Search

CN-121094247-B - Multi-data center calculation force-electric power cooperative scheduling and demand response reporting capacity optimization method

CN121094247BCN 121094247 BCN121094247 BCN 121094247BCN-121094247-B

Abstract

The invention provides a multi-data center calculation force-electric power cooperative scheduling and demand response reporting capacity optimizing method, which comprises the steps of firstly collecting multi-dimensional historical time sequence data of data centers, then predicting the working load capacity of each data center in each future single day period and the electric power market price of each data center in a future scheduling period by utilizing a machine learning algorithm, then constructing a multi-data center calculation force-electric power cooperative model which considers uncertainty and faces space-time coupling, wherein the cooperative model comprises a joint optimization framework of calculation force scheduling and electric power scheduling, takes actual profit maximization as a target, then converting the established cooperative model into a mixed integer linear programming problem and solving the mixed integer linear programming problem to obtain the optimal reporting capacity of each data center participating in demand response, and carrying out scheduling according to the task quantity of space migration and time migration, the charge and discharge power of an energy storage system and the renewable energy power generation capacity so as to realize the overall reporting capacity of the multi-data center.

Inventors

  • ZHAO YINGRU
  • LIU YUFEI
  • XIE SHAN
  • JING RUI
  • LIN JIAN

Assignees

  • 厦门大学

Dates

Publication Date
20260505
Application Date
20251110

Claims (10)

