Data center energy management method based on load flexibility mining

Abstract

The invention relates to a data center energy management method based on load flexibility mining, and the method comprises the following steps: S1, carrying out the modeling of the output characteristics of a fan and a photovoltaic panel, and building a wind-solar renewable energy output model; s2, establishing a data center load model according to the delay sensitivity difference of the data load; s3, modeling the energy consumption characteristics of the data center, and establishing the relation between the data center load and the data center load curve; and S4, taking the lowest power cost of the data center as an objective function, considering the energy supply constraint and the service level protocol constraint of the data center, establishing an economic energy management model of the data center, and solving to obtain an optimal load scheduling result and an optimal energy management scheme. The method can fully mine the time transfer characteristic of the data center load, consumes clean energy, and reduces the power cost of the data center.

Description

Technical Field

[0001] The present invention relates to the technical field of power system dispatching, and more specifically, to a data center energy management method based on load flexibility mining. Background Art

[0002] In recent decades, increasing carbon emissions have led to global climate change. The power industry produces electricity through fossil fuels, which generates a large amount of carbon emissions. Using renewable energy to produce electricity is an effective way to reduce carbon emissions. With the development of artificial intelligence, cloud computing and other fields, which has brought huge computing needs, the data center industry has developed rapidly, and the power load has grown rapidly. The huge energy consumption will generate large energy costs and carbon emissions, which will restrict the development of data centers. Therefore, realizing the low-carbon economic transformation of data centers is the key to the sustainable development of data centers.

[0003] Most existing methods focus on optimizing the subsystems of data centers, reducing energy consumption by optimizing the IT equipment system, cooling system, and network system of the data center. However, the electricity cost of the data center is affected by the price of the power grid. How to utilize the delay-tolerant load working mode of the data center, fully absorb renewable energy, and utilize peak and valley electricity prices to reduce electricity costs is the next step in the development of economic energy conservation and emission reduction in data centers. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a data center energy management method based on load flexibility mining, which can solve the problem that the existing data center energy management method cannot fully exploit the time transfer characteristics of the load, thereby resulting in higher electricity costs.

[0005] The technical solution adopted by the present invention to solve the technical problem is: constructing a data center energy management method based on load flexibility mining, comprising the following steps:

[0006] S1, model the output characteristics of wind turbines and photovoltaic panels, and establish a wind and solar renewable energy output model;

[0007] S2, establishes a data center load model based on the delay sensitivity differences of data loads;

[0008] S3, modeling the energy consumption characteristics of the data center and establishing the connection between the data center load and the data center load curve;

[0009] S4, taking the lowest electricity cost of the data center as the objective function, considering the energy supply constraints and service level agreement constraints of the data center, establishes an economic energy management model of the data center, and solves to obtain the optimal load scheduling results and energy management solutions.

[0010] According to the above scheme, in step S1, the fan output model is as follows:

[0011]

[0012] P wt,t =n wind,t (1-η wind )C i (w t ) (2)

[0013] Among them, constraint (1) represents the wind turbine output under ideal conditions, and constraint (2) is the corrected wind turbine output considering multiple factors such as wake effect, reliability and wind turbine efficiency.

[0014] C i (w) is the fan output, w t is the wind speed at time t, w in is the wind turbine cut-in wind speed, w rated is the rated wind speed of the fan, w out is the fan cut-out wind speed, η wind is the wake effect coefficient, n wind,t is the number of fans at time t, P wt, t is the wind power at time t, R is the maximum output power of the wind turbine;

[0015] The photovoltaic panel output model is as follows:

[0016]

[0017] sinβ=sinφsinδ+cosφcosδcosω (4)

[0018]

[0019] τ d =0.271-0.274τ b (7)

[0020] J b =J0τ b sinα (8)

[0021]

[0022] J 0t =J b +J d (10)

[0023] R t =cosθ / sinα (11)

[0024]

[0025] Among them, constraint (3) is the photovoltaic output, which uses the photovoltaic effect to convert light energy into electrical energy, constraints (4)-(10) are the total solar radiation intensity considering the air quality, direct solar radiation and diffuse solar radiation, constraint (11) is the ratio of the solar radiation intensity on the slope at the tth hour to the total radiation intensity on the photovoltaic panel, and constraint (12) is the total solar radiation intensity;

