Data center cluster low-carbon economic robust optimization method considering multiple types of loads

Abstract

The invention discloses a data center cluster low-carbon economic robust optimization method considering multiple types of loads, belongs to the technical field of smart power grid and energy management, and solves the problems of high energy consumption and low efficiency of an existing data center cluster management method. The method comprises the following steps: constructing a day-ahead optimization model and a real-time optimization model of a data center cluster considering multiple types of loads; constructing a target function of a day-ahead optimization model by taking the first time as a time interval and taking the minimum day-ahead total operation cost of the data center cluster as a target; the real-time optimization model is constructed by taking second time as a time interval, and comprises an upper-layer model for DCCO decision making and a lower-layer model for each DCP decision making; the day-ahead optimization model is solved, and a charging and discharging plan of the shared energy storage system with the first time as the time interval is obtained through calculation; and transmitting the charging and discharging plans of the shared energy storage system to a real-time optimization model, and solving the real-time optimization model so as to carry out real-time scheduling on each DCP.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart grid and energy management, and particularly relates to a low-carbon economic robust optimization method for data center cluster considering multiple types of loads. BACKGROUND

[0002] Traditional DCC (Data Center Cluster) energy management mainly optimizes single energy flow of electricity, ignores the multi-energy flow coupling characteristics of electricity, heat, cold, computing power, etc., and does not sufficiently tap the flexible response potential of multiple types of loads on the user side (temperature control equipment, interruptible tasks, shared energy storage systems, etc.), resulting in low energy efficiency and rigid supply-demand matching. In addition, the increasing penetration of renewable energy and the intensifying fluctuations in the electricity market further amplify the multiple uncertainties in DCC operation (such as the randomness of wind and solar power output and the heterogeneity of demand behavior), making it difficult for existing single-stage or deterministic optimization models to balance economic, low-carbon, and robustness objectives, and necessitating a breakthrough in the optimization bottleneck of multi-time scale coordination and multi-agent game.

[0003] DCC energy management involves the following aspects: in terms of multi-energy flow joint sharing modeling, existing energy dispatch optimization research lacks in-depth modeling of the multi-energy flow coupling and coordination mechanism of electricity, heat, cold, etc., and the spatiotemporal coupling characteristics of multi-energy flow have not been fully incorporated into the joint optimization framework, resulting in significant resource competition and efficiency loss in actual dispatch. In terms of distribution robustness optimization of uncertainty modeling, the core assumption relies on known or fixed probability distribution parameters (mean and variance of wind power prediction error), ignoring the fuzziness and time-varying nature of the probability distribution itself in actual scenarios. In terms of non-dynamic pricing mechanism of P2P transaction and user behavior incentive, existing point-to-point (P2P) energy transaction mechanisms mostly use static pricing rules (such as fixed percentage or cost allocation), failing to incorporate real-time market signals, user preferences, and energy supply-demand elasticity into a dynamic game framework. In terms of convergence and computational efficiency of real-time game, the convergence and uniqueness of equilibrium of the game are difficult to guarantee, especially when data center producers and consumers (DCPs) adopt distributed decision-making, which may lead to strategy conflicts and price volatility. Traditional centralized solution algorithms face the problem of computational delay, making it difficult to meet the timeliness requirements of real-time dispatch.

[0004] Therefore, how to design a more reasonable data center cluster energy management optimization method to reduce energy consumption and improve computational efficiency is a technical problem that needs to be solved at present. SUMMARY

[0005] In view of the above analysis, the embodiments of the present application aim to provide a low-carbon economic robust optimization method for data center cluster considering multiple types of loads, to solve the problems of high energy consumption and low efficiency existing in the existing data center cluster management method.

[0006] The present invention discloses a low-carbon economic robust optimization method for a data center cluster considering multiple types of loads, the method comprising: Construct a day-ahead optimization model and a real-time optimization model for a data center cluster that considers multiple load types. The objective function of the day-ahead optimization model is constructed with a first time interval and the goal of minimizing the total day-ahead operating cost of the data center cluster. The real-time optimization model is constructed with a second time interval and includes an upper-level model for DCCO decision-making and a lower-level model for each DCP decision-making. Solve the day-ahead optimization model to calculate the charge and discharge schedule of the shared energy storage system with the first time as the time interval; The shared energy storage system charging and discharging plan with the first time interval is transmitted to the real-time optimization model, and the real-time optimization model is solved to perform real-time scheduling of each DCP; Among them, the first time is greater than the second time.

