A double-layer scheduling method for multi-energy microgrid of data center based on virtual energy storage

Abstract

The application discloses a kind of based on virtual energy storage's data center multi-energy microgrid double-layer scheduling method, comprising: the model of multiple energy renewable energy system including wind power, photovoltaic, electrochemical energy storage and hydrogen energy storage is established;The calculated load is divided into high, medium and low priority, and the delay tolerance and user satisfaction quantification model is established, and the delayable task is modeled as virtual energy storage resource;Establish IT equipment and refrigeration equipment energy consumption model, form the scheduling relationship on time and space scale;The double-layer optimization model with minimum running cost in upper layer and minimum user comprehensive dissatisfaction in lower layer is constructed and solved.The present application realizes the joint optimization of economy and service quality through the layer-by-layer progressive mechanism of virtual energy storage and physical energy storage cooperation, energy consumption state feedback adjustment delay cost and optimization air supply temperature update energy consumption model, effectively reduces running cost and carbon emission.

Description

Technical Field

[0001] This invention relates to the field of power-load coordinated control technology, and in particular to a two-layer scheduling method for multi-energy microgrids in data centers based on virtual energy storage. Background Technology

[0002] The surge in AI computing power presents data centers with challenges in energy consumption and high carbon emissions. To reduce operating costs and ensure environmental benefits, the integration of data center microgrids with renewable energy is imperative. However, the intermittent nature of renewable energy sources limits their efficient utilization. Combining data centers with microgrids that integrate renewable energy, and optimizing scheduling to improve energy efficiency, reduce costs, and decrease carbon emissions, is a key path to promoting high-quality development of data center computing power.

[0003] Chinese Patent Publication No. CN120147060A discloses a two-layer deep reinforcement learning scheduling method for data center microgrids, comprising: constructing a two-layer multi-agent deep reinforcement learning framework containing a global layer and a local layer; the global layer allocates computing tasks through centralized deep reinforcement learning training; the local layer optimizes internal energy scheduling through distributed deep reinforcement learning; and a hierarchical multi-agent dual-delay deep deterministic policy gradient algorithm is combined to achieve cross-layer interaction and rapid convergence. The method also reduces non-renewable energy costs through spatiotemporal task adjustment and energy storage collaboration. However, this patent has the following problems: First, its virtual energy storage modeling lacks granularity, failing to finely classify high, medium, and low priorities according to task real-time requirements, and also failing to establish a quantitative relationship between latency tolerance and user satisfaction. Second, the patent does not involve long-term energy storage modeling for hydrogen energy storage systems, making it unable to support energy regulation across diurnal scales. Third, its two-layer reinforcement learning framework is a one-way parameter transfer mechanism, lacking a progressive closed-loop mechanism for adjusting latency costs through energy consumption status feedback and updating the energy consumption model with optimization results, making it difficult to achieve deep collaborative optimization of economic efficiency and service quality.

[0004] Chinese Patent Publication No. CN121663562A discloses a method and system for collaborative operation of a data center and a virtual power plant, comprising: constructing a two-layer collaborative optimization model for the data center and the virtual power plant, wherein the high-layer model aims to maximize the net profit of the virtual power plant, and the low-layer model aims to minimize the total operating cost of the multi-data center cluster; using a particle swarm optimization algorithm to solve the two-layer collaborative optimization model, encoding the decision variables of the high-layer and low-layer models together as the positions of particles in the particle swarm, and simultaneously optimizing the objective functions of the high-layer and low-layer models by iteratively updating the positions and velocities of the particles; and controlling the power of distributed new energy sources and energy storage in the virtual power plant based on the optimal collaborative scheduling scheme, and performing spatiotemporal migration scheduling for latency-sensitive and latency-tolerant workloads in the data center cluster. Therefore, the patent has the following problems: First, it only classifies workloads into two categories, delay-sensitive and delay-tolerant, which is too coarse and fails to establish a precise quantitative model of user satisfaction and latency. Second, the patent does not involve hydrogen energy storage systems and cannot fully utilize the inter-day regulation capability of hydrogen energy storage. Third, it uses particle swarm optimization algorithm to simultaneously optimize high and low level targets, but it does not establish a dynamic feedback mechanism for energy consumption status to task scheduling decisions, nor does it realize the dynamic adjustment of the cooling system's air supply temperature based on the optimization results, resulting in insufficient collaborative optimization capability of cooling and power linkage. Summary of the Invention

[0005] To address these issues, this invention provides a two-layer scheduling method for multi-energy microgrids in data centers based on virtual energy storage, which overcomes the problems of insufficient coordination between task scheduling and energy management, coarse granularity of virtual energy storage modeling, lack of hydrogen energy storage systems, and low coupling between cooling energy consumption and task scheduling in existing technologies.

[0006] To achieve the above objectives, this invention provides a two-layer scheduling method for multi-energy microgrids in data centers based on virtual energy storage, the method comprising:

[0007] Modeling of wind power, photovoltaic power and hydrogen energy storage batteries based on power generation data; and establishing a multi-energy renewable energy system model.

[0008] Based on the data load characteristics, a delay tolerance model and a user satisfaction quantification model are established to obtain the delayed task volume, and the multi-energy renewable energy system model is optimized based on the delayed task volume.

[0009] The energy consumption characteristics of the data center are modeled to obtain the energy consumption characteristic model and total energy consumption, and the coefficients of the latency tolerance model are adjusted according to the total energy consumption.