[0026] P pv，t is the photovoltaic output at time t, P stc is the rated output of photovoltaic power, J stc is the solar radiation intensity under standard conditions, α z is the temperature coefficient of photovoltaic power generation, T c / T STC is the test temperature in the real environment / standard environment, β is the solar altitude angle, φ is the latitude, δ is the solar declination angle, ω is the hour angle, τ b is the transparency coefficient of direct sunlight under clear sky conditions, M h is the atmospheric mass, h is the altitude, τ d is the transparency coefficient of diffuse solar radiation under clear sky conditions, J b is the direct solar radiation intensity, J d is the diffuse solar radiation intensity, J 0t is the total solar radiation intensity, ρ is the reflectivity of the ground, p / q is a parameter related to the atmospheric quality, R t The ratio of the solar radiation intensity on an inclined plane to the solar radiation intensity on a horizontal plane, J t is the total solar radiation intensity on the photovoltaic panel at time t, k t is the cleaning index at time t.

[0027] According to the above scheme, in step S2, the data center performs delay-tolerant data load during the period of low electricity price and sufficient renewable energy so as to adjust the power load of the data center;

[0028] Constraints (13)-(15) represent the processing time of delay-sensitive loads. The processing time of delay-sensitive loads is determined by the waiting time and working time of the data load. The waiting time of the workload is determined by the M / M / 1 queuing model, and the working time of the workload is calculated based on the service rate of the data center.

[0029]

[0030] Among them, T t int , are the processing time, waiting time, and working time of the delay-sensitive load at time t, is the number of delay-sensitive data loads processed at time t, is the number of delay-sensitive data payloads arriving at time t, n t is the number of servers working at time t;

[0031] Constraints (16)-(17) are the time transfer characteristics of delay-tolerant workloads, which only need to be processed before the specified deadline;

[0032]

[0033]

[0034] Among them, T t del is the processing time of the delay-tolerant load at time t, t wd is the maximum delay time for delay-tolerant loads, is the number of delay-tolerant loads transferred from time t to time τ, is the number of delay-tolerant loads processed by the data center at time t.

[0035] According to the above scheme, in step S3,

[0036] The energy consumption of the data center consists of the power consumption of IT equipment and the power consumption of cooling equipment. The IT equipment includes the power consumption of servers and the power consumption of data networks, which are all related to the number of servers working, as shown in constraints (18)-(21);

[0037] P t =P IT，t +P cool，t (18)

[0038] P IT，t =P ser,t +P net，t (19)

[0039] P ser，t =(α ser f+β ser ) t (20)

[0040] P net，t =α net n t +β net (twenty one)

[0041] Among them, P t is the power load of the data center at time t, P IT，t is the IT equipment power at time t, P cool，t is the refrigeration equipment power at time t, P ser，t is the server power at time t, P net，tis the data network power at time t, f is the operating frequency of the server, α ser / β ser / α net / β net The data related to the parameters and power load obtained for statistical analysis;

[0042] The number of servers that perform the work is calculated by the processing time of the data load and is limited by the total number of servers in the data center, as shown in constraints (22)-(24):

[0043]

[0044]

[0045] 0≤n t ≤n max (twenty four)

[0046] Among them, λ t is the number of workloads processed by the data center at time t, n max is the maximum number of servers, η ser is the efficiency factor for converting CPU frequency into the service rate of the server, T t is the processing time of the data load at time t;

[0047] Constraint (25) is the energy efficiency of the data center, which is the ratio of the total power consumption of the data center to the power consumption of IT equipment:

[0048]

[0049] Among them, εpu e Energy efficiency for data centers.

[0050] According to the above scheme, in step S4,

[0051] The objective function is to minimize the power cost of the data center, as shown in formula (26):

[0052]

[0053] Among them, C buy is the electricity cost of the data center, a t is the electricity price at time t, P t buy is the amount of electricity purchased by the power grid at time t, and T is the total time period of the dispatch cycle;

[0054] Service level agreement constraints (27)-(30) are contracts between IT users and data centers. They are the basic principles for scheduling workloads and specify the number and processing time of workloads. Constraint (27) limits the processing time of delay-sensitive workloads. Constraints (28)-(29) require that workload loss should be avoided during workload processing. Constraint (30) requires that delay-tolerant workloads should be executed within a specified time.