[0007] On the basis of the above solution, the present invention also makes the following improvements: Furthermore, the objective function of the day-ahead optimization model is expressed as: (1) in, express The electricity purchase cost for the period, Indicates that all gas boilers are Total fuel cost for the period, Indicates that the shared energy storage system The loss cost of charging and discharging during the period; is the worst-case risk value, is the risk aversion coefficient; Indicates the scheduling period of the day-ahead optimization model.

[0008] further, (2) in, express The grid electricity price for the time period, express Power purchased during the time period; (3) in, Indicates the The gas boiler in the DCP Gas consumption during the period, express Gas and electricity prices for the time period, Indicates the total number of DCPs; (4) wherein, denotes the loss cost coefficient of the shared energy storage system, , denote the charging power and discharging power of the shared energy storage system in time period, respectively.

[0009] Further, the decision variables of the day-ahead optimization model include , , , .

[0010] Further, the constraint conditions of the day-ahead optimization model include the electrical energy balance constraint, the thermal energy balance constraint, the SES operation constraint and the WCVaR risk constraint.

[0011] Further, the electrical energy balance constraint is expressed as: (5) wherein, denotes the net discharging power of the shared energy storage system in time period, ; denotes the wind-solar output of the RES in the th DCP in time period, i.e. the power generation of the renewable energy; denotes the power consumption of the IT server in the th DCP in time period; denotes the power of the cooling system in the th DCP in time period; denotes the load power of the application in the th DCP in time period.

[0012] Further, the second time is 15 minutes and the first time is 1 hour; The objective function is constructed with the objective of maximizing the DCCO revenue, denoted as: (6) wherein, denotes the transaction price of the shared energy storage system in time period, denotes the net discharging power of the shared energy storage system in time period; denotes the dispatching period of the real-time optimization model; The charging and discharging plan of the shared energy storage system with an hourly time interval , Transmitted to the upper model in the real-time optimization model to solve the hourly 、 ; is the decision variable of the upper model.

[0013] Furthermore, the upper model satisfies the IDR flexibility constraint.

[0014] Furthermore, each DCP makes its own decision and constructs an objective function with the goal of minimizing the real-time cost of the current DCP; The objective function of a DCP is expressed as: (7) (8) in, Indicates the DCP and shared energy storage system Trading power of the session; Indicates the DCPs in The trading power of the P2P market during the period; Indicates that the P2P market is The transaction price of the session; Indicates the DCP in Adjustment costs for time periods, such as task delay penalties or thermal comfort compensation; The decision variables of the lower model include 、 and , used to perform real-time scheduling of the corresponding DCP.

[0015] Furthermore, the lower-level model satisfies P2P transaction balance constraints, data load delay constraints, and thermal comfort constraints.

[0016] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: This paper addresses the challenges of high energy consumption, high carbon emissions, and the spatiotemporal mismatch between energy supply and demand in data center clusters by proposing a novel two-stage energy sharing model that integrates multi-load integrated demand response (IDR). The model aims to build an economically efficient, low-carbon, and resilient energy synergy system. This model utilizes a two-stage optimization architecture (day-ahead scheduling and real-time decision-making) and a dynamic coupling mechanism for multiple energy flows, integrating shared energy storage (SES), a hydrogen-heat cogeneration system, and a peer-to-peer (P2P) trading platform to address the bottleneck of heterogeneous energy regulation across time and space. The innovations achieved are as follows: (1) Multi-energy flow coordination scheduling, a unified optimization framework is constructed across the space-time coupling of energy flow. A dynamic optimization method based on "energy flow decoupling-coordinated matching" is proposed to break through the limitations of traditional single energy independent scheduling. By constructing a multi-energy flow energy hub model, the physical constraints and time scale differences (thermal inertia and instantaneous response of electric energy) of electric, thermal, cold and data flow are decoupled into hierarchical optimization problems, realizing dynamic balance across energy flow. The "virtual energy bus" mechanism is introduced, allowing different energies to be flexibly replaced in a unified space-time dimension according to priority, such as using the flexibility of data center computing load to adjust cooling demand, or linking waste heat recovery with peak valley electricity price, thereby tapping the global coordination potential in complex coupling.

[0017] (2) Real-time scheduling game, a distributed coordination and fast convergence equilibrium mechanism is established, a "master-slave game-distributed learning" hybrid architecture is designed to solve the conflict between game convergence and computational efficiency in large-scale real-time scheduling. For the uniqueness of Stackelberg game equilibrium, a guiding strategy based on marginal cost dynamic pricing is proposed to constrain the strategy space of game participants through price signals, avoiding the conflict divergence of distributed decision-making. At the same time, combined with the improved ADMM algorithm, the global optimization is decomposed into multiple parallel computing tasks, and the edge node is pre-trained with local strategies, and the cloud is aggregated in a lightweight manner, significantly reducing communication overhead. This architecture ensures the game equilibrium while achieving minute-level real-time response capability.