  1. 1. The capacity optimizing method for the collaborative dispatching and demand response declaration of the multi-data center computing power and the electric power is characterized by comprising the following steps: S1, data acquisition and prediction, namely acquiring multi-dimensional historical time sequence data of a data center, and predicting the working load capacity of each data center in a future single day and each period in a future scheduling period by using a machine learning algorithm And the price of the electric power market at the location of each data center ; S2, constructing a collaborative optimization model, namely constructing a multi-data center calculation force-power collaborative model which considers uncertainty and faces to space-time coupling based on the prediction data of the step S1 and the space isomerism of the time dynamic property and the geographic distribution of the calculation force workload of the data center, and combining the space-time characteristics of the power supply of the regional power grid, wherein the collaborative model comprises a joint optimization framework of a calculation force scheduling model and a power scheduling model, and aims at maximizing actual profit, the actual profit consists of demand response income, electricity purchasing cost and punishment cost, and the calculation formula is as follows: wherein C represents the actual profit of the data center; representing benefits obtained by the data center in participating in demand response; the electricity purchasing price of the data center i at the time t is shown; the purchase power of the data center i at the time t is represented; 、 、 Time-shift, space-shift, unprocessed penalty costs for the workload representing priority p; Indicating that data center i is transferred from time t to time t The task amount of the priority p processed at the moment; The task quantity with the priority p, which is transferred from the data center i to the data center j at the moment t, is represented; the task quantity with the priority p which is not processed by the data center i at the time t is represented; S2.1, constructing a computational power scheduling model, wherein the computational power scheduling model is used for managing the workload with different priorities, establishing scheduling constraint of time transfer and space transfer of the workload, and outputting task quantity 、 、 ; S2.2, constructing a power dispatching model, wherein the power dispatching model is used for managing power grid electricity purchasing, an energy storage system, local renewable energy sources and energy consumption of data center equipment, establishing power supply and demand balance constraint of the data center and outputting power grid electricity purchasing power Actual response load of data center participating in demand response ; S2.3, constructing a calculation model of actual benefits R of demand response, wherein the calculation formula of the benefits R actually obtained by the data center participating in demand response is as follows: In the formula, Representing the actual response load of the data center participating in the demand response, Representing a response time; Representing price patch coefficients; representing a response speed coefficient; Representing the subsidy price of the data center i; Considering uncertainty of model prediction, demand response reporting capacity To reserve a certain margin, the calculation formula is: In the formula, Representing a declared capacity correction factor; S3, model solving and strategy executing, namely converting the collaborative model established in the step S2 into a mixed integer linear programming problem and solving to obtain the optimal reporting capacity of each data center participating in demand response And the task amount of space migration and time migration, the charge and discharge power of the energy storage system and the energy generation amount of renewable energy sources, so as to generate detailed operation instructions including a calculation power space-time scheduling strategy, the charge and discharge strategy of the energy storage system and the renewable energy source allocation strategy, and each data center performs scheduling according to the instructions to participate in demand response.
  2. 2. The method for optimizing capacity of multi-data center power-power collaborative scheduling and demand response declaration according to claim 1, wherein in step S2.1, constructing a power-computing scheduling model includes the steps of: s2.1.1 dividing the workload into delay sensitive tasks and delay tolerant tasks according to the service types, and distributing different priorities Task priority set priority for data center workload = {0,1,2 In order to avoid overload of calculation force resources, the processing capacity of each priority task is constrained, the flexibility of the low priority task is considered, and the processing capacity of the low priority task is allowed to be slightly exceeded within the limit, and the formula is expressed as: In the formula, Representing data center i in The priority of time processing is the task quantity of p; Representing a maximum utilization of the data center server; representing the processing rate of a single server of the data center; representing data center i in The number of servers working at the moment; Representing the number of data centers; S2.1.2, establishing a time scheduling model Setting an allowable transition time upper limit Tmax, allowing a delay tolerant task to be transferred back and forth within the set maximum delay time Tmax, and meeting the task total conservation constraint that the task quantity which can be delayed by a data center cannot exceed the sum of the currently received task quantity and the transfer task quantity, wherein the task total conservation constraint is expressed as follows by a formula: In the formula, Indicating that data center i is transferred from time t to time t Task amount with priority of 0 for time processing; Indicating that data center i is transferred from time t to time t The priority of time processing is the task quantity of p; representing the task quantity of priority p received by the data center i from the moment t, wherein p is not equal to 0; representing the task amount with the priority of p transferred from the data center j to the data center i at the time t, wherein p is not equal to 0; s2.1.3, establishing a space scheduling model Setting the multiple data center set as ={1,2, N, where N is the number of data centers, allowing all priority tasks to be collected at a data center And (3) performing inter-geographic-position transfer distribution, and meeting the task transfer amount conservation constraint that the transfer-out task amount cannot exceed the sum of the currently received task amount and the transfer-in task amount, wherein the task transfer amount conservation constraint is expressed as the following formula: In the formula, Representation of The amount of tasks with priority p that are transferred from data center i to data center j at the moment, Representation of The task quantity with the priority of p is transferred from the data center j to the data center i at the moment; S2.1.4, meeting the total conservation constraint of tasks before and after the data center is scheduled, and expressing the total conservation constraint as follows: In the formula, 、 Respectively indicate that the data center i is in Task quantity with priority of p transferred by time and space scheduling at moment; representing data center i in The priority of time processing is the task quantity of p; Indicating that data center i is transferred from time t to time t The priority of time processing is the task quantity of p; representing data center i in The task quantity with the priority of p which is not processed at the moment; representing data center i in Task amount received at moment; Representing data center i slave Time of day transition to Task amount at time.