[0055]

[0056]

[0057] Among them, U int The maximum delay time for delay-sensitive loads, is the transfer amount of delay-tolerant data load at time t, is the number of delay-tolerant data loads on server σ at time t, is the number of delay-tolerant data loads waiting to be processed from server σ at time t, is the number of delay-tolerant data payloads that should be processed by the deadline, t1 is the deadline of the data payload, and t0 is the arrival time of the data payload;

[0058] Constraints (31)-(40) are the energy supply constraints of the data center. Constraints (31)-(35) are the energy storage charging and discharging constraints. Constraint (31) represents the state of charge constraint of the energy storage. Constraints (32)-(33) limit the maximum charging and discharging capacity of the energy storage. Constraint (34) indicates that the battery cannot be charged and discharged at the same time. Constraint (35) is the relationship between the energy storage capacity and the charging and discharging power. Constraint (36) limits the maximum and minimum power generation of photovoltaics. Constraint (37) is the power balance equation to ensure energy conservation in the data center. Constraint (38) limits the power purchase of the data center from the power grid. Constraint (39) limits the carbon dioxide emissions of the park. These emissions come from the electricity purchased from the power grid. Wind and solar power generation in the data center park will not produce carbon emissions. Constraint (39) is the backup capacity of the power system. Remember to change

[0059]

[0060] 0≤P cha，t ≤P max ξ cha，t (32)

[0061] 0≤P dch，t ≤p max ξ dch,t (33)

[0062] ξ cha,t +ξ dch，t ≤1 (34)

[0063]

[0064] P buy，t +P pv，t +P wt，t +ξ dch,t P dch,t =ξ cha，t P cha,t +P sell，t +P t (37)

[0065]

[0066] P pv，t +P wt，t +ξ dch，t P dch,t +P ＝buy，t ≥(1+α)P t (40)

[0067] in, is the minimum / maximum capacity of energy storage, p max is the maximum charging and discharging power of energy storage, E N is the rated capacity of energy storage, is the maximum / minimum output of photovoltaic, is the maximum amount of electricity purchased by the power grid, e ca is the maximum annual carbon emission, l is the carbon dioxide emission factor, α is the reserve capacity coefficient, is the energy storage capacity at time t, P dch，t / P cha，t is the energy storage charging and discharging power at time t, ξ dch，t / ξ cha，t is the energy storage charging and discharging state variable at time t, P sell，t is the amount of electricity sold to the grid at time t.

[0068] The data center energy management method based on load flexibility mining of the present invention has the following beneficial effects:

[0069] The present invention proposes a data center energy management solution that takes into account data load transferability, renewable energy, and dynamic electricity prices. Based on the external grid electricity price, wind and solar power output of renewable energy, and the storage capacity of energy storage, the solution fully exploits the time transfer characteristics of the data center load, optimizes the processing time of delay-tolerant data loads, and the energy storage charging and discharging process, absorbs clean energy, and effectively reduces the electricity cost of the data center. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0071] Figure 1 is a flow chart of a data center energy management method based on load flexibility mining of the present invention;

[0072] Figure 2 It is a schematic diagram of the processing time of the delay-tolerant load in scenario III and scenario IV of the data center energy management method based on load flexibility mining of the present invention;

[0073] Figure 3 The power load diagram of the data center in four cases of the present invention;

[0074] Figure 4 The power load diagram of the data center of the present invention; Figure 5 This is the optimal power supply diagram for the data center of the present invention. DETAILED DESCRIPTION

[0075] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described in detail with reference to the accompanying drawings.

[0076] like Figure 1 As shown, the data center energy management method based on load flexibility mining of the present invention comprises the following steps:

[0077] S1. Model the output characteristics of wind turbines and photovoltaic panels and establish a wind and solar renewable energy output model.

