Coupling system modeling method and collaborative optimization method of computing power network and power network

Abstract

The application relates to a kind of computing power network and power network coupling system modeling method and collaborative optimization method, modeling method includes: according to the actual equipment and topological structure of computing power network, the equivalent circuit model of computing power network is constructed, the equivalent circuit model includes computing power equipment, computing power bus and algorithmic power coupling device, computing power equipment includes computing power load, communication link and computing queue, each computing power equipment is accessed to computing power bus through computing power port, and algorithmic power coupling device is connected with computing power bus and power bus respectively;The equivalent circuit model of computing power network is coupled with power network, and the algorithmic power coupling system model is constructed, realizes that computing power resource is aggregated and then equivalent access different power grid node, and the network topology of computing power network and power network is converted into unified graph model.Compared with prior art, the modeling efficiency of large-scale implementation of algorithmic power fusion is improved, and the spatiotemporal flexibility of computing power network can be better stimulated to improve the ability of algorithmic power coupling system to consume new energy.

Description

Technical Field

[0001] This invention relates to the field of collaborative processing technology of computing networks and power networks, and in particular to a modeling method and a collaborative optimization method for coupled systems of computing networks and power networks. Background Technology

[0002] With the explosive growth of data centers in smart cities, smart grids, and autonomous driving, computing resources are evolving towards networking, distribution, and ubiquity. The surge in computing demand inevitably leads to massive energy consumption and carbon emissions, which will drive the "convergence and symbiosis" of computing power and power networks. This will enable the green upgrading of computing power through electricity and empower the digital transformation of the power industry through computing power. To achieve this goal, researchers have conducted studies on computing network modeling, collaborative optimization of computing-power coupled networks, and market mechanisms to promote computing-power convergence.

[0003] Computational network modeling is fundamental to research on the integration of computing and power. Existing literature largely employs algebraic modeling methods centered on computational workload, focusing on characterizing the spatiotemporal migration characteristics of the computational load to establish models of communication links, servers, and other equipment. These methods exhibit significant modeling inconsistencies, leading to problems such as a lack of standardized parameter collection and difficulties in integration with power grid models. To address these issues, some authors have attempted equivalent circuit modeling methods. For example, the paper "Electrical circuit analogy-based maximum latency calculation method of Internet data centers in power-communication network" uses a "circuit analogy" approach, defining concepts such as information resistance, information current, and information voltage. However, this method is primarily used to calculate the maximum latency of data centers. The paper "Aggregated model of datanetwork for the provision of demand response in generation and transmission expansion planning" uses a "virtual circuit" approach, transforming the computing network into a virtual power network composed of virtual loads and virtual generators. However, this method requires bottom-up, layer-by-layer model aggregation, a complex process, and is primarily suitable for power grid planning research.

[0004] In the area of ​​collaborative optimization of computation-power coupled networks, research can be divided into two categories based on whether or not the power grid structure is considered. The first category focuses on detailed modeling of the computational network, simplifying the power system to a single node or only retaining the electricity market environment. The drawback of this approach is its difficulty in reflecting the impact of the computational network on power flow and system supply-demand balance. The second category considers detailed power grid models and flow constraints, enabling accurate analysis of the interaction between the computational network and the power grid. However, this type of model often employs centralized optimization methods, which suffer from computational scalability issues and data center privacy protection problems for complex computation-power coupled networks.

[0005] Computing networks possess both spatiotemporal flexibility and, in particular, the unique ability to instantly migrate power loads without relying on the power grid. To incentivize computing networks to leverage this flexibility, researchers have proposed market mechanisms and profit-sharing strategies, such as node pricing, Nash bargaining, Shapley value, and master-slave game theory. Among these, node pricing, which provides real-time guidance for data centers to achieve spatiotemporal flexibility, has received widespread attention. However, node pricing mechanisms are currently difficult to apply to power distribution networks. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a modeling method and a collaborative optimization method for a coupled computing network and a power network, thereby achieving deep coupling between computing power and power networks.

[0007] The objective of this invention can be achieved through the following technical solutions: A method for modeling a coupled system of computing power networks and power networks, comprising: Based on the actual equipment and topology of the computing power network in the area to be processed, an equivalent circuit model of the computing power network is constructed. The equivalent circuit model includes computing power equipment, computing power bus, and computing-electrical coupling equipment. The computing power equipment includes computing power load, communication link, and computing queue. Each computing power equipment is connected to the computing power bus through a computing power port. The computing-electrical coupling equipment is connected to the computing power bus and the power bus respectively. The equivalent circuit model of the computing power network is coupled with the power network to construct a computing-power coupled system model. This enables the aggregation of computing resources and their equivalent access to different power grid nodes. The network topology of the computing power network and the power network is transformed into a unified graph model for unified scheduling of the computing power network and the power network.

