A method and system for data center and virtual power plant collaborative operation

Abstract

The present application relates to a kind of data center and virtual power plant collaborative operation method and system, including the two-layer collaborative optimization model of constructing data center and virtual power plant;Wherein, the two-layer system optimization model includes with virtual power plant net profit maximization as goal high-level model, and with the minimum as goal low-level model of multiple data center cluster total operating cost;Particle swarm optimization algorithm is used to solve the two-layer collaborative optimization model;Wherein, the decision variable of high, low layer model is collectively coded as the position of particle in particle swarm, the position and speed of particle are updated by iteration, the objective function of high, low layer model is simultaneously optimized, finally the global historical optimal position decoding is obtained, and the optimal collaborative scheduling scheme is obtained;Based on the optimal collaborative scheduling scheme, the power of distributed new energy and energy storage in virtual power plant is controlled, and the space-time migration scheduling of delay-sensitive and delay-tolerant workloads in data center cluster is carried out.

Description

Technical Field

[0001] This invention relates to the fields of power system operation, dispatch and control and smart grid technology, and in particular to a method and system for collaborative operation of a data center and a virtual power plant. Background Technology

[0002] With the booming development of the digital economy, data centers (DCs), as key infrastructure for data storage, computing power supply, and cloud services, have become the core engine for the digital, networked, and intelligent transformation of power grids, playing a crucial supporting role. However, the energy consumption problem of data centers is becoming increasingly prominent. Since the power consumption of data centers is strongly positively correlated with the amount of computing tasks processed, some non-real-time computing tasks can be flexibly scheduled through time delay and spatial migration. This spatiotemporal transfer mechanism based on the characteristics of computing power-power correlation essentially expands the traditional single-time-dimensional load shifting into a "time-space" dual-dimensional migration, making data centers a core flexible resource on the demand side. This characteristic makes data centers a highly valuable controllable resource unit in VPPs (Virtual Power Plants).

[0003] Currently, research on the participation of data centers in the optimized scheduling of virtual power plants has made initial progress. However, with the deepening of the power market mechanism and the high proportion of renewable energy integration, the collaborative operation of data centers and virtual power plants faces the following limitations:

[0004] (1) The dimensions of flexibility exploration are too singular. Existing results mostly focus on the time migration and spatial transfer characteristics of computing power tasks in data centers, while lacking a systematic analysis of the flexible adjustment potential of data centers and virtual power plants, resulting in an incomplete energy flexibility assessment system.

[0005] (2) Most models assume that the execution parameters of the computing power tasks in the data center are deterministic information, ignoring the impact of renewable energy output fluctuations, electricity market changes and task time parameter uncertainties on scheduling decisions, resulting in insufficient adaptability of optimization strategies in real uncertain environments;

[0006] (3) The interaction mechanism is weak. Most models simplify data centers as passive recipients of electricity, ignoring their reverse influence on regional electricity formation by adjusting electricity consumption behavior, and failing to fully release the creative ability of data centers as flexible demand-side resources in the electricity market.

[0007] Existing technologies have limitations such as a single dimension for flexibility mining, neglect of multiple uncertain factors, and weak interaction mechanisms. As a result, the optimization strategies for the collaborative operation of data centers and virtual power plants are not adaptable enough to real uncertain environments and fail to fully unleash the value creation capabilities of data centers as flexible demand-side resources in the electricity market. Summary of the Invention

[0008] Based on the above analysis, the embodiments of the present invention aim to provide a method and system for collaborative operation of data centers and virtual power plants, in order to solve the technical problems of low system operation flexibility and weak optimization strategy robustness caused by insufficient mining of the spatiotemporal coupling characteristics of computing power and power, poor adaptability to multiple uncertainties and weak interaction mechanism when existing data centers and virtual power plants are operating collaboratively.

[0009] This invention provides a method for collaborative operation of a data center and a virtual power plant, comprising the following steps:

[0010] A two-layer collaborative optimization model for data centers and virtual power plants is constructed. The two-layer system optimization model includes a high-level model aimed at maximizing the net profit of the virtual power plant and a low-level model aimed at minimizing the total operating cost of the multi-data center cluster.

[0011] The two-layer collaborative optimization model is solved using a particle swarm optimization algorithm. The decision variables of the high-level and low-level models are jointly encoded as the positions of particles in the particle swarm. By iteratively updating the positions and velocities of the particles, the objective functions of the high-level and low-level models are optimized synchronously. Finally, the output global historical optimal position is decoded to obtain the optimal collaborative scheduling scheme.

[0012] Based on the optimal collaborative scheduling scheme, the power of distributed new energy sources and energy storage in the virtual power plant is controlled, and the spatiotemporal migration scheduling of latency-sensitive and latency-tolerant workloads in the data center cluster is performed.

[0013] The present invention also discloses a collaborative operation system for data centers and virtual power plants, the system comprising a collaborative optimization modeling module M1, an intelligent optimization solution module M2, and a collaborative scheduling execution module M3;

[0014] The collaborative optimization modeling module M1 is used to construct a two-layer collaborative optimization model between the data center and the virtual power plant. The two-layer system optimization model includes a high-level model with the goal of maximizing the net profit of the virtual power plant and a low-level model with the goal of minimizing the total operating cost of the multi-data center cluster.

[0015] The intelligent optimization solution module M2 is used to solve the two-layer collaborative optimization model using the particle swarm optimization algorithm. The decision variables of the high-level and low-level models are jointly encoded as the positions of particles in the particle swarm. By iteratively updating the positions and velocities of the particles, the objective functions of the high-level and low-level models are optimized synchronously. Finally, the output global historical optimal position is decoded to obtain the optimal collaborative scheduling scheme.

[0016] The collaborative scheduling execution module M3 is used to control the power of distributed new energy sources and energy storage in the virtual power plant based on the optimal collaborative scheduling scheme, and to perform spatiotemporal migration scheduling of latency-sensitive and latency-tolerant workloads in the data center cluster.

[0017] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0018] 1. This invention addresses the problem of limited flexibility mining dimensions in the collaborative operation of data centers and virtual power plants in existing technologies. It proposes a multi-dimensional flexibility mining method based on the spatiotemporal coupling characteristics of "computing power-electricity". By constructing a demand response model that includes both time migration and spatial transfer dimensions, it overcomes the limitations of traditional methods that only focus on the time dimension of load shifting. It establishes a model that covers the time-transferability of computing tasks, achieving a deep match between data center load characteristics and virtual power plant adjustment needs.

[0019] 2. This invention addresses the shortcomings of existing models that neglect uncertainties and have weak market interaction mechanisms by establishing a master-slave game-theoretic collaborative optimization framework that considers multiple uncertainties. This framework innovatively combines dynamic electricity pricing mechanisms with game theory methods: high-level virtual power plants act as leaders, constructing stochastic optimization models to formulate time-of-use / node pricing strategies based on uncertainties such as renewable energy output forecasts and market price fluctuations; low-level data center clusters act as followers, achieving bidirectional interaction with the virtual power plants through load adjustment feedback. By establishing a closed-loop mechanism of price signal guidance and load response feedback, the value creation capability of data centers as flexible demand-side resources in the electricity market is fully unleashed.