[0078] The fan output model is as follows:

[0079]

[0080] P wt，t =n wind，t (1-η wind )C i (w t ) (2)

[0081] Among them, constraint (1) represents the wind turbine output under ideal conditions, and constraint (2) is the corrected wind turbine output considering multiple factors such as wake effect, reliability and wind turbine efficiency.

[0082] C i (w) is the fan output, w t is the wind speed at time t, w in is the wind turbine cut-in wind speed, w rated is the rated wind speed of the fan, w out is the fan cut-out wind speed, η wind is the wake effect coefficient, n wind，t is the number of fans at time t, P wt，t is the wind power at time t, and R is the maximum output power of the wind turbine.

[0083] The photovoltaic panel output model is as follows:

[0084]

[0085] sinβ=sinφsinδ+cosφcosδcosω (4)

[0086]

[0087] τ d =0.271-0.274τ b (7)

[0088] J b =J0τ b sinα (8)

[0089]

[0090] J 0t =J b +J d (10)

[0091] R t =cosθ / sinα (11)

[0092]

[0093] Among them, constraint (3) is the photovoltaic output, which uses the photovoltaic effect to convert light energy into electrical energy. Constraints (4)-(10) are the total solar radiation intensity taking into account the atmospheric quality, direct solar radiation and diffuse solar radiation. Constraint (11) is the ratio of the solar radiation intensity on the slope at the tth hour to the total radiation intensity on the photovoltaic panel. Constraint (12) is the total solar radiation intensity.

[0094] P pv,t is the photovoltaic output at time t, P stc is the rated output of photovoltaic power, J stc is the solar radiation intensity under standard conditions, α z is the temperature coefficient of photovoltaic power generation, T c / T STC is the test temperature in the real environment / standard environment. β is the solar altitude angle, φ is the latitude, δ is the solar declination angle, ω is the hour angle, τ b is the transparency coefficient of direct sunlight under clear sky conditions, M h is the atmospheric mass, h is the altitude, τ d is the transparency coefficient of diffuse solar radiation under clear sky conditions, J b is the direct solar radiation intensity, J d is the diffuse solar radiation intensity, J0t is the total solar radiation intensity, ρ is the reflectivity of the ground, p / q is a parameter related to the atmospheric quality, R t The ratio of the solar radiation intensity on an inclined plane to the solar radiation intensity on a horizontal plane, J t is the total solar radiation intensity on the photovoltaic panel at time t, k t is the cleaning index at time t.

[0095] S2. Establish a data center load model based on the delay sensitivity differences of data loads.

[0096] According to the different requirements for delay tolerance, the data load of the data center can be divided into delay-sensitive data load and delay-tolerant data load. Delay-sensitive data loads require real-time processing, such as web browsing, online video, game interaction, real-time communication and other user requests. Delay-tolerant data loads, such as data mining, digital image processing and program processing, only require to be completed within a certain request time. Therefore, the data center can execute delay-tolerant data loads during periods of low electricity prices and sufficient renewable energy to adjust the power load of the data center.

[0097] Constraints (13)-(15) represent the processing time of delay-sensitive workloads. The processing time of delay-sensitive workloads is mainly determined by the waiting time and working time of the data workload. The waiting time of the workload is determined by the M / M / 1 queuing model, and the working time of the workload is calculated based on the service rate of the data center.

[0098]

[0099] Among them, T t int , are the processing time, waiting time, and working time of the delay-sensitive load at time t, is the number of delay-sensitive data loads processed at time t, is the number of delay-sensitive data payloads arriving at time t, n t is the number of servers working at time t.

[0100] Constraints (16)-(17) are the time transfer characteristics of delay-tolerant loads, which only need to be processed before the specified deadline.

[0101]

[0102] Among them, T t del is the processing time of the delay-tolerant load at time t, t wd is the maximum delay time for delay-tolerant loads, is the number of delay-tolerant loads transferred from time t to time τ, is the number of delay-tolerant loads processed by the data center at time t.

[0103] S3. Model the energy consumption characteristics of the data center and establish the connection between the data center load and the data center load curve.

[0104] The energy consumption of a data center is mainly composed of IT equipment power consumption and cooling equipment power consumption. IT equipment includes server power consumption and data network power consumption, both of which are related to the number of servers working, as shown in constraints (18)-(21).