[0008] Furthermore, the computing load is analogous to the electrical load and modeled as a single-port device, with the corresponding set of port variables being... ,in, and These represent batch processing and interactive workloads, respectively. KThe total number of time periods; the calculation expressions for the batch processing and interactive load are: In the formula, for t Real-time batch processing load. for t Real-time interactive load, , For data centers t Total computation requests received during the time period; and These represent the percentages of batch processing and interactive workloads, respectively.

[0009] Furthermore, the communication link is analogous to a power transmission line and modeled as a two-port device, with the corresponding set of port variables being... , representing the batch and interactive loads on port 1 and port 2, respectively; For the communication links within the data center, the lossless transmission constraint is satisfied, and the calculation expression for the lossless transmission constraint is as follows: In the formula, and For port variables closer to the front end; For communication links that are remotely interconnected bidirectionally between data centers, the lossless transmission constraint, migration constraint, and bandwidth constraint must be satisfied; the expression for the migration constraint is: The expression for the bandwidth constraint is: In the formula, This is the upper limit of the bandwidth for a remote, two-way interconnected communication link.

[0010] Remote communication links require additional operational and maintenance costs. For data centers... l For each remote communication link, the following cost function is defined: In the formula, the coefficients This represents the cost factor for calculating the unit load space migration; This represents the fixed cost when the link is enabled.

[0011] Furthermore, the computing queue is analogous to energy storage; storing computing power requests in the queue is analogous to charging, and removing the server load from the queue and processing it is analogous to discharging. The computing queue is modeled as a single-port device, and the corresponding state equation is: In the formula, EThis represents the accumulated computational load in the computation queue. The computational load for the initial period. The computational load for the termination period; , Represent t Batch load entering and leaving the queue during a specific time period. for t Batch processing load at any given time. for t Interactive load at any given moment; The constraints satisfied by the computation queue include: In the formula, To calculate the maximum queuing time for the load; The expression for the cost function of the computation queue is: In the formula, To calculate the cost function of the queue, This is the penalty coefficient.

[0012] Furthermore, the computing-power coupling device is a server. In the computing-power coupling device model, servers of the same model within the data center are aggregated into a cluster, which is a server cluster. The model expression for the server cluster is: In the formula, for t Power consumption of the server cluster during the time period. and These are the peak power and standby power of a single server, respectively. The uptime rate of the server cluster. This refers to the computational load density of the server cluster. The constraints of the server cluster include: In the formula, express t Total computational load for processing each time period For data center power utilization efficiency, This refers to the total power, including auxiliary losses such as cooling. and These represent the upper and lower limits of server utilization. This represents the average processor utilization of the server cluster.

[0013] The cost function of the server cluster is defined as: In the formula: This is a server start-up and shutdown cost coefficient used to penalize frequent and large fluctuations in server uptime.

[0014] Furthermore, the computing power port and the power port are connected to the computing power bus and the power bus, respectively. The variables associated with the computing power port and the power port are divided into flow variables and potential variables. The flow variables include active power, reactive power, batch processing load, and interactive load. The flow variables satisfy the node conservation constraint, the expression of which is: In the formula, Port Stream variables; To connect to the bus n The port set; The potential variables include voltage and phase angle, and the potential variables satisfy a consistency constraint, the expression of which is: In the formula, To connect to the bus n The potential variable of the port.

[0015] Furthermore, the vertex set of the unified graph model is divided into a left set and a right set. The left set represents devices, including a set of computing power devices, a set of power devices, and a set of computing-power coupling devices. The right set represents nodes, including a set of computing power buses and a set of power buses. The edge set of the unified graph model includes computing power edge set and power edge set. Each edge connects to a device at one end and a node at the other end.