[0020] 3. This invention establishes a two-layer collaborative optimization model based on a master-slave game architecture. This model achieves the dual objectives of maximizing the aggregated revenue of virtual power plants and minimizing the energy costs of data centers through a dynamic pricing mechanism. The high-level model integrates the predicted output of distributed renewable energy and electricity market price signals, using the load adjustment of the low-level data center cluster as the decision variable. The low-level model establishes an optimization model that includes workload migration based on the spatiotemporal elasticity characteristics of the workload. The particle swarm optimization algorithm is used to achieve distributed solution of the two-layer model. While ensuring the service quality of data centers, it significantly improves the ability of virtual power plants to absorb a high proportion of renewable energy, providing technical support for building an integrated and collaborative new energy system of "source-grid-load-storage".

[0021] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0022] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0023] Figure 1 This is a flowchart of a method for collaborative operation of a data center and a virtual power plant according to an embodiment of the present invention;

[0024] Figure 2 This is a schematic diagram of a two-layer collaborative optimization scheduling model in an embodiment of the present invention;

[0025] Figure 3 This is a schematic diagram of the first-order equivalent thermal parameter model of a data center in an embodiment of the present invention;

[0026] Figure 4 This is a flowchart of the particle swarm optimization algorithm in an embodiment of the present invention;

[0027] Figure 5 This is a schematic diagram illustrating the number of latency-sensitive and latency-tolerant workload tasks in data center 1 at different times in an embodiment of the present invention;

[0028] Figure 6 This is a schematic diagram showing the number of latency-sensitive and latency-tolerant workload tasks in data center 2 at different times in this embodiment of the invention;

[0029] Figure 7 A schematic diagram of the optimization results of each unit of the virtual power plant in this embodiment of the invention;

[0030] Figure 8 A schematic diagram illustrating the spatial migration of data center workloads in this embodiment of the invention;

[0031] Figure 9 This is a schematic diagram of a data center and virtual power plant collaborative operation system module in an embodiment of the present invention. Detailed Implementation

[0032] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0033] From the perspective of "computing power-electricity" synergy, data centers can achieve dual value through virtual power plants: On the one hand, for data center operators, by actively participating in multi-regional electricity markets, they can fully utilize the differences in electricity prices under different time and space scenarios to optimize the allocation of computing resources, realizing the transformation from passive electricity receivers to active electricity users, and significantly reducing operating costs; on the other hand, as a high-energy-consuming but highly flexible load unit, the spatiotemporal adjustability of data centers can effectively solve the bottleneck problem of regional renewable energy consumption, balance the differences in electricity supply and demand through cross-regional load transfer, improve the grid's ability to accept high proportions of renewable energy, and provide key support for building an integrated and coordinated new energy system of "source-grid-load-storage". The data center optimized scheduling mode based on the virtual power plant platform not only meets the needs of energy structure transformation under the "dual carbon" goal, but also provides an innovative path for exploring the deep integration of the digital economy and the energy revolution.

[0034] This invention addresses the technical shortcomings of existing models in terms of flexibility assessment, uncertainty adaptation, and interaction mechanisms by establishing a master-slave game collaborative optimization framework that considers the spatiotemporal coupling characteristics of computing power and electricity.

[0035] A specific embodiment of the present invention discloses a method for collaborative operation of a data center and a virtual power plant, such as... Figure 1 As shown, it includes the following steps:

[0036] Step S1: Construct a two-layer collaborative optimization model for data centers and virtual power plants; wherein, the two-layer system optimization model includes a high-level model with the goal of maximizing the net profit of the virtual power plant, and a low-level model with the goal of minimizing the total operating cost of the multi-data center cluster;

[0037] Step S2: Solve the two-layer collaborative optimization model using the particle swarm optimization algorithm; wherein, the decision variables of the high-level and low-level models are jointly encoded as the positions of particles in the particle swarm, and the objective functions of the high-level and low-level models are simultaneously optimized by iteratively updating the positions and velocities of the particles, and finally the global historical optimal position is decoded to obtain the optimal collaborative scheduling scheme.

[0038] Step S3: Based on the optimal collaborative scheduling scheme, control the power of distributed new energy and energy storage in the virtual power plant, and perform spatiotemporal migration scheduling for latency-sensitive and latency-tolerant workloads in the data center cluster.

[0039] Step S1 includes steps S11-S12.

[0040] Step S11: Analysis of the two-layer collaborative operation framework of data center and virtual power plant.

[0041] This invention, from the perspective of data center clusters (including multiple data centers) participating as flexible loads in the aggregated operation of virtual power plants, constructs a two-layer collaborative optimization framework for data centers and virtual power plants, such as... Figure 2 As shown.

[0042] The system adopts a master-slave game strategy, with the high-level virtual power plant acting as the game leader. It integrates the predicted output of distributed renewable energy (such as wind power and photovoltaic) with market electricity price signals, and aims to maximize aggregate revenue by issuing incentive signals to member units through a dynamic pricing mechanism.

[0043] As a follower in the game, the low-level data center cluster responds to the price signals of the virtual power plant with the goal of minimizing energy costs based on the spatiotemporal elasticity of the workload, and participates in the global optimization of the virtual power plant through load response adjustment feedback.

[0044] In this invention, a data center cluster specifically refers to a collection of multiple geographically dispersed data centers interconnected through a high-speed communication network, which collaboratively participate in the aggregated operation of a virtual power plant under a unified scheduling strategy.

[0045] Data center computing tasks are divided into two categories based on their response characteristics:

[0046] (1) One type is offline tasks. Typical scenarios include large-scale batch computing such as log analysis, machine learning training, and video rendering. These tasks have high demand for computing, storage and communication resources, high energy consumption during execution, and users have a high tolerance for processing latency.

[0047] (2) Another type is online tasks, such as financial transactions, disaster warnings, video calls, etc., which need to be completed within milliseconds of latency. Timeout will directly affect service quality and even cause economic losses.

[0048] Offline tasks are further divided into two categories: periodic and non-periodic.

[0049] (1) Periodic tasks are executed repeatedly at fixed intervals (seconds / minutes / hours / days, etc.), and have the characteristics of consistent computational content, long life cycle and fixed execution sequence;

[0050] (2) Non-periodic tasks are submitted randomly by users, and their execution time and resource requirements are not fixed. They are usually scheduled to be processed during the server's idle period, which has higher latency flexibility.

[0051] While online tasks are latency-sensitive, they have low computational load, short execution time, and low resource consumption, and can be transferred to other data centers for processing via high-speed interconnect networks. Their workload distribution exhibits a clear diurnal fluctuation (high during the day, low at night). The differences between online and offline tasks in terms of resource requirements, latency characteristics, and scheduling potential provide a foundational framework for data centers to participate in the optimized scheduling of virtual power plants, enabling task-level flexibility exploration.