[0105] P t =P IT,t +P cool,t (18)

[0106] P IT，t =P ser,t +P net，t (19)

[0107] P ser,t =(α ser f+β ser ) t (20)

[0108] P net，t =α net n t +β net (twenty one)

[0109] Among them, P t is the power load of the data center at time t, P IT，t is the IT equipment power at time t, P cool，t is the refrigeration equipment power at time t, P ser，t is the server power at time t, P net，t is the data network power at time t, f is the operating frequency of the server, α ser / β ser / α net / β net The parameters obtained for statistical analysis are related to the data of power load.

[0110] The number of servers to work on is calculated by the processing time of the data load and is limited by the total number of servers in the data center, as shown in constraints (22)-(24).

[0111]

[0112] 0≤n t ≤n max (twenty four)

[0113] Among them, λ t is the number of workloads processed by the data center at time t, n max is the maximum number of servers, η ser is the efficiency factor for converting CPU frequency into the service rate of the server, T t is the processing time of the data load at time t.

[0114] Constraint (25) is the energy efficiency of the data center, which is the ratio of the total power consumption of the data center to the power consumption of IT equipment.

[0115]

[0116] Among them, ε pue Energy efficiency for data centers.

[0117] S4. Taking the lowest electricity cost of the data center as the objective function, considering the energy supply constraints and service level agreement constraints of the data center, an economic energy management model of the data center is established to obtain the optimal load scheduling results and energy management solutions.

[0118] As shown in formula (26), the objective function is to minimize the electricity cost of the data center.

[0119]

[0120] Among them, C buy is the electricity cost of the data center, a t is the electricity price at time t, P t buy is the amount of electricity purchased by the power grid at time t, and T is the total time period of the dispatch cycle.

[0121] SLA constraints (27)-(30) are contracts between IT users and data centers. They are the basic principles for scheduling workloads and specify the number and processing time of workloads. Constraint (27) limits the processing time of latency-sensitive workloads. Constraints (28)-(29) require that workload loss be avoided during workload processing. Constraint (30) requires that latency-tolerant workloads should be executed within a specified time.

[0122]

[0123]

[0124] Among them, U int The maximum delay time for delay-sensitive loads, is the transfer amount of delay-tolerant data load at time t, is the number of delay-tolerant data loads on server σ at time t, is the number of delay-tolerant data loads waiting to be processed from server σ at time t, is the number of delay-tolerant data payloads that should be processed by the deadline, t1 is the deadline of the data payload, and t0 is the arrival time of the data payload.

[0125] Constraints (31)-(40) are the energy supply constraints of the data center. Constraints (31)-(35) are energy storage charging and discharging constraints. Constraint (31) represents the state of charge constraint of the energy storage. Constraints (32)-(33) limit the maximum charging and discharging capacity of the energy storage. Constraint (34) indicates that the battery cannot be charged and discharged at the same time. Constraint (35) is the relationship between the energy storage power and the charging and discharging power. Constraint (36) limits the maximum and minimum power generation of photovoltaics. Constraint (37) is the power balance equation, which ensures the conservation of energy in the data center. Constraint (38) limits the purchase of electricity from the data center grid. Constraint (39) limits the carbon dioxide emissions of the park, which come from the electricity purchased from the grid. Wind and solar power generation in the data center park does not produce carbon emissions. Constraint (39) is the backup capacity of the power system. In order to improve the reliability of the power system, the power system needs to have a certain backup power capacity to cope with load fluctuations and failures of facilities such as photovoltaic energy storage, thereby improving the flexibility of the system. Remember to change

[0126]

[0127] 0≤P cha，t ≤P max ξ cha,t (32)

[0128] 0≤P dch，t ≤p max ξ dch，t (33)

[0129] ξ cha,t +ξ dch,t ≤1 (34)

[0130]

[0131] P buy，t +P pv，t +P wt，t +ξ dch,t P dch,t =ξ cha，t P cha,t +P sell，t +P t (37)

[0132]

[0133] P pv，t +P wt，t +ξdch,t P dch,t +P buy，t ≥(1+α)P t (40)