[0016] This invention also provides a method for the collaborative optimization of a coupled system of computing power networks and power networks, comprising: A model of a coupled computing network and a power network is obtained by using the above-described method for modeling such a system. With the goal of maximizing social welfare, a collaborative optimization problem of computing and power coupling system is constructed, and the power purchase of the distribution network and the computing power time migration load and computing power space migration load of each data center are determined by solving the problem. To achieve Pareto optimal redistribution of benefits, a logarithmic Nash social welfare function is defined, and a Nash negotiation problem is constructed. The electricity purchased by the distribution network and the computing power time migration load and computing power space migration load of each data center are substituted into the Nash negotiation problem to obtain a convex optimization problem with only the incentive price as the decision variable, thereby solving the problem to obtain the calculation result of the incentive price.

[0017] Furthermore, the expression for the cooperative optimization problem of the computational-electrical coupling system is as follows: In the formula, For the entire set of devices in the computer-electrical coupling system, Represents the cost function. This indicates the feasible domain for operation.

[0018] Furthermore, the expression for the Nash negotiation problem is: In the formula, For the Nash negotiation function, To incentivize price caps, To pay CSP i Spatial migration incentive price, and These refer to the utility of the distribution network and each CSP after the redistribution of benefits. and Respectively, power grid and CSP i Its utility when cooperation is not achieved, i.e., the point at which negotiations break down.

[0019] Compared with the prior art, the present invention has the following advantages: (1) This invention utilizes the equivalent circuit model of computing power network to solve the problem of model heterogeneity between computing power and power network, constructs a computing power-power network centered on the power grid, establishes a mapping relationship between the spatiotemporal flexibility of computing power and the flexibility of power regulation, improves the modeling efficiency of large-scale implementation of computing-power integration, and can better incentivize computing power network to give full play to its spatiotemporal flexibility and enhance the ability of computing-power coupling system to absorb new energy.

[0020] (2) Based on the computing power-power network, the present invention can derive a unified graph computing scheme for computing and power co-optimization, form a distributed algorithm with linear complexity, and naturally perform network segmentation according to the computing and power management interface, thereby improving the computing efficiency of the computing and power coupling system.

[0021] (3) The present invention constructs a revenue distribution scheme for computing power migration based on the Nash negotiation mechanism, which can stimulate the spatial flexibility of data centers without the implementation of node electricity prices, improve the ability of computing and power coupling systems to absorb new energy sources, and ensure a win-win situation for the power grid and data centers. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of an equivalent circuit model of a computing power network provided in an embodiment of the present invention; Figure 2 This is an original network topology diagram provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a unified graphical model of a computing-electrical coupling system provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a segmented network privacy computing mechanism provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0024] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0025] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0026] Example 1 like Figure 1 As shown, this embodiment provides a method for modeling a coupled system of computing power networks and power networks, including: S1: Based on the actual equipment and topology of the computing power network in the area to be processed, construct an equivalent circuit model of the computing power network. The equivalent circuit model includes computing power equipment, computing power bus and computing-electrical coupling equipment. The computing power equipment includes computing power load, communication link and computing queue. Each computing power equipment is connected to the computing power bus through the computing power port. The computing-electrical coupling equipment is connected to the computing power bus and the power bus respectively. S2: Based on the equivalent circuit model of the computing power network and its coupling with the power network, construct a computing-power coupling system model to realize the equivalent access of the aggregated computing power resources to different power grid nodes, and transform the network topology of the computing power network and the power network into a unified graph model for unified scheduling of the computing power network and the power network.

[0027] The details are as follows: 1. Equivalent circuit model of computing power network Currently, computing power and power networks differ significantly in their mathematical representations, and data center modeling methods vary. This model heterogeneity and lack of standardization hinder the large-scale implementation of computing-power convergence. To address this, this invention proposes an equivalent circuit model for computing power networks, such as... Figure 1As shown, this model is a simplified and pseudo-circuit representation of the actual devices and topology in a computing power network. It aims to clearly quantify and map the spatiotemporal flexibility of the computing power network into the flexibility of power regulation, and to form an intuitive and standardized representation scheme for the computing power model.

[0028] exist Figure 1 In this invention, the main components of a computing power network are abstracted as computing power devices, a computing power bus, and computing-power coupling devices. The computing power devices include computing load, communication links, and computing queues. Each computing power device is connected to the computing power bus through a computing power port; the computing-power coupling device refers to a server, which serves as both a computing power port and a power port. A single cloud service provider (CSP) can operate multiple data centers distributed across different geographical locations and interconnected by communication links.