  3. 3. The method for optimizing capacity for collaborative power-power scheduling and demand response reporting in a multi-data center as set forth in claim 1, wherein in step S2.2, constructing a power scheduling model comprises the steps of: s2.2.1 building an energy storage system model The capacity of the energy storage system is determined by the capacity at the previous moment and the charge and discharge power, and the energy storage system model is as follows: In the formula, Representing the capacity of the energy storage system t of the data center i at the moment; 、 Respectively representing the charge and discharge efficiency of the energy storage system; 、 respectively representing the charge and discharge power of the energy storage system; 、 Respectively representing the minimum value and the maximum value of the charge state of the energy storage system; the rated capacity of the energy storage system of the data center i; 、 Respectively representing the maximum charge and discharge power of the energy storage system; s2.2.2, establishing a wind power generation system model The wind power generation power is determined by the rated capacity of a wind power generation system and a wind power coefficient, the wind power coefficient is related to wind speed, wind wheel diameter and wind energy utilization efficiency, and the wind power generation system model is as follows: In the formula, Representing wind power generation power of the data center i at the time t; Representing the rated capacity of the wind power generation system of the data center i, which is a known quantity; the wind power coefficient of the data center i at the time t is represented; s2.2.3 building a photovoltaic power generation system model The photovoltaic power generation power is related to the nominal power, the total solar radiance and the comprehensive utilization efficiency, and the photovoltaic power generation system model is as follows: In the formula, The photovoltaic power generation power of the data center i at the time t is represented; Representing the comprehensive utilization efficiency of the solar energy system, which is a known quantity; Representing the nominal power of the photovoltaic power generation system of the data center i; Representing the total solar radiation degree actually received by the data center i at the moment t; Representing irradiance under standard test conditions, and taking 1kW/m <2 >; S2.2.4, using power supply using efficiency as energy efficiency evaluation index In order to measure the energy efficiency of the data center, the power supply utilization efficiency PUE is used as an energy efficiency evaluation index, and the ratio of the total energy consumption of the data center to the energy consumption of the server equipment is defined as the following formula: In the formula, Indicating the power supply use efficiency; representing the total energy consumption of the data center i at the time t; Representing the energy consumption of server equipment of the data center i at the time t; s2.2.5, building an equipment energy consumption model The energy consumption of the server equipment is related to the utilization rate and the electric power, and the energy consumption of the data center server equipment is as follows: In the formula, 、 The power consumption of a single server device of the data center i in the full load state and the standby state is respectively represented; representing the number of full servers of the data center i at the time t; Representing the total number of servers configured by the data center i; S2.2.6 building a total energy consumption and electric power balance model The power supply and demand power of the data center is balanced in real time, and the power supply and demand power is expressed as follows: In the formula, The purchase power of the data center i at the time t is represented; The photovoltaic power generation power of the data center i at the time t is represented; representing the total energy consumption of the data center i at the time t; Representing wind power generation power of the data center i at the time t; 、 respectively representing the charge and discharge power of the energy storage system; S2.2.7, establishing a demand response load calculation model The data center participates in demand response, actual response load delta The calculation formula of (2) is as follows: In the formula, Representing the actual response load of the data center i at the time t; Indicating the electrical load baseline of data center i at time t.
  4. 4. The method for optimizing capacity for collaborative power-power scheduling and demand response reporting in a multi-data center as set forth in claim 1, further comprising the steps of, in order to improve robustness of the collaborative model, in step S2: S2.4, introducing conditional risk value CVaR to generate anti-risk scheduling strategy Considering predicted workload And electric market price Setting a prediction error, and generating a scene set S by a Monte Carlo simulation method, wherein the number of scenes S in the set is more than 500, and the calculation formula of the actual value under the scenes S is expressed as follows: In the formula, The electricity purchasing price of the data center i at the time t under the scene s is shown; A random error sample of the purchase price of the data center i at the moment t under the scene s is shown; Representing the task quantity with the priority p received by the data center i from the moment t under the scene s; A random error sample representing the task quantity with the priority of p received by the data center i from the moment t under the scene s; assigning probabilities to scenes in scene set S Expressed by the formula: In the formula, Representing the probability under the electricity price scene s; Representing the probability under the workload scenario s; CVaR is formulated as: In the formula, Representing a loss function, defined as , Indicating an excess loss of the device, Representing a maximum acceptable loss; after CVaR is introduced, profit and high loss risk are required to be balanced, and the objective function of the coordination model is modified as follows: In the formula, A probability representing scene s; Representing actual profit under scene s; representing a risk aversion coefficient; Representing a maximum acceptable loss; Representing a confidence level; Indicating excess loss.
  5. 5. The method for optimizing the capacity of the multi-data center power-power collaborative scheduling and demand response declaration according to claim 1, wherein in step S2.3, a calculation formula of an actual response load accounting for an invited response amount ratio S DR is as follows: according to a regional power demand response embodiment, parameters ζ, S DR , 、 Wherein the price patch coefficient ζ is related to S DR , expressed as: the response speed coefficient v is related to Δt and expressed as: When each electric power network formally issues a demand response offer, the model parameters are further corrected by combining the issued demand response mode, scale, time period, area range and incentive price information, further in the subsequent step S3, a more accurate output value is obtained through model solving, and the data center is invited to fill response information according to the demand response mode, the scale, the time period, the area range and the incentive price information.
  6. 6. The method for optimizing capacity for collaborative power-power scheduling and demand response reporting in a data center of claim 1, wherein in step S1, the multi-dimensional historical time series data of the data center includes server operating states including power, CPU utilization and server processing speed in server standby and full states, workload data and power market price data.
  7. 7. The method for optimizing the declared capacity for collaborative power-power dispatching and demand response of a multi-data center as claimed in claim 1, wherein the declared capacity correction factor k is 0.85-0.95 in step S2.3.
  8. 8. The method for optimizing capacity of multi-data center power-power cooperative scheduling and demand response declaration according to claim 1, wherein in step S3, after historical data is imported and each parameter is assigned, a coordination model is converted into a mixed integer linear programming solution, and a Gurobi solver is used for computing the solution model to obtain an optimal capacity declaration value of data center participation demand response Task amount of spatial migration and temporal migration of each data center And Charging and discharging power of energy storage system And Wind power generation And photovoltaic power generation 。
  9. 9. The method for optimizing capacity of multi-data center power-power collaborative scheduling and demand response declaration according to claim 1, wherein in step S1, model training is performed by adopting LSTM, data with history of 30-60 days is taken as input during model prediction, work load amount and electricity price data are taken as prediction targets, a scheduling period range of a data center is considered as a single day, 96 periods are divided every day, the duration of each period is Δt=15 min, namely T= { 1,2, ,96}。
  10. 10. The method for optimizing capacity of multi-data center power-power collaborative scheduling and demand response declaration according to claim 1, wherein in step S1, after multi-dimensional historical time series data of a data center are collected, preprocessing of supplementing missing values, eliminating abnormal values and normalizing is performed on original historical data, and then machine learning model is utilized for prediction.

