the iterative sequence.

[0123] This invention reveals the substitution and complementarity effects between shared energy storage and P2P electricity trading, achieving global optimization of spatiotemporal resources within a region through coordinated scheduling. P2P electricity trading, through spatial energy complementarity, solves the spatiotemporal heterogeneity problem of wind and solar power output from different computing centers; shared energy storage, through time-domain energy transfer, smooths out intraday fluctuations in renewable energy output and computing load, providing capacity support for P2P trading and improving trading matching efficiency.

[0124] Example 2: Based on the same inventive concept, this invention also provides a computing and power collaborative scheduling system for regional computing center clusters, comprising: The model building module is used to construct local computing and electricity coupling feasible domains and transaction models of multiple transaction types based on the operational capabilities of each computing power center, P2P transaction behavior between each computing power center, leasing behavior of each computing power center to shared energy storage power stations, and electricity purchase and sale behavior of each computing power center to the main grid. The local computing and electricity coupling feasible domains are coupled with each transaction model to construct a generalized Nash game model with each computing power center as the main player. The reconstructed solution module is used to construct an augmented Lagrange function based on the generalized Nash game model by introducing Lagrange multipliers, and to establish a distributed solution system of equations based on the KKT optimality conditions and auxiliary variables. The system of equations is solved iteratively with distributed privacy protection using Paillier encryption combined with the ADMM algorithm to obtain the optimal collaborative scheduling plan for the day. The regional computing center cluster is then scheduled based on the optimal collaborative scheduling plan.

[0125] The model building module includes: The feasible domain construction unit is used to calculate the task arrival constraints, offline task time shift constraints, server processing capacity constraints, and DVFS tier power consumption of each computing power center based on the computing power task scheduling capacity and server scheduling capacity of each computing power center, and to form the local computing power coupling feasible domain of each computing power center based on the task arrival constraints, offline task time shift constraints, server processing capacity constraints, and DVFS tier power consumption. The private energy storage model construction unit is used to calculate the private energy storage usage cost of each computing center based on the charging and discharging power of the private energy storage configured in each computing center during the scheduling period, and to construct a private energy storage model with the private energy storage usage cost as the objective function. The shared energy storage capacity leasing model construction unit is used to calculate the shared energy storage leasing cost and shared energy storage usage cost of each computing power center based on the leasing behavior of each computing power center to the shared energy storage power station, according to the leasing capacity of the computing power center and the competing computing power center to the shared energy storage power station and the charging and discharging power of the shared energy storage power station during the scheduling period. The shared energy storage leasing cost and shared energy storage usage cost are used as objective functions, and the shared energy storage charging and discharging power constraint, the shared energy storage charged energy constraint and the shared energy storage total capacity constraint are used as constraints to construct the shared energy storage capacity leasing model. The P2P directed transaction model construction unit is used to calculate the full-cycle cost of P2P transactions based on the P2P transaction behavior between various computing power centers, according to the line usage cost and transaction electricity price between different computing power centers, and to construct a P2P directed transaction model with the full-cycle cost of P2P transactions as the objective function and the P2P bilateral power conservation constraint as the constraint condition. The main grid power purchase and sale model construction unit is used to calculate the main grid power purchase and sale cost based on the power purchase and sale behavior of each computing power center to the main grid, and to construct the main grid power purchase and sale model with the main grid power purchase and sale cost as the objective function and the power purchase and sale constraint as the constraint condition. The generalized Nash game model construction unit is used to extract local decision variables and coupled decision variables for each computing power center based on the coupling relationship between the local computing power coupled feasible domain and each model. The unit calculates the full-cycle operating cost of each computing power center by summing the costs determined by the local decision variables and the costs determined by the coupled decision variables. Each computing power center is taken as a game subject, and the optimization objective function is to minimize the full-cycle operating cost of each game subject. The unit constructs a generalized Nash game model that is coupled and related to each game subject and the local computing power coupled feasible domain.

[0126] The reconstruction solution module includes: The reconstruction unit is used to construct an augmented Lagrange function by introducing Lagrange multipliers based on the generalized Nash game model, and to establish a distributed solution system based on the KKT optimality conditions; it introduces auxiliary variables to rewrite the P2P bilateral power conservation constraints and the shared energy storage total capacity constraints into consistency constraints with the original coupling variables, and reconstructs the equation system into a distributed solution system with a two-block structure adapted to the ADMM algorithm; The solution unit initializes the iteration parameters and generates a random mask factor in each iteration. Based on the random mask factor and the original coupled variables and auxiliary variables, a consistent intermediate quantity is constructed. Paillier encryption is used to perform encrypted collaborative computation on the consistent intermediate quantity. The original coupled variables, auxiliary variables, and dual variables are iteratively updated sequentially under encrypted conditions, and the original residual and dual residual are calculated. Iteration convergence is determined based on the original residual and dual residual, and the parameters corresponding to the iteration convergence are used as the intraday optimal scheduling parameters. The intraday optimal collaborative scheduling plan is obtained based on the intraday optimal scheduling parameters.