[0134] in, is the minimum / maximum capacity of energy storage, p max is the maximum charging and discharging power of energy storage, E N is the rated capacity of the energy storage, is the maximum / minimum output of photovoltaic, is the maximum amount of electricity purchased by the power grid, e ca is the maximum annual carbon emission, l is the carbon dioxide emission factor, α is the reserve capacity coefficient, is the energy storage capacity at time t, P dch，t / P cha，t is the energy storage charging and discharging power at time t, ξ dch，t / ξ cha，t is the energy storage charging and discharging state variable at time t, P sell，t is the amount of electricity sold to the grid at time t.

[0135] Example

[0136] Four scenarios are designed to illustrate the effectiveness of the proposed method:

[0137] Scenario I: The data center processes the data load in a timely manner without considering energy storage.

[0138] Scenario II: Data centers process data loads in a timely manner and consider energy storage.

[0139] Scenario III: The data center processes delay-sensitive loads in a timely manner and flexibly schedules delay-tolerant loads without considering energy storage.

[0140] Scenario IV: Data centers process delay-sensitive loads in a timely manner, flexibly schedule delay-tolerant loads, and consider energy storage.

[0141] Table 1 compares the electricity cost, electricity purchased from the grid, and the number of servers working under different scenarios. Scenario I has the highest electricity cost because it does not consider energy storage. In this scenario, the lack of flexibility prevents cheaper electricity from being purchased from the grid. In Scenario II, energy storage is considered to add flexibility to the power system. Therefore, in the second scenario, energy storage helps to store cheaper electricity purchased from the grid, so the electricity cost is reduced. In Scenario III, data centers can flexibly schedule delay-tolerant workloads, which enhances the flexibility of the power system. Compared with Scenario II, Scenario III has lower electricity costs and electricity purchased from the grid. Delay-tolerant loads are processed during periods of low electricity prices and abundant renewable energy, resulting in a higher number of active servers at a specific time. Scenario IV has a lower electricity cost compared to Scenario III. In this scenario, the increased system flexibility allows it to purchase more electricity when electricity prices are low, so the amount of electricity purchased from the grid is larger at this time.

[0142] Table 1 Operation results in four scenarios

[0143]

[0144] Figure 1 The figure shows the processing time of the delay-tolerant workload in Case III and Case IV. In Scenario III, the delay-tolerant workload remains relatively stable throughout the scheduling period. In Scenario IV, the delay-tolerant workload varies greatly throughout the scheduling period. In hours 13-14 and 23-24, the data center processes a large amount of delay-tolerant workload due to low electricity prices and abundant renewable energy. In hours 6-8 and 19-22, due to high electricity prices and insufficient renewable energy, less delay-tolerant workload is processed.

[0145] Figure 2 The power load of the data center in four scenarios is described. The load curves of scenarios I and II are similar, and the power consumption curves of scenarios III and IV are similar. The difference in power load between scenarios I and II, and scenarios III and IV is caused by battery charging and discharging. Because the power load changes according to the change in the processing time of the delay-tolerant load, there is a significant difference in the power load between scenarios I and II compared with scenarios III and IV.

[0146] like Figure 3 As shown in the figure, the power load of the data center includes IT equipment load, cooling equipment load and energy storage charging. The order of power load is: IT equipment load, cooling equipment load, energy storage discharge. The data center requires a lot of energy to maintain the operation of IT equipment. The cooling equipment also consumes a lot of energy to maintain a constant temperature.

[0147] like Figure 4As shown, the data center is powered by the grid, energy storage, and renewable energy. The power supply is ranked as follows: wind power generation 20903kWh, grid-purchased electricity 20274kWh, photovoltaic power supply 4905kWh, and energy storage 1089kWh. The renewable energy system is used to supply power first, and when renewable energy is insufficient, electricity is purchased from the grid. During hours 1-5, 17, 23, and 24, the energy storage is charged due to low electricity prices. During hours 19-22, the energy storage is discharged due to high electricity prices. Energy storage can economically shift electricity from renewable energy and grid-purchased electricity from low demand / price periods to high demand / price periods.