[0052] Data center computing tasks exhibit significant potential for power flexibility in both spatiotemporal dimensions. Offline tasks, due to their long execution times and high user tolerance, can be flexibly scheduled in the time dimension by setting deadlines: the scheduling system dynamically adjusts the execution plan based on task execution time estimates and priority ranking, combined with server resource availability, allowing for a certain delay while ensuring completion before the deadline, providing adjustment space for time-scale coordination such as power grid peak shaving and frequency regulation. Although online tasks are latency-sensitive, relying on redundant data storage and high-speed interconnection technology, spatial transfer can be achieved without significantly affecting bandwidth and computing continuity. Overloaded tasks can be quickly taken over by backup servers, solving the problem of uneven resource allocation. The spatial flexibility brought by its massive workload can support cross-regional power balancing. The flexibility of these two types of tasks is aggregated into the total power load through the additive nature of computing resource consumption, jointly constituting the spatiotemporal coupling adjustment capability of the data center as a virtual power plant unit. This optimizes power load distribution while ensuring service quality, providing key support for "computing power-power" coordination.

[0053] Data center modeling is as follows:

[0054] As highly complex physical infrastructure, data centers exhibit significantly different energy consumption structures compared to traditional electricity users. Core energy consumption is concentrated in two main areas: IT equipment and cooling systems. IT equipment, acting as the brain of the data center, encompasses servers, storage, and network devices. Through virtualization technology, it pools and dynamically schedules computing and storage resources, supporting centralized data processing and real-time interaction. To meet the high-reliability, high-capacity data processing demands brought about by the rapid development of the internet, IT equipment needs to operate 24 / 7, accounting for up to 45% of energy consumption. Simultaneously, the heat generated during operation must be continuously dissipated through cooling systems; otherwise, equipment efficiency will decline, or even system crashes may occur.

[0055] The refrigeration system plays a crucial role in maintaining the stable operating environment of IT equipment, accounting for approximately 40% of energy consumption, second only to the IT equipment itself. Traditional mechanical refrigeration mainly uses air-cooled direct expansion systems and water-cooled chilled water systems. Air-cooled direct expansion systems are gradually being replaced by water-cooled refrigeration systems due to their lower coefficient of performance (COP). Water-cooled refrigeration systems provide the foundation for the application of natural cooling technology, using outdoor low-temperature cooling water to shut down refrigeration units in winter, thereby reducing energy consumption.

[0056] Natural cooling technology is further subdivided into three categories: water-side, air-side, and heat pipe. Water-side relies on the circulation of cooling water for heat exchange, air-side achieves indirect cooling by introducing outdoor cold air or air heat exchangers, and heat pipe completely eliminates mechanical work and uses phase change heat transfer to achieve power-free cooling.

[0057] (1) Energy consumption modeling of IT systems

[0058] As a high-energy-consuming physical facility, the key to energy consumption optimization in data centers lies in accurately modeling the power consumption characteristics of core equipment. IT equipment, as the main energy consumer in data centers (accounting for approximately 45% of total energy consumption), has its adjustability directly determining the overall potential for energy efficiency improvement.

[0059] Current mainstream IT equipment energy consumption modeling adopts a power model based on Dynamic Voltage Frequency Scaling (DVFS) technology. This model achieves a refined characterization of server energy consumption by quantifying the dynamic relationship between CPU operating frequency, voltage, and server energy consumption, as shown below:

[0060] Formula (1)

[0061] in, This refers to the server's power consumption when the CPU is idle. This is the baseline power (the baseline power consumption of the CPU in idle state, i.e., the baseline power consumption of the server as a whole in idle state). It is a constant; This refers to the CPU's operating frequency; This refers to the CPU's operating voltage. The workload allocated to a single server per unit of time; This refers to the processing capacity of a single server.

[0062] There is a linear relationship between CPU operating frequency and operating voltage, that is, the operating voltage and operating frequency approximately satisfy... ∝ The linear dependence of equation (1) can be rewritten as follows:

[0063] Formula (2)

[0064] in, is the dynamic power coefficient, and is a constant.

[0065] For example, an Intel Xeon E5-2600 v3 series processor with a dynamic power factor is used. Values Scale; for x86 architecture servers, Values ​​in to between; The value is determined by the specific processor's hardware characteristics and is obtained through technical manuals or experimental calibration for different server models.

[0066] The core advantage of the power model based on dynamic voltage and frequency regulation technology lies in its close alignment with actual operating scenarios. The server can dynamically adjust the CPU frequency according to the real-time load, thereby flexibly controlling the energy consumption level. However, the model has a high computational complexity, requiring a trade-off between accuracy and efficiency.

[0067] (2) Energy consumption modeling of refrigeration system

[0068] As the second largest energy consumer in data centers (accounting for approximately 40% of total energy consumption), cooling systems are primarily modeled using equivalent thermodynamic models. This involves coupling the indoor and outdoor environments (including heat generation from IT equipment, building heat exchange, lighting, and occupant heat dissipation) with the cooling system's cooling capacity to construct dynamic equations for heat transfer. While third-order equivalent thermodynamic models can accurately describe the complex changes in indoor and wall surface temperatures, the calculation process is cumbersome. In practical applications, simplified first-order models are often used. By ignoring the temperature differences between the inside and outside of the walls, the computational load is significantly reduced, achieving efficient energy consumption assessment while maintaining basic accuracy. Therefore, this has become the mainstream choice for current cooling system modeling.

[0069] like Figure 3 As shown, the first-order equivalent thermal parameter model of the data center thermodynamic process establishes a dynamic correlation mechanism between the indoor temperature field and the server heat load and the cooling output of the cooling system by equating environmental parameters and building characteristics with resistors and capacitors in a circuit.

[0070] The server's power consumption is approximately equivalent to its heat load output, as shown below:

[0071]

[0072] Formula (3)

[0073] in, For data centers Indoor temperature at any time, by The state calculated at time 1 The indoor temperature at any given time; , Data centers respectively Real-time indoor and outdoor temperatures; For the equivalent heat capacity of the data center, The equivalent combined thermal resistance of the data center building enclosure structure is equivalent to Figure 3 In ; The sampling time interval; For data centers The net heat power at any given time indicates that the room is receiving net heat and the temperature is rising, while a negative value indicates that the room is losing net heat and the temperature is falling. and Data centers respectively The waste heat power (mainly from the electrical power consumption of IT equipment servers) and cooling power (the cooling capacity provided by the cooling system, which is an adjustable control variable) at any given time. The heat output of personnel and lighting systems. Indoor air density in data centers; For the indoor volume of the data center; The specific heat capacity of air; Total mass of the server; The server's specific heat capacity; Total mass of the frame; This is the specific heat capacity of the frame.

[0074] The first formula in formula (3) is the core dynamic equation, which predicts the indoor temperature at the next moment based on the law of conservation of energy.

[0075] The second formula in formula (3) is the net heat power equation. A value greater than 0 indicates insufficient net heat gain indoors, resulting in insufficient cooling and a rise in temperature; otherwise... A value less than 0 indicates net heat loss indoors, excessive cooling, and a drop in temperature.

[0076] Figure 3 In The thermal resistance of the wall;

[0077] Compared to traditional analytical frameworks based on Power Usage Effectiveness (PUE), the equivalent thermal parameter model has significant advantages in terms of system analytical depth. This model establishes an energy conservation relationship through the first law of thermodynamics, reflecting the dynamic coupling characteristics of the thermal power of IT equipment and the cooling supply of the cooling system, thus providing a more refined analytical tool for energy efficiency optimization within data centers.