[0029] Computing workloads can be categorized into batch processing and interactive processing, offering flexibility in time-shifting and spatial migration, respectively. When a data center front-end receives a computing power request, if it's a batch processing workload, it's handled locally; if it's an interactive workload, it can be migrated to other data centers within the same CSP for processing as needed. To reflect this computing power scheduling function, Figure 1 The system sets up a front-end computing power bus for cross-space allocation of computing resources and a back-end computing power bus for supporting local caching of computing tasks. However, it should be noted that the equivalent circuit model of this invention is only used to describe the external capabilities and business constraints of the computing power network and does not involve the actual computing power scheduling algorithm.

[0030] 1.1 Computing Power Equipment Model The power ports of the simulated power equipment are defined, with each computing device containing at least one computing port. For any computing device... d This invention adopts an object-oriented modeling approach, using... This represents its set of port variables, and specifies that the incoming device symbol is positive; using Its cost function is expressed by... This represents its feasible operating domain. The models for each computing device are as follows.

[0031] 1) Computational load can be compared to electrical load, and is a single-port device. Its port variable set is as follows: ,in K The total number of time periods is 24 in this invention; , These represent batch processing and interactive workloads, respectively (subscripts T and S represent temporal and spatial flexibility, respectively): (1) In the formula: ; For data centers t Total computation requests received during the time period; , This represents the proportion of the two types of load.

[0032] 2) The communication link can be compared to a power transmission line, and is a two-port device. The set of port variables is... , representing the batch and interactive loads on port 1 and port 2, respectively.

[0033] For internal communication links within the data center, let the port variable closest to the front end be... , It satisfies the following lossless transmission constraints: (2) Data centers are interconnected bidirectionally via remote communication links. In addition to meeting the requirement of lossless transmission, the following constraints must also be met: (3) In the formula: the first constraint ensures that the batch processing load does not migrate across nodes; B This represents the upper limit of bandwidth for remote communication links.

[0034] Remote communication links require additional operational and maintenance costs. For data centers... l For each remote communication link, the following cost function is defined: (4) Where: coefficient This represents the cost factor for calculating the unit load space migration; This represents the fixed cost when the link is enabled.

[0035] 3) The computation queue can be likened to energy storage: "charging" represents storing computing power requests in the queue, while "discharging" represents the server retrieving the load from the queue and processing it. This invention models the computation queue as a single-port device, with the port variable set as follows: Note that, unlike energy storage, the computational queue model of this invention allows the aforementioned "charging" and "discharging" processes to occur simultaneously.

[0036] Based on the above description, the state equation of the queue is calculated as follows: (5) In the formula: E This represents the accumulated computational load in the computation queue; , Represent t The batch load entering and leaving the queue during a given time period; the last constraint ensures that only batch loads can enter the queue.

[0037] To ensure service quality, the computing queue must also meet the following requirements: (6) In the formula: To calculate the maximum queuing time for the load, the intuitive meaning of the above formula is: any... Tasks that enter the queue at a later time or earlier, t Everything should be dealt with at all times.

[0038] In addition to the constraints mentioned above, to improve user experience, this invention also defines a cost function for calculating the queue in the form of a penalty term: (7) In the formula: This is the penalty coefficient.

[0039] 1.2 Computing Power-Power Coupling Equipment Model This invention aggregates servers of the same model within a data center into a cluster. Without loss of generality, it is assumed that the data center contains only one cluster. As a coupling device, the server cluster has both computing power and power ports. The set of computing power port variables is as follows: The power port variable set is , respectively, represent active power, reactive power, voltage, and phase angle.

[0040] Define the server cluster's uptime as follows: (8) In the formula: and These represent the number of active servers and the total number of servers, respectively.

[0041] Define the computational load density of the server cluster as: (9) In the formula: express t The total computational load processed in a given time period, including both batch and interactive workloads; This is for single-server processing capacity.

[0042] Define the average processor utilization of the cluster as: (10) The relationship between power and computing power in a server cluster can be expressed as: (11) In the formula: for t Power consumption of the server cluster during the specified time period; , These represent the peak power and standby power of a single server, respectively.

[0043] Will , Substituting into equation (11), the computing power-electricity relationship model of the server cluster of the present invention is obtained as follows: (12) For large-scale clusters, the startup rate in the above formula can be used as an example. Relaxation is done by using continuous variables to eliminate integer variables.

[0044] Other constraints are as follows: (13) In the formula: For data center power usage efficiency (PUE). This refers to the total power, including auxiliary losses such as cooling. and These represent the upper and lower limits of server utilization.