[0127] This invention accurately characterizes the coupled competitive relationships among multiple entities, achieves global optimization of system operating costs, deeply integrates the computing power-electricity operation mechanism, and fully explores the multi-dimensional flexible adjustment potential of computing power centers; it constructs a dual privacy protection mechanism to achieve a synergistic balance between solution efficiency and privacy security; and it realizes the synergistic complementarity between shared energy storage and P2P transactions to improve the efficiency of regional resource utilization.

[0128] Example 3: like Figure 4 As shown, the present invention also provides an electronic device, which may be a computer device, a microcontroller device, a smart mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, processor, and transceiver component are connected via a bus; the memory can be used to store executable programs, and an exemplary executable program may include instructions; the processor is used to execute the instructions stored in the memory. The memory can also be used to store data, which can be accessed and / or modified when instructions are executed.

[0129] The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, and it is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the storage medium to realize the corresponding method flow or corresponding function, so as to realize the steps of the computing and power collaborative scheduling method for a regional computing center cluster in the above embodiment.

[0130] Example 4: Based on the same inventive concept, this invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory). This readable storage medium is a memory device within an electronic device used to store programs and data. It is understood that the storage medium here can include both built-in storage media within the electronic device and extended storage media supported by the electronic device. The storage medium provides storage space, which stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more executable programs (including program code). It should be noted that the storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. Loading and executing one or more instructions stored in the storage medium by the processor can implement the steps of a computing-electricity collaborative scheduling method for regional computing center clusters as described in the above embodiments.

[0131] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0132] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0133] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0134] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0135] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A computing-power collaborative scheduling method for regional computing power center clusters, applied to a regional computing-power collaborative interaction architecture mainly composed of P2P trading centers, shared energy storage power stations, and multiple computing power centers, characterized in that... include: Based on the operational capabilities of each computing center, the P2P transaction behavior between each computing center, the leasing behavior of each computing center to shared energy storage power stations, and the electricity purchase and sale behavior of each computing center to the main grid, a local computing-electricity coupling feasible domain and a transaction model of multiple transaction types are constructed respectively. The local computing-electricity coupling feasible domain is coupled with each transaction model to construct a generalized Nash game model with each computing center as the main player. Based on the generalized Nash game model, an augmented Lagrange function is constructed by introducing Lagrange multipliers, and a distributed solution system of equations is established based on the KKT optimality condition and auxiliary variables. The system of equations is solved iteratively with distributed privacy protection using Paillier encryption combined with the ADMM algorithm to obtain the optimal collaborative scheduling plan for the day. The regional computing center cluster is then scheduled based on the optimal collaborative scheduling plan.

2. The method according to claim 1, characterized in that, The transaction models include private energy storage model, shared energy storage capacity leasing model, P2P directed transaction model, and main grid power purchase and sales model.

3. The method according to claim 2, characterized in that, Based on the operational capabilities of each computing center, P2P transactions between computing centers, leasing activities from shared energy storage power stations by each computing center, and electricity purchase and sale activities from the main grid by each computing center, a local computing-power coupling feasible domain and a transaction model for multiple transaction types are constructed, including: Based on the computing power task scheduling capability and server scheduling capability of each computing power center, calculate the task arrival constraint, offline task time shift constraint, server processing capability constraint and DVFS level power consumption of each computing power center. Based on the task arrival constraint, offline task time shift constraint, server processing capability constraint and DVFS level power consumption, form the local computing power coupling feasible domain of each computing power center. The private energy storage usage cost of each computing center is calculated based on the charging and discharging power of the private energy storage configured in each computing center during the scheduling period, and a private energy storage model is constructed with the private energy storage usage cost as the objective function. Based on the leasing behavior of each computing power center to the shared energy storage power station, and according to the leasing capacity of the computing power center and the competing computing power center to the shared energy storage power station and the charging and discharging power of the shared energy storage power station during the scheduling period, the shared energy storage leasing cost and shared energy storage usage cost of each computing power center are calculated respectively. Taking the shared energy storage leasing cost and shared energy storage usage cost as the objective function, and the shared energy storage charging and discharging power constraint, the shared energy storage charged energy constraint and the shared energy storage total capacity constraint as the constraint conditions, a shared energy storage capacity leasing model is constructed. Based on the P2P transaction behavior between various computing power centers, the full-cycle cost of P2P transactions is calculated according to the line usage cost and transaction electricity price between different computing power centers. The full-cycle cost of P2P transactions is used as the objective function and the P2P bilateral power conservation constraint is used as the constraint condition to construct a P2P directed transaction model. Based on the electricity purchase and sale behavior of each computing center to the main grid, the electricity purchase and sale cost of the main grid is calculated according to the electricity purchased and sold by each computing center to the main grid. The main grid electricity purchase and sale model is constructed with the electricity purchase and sale cost as the objective function and the electricity purchase and sale power constraint as the constraint condition.