[0078] Data center's own energy consumption The main source of power consumption is the IT (Information Technology) equipment responsible for data processing and interactive communication. Refrigeration system and energy consumption of power distribution equipment Furthermore, there is a strong correlation between the energy consumption of cooling equipment and power distribution equipment and the power consumption of IT equipment. (Data Center) The data load energy consumption model is shown below:

[0079] Formula (4)

[0080] in, Index for data centers; The power utilization efficiency of a data center is a key indicator for measuring its energy efficiency. For data centers exist Total active power load during the time period; Data Center exist Energy consumption of IT equipment, cooling system, and power distribution equipment during different time periods; This represents the typical PUE value for a data center.

[0081] IT equipment energy consumption As shown below:

[0082] Formula (5)

[0083] in, The total number of server types. Index for server type; Indexed by time period; For data centers During the period The The number of servers running online. For data centers During the period The Real-time power consumption of a single server.

[0084] II. Virtual power plant unit modeling and analysis are as follows:

[0085] (1) Distributed new energy sources, including wind turbines and photovoltaics.

[0086] 1) Wind turbine model

[0087] Wind turbines differ from traditional generators because they are highly sensitive to changes in wind speed. The randomness of wind speed variations is characterized by random functions for analyzing and simulating wind power generation. Wind speed follows a Weibull distribution, with the following probability distribution:

[0088] Formula (6)

[0089] in, Wind speed; and These are the shape parameter and the distribution scale, respectively.

[0090] Wind power generation and wind speed The output relationship is shown in equation (7).

[0091] Formula (7)

[0092] in, For the set of all wind turbine nodes; For wind turbine nodes The predicted output power; To cut in wind speed; To cut off the wind speed; Rated wind speed; This is the rated maximum power of the fan.

[0093] As a typical renewable energy source, wind power generation exhibits significant intermittent and nonlinear characteristics in its output. The operation of wind power systems is constrained by a dual threshold wind speed constraint.

[0094] When the wind speed is below the cut-in threshold, the wind turbine cannot overcome mechanical resistance to complete its rotation, and the output power drops to zero. When the wind speed exceeds the rated threshold, the pitch control system activates its protection mechanism, adjusting the blade angle of attack to disconnect the unit from the grid, at which point the power generation also drops to zero. This non-linear power output characteristic requires wind turbine units to be equipped with energy storage devices or operate in conjunction with other power sources to achieve smooth regulation of power output. Therefore, in actual operation, these factors need to be fully considered to ensure optimal power generation efficiency and output power.

[0095] 2) Photovoltaic model

[0096] Solar photovoltaic power generation, like wind power, is a widely distributed and pollution-free clean resource with high renewability and environmental friendliness.

[0097] Photovoltaic (PV) power generation systems directly convert light energy into electrical energy through the photovoltaic effect. Their core components include photovoltaic arrays, DC / DC converters, and grid-connected inverters. Photovoltaic cells are connected in series and parallel to form power generation units, and multiple units are further combined to form photovoltaic arrays with power expansion capabilities. Light intensity is a key input variable, and its probability distribution characteristics directly affect the accuracy of power generation prediction. Compared to the mechanical energy conversion path of wind power generation, the direct current (DC) characteristics of photovoltaic power generation necessitate the additional configuration of inverter devices at the grid connection stage, which also leads to a higher complexity in power quality control compared to wind power generation.

[0098] Studies have shown that, over a finite timescale, the intensity of solar radiation can approximately follow a Beta distribution, as shown below:

[0099] Formula (8)

[0100] Where Γ() is the gamma function, α and βLet be the shape parameter of the Beta distribution. r and r max This refers to the light intensity and maximum intensity during this period.

[0101] The relationship between light intensity and photovoltaic power generation is shown in equation (9).

[0102] Formula (9)

[0103] in, Photovoltaic power generation capacity, Indicates the rated power of photovoltaic power generation. Light intensity, Rated light intensity; for example, The value is 1000 .

[0104] (2) Energy storage model

[0105] Power system energy storage technology, as a key support for energy transition, encompasses various technical routes such as mechanical energy storage, chemical energy storage, and electromagnetic energy storage. Its core function lies in achieving power balance and system dispatch optimization through the spatiotemporal transfer of electrical energy. The introduction of energy storage systems can significantly improve the stability, reliability, and flexibility of power grid operation, and it is particularly strategically valuable in mitigating the challenges brought about by the large-scale grid integration of new energy sources.

[0106] The intermittent and fluctuating nature of renewable energy generation makes it difficult to naturally match its output curve with load demand. Energy storage batteries, through charge and discharge control, can flexibly adjust and store energy from renewable energy sources such as wind and solar power, effectively smoothing out short-term power fluctuations. During peak load periods, they can release stored energy to alleviate grid pressure and ensure the continuity of power supply. This two-way adjustment capability not only promotes the consumption of renewable energy but also enhances the power system's anti-interference capability and operational resilience by providing rapid response regulation capacity. Therefore, the use of energy storage batteries not only helps promote the consumption of renewable energy but also improves the flexibility and reliability of the power system.

[0107] This invention considers battery energy storage as a controllable resource. The battery capacity is expressed using the state of charge (SOC) to represent the output characteristics of the energy storage device. The output of the energy storage device during charging and discharging can be represented as follows:

[0108] During charging:

[0109] Formula (10)

[0110] During discharge:

[0111] Formula (11)

[0112] in, for The state of charge of the energy storage device at any time; The sampling interval; The state of charge of the energy storage device at the next moment; and They are respectively Real-time charging and discharging power at any given moment; and These represent the charge and discharge efficiency of the energy storage device; E H This refers to the rated capacity of the energy storage device.

[0113] (3) Gas turbine

[0114] In a virtual power plant architecture, the gas turbine serves as the core controllable unit, and its operational characteristics need to be deeply integrated into a multi-energy collaborative optimization system. Compared to the single-source positioning in traditional distribution networks, virtual power plants aggregate gas turbines with wind and solar distributed energy to construct a multi-energy complementary system, achieving coordinated control of source-storage-load. The gas turbine, with its rapid dynamic response capability, becomes a key regulating resource for mitigating renewable energy fluctuations and ensuring system power balance. Its technical characteristics are mainly reflected in the following dimensions:

[0115] First, it adopts gas-steam combined cycle technology to achieve cascaded energy utilization through waste heat recovery, thereby improving the overall power generation efficiency by more than 30% compared with traditional units;

[0116] Secondly, through the thermoelectric decoupling control strategy, it can simultaneously respond to electrical load and thermal load demands, and realize the cascade utilization of energy.

[0117] Third, as a synchronous generator, it has inertial response capability and can provide second-level frequency regulation services for virtual power plants. Its spinning standby state can respond to system frequency fluctuations in real time.

[0118] Against the backdrop of energy structure transformation, gas turbines, with their peak-shaving and frequency regulation capabilities and low-carbon emission characteristics, will play an important transitional role in new power systems. The virtual power plant coordinated control model constructed in this patent does not yet incorporate the dynamic coupling characteristics of heating and cooling loads; the current analytical framework simplifies the gas turbine as a pure electric distributed power source.

[0119] The output power of this distributed power source must be limited by the unit capacity, and the constraints are as follows:

[0120] Formula (12)

[0121] in, For gas turbine units Constant effort and These are the lower and upper limits of the output of the gas turbine unit, respectively.