[0045] This invention defines the cost function of a server cluster as: (14) In the formula: This is a server start-up and shutdown cost coefficient used to penalize frequent and large fluctuations in server uptime.

[0046] 1.3 Computing Power and Power Bus Model The computing power and power ports are connected to the computing power and power buses respectively, and are used for the collection and distribution of information and energy. To facilitate unified modeling of computing power and power, the variables associated with these two types of ports are divided into... Flow variables and potential variables. Among them, power... P , Q and computing load , All are flow variables and all satisfy the node conservation constraint: (15) In the formula: Port Stream variables; To connect to the bus n The port set.

[0047] Voltage V and phase angle θ As potential variables, they all satisfy the consistency constraint: (16) In the formula: To connect to the bus n The potential variable of the port.

[0048] 2. Joint modeling of computation-electrical coupled systems Computing power and power networks are interconnected at multiple points. Because computing power can be spatially migrated, it is equivalent to establishing "virtual lines" between data centers. Thus, a single data center is no longer an independent load, requiring modeling and optimization research from the perspective of the convergence of the two networks.

[0049] 2.1 Unified graphical model of computation-electric coupling system This invention proposes a unified graph model representation method for computing power and power networks. For ease of explanation, Figure 2 A simple computation-electric coupling system is given. Figure 3 Its unified graphical model has the following characteristics: 1) The vertex set of a graph can be divided into two subsets, left and right. The left subset... Representative equipment, including computing power equipment sets Power equipment collection and computer-coupled device set Right set Representative nodes, containing computing power bus sets and power busbar .

[0050] 2) Edge set of a graph Includes computing power edge set and power edge collection Every edge must connect to a device at one end and a node at the other. (See diagram for example, a computing edge.) - ) and power edge (as shown in the figure) - These are used to exchange information flow and energy flow, respectively.

[0051] The aforementioned graph model properties allow edges to be grouped either by device or by node. This invention uses... Indicates access device d The edge set, i.e., the device d port set; using Indicates access node n The edge set, i.e., the nodes n The port set on.

[0052] Thus, this invention has achieved unified modeling of computing and electrical systems, which are two very different types of systems, at the device and network levels respectively, and has built a "computing power-electricity unified network" with the power grid as the center.

[0053] Example 2 This embodiment provides a collaborative optimization method for a coupled system of computing power network and power network, including: A modeling method for a coupled computing network and a power network, as described in Example 1, is used to obtain a model of a computing-power coupled device. With the goal of maximizing social welfare, a collaborative optimization problem of computing and power coupling system is constructed, and the power purchase of the distribution network and the computing power time migration load and computing power space migration load of each data center are determined by solving the problem. To achieve Pareto optimal redistribution of benefits, a logarithmic Nash social welfare function is defined, and a Nash negotiation problem is constructed. The electricity purchased by the distribution network and the computing power time migration load and computing power space migration load of each data center are substituted into the Nash negotiation problem to obtain a convex optimization problem with only the incentive price as the decision variable, thereby solving the problem to obtain the calculation result of the incentive price.

[0054] The following is a detailed description: 1. Cooperative optimization of computation-electrical coupling systems For electrical equipment such as transmission lines, energy storage, and loads, the same applies. Representing the cost function, using This indicates the feasible domain for operation.

[0055] The main power grid to which the root node of the distribution network is connected is regarded as an equivalent power source device named root, and its cost function is: (17) In the formula: The price at which electricity is purchased from the main power grid. Power purchased; dispatch interval ; and These are the carbon footprint factor and carbon emission baseline value for coal-fired power units, respectively. This refers to the carbon price.

[0056] With the goal of maximizing social welfare, based on the aforementioned unified graph model, the cooperative optimization problem of computation-electrical coupling systems can be expressed as: (18) In the formula: It refers to the set of all devices in the computer-coupled system.

[0057] The business constraints and spatiotemporal flexibility of computing power networks have been fully embedded (18). At the same time, the above formula eliminates the differences between computing power and power networks in form, and analysis and calculation can be completed under a unified framework.

[0058] 2. Distributed Cooperative Optimization of Computation-Electrical Coupled Systems Problem (18) contains three types of coupling constraints: first, the computational-electrical coupling relationship of equation (12) exists in the feasible domain of the server cluster; second, the network constraints of computational power and power in equations (15)-(16); and third, the network constraints of computational power and power respectively. Defined in KIn each optimization time period, there are cross-time period constraints. Directly solving this problem presents the privacy and scalability issues mentioned earlier. Based on the decoupling approach of the alternating direction method of multipliers (ADMM), this invention derives and implements a distributed collaborative optimization method for computationally coupled systems.