[0010] With the objective function of minimizing data center operating costs, constraints are set for energy supply balance, IT load balance, and minimum user satisfaction. An economic scheduling and management model is established to obtain the optimal load scheduling results, energy supply scheme, and optimal cooling system supply air temperature. The energy consumption characteristic model is then updated with parameters based on the optimal cooling system supply air temperature.

[0011] Furthermore, the multi-energy renewable energy system model includes: establishing a piecewise function-based wind power generation model based on real-time wind speed and wind turbine operating parameters, which is used to describe the power output characteristics of the wind turbine in different wind speed ranges.

[0012] Furthermore, the multi-energy renewable energy system model includes: establishing a photovoltaic power generation model based on solar irradiance, ambient temperature, and photovoltaic module performance parameters, wherein solar irradiance is described using a cosine function of sunrise and sunset times, and the photovoltaic cell temperature considers the coupling effect of irradiance absorption and ambient temperature.

[0013] Furthermore, the multi-energy renewable energy system model includes: establishing an energy storage system update model based on the energy storage system's charging and discharging efficiency, self-discharge rate, and rated power, and introducing 0-1 variables to constrain the mutual exclusivity of charging and discharging states;

[0014] Based on the energy conversion efficiency and loss characteristics of electrolyzers, hydrogen storage tanks, and fuel cells, an energy state update model for the entire process of hydrogen production in electrolyzers, dynamic balancing of hydrogen storage tanks, and power generation in fuel cells is established.

[0015] Furthermore, the establishment of the latency tolerance model and the user satisfaction quantification model specifically includes: the load of different data can be divided into high-priority interactive load and medium- and low-priority batch processing load, and their latency tolerance characteristic models are established respectively.

[0016] Furthermore, the delay tolerance characteristic model includes:

[0017] For high-priority interactive loads, a dual-channel processing model of local processing and remote migration is established as a high-priority load latency tolerance model. 0-1 variables and network quality adjustment factors are used to constrain the remote migration capability to ensure the real-time response requirements of the load.

[0018] For medium-priority batch processing loads, a backlog queue dynamic update model is established as a latency tolerance characteristic model for medium-priority loads. The average waiting time is used to describe the task processing latency, and the immediate processing cost and dynamic latency cost are defined. A cost ratio function is introduced as the basis for scheduling decisions.

[0019] For low-priority batch processing loads, a backlog queue and average waiting time model are established as a latency tolerance characteristic model for low-priority loads. A time decay factor is introduced into the dynamic latency cost to encourage the processing of backlog tasks as early as possible within the scheduling cycle.

[0020] The delay task quantity calculated based on the delay tolerance characteristics of medium and low priority loads is used as the state of charge constraint for the energy storage system update model.

[0021] Furthermore, when modeling the energy consumption characteristics of a data center, an energy consumption characteristic model is established based on the data center's operating status. This energy consumption characteristic model includes a load energy consumption model and a cooling system energy consumption model, resulting in total energy consumption.

[0022] Furthermore, when adjusting the coefficients of the delay tolerance model based on the total energy consumption, the total energy consumption is used as a feedback parameter to dynamically adjust the delay cost coefficient of the delay tolerance model in step S2: when the total energy consumption is close to the upper limit of the data center power supply, the delay cost coefficient of medium and low priority tasks is increased, prompting the scheduling system to postpone the tasks to low load periods or migrate them to off-site processing.

[0023] Furthermore, when establishing an economic scheduling and management model for the operating costs of data centers, an objective function for minimizing data center operating costs is constructed. The total cost is decomposed into the energy cost of the data center microgrid, the battery loss cost of energy storage equipment, the delay compensation cost of each priority task, and the cost of the hydrogen energy storage system. Balance constraints are established, and a lower-level optimization constraint with the goal of minimizing satisfaction is constructed to obtain a comprehensive dissatisfaction index.

[0024] Furthermore, when updating the parameters of the energy consumption characteristic model, the energy consumption of the data center load and the energy consumption of the cooling equipment are substituted into the balance constraints to obtain the optimal cooling system supply air temperature, and the optimal cooling system supply air temperature is used as the supply air temperature of the cooling equipment and input into the cooling system energy consumption model.

[0025] Compared with existing technologies, the beneficial effects of this invention are as follows: First, by dividing the computing load into three priorities—high, medium, and low—and establishing a delay tolerance model and a user satisfaction quantification model respectively, deferred tasks are modeled as virtual energy storage resources with time-shifting capabilities. These resources work in conjunction with physical energy storage such as electrochemical energy storage and hydrogen energy storage, enabling data centers to process non-urgent tasks during periods of low electricity prices or abundant renewable energy, and postpone processing during periods of power shortage, effectively smoothing the load curve and improving the renewable energy absorption capacity. Second, a comprehensive model of a multi-energy renewable energy system, including wind power, photovoltaics, electrochemical energy storage, and hydrogen energy storage, is established. In particular, the entire process of electrolyzers, hydrogen storage tanks, and fuel cells is modeled in detail, providing technical support for energy regulation on a trans-diurnal scale. Third, a two-layer optimization model is constructed, with the upper layer aiming at minimizing operating costs and the lower layer aiming at minimizing overall user dissatisfaction. Energy costs, energy storage loss costs, delay compensation costs, and hydrogen energy storage system costs are uniformly incorporated into the optimization objective, and the service quality minimum is guaranteed through satisfaction constraints. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the two-layer scheduling method for a data center multi-energy microgrid based on virtual energy storage in this embodiment.