[0122] Step S12: Construct a two-layer collaborative optimization model for the data center and the virtual power plant.

[0123] A high-level model is constructed with the goal of maximizing the net profit of the virtual power plant. This model primarily includes electricity sales revenue, electricity purchase cost, gas turbine power generation cost, distributed renewable energy cost, energy storage charging and discharging cost, and grid loss cost. The objective function and constraints of the high-level model will be described in detail below.

[0124] The objective function of the high-level model is as follows:

[0125] Formula (13)

[0126] in, Net profit of virtual power plants For the electricity sales revenue of the virtual power plant, For the cost of purchasing electricity for virtual power plants, For the cost of gas turbine power generation, The costs of operation and maintenance of distributed renewable energy sources, as well as the costs of curtailment penalties, For energy storage maintenance costs, For network loss costs, The costs include wind power operation and maintenance, as well as the costs of curtailment penalties. The costs include photovoltaic operation and maintenance costs, as well as the costs of curtailment penalties. Number of scheduling periods; This represents the number of substation nodes. express Electricity purchase price for a given period of time; , The first Each substation node is located at Power purchased during specific time periods and its adjustment amounts; This represents the number of gas turbine nodes. This refers to the unit cost of gas turbine power generation. No. Each gas turbine node at Power generation during a given time period; This represents the number of wind turbine nodes. and These are the marginal cost of electricity generated per unit by the wind turbine and the cost of wind curtailment penalty, respectively. , The first Each wind turbine node is at Power generation and power curtailment during a given time period; This represents the number of photovoltaic nodes. and These are the marginal cost of photovoltaic power generation per unit and the cost of curtailment penalty, respectively. , The first Each wind turbine node is at Power generation and power curtailment during a given time period; This represents the number of energy storage nodes. Maintenance costs for energy storage units; , The first One energy storage node in Real-time charging and discharging power during a given time period; , These are the charging and discharging efficiencies of the energy storage device;

[0127] , , , , and These are decision variables for high-level models.

[0128] The revenue from electricity sales by the virtual power plant is shown below:

[0129] Formula (14)

[0130] in, This represents the number of load nodes. , Representing load nodes Place Real-time electricity price and benchmark electricity price at any given time; , Representing load nodes exist The total active power of the data center during the time period, and the active power of other loads excluding the data center.

[0131] Network loss costs are shown below:

[0132] Formula (15)

[0133] in, This represents the number of load nodes. Indicates the unit cost of network loss; Represents a node Place Real-time network loss power.

[0134] (2) Constraints:

[0135] The constraints of the high-level model include:

[0136] The constraints based on DisrFlow include power balance constraints, line transmission capacity constraints, upper and lower limits of safe operation of node voltage, upper and lower limits of gas turbine output and ramp rate constraints, output constraints of distributed new energy sources, and state of charge range and charging and discharging power constraints of energy storage systems.

[0137] 1) Power balance constraints based on DisrFlow power flow

[0138] Formula (16)

[0139] in, , For each system Time Node Active and reactive loads; and For each system Time Node and nodes The voltage amplitude; and They are nodes and nodes The electrical conductance and susceptance between them; For the system Time Node and nodes The voltage phase difference between them.

[0140] (2) Line capacity constraints

[0141] Formula (17)

[0142] in, , Representing branch roads The active and reactive transmission capacity, , Representing branch roads Maximum active and reactive transmission capacity.

[0143] (3) Voltage and electrical safety constraints

[0144] Formula (18)

[0145] in, Indicates at time Flowing through the branch road The current; , Representing branch roads The minimum and maximum currents that are allowed to pass through; Indicates at time node The voltage amplitude; , Representing nodes respectively The upper and lower limits of the allowable voltage amplitude.

[0146] (4) Gas turbine constraints, as shown below:

[0147] Formula (19)

[0148] in, , This represents the upper limit of the active and reactive power output of the gas turbine; 、 Representing nodes respectively exist and The power generation capacity of the gas turbine at any given moment; and These represent the maximum downward and upward ramp power of the gas turbine, respectively.

[0149] (5) Output constraints of distributed renewable energy sources, including wind turbine and photovoltaic output constraints, are as follows:

[0150] Formula (20)

[0151] in, , Wind turbine nodes exist Output and maximum output for each time period; , Photovoltaic nodes exist Output and output limit for each time period.

[0152] (6) Energy storage operation constraints are as follows:

[0153] Formula (21)

[0154] in, This represents the maximum allowable charge and discharge power for energy storage. This refers to the capacity of energy storage at the initial moment of scheduling; and These are the minimum and maximum allowable remaining capacity of energy storage, respectively. , The first One energy storage node in Real-time charging and discharging power during a given time period; , These represent the charging and discharging efficiency of the energy storage device.

[0155] (7) Electricity price constraints, as shown below:

[0156] Formula (22)

[0157] in, For data centers exist Real-time electricity prices for different time periods; and These represent the minimum and maximum retail electricity price coefficients, respectively. This is the benchmark electricity price.

[0158] (8) Electricity purchase constraints, as shown below:

[0159] Formula (23)

[0160] in, and They represent Active and reactive power purchased during specific time periods; , These represent the upper limits for active and reactive power in electricity purchases, respectively.

[0161] (9) Node power balance constraints, as shown below:

[0162] Formula (24)

[0163] in, Indicates a branch With nodes The first node; Indicates a branch With nodes It is the last node; and Branch roads The active and reactive power; branch road Active power loss; branch road Line resistance; branch circuit Reactive power loss on branch road The resistance of the line.

[0164] A low-level model is constructed with the aim of minimizing the economic cost of the data center, primarily the overall operating cost across multiple data centers. Decision variables include workload migration and auxiliary variables associated with workload. The objective function and constraints of the low-level model are as follows.

[0165] 1) Objective function

[0166] The objective function of the low-level model is as follows:

[0167] Formula (25)

[0168] in, The total operating cost of the data center cluster. Number of scheduling periods For scheduling period index, Total number of data centers Index for data centers; For data centers exist Real-time electricity prices for different time periods; For data centers exist Total active power load during the time period; Data Center exist Energy consumption of IT equipment, cooling system, and power distribution equipment during different time periods; This represents the typical PUE value for a data center. This is the ratio of power consumption of IT equipment to power consumption of servers. For data centers exist Total power consumption of all types of servers during the time period; For data centers middle Number of servers of this type; For data centers Mid-table Type server in Power consumption during a given period.

[0169] 2) Constraints

[0170] The data center model is consistent with the DVFS-based model mentioned earlier, and the constraints include:

[0171] 1) Based on PUE, the energy consumption model for DC data load is as follows:

[0172] Formula (26)

[0173] in, This represents the typical PUE value for a data center; This represents the ratio of power consumption of IT equipment to power consumption of servers. Indicates data center exist Total power consumption of all types of servers at any given time; Indicates data center middle Number of servers of this type.

[0174] In summary, the server power consumption can be further expressed as:

[0175] Formula (27)

[0176] in, This represents the CPU's power consumption coefficient. Indicates data center middle Type of server The time period is in the first File operating frequency.

[0177] 2) Calculate load constraints

[0178] Because data center computing loads are flexible and adjustable in time and space, the computing loads to be processed in the data center can be transferred to nodes or time periods with lower electricity prices based on their own latency sensitivity, thereby minimizing their own operating costs.