Description

Multi-data center calculation force-electric power cooperative scheduling and demand response reporting capacity optimization method Technical Field The invention belongs to the technical field of data center energy management, and particularly relates to a multi-data center calculation force-electric power cooperative scheduling and demand response reporting capacity optimization method. Background The data center is used as the core of the computing infrastructure, and the problems of high energy consumption and high carbon emission and insufficient coordination of computing power and power resources are increasingly highlighted. The traditional data center usually adopts an operation mode that power dispatching and calculation power dispatching are mutually independent, wherein the power dispatching is mainly based on peak-valley electricity price and power grid supply and demand conditions, and the calculation power dispatching is focused on local optimization of task delay and resource utilization rate. Most of the existing researches fail to fully consider the difference of the two operation mechanisms, so that the calculation power and the electric power resources are difficult to realize the global optimal configuration of the system level. With the wide popularization of the demand response mechanism in various industries, the data center shows stronger participation potential by virtue of the load adjustable characteristic. However, most data centers are passively participated in the demand response by taking a single data center as a unit, and lack of reporting capacity overall planning across the data centers, so that the mismatch problems of reporting redundancy of part of the data centers and insufficient capacity of part of the data centers often occur. The existing research ignores core requirements such as cross-center capacity dynamic adjustment, power cross-province transfer and the like, and a collaborative optimization model covering the whole chain is not established yet. Disclosure of Invention The invention aims to provide a multi-data center calculation force-electric power cooperative scheduling and demand response reporting capacity optimization method, which realizes overall optimization of multi-data center demand response reporting capacity through an uncertainty-oriented time-space coupled multi-data center calculation force-electric power cooperative model, avoids capacity redundancy or deficiency, obviously reduces operation cost and relieves peak pressure of an electric network. In order to achieve the above purpose, the solution of the present invention is to provide a capacity optimization method for collaborative power-power dispatching and demand response reporting of a multi-data center, comprising the following steps: S1, data acquisition and prediction, namely acquiring multi-dimensional historical time sequence data of a data center, and predicting the working load capacity of each data center in a future single day and each period in a future scheduling period by using a machine learning algorithm And the price of the electric power market at the location of each data center; S2, constructing a collaborative optimization model, namely constructing a multi-data center calculation force-power collaborative model which considers uncertainty and faces to space-time coupling based on the prediction data of the step S1 and the space isomerism of the time dynamic property and the geographic distribution of the calculation force workload of the data center, and combining the space-time characteristics of the power supply of the regional power grid, wherein the collaborative model comprises a joint optimization framework of a calculation force scheduling model and a power scheduling model, and aims at maximizing actual profit, the actual profit consists of demand response income, electricity purchasing cost and punishment cost, and the calculation formula is as follows: wherein C represents the actual profit of the data center; representing benefits obtained by the data center in participating in demand response; the electricity purchasing price of the data center i at the time t is shown; the purchase power of the data center i at the time t is represented; 、、 Time-shift, space-shift, unprocessed penalty costs for the workload representing priority p; Indicating that data center i is transferred from time t to time t The task amount of the priority p processed at the moment; The task quantity with the priority p, which is transferred from the data center i to the data center j at the moment t, is represented; the task quantity with the priority p which is not processed by the data center i at the time t is represented; S2.1, constructing a computational power scheduling model, wherein the computational power scheduling model is used for managing the workload with different priorities, establishing scheduling constraint of time transfer and space transfer of the worklo