[0027] Figure 2 A flowchart illustrating the method for establishing a multi-energy renewable energy system model in this embodiment;

[0028] Figure 3 This is a flowchart illustrating the method for establishing the delay tolerance model and the user satisfaction quantification model in this embodiment. Detailed Implementation

[0029] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0030] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0031] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0032] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0033] Example 1

[0034] Please see Figure 1 As shown, this is a flowchart illustrating the two-layer scheduling method for a data center multi-energy microgrid based on virtual energy storage in this embodiment. The method includes:

[0035] Step S1: Model wind power generation, photovoltaic power generation and hydrogen energy storage battery based on power generation data to establish a multi-energy renewable energy system model;

[0036] Step S2: Based on the data load characteristics, establish a delay tolerance model and a user satisfaction quantification model to obtain the delayed task volume, and optimize the multi-energy renewable energy system model based on the delayed task volume.

[0037] Step S3: Model the energy consumption characteristics of the data center to obtain the energy consumption characteristic model and total energy consumption, and adjust the coefficients of the latency tolerance model according to the total energy consumption.

[0038] Step S4: With the lowest data center operating cost as the objective function, constraints are set for energy supply balance, IT load balance, and minimum user satisfaction. An economic scheduling management model is established to obtain the optimal load scheduling result, energy supply scheme, and optimal cooling system supply air temperature. The energy consumption characteristic model is then updated with parameters based on the optimal cooling system supply air temperature.

[0039] In this implementation, the dual-layer scheduling method for multi-energy microgrids in data centers based on virtual energy storage is applied to data center microgrids. The beneficial effects of this invention are as follows: First, by dividing the computing load into three priorities—high, medium, and low—and establishing a delay tolerance model and a user satisfaction quantification model respectively, deferred tasks are modeled as virtual energy storage resources with time-shifting capabilities. These resources are then coordinated with physical energy storage such as electrochemical energy storage and hydrogen energy storage, enabling data centers to process non-urgent tasks during periods of low electricity prices or abundant renewable energy, and to postpone processing during periods of power shortage, effectively smoothing the load curve and improving the renewable energy absorption capacity. Second, a comprehensive model of a multi-energy renewable energy system, including wind power, photovoltaics, electrochemical energy storage, and hydrogen energy storage, is established. In particular, the entire process of electrolyzers, hydrogen storage tanks, and fuel cells is modeled in detail, providing technical support for energy regulation on a trans-daily scale. Third, a dual-layer optimization model is constructed with the upper layer aiming at minimizing operating costs and the lower layer aiming at minimizing overall user dissatisfaction. Energy costs, energy storage loss costs, delay compensation costs, and hydrogen energy storage system costs are uniformly incorporated into the optimization objective, and the service quality lower limit is guaranteed through satisfaction constraints.

[0040] Example 2

[0041] Please see Figure 2 As shown, this is the method for establishing a multi-energy renewable energy system model in this embodiment:

[0042] In this embodiment, as an optional implementation, the establishment of a multi-energy renewable energy system model in step S1 specifically includes:

[0043] Based on real-time wind speed and wind turbine operating parameters, a piecewise function-form wind power generation model is established to describe the power output characteristics of the wind turbine in different wind speed ranges.

[0044] A photovoltaic power generation model is established based on solar irradiance, ambient temperature, and photovoltaic module performance parameters. Solar irradiance is described by a cosine function of sunrise and sunset times, and the photovoltaic cell temperature takes into account the coupling effect of irradiance absorption and ambient temperature.

[0045] Based on the charging and discharging efficiency, self-discharge rate, and rated power of the energy storage system, an energy storage system update model is established, and 0-1 variables are introduced to constrain the mutual exclusivity of the charging and discharging states.

[0046] Based on the energy conversion efficiency and loss characteristics of electrolyzers, hydrogen storage tanks, and fuel cells, an energy state update model for the entire process of hydrogen production in electrolyzers, dynamic balancing of hydrogen storage tanks, and power generation in fuel cells is established.

[0047] In this embodiment, as another optional implementation, the wind power generation model in step S1 is as follows:

[0048]

[0049] Among them, constraints This represents the wind power generation capacity under ideal conditions;

[0050] Typical scenario Downdraft fan Output power at any moment For wind turbine Wind speed at any moment The cut-in wind speed for the fan. To cut off the wind speed for the fan. The wind speed at which the fan reaches its rated power. This refers to the rated power of the fan.

[0051] In this embodiment, as another optional implementation, the photovoltaic power generation model in step S1 is as follows:

[0052]

[0053] Among them, constraints It is a photovoltaic power generation model that uses solar energy to convert into electrical energy, with constraints. It is the solar irradiance that varies with sunrise and sunset times, constraining It is a model of the temperature change of photovoltaic cells;

[0054] for The output power of the photovoltaic power generation system at all times for Solar irradiance at any time Irradiance under standard test conditions. Rated output power under standard test conditions. For power temperature coefficient, for Ambient temperature at all times The reference temperature is under standard test conditions. For overall system efficiency, Peak irradiance, This is a weather correction factor that reflects the effects of cloud cover, atmospheric transparency, etc. for Photovoltaic cell temperature at all times For battery photoelectric conversion efficiency, The heat dissipation coefficient reflects the battery's ability to dissipate heat to the environment. is the irradiation absorption coefficient.