[0179] The computational load constraints of the low-level model are as follows:

[0180] Formula (28)

[0181] in, , Data centers before and after scheduling exist The number of loads processed within a given time period; , Data centers before and after scheduling exist The amount of latency-sensitive load space migration during the time period; , Data centers before and after scheduling exist Delay-tolerant load migration load within a time period; , Data Center exist Power consumption of loads before and after the dispatch within a time period; This refers to the number of servers within the data center. , Data Center The idle power consumption and full load power consumption of a single server; This refers to the server's processing speed. For data centers Available computing resources reserved ratio; For data centers The proportion of available computing resources; For data centers Total computing resources within;

[0182] and These are the decision variables for the low-level model.

[0183] Latency-sensitive loads have high latency requirements and can only track low electricity prices through spatial migration. The total latency-sensitive load across multiple data centers remains unchanged before and after flexible scheduling. Latency-tolerant loads can track low electricity prices through both spatial and temporal migration.

[0184] The constraints on computing load before and after flexible scheduling in the data center are as follows:

[0185] Formula (29)

[0186] in, and Data Center During the period The number of computationally sensitive loads in pre- and post-processing can be flexibly adjusted. The maximum delay time for a delay-tolerant load; and Data Center During the period The computational tolerance load can be flexibly adjusted before and after processing.

[0187] The purpose of step S1 is to construct a master-slave game two-layer optimization model that integrates the spatiotemporal coupling characteristics of "computing power-electricity", providing an accurate mathematical representation and optimization framework for the collaboration between data centers and virtual power plants.

[0188] Step S2, specifically.

[0189] The process of solving the two-layer collaborative optimization model using the particle swarm optimization algorithm includes the following iterative process:

[0190] Initialize the particle swarm, where the position vector of each particle simultaneously encodes the decision variables of the high-level model and the decision variables of the low-level model;

[0191] For each particle in the current particle swarm, the fitness is calculated using the following master-slave game interaction:

[0192] (1) The virtual power plant, as the leader of the game, releases the decision variables of the high-level model in the particle position as dynamic electricity price signals;

[0193] (2) As a follower in the game, the data center uses the decision variables of the low-level model in the particle position as the workload spatiotemporal migration of the data center under the dynamic electricity price signal.

[0194] (3) Based on the dynamic electricity price signal and the spatiotemporal migration, calculate the net profit of the virtual power plant and the operating cost of the data center, and obtain the fitness value of the particle by weighted summation of the net profit of the virtual power plant and the operating cost of the data center;

[0195] Based on the fitness values ​​of all particles, update the individual historical best position of each particle and the global historical best position of the particle swarm.

[0196] Based on the individual historical best position and the global historical best position, the velocity and position vector of each particle are updated; until the preset iteration termination condition is reached, the final global historical best position is decoded as the optimal collaborative scheduling scheme for the virtual power plant and data center.

[0197] To address the nonlinear coupling characteristics of the two-layer data center-virtual power plant model, a particle swarm optimization (PSO) algorithm is employed for collaborative optimization. Among numerous biomimetic intelligent algorithms, PSO stands out for its fast convergence speed and simple algorithm. It simulates bird predation behavior, leveraging the cooperation and sharing among individuals within a group to achieve the optimal solution to the problem in the solution space.

[0198] In the particle swarm optimization algorithm, particles are used to simulate birds in a flock. Each particle has two attributes: velocity and position. The velocity attribute represents the direction and distance the particle will move in the next iteration, while the position attribute represents a feasible solution to the current problem.

[0199] The particle velocity and formula are as follows:

[0200] Formula (30)

[0201] in, The particle velocity; This represents the current iteration number; Inertial weights; , These represent individual learning factors and group learning factors, respectively, indicating particle... The trend of convergence between the optimal position of an individual and the optimal position of the group; , To increase the randomness of the learning factor, the random numbers are uniformly distributed on [0,1]. Represented as the first of The position of each particle; For the first The velocity of each particle; For the first The optimal position or individual optimal solution for each particle; The optimal position found by the particle swarm search can also be called the swarm optimal solution.

[0202] This invention employs a particle swarm optimization algorithm to solve the problem. The optimization variables include wind turbines, photovoltaics, gas turbines, energy storage, purchased power, and data center load shifting. The fitness value is to minimize the operating cost of the virtual power plant. The solution flowchart is shown below. Figure 4 As shown.

[0203] (1) Initialization: Initialize the particle swarm parameters, including particle swarm size, particle dimension, number of iterations, inertia weight, learning factor, iteration step size range, etc., calculate the optimal fitness value of each particle, and randomly initialize the position and velocity of each particle.

[0204] (2) Update the velocity and position of each particle based on the formula;

[0205] (3) Recalculate the fitness value for each particle;

[0206] (4) Update the individual historical best fitness and position of each particle;

[0207] (5) Update the population's historical best fitness and position;

[0208] (6) Update the relevant parameters and determine whether the maximum number of iterations has been reached. If not, repeat steps (2)-(6). If the number of iterations has been reached, output the optimal position of the population, which is the optimal solution to the problem.

[0209] The particle position vector is shown below:

[0210] Formula (31)

[0211] in, For the first The position of each particle.

[0212] Particle position refers to a set of feasible scheduling schemes, and particle velocity represents the adjustment direction and magnitude of the scheduling scheme, which is dynamically corrected by the objective function gradient and constraints.

[0213] Individual optimal solution refers to the historical optimal scheduling scheme for a single particle, while swarm optimal solution represents the globally optimal scheduling scheme, which is the combination of decision variables that brings the two-level objective functions into equilibrium.

[0214] Construct the fitness function for the particles and calculate their fitness values;

[0215] The fitness function is as follows:

[0216] Formula (32)

[0217] in, , These are the preset normalized weight coefficients, and .

[0218] Step S2 serves to establish a collaborative solution mechanism for synchronously optimizing the objective of a two-layer model by embedding master-slave game interaction into particle swarm iteration, transforming complex non-convex problems into an intelligent search process to efficiently obtain a globally balanced scheduling scheme.

[0219] Step S3, specifically.

[0220] Based on the optimal collaborative scheduling scheme, the power of distributed new energy sources and energy storage in the virtual power plant is controlled, and the spatiotemporal migration scheduling of latency-sensitive and latency-tolerant workloads in the data center cluster is performed.

[0221] The spatiotemporal migration scheduling of latency-sensitive and latency-tolerant workloads within the data center cluster includes:

[0222] The latency-sensitive workloads are constrained to perform spatial transfers between data centers at any time during the scheduling period, while the total amount of processing across all data centers remains constant.

[0223] The latency-tolerant workloads are constrained to perform time-shifting within the data center or spatial shifting between data centers within the maximum allowable latency time, while the total processing volume across all data centers remains unchanged.

[0224] Step S3 transforms the collaborative scheduling scheme obtained from high-level optimization into precise control commands for the physical system, thereby enabling proactive adjustment of the source-storage power on the virtual power plant side and spatiotemporal reconstruction of the computing load on the data center side, completing the closed loop from decision-making to execution.