[0059] For equation (18), firstly, a port variable is introduced. x mirror variables z According to the properties of the graphical model of the present invention, x Group by device z Group by node and define the following indicator function to represent network constraints: (19) In the formula: The node constraints are determined by equations (15) and (16).

[0060] To simplify the representation, define the device. d The extended cost function is: (20) This style (18) can be rewritten in the standard ADMM form: (twenty one) In the formula: the third term in the objective function is an augmentation term, used to improve the convergence of the algorithm. This is a penalty factor. For each component of the Lagrange multiplier, the component corresponds to the node conservation constraint of equation (15) and the node consistency constraint of equation (16).

[0061] For all feasible solutions that satisfy the network constraints, the second and third terms of the objective function are both 0. Therefore, problem (21) is equivalent to the original problem (18).

[0062] Problem (21) can be solved iteratively by following these steps: Step 1) Left set Each device in d Parallel solution of the following optimization problem: (twenty two) In the formula: k This represents the number of iterations. and These are the multiplier vectors for the flow variable and the potential variable, respectively; the meanings of variables with the - superscript are as follows.

[0063] Step 2) Right set Each node in nThe multiplier vectors are updated in parallel. The streaming variables are updated as follows: (twenty three) In the formula: , For access nodes n The average value of all port flow variables will be substituted into equation (22) in the next iteration; For access nodes n The number of ports.

[0064] Potential variables are updated as follows: (twenty four) In the formula: , For access nodes n The average value of all port potential variables will also be substituted into equation (22) in the next iteration.

[0065] Step 3) If the iteration does not converge, return to step 1; otherwise, the calculation ends.

[0066] contrast Figure 3 As can be seen from the graph model, the above calculations are performed with vertices as the center and near-end information exchange is done entirely along the edges: In step 1, each device only receives the mean from its neighboring nodes. In step 2, each node only collects port variables from adjacent devices. and its multipliers Used to calculate the average. Through multiple iterations, the computing nodes and power nodes are gradually driven to achieve a balance between computing power and energy.

[0067] 3. Privacy computation in computer-coupled systems Collaborative optimization requires the computing power network to provide its equivalent circuit model. Although this model hides the physical structural details of the computing power network through simplification and equivalence, the CSP may still be unwilling to provide it to the public.

[0068] In this situation, the present invention can naturally protect privacy by segmenting the network. Figure 4 For example, this invention uses the power supply line connecting the data center to the power grid as the dividing point, naturally separating the computing network of each CSP from the power network for deployment and computation. At this time, the only change in steps 1 and 2 of the previous section is that each CSP and the power grid compute equations (22)-(24) in parallel on their respective computers. In the iteration, only limited information on the computing and power network division surfaces needs to be exchanged, thereby achieving privacy computing.

[0069] 4. Market mechanisms to promote the migration of computing power space To reduce operating costs, data centers spontaneously leverage their time flexibility to migrate computing loads to periods with lower electricity prices. However, spatial migration incurs additional communication and maintenance costs, computing power scheduling costs, and bandwidth latency risks. Without effective market incentives, the spatial flexibility of data centers cannot be fully utilized. To address this issue, this section first discusses the distribution locational marginal price (DLMP) mechanism, and then proposes a spatial migration incentive mechanism based on Nash negotiations.

[0070] 4.1 Distribution Network Node Pricing Mechanism The utility function of the distribution network is defined as follows: (25) In the formula: The power purchase cost of the distribution network is given by equation (17). This refers to the electricity sales price on the distribution network. I For the CSP set connected to the distribution network, I i For CSP i A collection of data centers For CSP i Central Data Center j for The power of the power supply line (PSL); The remaining load power in the distribution network is considered as non-adjustable load in this invention.

[0071] Define CSP i The utility function is: (26) In the formula: the first term on the right side of the equation is the electricity purchase cost of the CSP, and the remaining terms are the communication link space migration cost, computing queue delay cost and server start-up and shutdown cost as defined by equations (4), (7) and (14), respectively. L ( ij ) indicates CSP i Data Center j A set of remote communication links.

[0072] The total social welfare of the entire computational-electrical coupling system is: (27) In the formula: The utility of the remaining loads in the distribution network is represented by a negative electricity cost.