[0055] In this embodiment, as an optional example, in step S1, the energy storage system update model includes an energy state update model for an electrochemical energy storage system and an energy state update model for a hydrogen energy storage system. The energy state update model for the electrochemical energy storage system is as follows:

[0056]

[0057] Among them, constraints This considers the chemical energy storage power under the condition of charge / discharge state update, and is constrained. - Constraints on the charge and discharge states of chemical energy storage;

[0058] In the scene Down The energy stored in an electrochemical energy storage system at any time In the scene Down The energy stored in an electrochemical energy storage system at any time The self-discharge rate of the energy storage system. , Scenes Down Real-time energy storage system charging and discharging efficiency , Scenes Down The charging and discharging power of the energy storage system at all times. For unit scheduling duration, The rated charge and discharge power of the energy storage system. , For 0-1 variables, =1, the energy storage system is in a charging state. =1, the energy storage system is in a discharging state, and charging and discharging cannot occur simultaneously;

[0059] The energy state update model of the hydrogen energy storage system is as follows:

[0060]

[0061] in, This represents the actual hydrogen production power of the electrolyzer. This refers to the input electrical power of the electrolytic cell. The energy conversion efficiency of the electrolytic cell, The hydrogen production of the electrolyzer, This refers to the calorific value of hydrogen. This refers to the self-loss rate of the electrolytic cell. for The amount of hydrogen stored in the hydrogen tank at any given time. For the maximum hydrogen storage in the hydrogen tank, The minimum safe hydrogen storage capacity of a hydrogen storage tank. For the output electrical power of the fuel cell, For the power generation efficiency of fuel cells, The self-loss rate of the hydrogen storage system, The amount of hydrogen consumed by the fuel cell.

[0062] In this implementation, the real-time wind speed and wind turbine operating parameters refer to the instantaneous wind speed at the height of the wind turbine hub. The operating parameters of the wind turbine include the cut-in wind speed. Cut-off wind speed Rated wind speed and rated power of the fan The aforementioned wind speed data is collected in real time by an anemometer installed on the top of the wind turbine nacelle or on a wind measurement tower, and various thresholds are determined in conjunction with the technical parameter tables provided by the wind turbine manufacturer. The solar irradiance, ambient temperature, and photovoltaic module performance parameters refer to parameters including solar irradiance. The instantaneous irradiance intensity refers to the intensity measured in real time on a horizontal plane or the tilted surface of a photovoltaic array using an irradiometer; ambient temperature. Data is collected via a temperature sensor; photovoltaic module performance parameters include irradiance under standard test conditions. Rated output power Power temperature coefficient Reference temperature Overall system efficiency Peak irradiance Weather correction factor Battery photoelectric conversion efficiency Heat dissipation coefficient and radiation absorption coefficient The above parameters are mainly derived from the specifications provided by photovoltaic module manufacturers or obtained through on-site calibration experiments. The weather correction factor can be determined based on statistical analysis of local historical meteorological data. The charge / discharge efficiency, self-discharge rate, and rated power of the energy storage system refer to the charging efficiency of the electrochemical energy storage system. Discharge efficiency Self-discharge rate The battery management system (BMS) monitors battery voltage, current, temperature, and other state data in real time, and obtains this data by combining it with battery characteristic curves provided by the manufacturer or through periodic charge-discharge experiments; the state of charge (SOC) of the energy storage system is thus determined. The energy conversion efficiency and loss characteristics of the electrolyzer, hydrogen storage tank, and fuel cell are estimated in real time by the BMS using algorithms such as the ampere-hour integral method. These characteristics include the energy conversion efficiency of the electrolyzer. and self-loss rate Hydrogen storage self-loss rate of hydrogen storage tank and the upper and lower limits of safe hydrogen storage capacity , and the power generation efficiency of its fuel cells All parameters, including hydrogen production, were obtained from the technical manuals provided by the equipment manufacturer. and consumption The amount of hydrogen stored in the hydrogen storage tank can be monitored in real time by a flow meter installed on the hydrogen pipeline. It is calculated using pressure and temperature sensors combined with the gas state equation.

[0063] In this implementation, by establishing high-precision models for wind power generation, photovoltaic power generation, electrochemical energy storage, and hydrogen energy storage, this invention can accurately describe the volatility and intermittency of renewable energy output, as well as the dynamic response characteristics of various energy storage media. This provides reliable energy-side boundary conditions for subsequent steps, effectively supports the synergistic optimization of virtual and physical energy storage, and lays a solid model foundation for reducing data center operating costs and improving renewable energy absorption rates.

[0064] Example 3

[0065] Please see Figure 3 As shown, this is the method for establishing the delay tolerance model and the user satisfaction quantification model in this embodiment:

[0066] In this implementation, as an optional embodiment, in step S2, a latency tolerance model and a user satisfaction quantification model are established, specifically including: the load of different data can be divided into high-priority interactive load and medium and low-priority batch processing load, and their latency tolerance characteristic models are established respectively.

[0067] For high-priority interactive loads, a dual-channel processing model of local processing and remote migration is established as a high-priority load latency tolerance model, and 0-1 variables and network quality adjustment factors are used to constrain remote migration capabilities.

[0068] For medium-priority batch processing loads, a backlog queue dynamic update model is established as a latency tolerance characteristic model for medium-priority loads. The average waiting time is used to describe the task processing latency, and the immediate processing cost and dynamic latency cost are defined. A cost ratio function is introduced as the basis for scheduling decisions.

[0069] For low-priority batch processing loads, a backlog queue and average waiting time model are established as a latency tolerance characteristic model for low-priority loads. A time decay factor is introduced into the dynamic latency cost to encourage the processing of backlog tasks as early as possible within the scheduling cycle.