[0225] This invention uses a virtual power plant and an enterprise-type data center in East China as simulation examples. The system includes one wind turbine, one photovoltaic power unit, one gas turbine, and an energy storage power station, as well as two data centers. Each data center has 4000 servers, and it is assumed that the servers in each data center are of the same model. Fuzzy-processed data from a typical day in one of the data centers is selected for optimization and analysis, with a time step of 1 hour and a scheduling cycle of 1 day (24 hours).

[0226] (1) Data foundation, as shown in Table 1-4, consists of relevant parameters. The table is divided into sections for system correlation coefficients, gas turbine related parameters, data center correlation coefficients, and energy storage related parameters.

[0227] Table 1 System-related parameters

[0228]

[0229] Table 2 Relevant parameters of the gas turbine

[0230]

[0231] Table 3. Correlation coefficient parameters of data centers

[0232]

[0233] Table 4 Energy Storage Related Parameters

[0234]

[0235] The number of latency-sensitive and latency-tolerant workloads in Data Center 1 and Data Center 2 at different times on a typical day are as follows: Figure 5 and Figure 6 As shown.

[0236] (2) Two-layer optimization results: First, the effectiveness of the two-layer collaborative optimization scheduling model for the interaction between multiple data centers and virtual power plants is analyzed. The following two calculation cases are constructed for comparative analysis. The cost and revenue of virtual power plants and the operating cost of data centers under each calculation case are shown in Table 5.

[0237] Scenario 1: Fixed electricity price;

[0238] Scenario 2: Using time-of-use electricity pricing (while also considering the spatiotemporal shift characteristics of data center workloads)

[0239] The optimization results of each unit in the virtual power plant in Scenario 2 are as follows: Figure 7 As shown, the energy storage battery is charged at night and during the midday peak of wind and solar power output, mainly concentrated between 0:00-3:00 and 12:00-14:00. When the energy storage battery's discharge power is insufficient to meet the electricity demand, the gas turbine and energy storage will supplement the power together.

[0240] based on Figure 8 The results show that, influenced by real-time node electricity prices, workload migration occurs at every stage. This is because, on the one hand, the electricity prices vary across nodes where data centers are located. Therefore, from the perspective of minimizing operating costs, data centers will choose the optimal workload migration amount for each time period.

[0241] Table 5. Cost and Benefit Analysis of Virtual Power Plants and Data Centers

[0242]

[0243] A comparison of operational data under two typical scenarios reveals that the real-time node pricing mechanism can simultaneously achieve the dual optimization goals of improving the operational efficiency of virtual power plants and reducing data center operating costs, while the fixed pricing model exhibits the opposite trend. This is because the traditional fixed pricing system, lacking price signal guidance, struggles to effectively stimulate the demand response potential of data centers, resulting in limited resource allocation flexibility. Compared to fixed pricing, time-of-use pricing strategies have demonstrated significant improvements: by guiding data centers to utilize the time-shifting characteristics of latency-tolerant loads for demand response, the daily net profit of virtual power plants increased by 227.01 yuan, and both their electricity purchase costs and gas turbine power generation costs were effectively reduced.

[0244] Further analysis revealed that the real-time node electricity pricing mechanism exhibits superior economic efficiency and system adaptability. This mechanism not only further enhances the daily profits of virtual power plants but also reduces data center operating costs by 17.66 yuan. Its core advantage lies in its perfect alignment with the spatiotemporal coupling characteristics of data center loads: data center operators can respond to demand through workload time migration and perform spatial migration based on electricity price differences between different nodes during the same period, leveraging the differences in energy consumption characteristics of servers across data centers. This two-way interactive mechanism based on dynamic electricity price signals constructs a new model for collaborative optimization between data centers and virtual power plants, achieving the dual goals of improving resource allocation efficiency and enhancing the absorption capacity of wind and solar power renewable energy.

[0245] Example 2:

[0246] A specific embodiment of the present invention discloses a data center and virtual power plant collaborative operation system, thereby realizing the data center and virtual power plant collaborative operation method in Embodiment 1. The specific implementation of each module is described in the corresponding description in Embodiment 1.

[0247] like Figure 9 As shown, a collaborative operation system for a data center and a virtual power plant is disclosed. The system includes a collaborative optimization modeling module M1, an intelligent optimization solution module M2, and a collaborative scheduling execution module M3.

[0248] The collaborative optimization modeling module M1 is used to construct a two-layer collaborative optimization model between the data center and the virtual power plant. The two-layer system optimization model includes a high-level model with the goal of maximizing the net profit of the virtual power plant and a low-level model with the goal of minimizing the total operating cost of the multi-data center cluster.

[0249] The intelligent optimization solution module M2 is used to solve the two-layer collaborative optimization model using the particle swarm optimization algorithm. The decision variables of the high-level and low-level models are jointly encoded as the positions of particles in the particle swarm. By iteratively updating the positions and velocities of the particles, the objective functions of the high-level and low-level models are optimized synchronously. Finally, the output global historical optimal position is decoded to obtain the optimal collaborative scheduling scheme.

[0250] The collaborative scheduling execution module M3 is used to control the power of distributed new energy sources and energy storage in the virtual power plant based on the optimal collaborative scheduling scheme, and to perform spatiotemporal migration scheduling of latency-sensitive and latency-tolerant workloads in the data center cluster.

[0251] Since the system in this embodiment and the method in Embodiment 1 are related and can be referenced from each other, this description is redundant and will not be repeated here. Because this system embodiment shares the same principle as the above method embodiment, it also possesses the corresponding technical effects of the above method embodiment.

[0252] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0253] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for collaborative operation of a data center and a virtual power plant, characterized in that, include: A two-layer collaborative optimization model for data centers and virtual power plants is constructed. The two-layer collaborative optimization model includes a high-level model with the goal of maximizing the net profit of the virtual power plant and a low-level model with the goal of minimizing the total operating cost of the multi-data center cluster. The high-level model integrates the predicted output of distributed renewable energy and electricity market price signals, and uses the load adjustment of the multi-data center cluster in the low-level model as the decision variable. The two-layer collaborative optimization model is solved using a particle swarm optimization algorithm. The decision variables of the high-level and low-level models are jointly encoded as the positions of particles in the particle swarm. By iteratively updating the positions and velocities of the particles, the objective functions of the high-level and low-level models are optimized synchronously. Finally, the output global historical optimal position is decoded to obtain the optimal collaborative scheduling scheme. The process of solving the two-layer collaborative optimization model using the particle swarm optimization algorithm includes the following iterative process: Initialize the particle swarm, where the position vector of each particle simultaneously encodes the decision variables of the high-level model and the decision variables of the low-level model; For each particle in the current particle swarm, the fitness is calculated using the following master-slave game interaction: (1) The virtual power plant, as the leader of the game, releases the decision variables of the high-level model in the particle position as dynamic electricity price signals; (2) As a follower in the game, the data center uses the decision variables of the low-level model in the particle position as the workload spatiotemporal migration of the data center under the dynamic electricity price signal. (3) Based on the dynamic electricity price signal and the spatiotemporal migration, calculate the net profit of the virtual power plant and the operating cost of the data center, and obtain the fitness value of the particle by weighted summation of the net profit of the virtual power plant and the operating cost of the data center; Based on the fitness values ​​of all particles, update the individual historical best position of each particle and the global historical best position of the particle swarm. Based on the individual historical best position and the global historical best position, update the velocity and position vector of each particle; until the preset iteration termination condition is reached, decode the final global historical best position as the optimal collaborative scheduling scheme for the virtual power plant and data center. The fitness function is as follows: in, Net profit of virtual power plants The total operating cost of the data center cluster. , These are the preset normalized weight coefficients, and ; Based on the optimal collaborative scheduling scheme, the power of distributed new energy sources and energy storage in the virtual power plant is controlled, and the time-space migration scheduling of latency-sensitive and latency-tolerant workloads in the data center cluster is performed. The latency-sensitive workloads are constrained to perform spatial transfers between data centers at any time during the scheduling period, while the total amount of processing across all data centers remains constant. The latency-tolerant workloads are constrained to perform time-shifting within the data center or spatial shifting between data centers within the maximum allowable latency time, while the total processing volume across all data centers remains unchanged.