[0073] In the above formula, the purchase and sale of electricity between the distribution network and the load have been mutually offset, so the objective function of equation (27) is consistent with that of problem (18). Therefore, the graph computation framework proposed in the previous section can be directly used to solve DLMP. In this invention, the Lagrange multipliers corresponding to the power conservation constraints of each power node in equation (21) are used as DLMP.

[0074] 4.2 Nash Negotiation Mechanism To incentivize spatial flexibility in data centers without the implementation of nodal pricing, this invention proposes the following spatial migration incentive mechanism based on Nash negotiations: (28) In the formula: To pay CSP i Spatial migration incentive price; For CSP i exist t Total computing power spatial migration load during the time period: (29) In the formula: L ( i ) is CSP i A set of remote communication links.

[0075] Equation (28) represents the distribution network relinquishing a portion of its profits to the CSP. and These represent the utility of the distribution network and each CSP after the redistribution of benefits.

[0076] To quantify the value of cooperation, define , Respectively, power grid and CSP i The utility at which cooperation is not reached, i.e., the point of breakdown in negotiations. This invention uses the optimal utility of all parties when the data center only utilizes time flexibility as the point of breakdown in negotiations. Therefore, the cooperation value for the power grid and each CSP can be quantified as follows: and .

[0077] To achieve Pareto optimal redistribution of benefits, this invention defines the following logarithmic form of the Nash social welfare function: (30) Then, the following Nash negotiation problem is established: (31) In the formula: the objective function aims to keep the utility of each cooperating party as far away as possible from the point of negotiation breakdown; the constraints in the first row... The first line sets a price ceiling to incentivize price increases; the second line constraint is used to ensure the assumption of individual rationality.

[0078] However, problem (31) attempts to address both quantification (determining spatial migration, etc.) and pricing (determining incentive prices) simultaneously. This problem, which makes it a difficult non-convex problem, is solved by decomposing problem (31) into two sequential subproblems for approximate solution: 1) Quantitative sub-problems: First, the graph calculation method of this invention is used to solve problem (18). At this stage, the distribution of cooperative benefits is not considered. Instead, the goal is to maximize social welfare and determine decision variables such as the electricity purchased by the distribution network and the temporal and spatial migration of each data center.

[0079] 2) Pricing Sub-problem: Then substitute the above quantitative results into equation (31). At this point, the problem is simplified to using only the incentive price. This is a convex optimization problem with decision variables.

[0080] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for modeling a coupled system of computing power network and power network, characterized in that, include: Based on the actual equipment and topology of the computing power network in the area to be processed, an equivalent circuit model of the computing power network is constructed. The equivalent circuit model includes computing power equipment, computing power bus, and computing-electrical coupling equipment. The computing power equipment includes computing power load, communication link, and computing queue. Each computing power equipment is connected to the computing power bus through a computing power port. The computing-electrical coupling equipment is connected to the computing power bus and the power bus respectively. The equivalent circuit model of the computing power network is coupled with the power network to construct a computing-power coupled system model. This enables the aggregation of computing resources and their equivalent access to different power grid nodes. The network topology of the computing power network and the power network is transformed into a unified graph model for unified scheduling of the computing power network and the power network.

2. The method for modeling a coupled system of computing power network and power network according to claim 1, characterized in that, The computing load is analogous to the electrical load and modeled as a single-port device, with the corresponding set of port variables being: ,in, and These represent batch processing and interactive workloads, respectively. K The total number of time periods; the calculation expressions for the batch processing and interactive load are: In the formula, for t Real-time batch processing load. for t Real-time interactive load, , For data centers t Total computation requests received during the time period; and These represent the percentages of batch processing and interactive workloads, respectively.

3. The method for modeling a coupled system of computing power network and power network according to claim 1, characterized in that, The communication link is analogous to a power transmission line and modeled as a two-port device, with the corresponding set of port variables being: , representing the batch and interactive loads on port 1 and port 2, respectively; For the communication links within the data center, the lossless transmission constraint is satisfied, and the calculation expression for the lossless transmission constraint is as follows: In the formula, and For port variables closer to the front end; For communication links that are remotely interconnected bidirectionally between data centers, the lossless transmission constraint, migration constraint, and bandwidth constraint must be satisfied; the expression for the migration constraint is: The expression for the bandwidth constraint is: In the formula, The upper limit of bandwidth for a remote two-way interconnected communication link; For remote communication links, additional operation and maintenance costs need to be set up, and the data center's first l For each remote communication link, the following cost function is defined: In the formula, the coefficients This represents the cost factor for calculating the unit load space migration; This represents the fixed cost when the link is enabled.