[0148] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A data center energy management method based on load flexibility mining, characterized in that: The following steps are involved: S1, model the output characteristics of wind turbines and photovoltaic panels, and establish a wind and solar renewable energy output model; S2, establishes a data center load model based on the delay sensitivity differences of data loads; S3, modeling the energy consumption characteristics of the data center and establishing the connection between the data center load and the data center load curve; S4, taking the lowest electricity cost of the data center as the objective function, considering the energy supply constraints and service level agreement constraints of the data center, establishes an economic energy management model of the data center, and solves to obtain the optimal load scheduling results and energy management solutions.

2. The data center energy management method based on load flexibility mining according to claim 1 is characterized in that: In step S1, the fan output model is as follows: P wt,t =n wind，t (1st) wind )C i (W t ) (2) Among them, constraint (1) represents the wind turbine output under ideal conditions, and constraint (2) represents the corrected wind turbine output considering multiple factors such as wake effect, reliability and wind turbine efficiency. C i (w) is the fan output, w t is the wind speed at time t, w in is the wind turbine cut-in wind speed, w rated is the rated wind speed of the fan, w out is the fan cut-out wind speed, η wind is the wake effect coefficient, n wind,t is the number of fans at time t, P wt，t is the wind power at time t, R is the maximum output power of the wind turbine; The photovoltaic panel output model is as follows: sinβ=sinφsinδ+cosφcosδcosω (4) t d =0.271-0.274t b (7) J b =J0τ b China (8) I 0t =J b +J d (10) R t =cosθ / sin α(11) Among them, constraint (3) is the photovoltaic output, which uses the photovoltaic effect to convert light energy into electrical energy, constraints (4)-(10) are the total solar radiation intensity considering the air quality, direct solar radiation and diffuse solar radiation, constraint (11) is the ratio of the solar radiation intensity on the slope at the tth hour to the total radiation intensity on the photovoltaic panel, and constraint (12) is the total solar radiation intensity; P pv,t is the photovoltaic output at time t, P stc is the rated output of photovoltaic power, J stc is the solar radiation intensity under standard conditions, α z is the temperature coefficient of photovoltaic power generation, T c / T STC is the test temperature in the real environment / standard environment, β is the solar altitude angle, φ is the latitude, δ is the solar declination angle, ω is the hour angle, τ b is the transparency coefficient of direct sunlight under clear sky conditions, M h is the atmospheric mass, h is the altitude, τ d is the transparency coefficient of diffuse solar radiation under clear sky conditions, J b is the direct solar radiation intensity, J d is the diffuse solar radiation intensity, J 0t is the total solar radiation intensity, ρ is the reflectivity of the ground, p / q is a parameter related to the atmospheric quality, R t The ratio of the solar radiation intensity on an inclined plane to the solar radiation intensity on a horizontal plane, J t is the total solar radiation intensity on the photovoltaic panel at time t, k t is the cleaning index at time t.

3. The data center energy management method based on load flexibility mining according to claim 1 is characterized in that: In step S2, the data center executes delay-tolerant data load during a period of low electricity price and sufficient renewable energy so as to adjust the power load of the data center; Constraints (13)-(15) represent the processing time of delay-sensitive loads. The processing time of delay-sensitive loads is determined by the waiting time and working time of the data load. The waiting time of the workload is determined by the M / M / 1 queuing model, and the working time of the workload is calculated according to the service rate of the data center. in, are the processing time, waiting time, and working time of the delay-sensitive load at time t, is the number of delay-sensitive data loads processed at time t, is the number of delay-sensitive data payloads arriving at time t, n t is the number of servers working at time t; Constraints (16)-(17) are the time transfer characteristics of delay-tolerant workloads, which only need to be processed before the specified deadline; in, is the processing time of the delay-tolerant load at time t, t wd is the maximum delay time for delay-tolerant loads, is the number of delay-tolerant loads transferred from time t to time τ, is the number of delay-tolerant loads processed by the data center at time t.