2. The data center and virtual power plant collaborative operation method according to claim 1, characterized in that, The objective function of the high-level model is as follows: in, For the electricity sales revenue of the virtual power plant, For the cost of purchasing electricity for virtual power plants, For the cost of gas turbine power generation, The costs of operation and maintenance of distributed renewable energy sources, as well as the costs of curtailment penalties, For energy storage maintenance costs, For network loss costs, The costs include wind power operation and maintenance, as well as the costs of curtailment penalties. The costs include photovoltaic operation and maintenance costs, as well as the costs of curtailment penalties. Number of scheduling periods; This represents the number of nodes in the substation. express Electricity purchase price for a given period of time; , The first Each substation node is located at Power purchased during specific time periods and its adjustment amounts; This represents the number of gas turbine nodes. This refers to the unit cost of gas turbine power generation. No. Each gas turbine node at Power generation during a given time period; This represents the number of wind turbine nodes. and These are the marginal cost of electricity generated per unit by the wind turbine and the cost of wind curtailment penalty, respectively. , The first Each wind turbine node is at Power generation and power curtailment during a given time period; This represents the number of photovoltaic nodes. and These are the marginal cost of photovoltaic power generation per unit and the cost of curtailment penalty, respectively. , The first Each wind turbine node is at Power generation and power curtailment during a given time period; This represents the number of energy storage nodes. Maintenance costs for energy storage units; , The first One energy storage node in Real-time charging and discharging power during a given period; , These are the charging and discharging efficiencies of the energy storage device; , , , , and These are decision variables for high-level models.

3. The data center and virtual power plant collaborative operation method according to claim 2, characterized in that, The constraints of the high-level model include: The constraints based on DisrFlow include power balance constraints, line transmission capacity constraints, upper and lower limits of safe operation of node voltage, upper and lower limits of gas turbine output and ramp rate constraints, output constraints of distributed new energy sources, and state of charge range and charging and discharging power constraints of energy storage systems.

4. The data center and virtual power plant collaborative operation method according to claim 3, characterized in that, The objective function of the low-level model is as follows: in, Number of scheduling periods For scheduling period index, Total number of data centers Index for data centers; For data centers exist Real-time electricity prices for different time periods; For data centers exist Total active power load during the time period; Data Center exist Energy consumption of IT equipment, cooling system, and power distribution equipment during different time periods; This represents the typical PUE value for a data center. This is the ratio of power consumption of IT equipment to power consumption of servers. For data centers exist Total power consumption of all types of servers during the time period; For data centers middle Number of servers of this type; For data centers Mid-table Type server in Power consumption during a given period.

5. The data center and virtual power plant collaborative operation method according to claim 4, characterized in that, The computational load constraints of the low-level model are as follows: in, , Data centers before and after scheduling exist The number of loads processed within a given time period; , Data centers before and after scheduling exist The amount of latency-sensitive load space migration during the time period; , Data centers before and after scheduling exist Delay-tolerant load migration load within a time period; , Data Center exist Power consumption of loads before and after the dispatch within a time period; For data centers Number of internal servers; , Data Center The idle power consumption and full load power consumption of a single server; This refers to the server's processing speed. For data centers Available computing resources reserved ratio; For data centers The proportion of available computing resources; For data centers Total computing resources within; and These are the decision variables for the low-level model.

6. The data center and virtual power plant collaborative operation method according to claim 5, characterized in that, The particle position vector is shown below: in, For the first The position of each particle.

7. A collaborative operation system for data centers and virtual power plants, characterized in that, The system includes a collaborative optimization modeling module M1, an intelligent optimization solution module M2, and a collaborative scheduling and execution module M3. The collaborative optimization modeling module M1 is used to construct a two-layer collaborative optimization model for data centers and virtual power plants. The two-layer collaborative optimization model includes a high-level model with the goal of maximizing the net profit of the virtual power plant and a low-level model with the goal of minimizing the total operating cost of the multi-data center cluster. The high-level model integrates the predicted output of distributed renewable energy and electricity market price signals, and uses the load adjustment of the multi-data center cluster in the low-level model as the decision variable. The intelligent optimization solution module M2 is used to solve the two-layer collaborative optimization model using the particle swarm optimization algorithm. The decision variables of the high-level and low-level models are jointly encoded as the positions of particles in the particle swarm. By iteratively updating the positions and velocities of the particles, the objective functions of the high-level and low-level models are optimized synchronously. Finally, the output global historical optimal position is decoded to obtain the optimal collaborative scheduling scheme. The process of solving the two-layer collaborative optimization model using the particle swarm optimization algorithm includes the following iterative process: Initialize the particle swarm, where the position vector of each particle simultaneously encodes the decision variables of the high-level model and the decision variables of the low-level model; For each particle in the current particle swarm, the fitness is calculated using the following master-slave game interaction: (1) The virtual power plant, as the leader of the game, releases the decision variables of the high-level model in the particle position as dynamic electricity price signals; (2) As a follower in the game, the data center uses the decision variables of the low-level model in the particle position as the workload spatiotemporal migration of the data center under the dynamic electricity price signal. (3) Based on the dynamic electricity price signal and the spatiotemporal migration, calculate the net profit of the virtual power plant and the operating cost of the data center, and obtain the fitness value of the particle by weighted summation of the net profit of the virtual power plant and the operating cost of the data center; Based on the fitness values ​​of all particles, update the individual historical best position of each particle and the global historical best position of the particle swarm. Based on the individual historical best position and the global historical best position, update the velocity and position vector of each particle; until the preset iteration termination condition is reached, decode the final global historical best position as the optimal collaborative scheduling scheme for the virtual power plant and data center. The fitness function is as follows: in, Net profit of virtual power plants The total operating cost of the data center cluster. , These are the preset normalized weight coefficients, and ; The collaborative scheduling execution module M3 is used to control the power of distributed new energy and energy storage in the virtual power plant based on the optimal collaborative scheduling scheme, and to perform spatiotemporal migration scheduling of latency-sensitive and latency-tolerant workloads in the data center cluster. The latency-sensitive workloads are constrained to perform spatial transfers between data centers at any time during the scheduling period, while the total amount of processing across all data centers remains constant. The latency-tolerant workloads are constrained to perform time-shifting within the data center or spatial shifting between data centers within the maximum allowable latency time, while the total processing volume across all data centers remains unchanged.

Images