4. The method for modeling a coupled system of computing power network and power network according to claim 1, characterized in that, The computing queue is analogous to energy storage; storing computing power requests in the queue is analogous to charging; and removing the load from the queue and processing it is analogous to discharging. The computing queue is modeled as a single-port device, and the corresponding state equation is: In the formula, E This represents the accumulated computational load in the computation queue. The computational load for the initial period. The computational load for the termination period; , Represent t Batch load entering and leaving the queue during a specific time period. for t Batch processing load at any given time. for t Interactive load at any given moment; The last constraint ensures that only batch processing loads can enter the queue; The constraints satisfied by the computation queue include: In the formula, To calculate the maximum queuing time for the load; The expression for the cost function of the computation queue is: In the formula, To calculate the cost function of the queue, This is the penalty coefficient.

5. The method for modeling a coupled system of computing power network and power network according to claim 1, characterized in that, The computing-power coupling device is a server. In the computing-power coupling device model, servers of the same model in the data center are aggregated into a cluster as a server cluster. The model expression for the server cluster is: In the formula, for t Power consumption of the server cluster during the time period. and These are the peak power and standby power of a single server, respectively. The uptime rate of the server cluster. This refers to the computational load density of the server cluster. The constraints of the server cluster include: In the formula, express t Total computational load for processing each time period For data center power utilization efficiency, This refers to the total power, including auxiliary losses such as cooling. and These represent the upper and lower limits of server utilization. This represents the average processor utilization of the server cluster. The cost function of the server cluster is defined as: In the formula: This is a server start-up and shutdown cost coefficient used to penalize frequent and large fluctuations in server uptime.

6. The method for modeling a coupled system of computing power network and power network according to claim 1, characterized in that, The computing power port and power port are connected to the computing power bus and power bus, respectively. The variables associated with the computing power port and power port are divided into flow variables and potential variables. The flow variables include active power, reactive power, batch processing load, and interactive load. The flow variables satisfy the node conservation constraint, the expression of which is: In the formula, Port Stream variables; To connect to the bus n The port set; The potential variables include voltage and phase angle, and the potential variables satisfy a consistency constraint, the expression of which is: In the formula, To connect to the bus n The potential variable of the port.

7. The method for modeling a coupled system of computing power network and power network according to claim 1, characterized in that, The unified graph model's vertex set is divided into a left set and a right set. The left set represents devices, including a computing power device set, a power device set, and a computing-power coupling device set. The right set represents nodes, including a computing power bus set and a power bus set. The edge set of the unified graph model includes computing power edge set and power edge set. Each edge connects to a device at one end and a node at the other end.

8. A collaborative optimization method for a coupled system of computing power network and power network, characterized in that, include: A modeling method for a coupled computing network and a power network as described in any one of claims 1-7 is used to obtain a model of a computing-power coupled device. With the goal of maximizing social welfare, a collaborative optimization problem of computing and power coupling system is constructed, and the power purchase of the distribution network and the computing power time migration load and computing power space migration load of each data center are determined by solving the problem. To achieve Pareto optimal redistribution of benefits, a logarithmic Nash social welfare function is defined, and a Nash negotiation problem is constructed. The electricity purchased by the distribution network and the computing power time migration load and computing power space migration load of each data center are substituted into the Nash negotiation problem to obtain a convex optimization problem with only the incentive price as the decision variable, thereby solving the problem to obtain the calculation result of the incentive price.

9. The collaborative optimization method for a coupled system of computing power network and power network according to claim 8, characterized in that, The expression for the collaborative optimization problem of the computational-electrical coupling system is: In the formula, For the entire set of devices in the computer-electrical coupling system, Represents the cost function. This indicates the feasible domain for operation.

10. The collaborative optimization method for a coupled system of computing power network and power network according to claim 8, characterized in that, The expression for the Nash negotiation problem is: In the formula, For the Nash negotiation function, To incentivize price caps, To pay CSP i Spatial migration incentive price, and These refer to the utility of the distribution network and each CSP after the redistribution of benefits. and Respectively, power grid and CSP i Its utility when cooperation is not achieved, i.e., the point at which negotiations break down.

Images