4. The data center energy management method based on load flexibility mining according to claim 1 is characterized in that: In the step S3, The energy consumption of the data center consists of the power consumption of IT equipment and the power consumption of cooling equipment. The IT equipment includes the power consumption of servers and the power consumption of data networks, which are all related to the number of servers working, as shown in constraints (18)-(21); P t =P IT，t +P cool，t (18) P IT，t =P ser,t +P net，t (19) P ser,t =(a ser f+b ser )n t (20) P net,t =a net n t +b net (21) Among them, P t is the power load of the data center at time t, P IT，t is the IT equipment power at time t, P cool，t is the power of the refrigeration equipment at time t, P ser，t is the server power at time t, P net，t is the data network power at time t, f is the operating frequency of the server, α ser / β ser / α net / β net The data related to the parameters and power load obtained for statistical analysis; The number of servers’ jobs is calculated by the processing time of the data load and is limited by the total number of servers in the data center, as shown in constraints (22)-(24); 0≤n t ≤n max (24) Among them, λ t is the number of workloads processed by the data center at time t, n max is the maximum number of servers, η ser is the efficiency factor for converting CPU frequency into the service rate of the server, T t is the processing time of the data load at time t; Constraint (25) is the energy efficiency of the data center, which is the ratio of the total power consumption of the data center to the power consumption of IT equipment; Among them, ε pue Energy efficiency for data centers.

5. The data center energy management method based on load flexibility mining according to claim 1 is characterized in that: In the step S4, The objective function is to minimize the power cost of the data center, as shown in formula (26): Among them, C buy is the electricity cost of the data center, a t is the electricity price at time t, P t buy is the amount of electricity purchased by the power grid at time t, and T is the total time period of the dispatch cycle; Service level agreement constraints (27)-(30) are contracts between IT users and data centers. They are the basic principles for scheduling workloads and specify the number and processing time of workloads. Constraint (27) limits the processing time of delay-sensitive workloads. Constraints (28)-(29) require that workload loss should be avoided during workload processing. Constraint (30) requires that delay-tolerant workloads should be executed within a specified time. Among them, U int The maximum delay time for delay-sensitive loads, is the transfer amount of delay-tolerant data load at time t, is the number of delay-tolerant data loads on server σ at time t, is the number of delay-tolerant data loads waiting to be processed from server σ at time t, is the number of delay-tolerant data payloads that should be processed by the deadline, t1 is the deadline of the data payload, and t0 is the arrival time of the data payload; Constraints (31)-(40) are the energy supply constraints of the data center. Constraints (31)-(35) are the energy storage charging and discharging constraints. Constraint (31) represents the state of charge constraint of the energy storage. Constraints (32)-(33) limit the maximum charging and discharging capacity of the energy storage. Constraint (34) indicates that the battery cannot be charged and discharged at the same time. Constraint (35) is the relationship between the energy storage capacity and the charging and discharging power. Constraint (36) limits the maximum and minimum power generation of photovoltaics. Constraint (37) is the power balance equation to ensure energy conservation in the data center. Constraint (38) limits the power purchase of the data center from the power grid. Constraint (39) limits the carbon dioxide emissions of the park. These emissions come from the electricity purchased from the power grid. Wind and solar power generation in the data center park will not produce carbon emissions. Constraint (39) is the backup capacity of the power system, which is expressed as: 0≤P cha，t ≤P max x cha，t (32) 0≤P dch，t ≤p max x dch，t (33) x cha,t +ξ dch,t ≤1 (34) P buy，t +P pv，t +P wt，t +ξ dch,t P dch,t =ξ cha，t P cha,t +P sell，t +P t (37) P pv，t +P wt，t +ξ dch,t P dch,t +P buy，t ≥(1+α)P t (40) in, is the minimum / maximum capacity of energy storage, p max is the maximum charging and discharging power of energy storage, E N is the rated capacity of energy storage, is the maximum / minimum output of photovoltaic, is the maximum amount of electricity purchased by the power grid, e ca is the maximum annual carbon emission, l is the carbon dioxide emission factor, α is the reserve capacity coefficient, is the energy storage capacity at time t, P dch,t / P cha,t is the energy storage charging and discharging power at time t, ξ dch,t / ξ cha,t is the energy storage charging and discharging state variable at time t, P sellt is the amount of electricity sold to the grid at time t.

Images