[0070] Establish a quantitative model for user satisfaction under medium and low priority loads, set a basic satisfaction level and a decay coefficient, use a minimum satisfaction constraint to ensure the lower limit of service quality, and optionally use a piecewise nonlinear satisfaction function to describe the sharp decline in satisfaction after timeout.

[0071] The delay task quantity calculated based on the delay tolerance characteristics of medium and low priority loads is used as the state of charge constraint for the energy storage system update model in step S1.

[0072] In this embodiment, as another optional implementation, the high-priority load delay tolerance characteristic model in step S2 is as follows:

[0073]

[0074] in, For local processing of high-priority power; For off-site processing of high-priority power; For high-priority arrivals; For local server capacity; For network transmission efficiency This refers to the available network bandwidth. for Moment-based network quality adjustment factor; This is a function indicating the nighttime period.

[0075] In this embodiment, as another optional implementation, the medium-priority load delay tolerance characteristic model in step S2 is as follows:

[0076]

[0077] in, for The amount of priority backlogged tasks at any given time; for The amount of priority backlogged tasks at any given time; for The number of priority tasks arriving at a given time; for Actual processing power of priority at any given moment; for Average waiting time for priority levels at any given moment; for Average waiting time for priority levels at any given moment; To minimize the value and prevent division by zero errors; for The cost of immediate processing of priority items at any given moment; for Dynamic delay cost of priority at any given moment; This is the basic delay cost coefficient for medium priority. The maximum waiting time for medium priority; for Priority cost ratio at any given time; for Prioritize user satisfaction at any given moment; This is the penalty coefficient for medium-priority satisfaction.

[0078] In this embodiment, as another optional implementation, the low-priority load delay tolerance characteristic model in step S2 is as follows:

[0079]

[0080] in, for Low-priority backlog of tasks at all times; for Low-priority backlog of tasks at all times; for The arrival rate of low-priority tasks at any given time; for Actual processing power of low-priority tasks at any given time; for Average waiting time for low-priority time; for Average waiting time for low-priority time; for Low-priority dynamic delay cost at any time; This is the low-priority basic delay cost coefficient; Maximum waiting time for low priority; for Low priority cost ratio at any time; for Low-priority user satisfaction at all times; This is a low-priority satisfaction penalty coefficient.

[0081] In this implementation, as an optional embodiment, the user satisfaction quantification model for the medium and low priority loads in step S2 is as follows:

[0082]

[0083] in, Medium priority Satisfaction at any given moment; Low priority Satisfaction at any given moment; =0.95, =0.85 is the baseline satisfaction level; =0.6, =0.4 is the satisfaction decay coefficient; This represents the average waiting time. =8h, =16h is the maximum acceptable waiting time; The threshold for minimum satisfaction; The decrease in satisfaction is quantified as economic cost; the latency of batch processing load is also considered.

[0084] In this embodiment, as an alternative implementation, the state of charge constraint in step S2 is as follows:

[0085]

[0086] in, This is the standard standby capacity. As an impact factor, when the amount of delayed tasks is large, the energy storage system needs to reserve more capacity to cope with the additional load impact brought about by the delayed task processing and prevent insufficient power supply.

[0087] In this embodiment, the dual-channel processing model of local processing and remote migration refers to a high-priority interactive load simultaneously having two processing channels: local data center processing and processing via wide area network migration to other data centers, as constrained. As shown, local processing power is constrained by the capacity of the local server, such as... As shown, the power of remote migration is jointly constrained by network transmission efficiency, available bandwidth, and network quality adjustment factors, such as the constraints. As shown, the 0-1 variables and network quality adjustment factors refer to the mutual exclusivity of the 0-1 variables used to characterize the charging and discharging states of the energy storage system. When =1, the energy storage system is in a charging state. When =1, it is in a discharge state; both cannot be 1 simultaneously, as per the constraint. - As shown; Network quality adjustment factor Used to adjust for the actual available bandwidth during off-site migration, such as constraints. As shown, the value is 1.25 during the night and 1.0 during the day, reflecting the time-varying characteristics of network load. The backlog queue dynamic update model refers to the dynamic update of medium-priority tasks. Backlog of tasks at any time Equal to the backlog at the previous moment Adding the newly arrived task volume at this moment Subtract the actual processing power at this moment Such as constraints As shown, the average waiting time refers to the time taken for medium-priority tasks to complete. Average waiting time of time When there were backlogged tasks in the previous time step, the result is equal to the product of the average waiting time and the number of backlogged tasks in the previous time step, plus 1, divided by the sum of the number of backlogged tasks and the minimum value, as constrained. As shown, this is used to quantify the latency of a task. The immediate processing cost and dynamic latency cost refer to the immediate processing cost. Equal to the current grid electricity price, if constrained As shown; Dynamic delay cost It equals the base delay cost factor multiplied by the ratio of average waiting time to maximum waiting time, as per constraints. As shown, this model measures the economic losses caused by delayed processing. The backlog queue and average waiting time model refers to a method for low-priority tasks to describe backlog updates and average waiting time calculations in a similar manner to medium-priority tasks, such as constraints. - As shown, however, a time decay factor is introduced to adjust the calculation method of delay cost. The time decay factor refers to the correction coefficient in the calculation of low-priority delay costs, such as constraints. As shown, backlogged tasks are encouraged to be processed as early as possible at the end of the scheduling cycle to avoid indefinite delays. The basic satisfaction and decay coefficient refer to the basic satisfaction of medium-priority tasks. Take 0.95, low-priority basic satisfaction. Set to 0.85; Medium priority satisfaction decay coefficient Set to 0.6, the satisfaction decay coefficient for low priority levels. A value of 0.4 is used to describe the rate at which satisfaction decreases as waiting time increases, as in constraints. - As shown, the minimum satisfaction level refers to the lower limit constraint that the satisfaction level of medium and low priority tasks must meet. To ensure that the quality of user service is not lower than the basic level, such as constraints As shown, when the satisfaction level of any priority is lower than the threshold, the scheduling scheme is not feasible. The piecewise nonlinear satisfaction function refers to describing the relationship between satisfaction and waiting time in a piecewise form, such as the constraint. As shown, the satisfaction level after timeout refers to the satisfaction level when the task waiting time exceeds the maximum acceptable waiting time. Afterwards, user satisfaction dropped sharply, such as due to constraints. As shown in the second half, the rate of decline in satisfaction is from and This joint decision reflects users' strong dissatisfaction with severe delays. The aforementioned state of charge constraint refers to the requirement that the stored energy of the electrochemical energy storage system must meet upper and lower limits, while also being limited by charge / discharge efficiency, self-discharge rate, and rated power. - As shown, ensure that energy storage devices operate within safe limits.