4. The method according to claim 3, characterized in that, The process of coupling the local computing power coupled feasible region with each transaction model to construct a generalized Nash game model with each computing power center as the main player includes: Based on the local computing power coupling feasible domain and the coupling relationship between each model, local decision variables and coupled decision variables of each computing power center are extracted. The full-cycle operating cost of each computing power center is obtained by summing the cost determined by the local decision variables and the cost determined by the coupled decision variables. Each computing power center is taken as a game subject, and the optimization objective function is to minimize the full-cycle operating cost of each game subject. A generalized Nash game model is constructed that couples the game subjects with the local computing power coupled feasible domain.

5. The method according to claim 4, characterized in that, The process involves constructing an augmented Lagrange function based on the generalized Nash game model, introducing Lagrange multipliers, and establishing a distributed solution system of equations based on KKT optimality conditions and auxiliary variables, including: Based on the generalized Nash game model, an augmented Lagrange function is constructed by introducing Lagrange multipliers, and a distributed solution system of equations is established based on the KKT optimality condition. By introducing auxiliary variables, the P2P bilateral power conservation constraints and the shared total energy storage capacity constraints are rewritten as consistency constraints with the original coupling variables, and the equation set is reconstructed into a distributed solution equation set with a two-block structure adapted to the ADMM algorithm.

6. The method according to claim 5, characterized in that, The process of using Paillier encryption combined with the ADMM algorithm to perform distributed privacy-preserving iterative solution of the equations to obtain the intraday optimal cooperative scheduling plan includes: Initialize the iteration parameters and generate a random mask factor in each iteration; Based on the random masking factor, the original coupling variable and the auxiliary variable, a consistent intermediate quantity is constructed, and the consistent intermediate quantity is then co-computed with Paillier encryption using ciphertext. Based on the encrypted state, the original coupled variables, auxiliary variables and dual variables are iteratively updated sequentially, and the original residuals and dual residuals are calculated. Based on the original residual and the dual residual, the iterative convergence is determined, and the parameters corresponding to the iterative convergence are taken as the intraday optimal scheduling parameters. The optimal intraday collaborative scheduling plan is obtained based on the intraday optimal scheduling parameters.

7. The method according to claim 6, characterized in that, The collaborative scheduling plan includes: local offline task time-shifting plan, server DVFS allocation plan, private energy storage charging and discharging plan, shared energy storage charging and discharging plan, P2P transaction plan, main grid power purchase and sales plan, and shared energy storage total capacity allocation and execution plan.

8. A computing and power collaborative scheduling system for regional computing center clusters, characterized in that, include: The model building module is used to construct local computing and electricity coupling feasible domains and transaction models of multiple transaction types based on the operational capabilities of each computing power center, P2P transaction behavior between each computing power center, leasing behavior of each computing power center to shared energy storage power stations, and electricity purchase and sale behavior of each computing power center to the main grid. The local computing and electricity coupling feasible domains are coupled with each transaction model to construct a generalized Nash game model with each computing power center as the main player. The reconstructed solution module is used to construct an augmented Lagrange function based on the generalized Nash game model by introducing Lagrange multipliers, and to establish a distributed solution system of equations based on the KKT optimality conditions and auxiliary variables. The system of equations is solved iteratively with distributed privacy protection using Paillier encryption combined with the ADMM algorithm to obtain the optimal collaborative scheduling plan for the day. The regional computing center cluster is then scheduled based on the optimal collaborative scheduling plan.

9. An electronic device, characterized in that, include: At least one processor and memory; The memory and processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, a computing and power collaborative scheduling method for regional computing center clusters as described in any one of claims 1 to 7 is implemented.

10. A readable storage medium, characterized in that, It contains an execution program, which, when executed, implements a computing and power collaborative scheduling method for regional computing center clusters as described in any one of claims 1 to 7.

Images