[0088] In this embodiment, by dividing the computing load into three priorities—high, medium, and low—and establishing a latency tolerance model and a user satisfaction quantification model respectively, the present invention can accurately identify delayable tasks and model them as virtual energy storage resources with time-shifting capabilities. These virtual energy storage resources work in conjunction with the physical energy storage system in step S1, enabling the data center to process non-urgent tasks during periods of low electricity prices or abundant renewable energy, and postpone processing during periods of high electricity prices or power shortages. This smooths the load curve and reduces operating costs. At the same time, through the constraints of the satisfaction model, the present invention effectively ensures the quality of user service, avoids the experience loss caused by excessive latency, and achieves a balance between economic efficiency and service quality.

[0089] Example 4

[0090] In this embodiment, as an optional example, in step S3, an energy consumption characteristic model is established based on the operating status of the data center. The energy consumption characteristic model includes a load energy consumption model and a cooling system energy consumption model to obtain the total energy consumption.

[0091] Using the total energy consumption as a feedback parameter, the delay cost coefficient of the delay tolerance model in step S2 is dynamically adjusted: when the total energy consumption is close to the upper limit of the data center power supply, the delay cost coefficient of medium and low priority tasks is increased, prompting the scheduling system to postpone the tasks to low load periods or migrate them to off-site processing.

[0092] In this embodiment, as another optional implementation, the load energy consumption model is as follows:

[0093]

[0094] in, Data center load energy consumption; This refers to the CPU's power consumption. This refers to the energy consumption of memory. This refers to the power consumption of the hard drive. For the network card's power consumption; for Energy consumption of the equipment.

[0095] In this embodiment, as another optional implementation, the energy consumption model of the refrigeration system is as follows:

[0096]

[0097] in, for Energy consumption of the cooling equipment corresponding to the server at any given time; This refers to the air supply temperature of the refrigeration equipment.

[0098] In this embodiment, as another optional example, the delay cost coefficient of the delay tolerance model in the state adjustment step S2 is as follows:

[0099]

[0100] in, For data centers Energy load rate at any given time; The rated power supply capacity for the data center; The adjusted medium-priority base delay cost coefficient; The adjusted low-priority base delay cost coefficient; This is a dynamically adjusted function; This is the energy consumption early warning threshold.

[0101] In this embodiment, the operating status of the data center refers to the data center's... Real-time operating status, including actual power consumption of IT equipment. Power consumption of refrigeration equipment and air supply temperature Key operating parameters, such as those mentioned above, collectively determine the total energy consumption level of a data center, and constraints... - As shown, the feedback parameter refers to the total energy consumption calculated in step S3. As a feedback signal, the latency cost coefficient of the latency tolerance model in step S2 is dynamically adjusted. The low-load period refers to the time interval during which the total energy consumption of the data center is lower than a certain percentage of the rated power supply capacity. This typically corresponds to off-peak electricity prices at night or periods of renewable energy surplus. Processing non-urgent tasks during this period can effectively reduce operating costs. Migrating to off-site processing refers to transmitting high-priority interactive loads to other data centers via a wide area network for processing when the total energy consumption of the local data center approaches its power supply limit, as constrained. As shown, the upper limit of power supply for the data center refers to the upper limit of power supply for the data center microgrid system. The maximum safe power supply capacity at any given time includes the sum of electricity purchased from the grid, renewable energy generation, energy storage discharge, and fuel cell output. When the total energy consumption approaches this upper limit, a load adjustment mechanism needs to be activated. The energy consumption warning threshold is... This refers to the critical value used to determine whether a data center is operating under high load; when the energy consumption load rate... When the threshold is exceeded, dynamic adjustment of the delay cost coefficient is initiated. This embodiment does not limit the specific setting of the energy consumption warning threshold. Those skilled in the art can set it according to actual conditions. For example, based on the historical operating data of the data center, the energy consumption warning threshold can be set to 80% of the rated power supply capacity, and the delay cost coefficient can be increased when the load rate exceeds 80%; or based on the time-of-use electricity price structure of the power grid, the energy consumption warning threshold corresponding to the peak electricity price period can be lowered to 70% to trigger load adjustment earlier; or based on the predicted power of renewable energy, the energy consumption warning threshold for the period when renewable energy output is insufficient can be set to 75% to address the power supply gap in advance.

[0102] In this embodiment, by establishing a refined energy consumption model for IT equipment and cooling equipment, the present invention achieves accurate calculation and spatiotemporal coupling analysis of the total energy consumption of the data center. The total energy consumption is used as a feedback parameter to dynamically adjust the delay cost coefficient in step S2, forming a closed-loop control mechanism for energy consumption status and task scheduling. When the total energy consumption approaches the power supply limit, the delay cost coefficient of medium and low priority tasks is automatically increased, prompting the scheduling system to postpone non-urgent tasks to low-load periods or migrate them to other locations for processing, thereby avoiding power overload and reducing electricity costs under peak electricity prices. At the same time, the energy consumption model in step S3 provides an accurate energy consumption benchmark for the optimization solution in step S4, enabling reliable input data for the optimization adjustment of the cooling system's air supply temperature, and ultimately achieving a synergistic improvement in the energy efficiency and economy of the data center.

[0103] Example 5

[0104] In this implementation, as an optional embodiment, in step S4, an objective function for minimizing the data center operating cost is constructed, and the total cost is decomposed into the data center microgrid energy cost, energy storage device battery loss cost, delay compensation cost for each priority task, and hydrogen energy storage system cost. Balance constraints are established, and lower-level optimization constraints with the goal of minimizing satisfaction are constructed.

[0105] Substituting the data center load energy consumption and the cooling equipment energy consumption into the balance constraints, the optimal cooling system supply air temperature is obtained, and this optimal cooling system supply air temperature is then used as the supply air temperature of the cooling equipment in the cooling system energy consumption model. middle.

[0106] In this embodiment, as another optional implementation, the objective function for minimizing the data center operating cost is as follows:

[0107]

[0108] in, Energy costs for data center microgrids; Cost of battery wear and tear for energy storage devices; Cost compensation for delays in each priority task; Cost of hydrogen energy storage systems.

[0109] In this embodiment, as another optional implementation, the total cost is as follows:

[0110]

[0111] in, Power purchased from the power grid; Photovoltaic power generation capacity; This refers to the power output of wind power generation. For diesel generator power; Battery charging power This refers to the battery discharge power. The electricity price for a time-varying power grid; For photovoltaic costs; For wind power costs; Cost of diesel generators; For battery wear and tear costs; For medium-priority tasks, the dynamic compensation coefficient is used. For low-priority tasks, a dynamic compensation coefficient is used. Delaying the workload due to low priority tasks; The amount of delayed tasks is for medium-priority tasks; This is the unit power operating cost coefficient for the electrolytic cell; This represents the unit power operating cost coefficient for fuel cells. This is the unit capacity operating cost coefficient for the hydrogen storage system. The decline in satisfaction is quantified as an economic cost.

[0112] In this embodiment, as another optional implementation, the balance constraint is as follows:

[0113]

[0114] in This represents the actual power consumption of data center equipment. Power consumption of data center cooling equipment systems; This represents the actual hydrogen production power of the dehydrogenation tank; For the energy consumption of other auxiliary equipment; Power purchased from the power grid; Photovoltaic power generation; This refers to the power output of wind power generation. For diesel generator power; Battery charging power; This refers to the battery discharge power. This refers to the output electrical power of the fuel cell.

[0115] In this implementation, as an alternative embodiment, the lower-level optimization constraint with the goal of minimizing satisfaction is as follows:

[0116]

[0117] in, To provide a comprehensive index of dissatisfaction; Priority At any moment User satisfaction; To delay the workload; This is the average delay time; This represents the penalty coefficient for delay time.

[0118] In this embodiment, the optimal supply air temperature of the cooling system refers to the optimal setpoint value of the supply air temperature of the cooling system obtained by solving the balance constraints in step S4. This temperature value is within the allowable range of the server's air intake temperature, typically 18℃ to 27℃. Adjusting the supply air temperature affects the energy consumption of the cooling equipment, such as the constraints. As shown, the delay time penalty coefficient This refers to the weighting parameters in the overall user dissatisfaction objective function used to balance the loss of satisfaction and the amount of delayed tasks. Such as constraints As shown, this coefficient reflects the degree of impact of latency on user experience. This embodiment does not limit the specific setting of the latency penalty coefficient. Those skilled in the art can set it according to the actual situation. For example, according to the service level agreement requirements of the data center, the latency penalty coefficient can be set to 0.5 to reflect the equal importance of delayed task volume and satisfaction loss; or according to user survey data on latency sensitivity, the latency penalty coefficient can be set to 0.8 to strengthen the penalty for task delay; or according to the correlation analysis between user complaint rate and latency in historical operation data, the latency penalty coefficient can be set to 0.3 to reflect the actual user tolerance; or according to the difference in task priority, the latency penalty coefficient for medium priority tasks can be set to 0.6, and for low priority tasks to 0.2, reflecting the difference in latency tolerance for different priority tasks.

[0119] In this embodiment, by constructing an upper-level optimization model aimed at minimizing operating costs and a lower-level optimization model aimed at minimizing overall user dissatisfaction, the present invention achieves synergistic optimization of economic efficiency and service quality. The upper-level model incorporates energy costs, energy storage loss costs, latency compensation costs, and hydrogen energy storage system costs into the optimization objective, and ensures stable operation of the data center through power balance constraints. The lower-level model weights and sums user satisfaction loss and latency task load, and ensures a lower limit for service quality through satisfaction constraints.

[0120] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A two-layer scheduling method for multi-energy microgrids in data centers based on virtual energy storage, characterized in that, The method includes: Modeling of wind power, photovoltaic power and hydrogen energy storage batteries based on power generation data; and establishing a multi-energy renewable energy system model. Based on the data load characteristics, a delay tolerance model and a user satisfaction quantification model are established to obtain the delayed task volume, and the multi-energy renewable energy system model is optimized based on the delayed task volume. The energy consumption characteristics of the data center are modeled to obtain the energy consumption characteristic model and total energy consumption, and the coefficients of the latency tolerance model are adjusted according to the total energy consumption. With the objective function of minimizing data center operating costs, constraints are set for energy supply balance, IT load balance, and minimum user satisfaction. An economic scheduling and management model is established to obtain the optimal load scheduling results, energy supply scheme, and optimal cooling system supply air temperature. The energy consumption characteristic model is then updated with parameters based on the optimal cooling system supply air temperature.

2. The two-layer scheduling method for data center multi-energy microgrids based on virtual energy storage according to claim 1, characterized in that, The multi-energy renewable energy system model includes: establishing a piecewise function-based wind power generation model based on real-time wind speed and wind turbine operating parameters, which is used to describe the power output characteristics of the wind turbine in different wind speed ranges.

3. The two-layer scheduling method for data center multi-energy microgrids based on virtual energy storage according to claim 1, characterized in that, The multi-energy renewable energy system model includes: establishing a photovoltaic power generation model based on solar irradiance, ambient temperature, and photovoltaic module performance parameters, wherein solar irradiance is described by a cosine function of sunrise and sunset times, and the photovoltaic cell temperature takes into account the coupling effect of irradiance absorption and ambient temperature.

4. The two-layer scheduling method for data center multi-energy microgrids based on virtual energy storage according to claim 1, characterized in that, The multi-energy renewable energy system model includes: establishing an energy storage system update model based on the energy storage system's charging and discharging efficiency, self-discharge rate, and rated power, and introducing 0-1 variables to constrain the mutual exclusivity of charging and discharging states; Based on the energy conversion efficiency and loss characteristics of electrolyzers, hydrogen storage tanks, and fuel cells, an energy state update model for the entire process of hydrogen production in electrolyzers, dynamic balancing of hydrogen storage tanks, and power generation in fuel cells is established.

5. The two-layer scheduling method for data center multi-energy microgrids based on virtual energy storage according to claim 1, characterized in that, The establishment of the latency tolerance model and the user satisfaction quantification model specifically includes: the load of different data can be divided into high-priority interactive load and medium and low-priority batch processing load, and a latency tolerance characteristic model is established for each.

6. The two-layer scheduling method for data center multi-energy microgrids based on virtual energy storage according to claim 5, characterized in that, The delay tolerance characteristic model includes: For high-priority interactive loads, a dual-channel processing model of local processing and remote migration is established as a high-priority load latency tolerance model. 0-1 variables and network quality adjustment factors are used to constrain the remote migration capability to ensure the real-time response requirements of the load. For medium-priority batch processing loads, a backlog queue dynamic update model is established as a latency tolerance characteristic model for medium-priority loads. The average waiting time is used to describe the task processing latency, and the immediate processing cost and dynamic latency cost are defined. A cost ratio function is introduced as the basis for scheduling decisions. For low-priority batch processing loads, a backlog queue and average waiting time model are established as a latency tolerance characteristic model for low-priority loads. A time decay factor is introduced into the dynamic latency cost to encourage the processing of backlog tasks as early as possible within the scheduling cycle. The delay task quantity calculated based on the delay tolerance characteristics of medium and low priority loads is used as the state of charge constraint for the energy storage system update model.

7. The two-layer scheduling method for data center multi-energy microgrids based on virtual energy storage according to claim 1, characterized in that, When modeling the energy consumption characteristics of a data center, an energy consumption characteristic model is established based on the data center's operating status. The energy consumption characteristic model includes a load energy consumption model and a cooling system energy consumption model to obtain the total energy consumption.

8. The two-layer scheduling method for data center multi-energy microgrids based on virtual energy storage according to claim 7, characterized in that, When adjusting the coefficients of the delay tolerance model based on total energy consumption, the total energy consumption is used as a feedback parameter to dynamically adjust the delay cost coefficient of the delay tolerance model: when the total energy consumption is close to the upper limit of the data center power supply, the delay cost coefficient of medium and low priority tasks is increased, prompting the scheduling system to postpone the tasks to low load periods or migrate them to off-site processing.

9. The two-layer scheduling method for data center multi-energy microgrids based on virtual energy storage according to claim 1, characterized in that, When establishing the economic scheduling and management model for the operating cost of the data center, an objective function for minimizing the operating cost of the data center is constructed. The total cost is decomposed into the energy cost of the data center microgrid, the battery loss cost of the energy storage equipment, the delay compensation cost of each priority task, and the cost of the hydrogen energy storage system. Balance constraints are established, and a lower-level optimization constraint with the goal of minimizing satisfaction is constructed to obtain a comprehensive dissatisfaction index.

10. The two-layer scheduling method for data center multi-energy microgrids based on virtual energy storage according to claim 9, characterized in that, When updating the parameters of the energy consumption characteristic model, the energy consumption of the data center load and the energy consumption of the cooling equipment are substituted into the balance constraints to obtain the optimal cooling system supply air temperature, and the optimal cooling system supply air temperature is used as the supply air temperature of the cooling equipment and input into the cooling system energy consumption model.

Images