[0018] (3) Comprehensive demand response mechanism, user behavior guidance and flexible resource coordination are realized. A demand response system based on "incentive compatibility-flexible adaptation" is constructed to solve the dual problems of insufficient user participation and uncertain behavior. On the one hand, through dynamic internal electricity price and external market linkage design, the user's energy saving benefits are bound with the system's economic target, and the price lever is used to drive active response; on the other hand, based on user behavior feature clustering analysis, "rigid-flexible-interruptible" load levels are divided, and response strategies (task delay fault tolerance threshold, temperature control comfort flexibility interval) are designed differently, so that the optimization target and user preference are dynamically adapted, and the user is transformed from a passive participant to an active collaborator, releasing the flexible potential of demand side resources.

[0019] In the present invention, the above technical solutions can be combined with each other to realize more preferred combination schemes. Other features and advantages of the present invention will be described in the subsequent specification, and some advantages will become apparent from the specification, or will be understood by implementing the present invention. The purpose and other advantages of the present invention can be achieved and obtained from the specific content indicated in the specification and the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and are used to explain the principles of the application, but not to limit the scope thereof. Figure 1 A flow chart of the low-carbon economic robust optimization method for a data center cluster considering multiple types of loads is provided for Embodiment 1 of the application. Figure 2 A structural schematic diagram of the DCC energy sharing framework is provided for the embodiment of the application. Figure 3 A specific structure of the DCP is provided for the embodiment of the application. Figure 4 The decision-making process of the DCC framework provided for the embodiment of the application adopts a two-stage collaborative optimization mechanism. Figure 5 A flow chart of the optimization algorithm is provided for the embodiment of the application. DETAILED DESCRIPTION

[0021] The preferred embodiments of the application are specifically described below in conjunction with the accompanying drawings, which form a part of this application and are used to explain the principles of the embodiments of the application and are not used to limit the scope of the application.

[0022] One specific embodiment of the application discloses a low-carbon economic robust optimization method for a data center cluster considering multiple types of loads, a flow chart of which is shown in Figure 1 as follows.

[0023] Step S1: constructing a day-ahead optimization model and a real-time optimization model of the data center cluster considering multiple types of loads; wherein a target function of the day-ahead optimization model is constructed with the first time as a time interval and the minimum total operation cost of the data center cluster as a target; the real-time optimization model is constructed with the second time as a time interval, including an upper model for DCCO decision-making and a lower model for DCP decision-making; Step S2: solving the day-ahead optimization model to calculate the charging and discharging plan of the shared energy storage system with the first time as a time interval; Step S3: transmitting the charging and discharging plan of the shared energy storage system with the first time as a time interval to the real-time optimization model to solve the real-time optimization model and to perform real-time scheduling for each DCP; Wherein the first time is greater than the second time.

[0024] The DCC energy sharing framework provided in this embodiment is shown in Figure 2As shown, the demand response coordination operator (DCCO) and multiple distributed production and consumption units (DCPs) are composed, and the electricity, heat, and gas multi-energy complementary distributed energy sharing is realized through a hierarchical collaborative mechanism. The DCC energy sharing framework can realize dynamic adjustment and energy buffering through SES (shared energy storage system), reduce the direct interaction cost with the power grid and heat network through the P2P transaction mechanism, thereby improving the economy and efficiency of the system. In addition, the framework also realizes the complementation and global optimization distribution of electric energy, heat energy and gas energy through multi-energy flow coupling. Figure 2 Among the producers and consumers, the RES (renewable energy system) provides renewable energy for the DCP, and the DCP forms an integrated energy system of "power-storage-use" by integrating these renewable energy, and obtains electric energy and heat energy through the P2P transaction mechanism, thereby reducing the direct interaction cost with the power grid and heat network.

[0025] The DCCO plays a triple role of energy dispatching center, P2P transaction agent and sharing management platform: first, the SES dynamically adjusts the unbalanced power of the DCP electric P2P transaction, and the charging and discharging price is between the power grid price interval, which can not only alleviate the fluctuation of the power grid but also ensure the economy, and can also dynamically trade with the power grid according to the SES charging and discharging plan and the DCP energy gap to maintain system balance; second, as a heat P2P transaction intermediary, the unbalanced heat between DCPs is handed over to the heat network for unified smoothing processing, realizing the global optimization distribution of heat energy; third, the DCCO proxies the DCP to purchase natural gas from the natural gas network (GN) to ensure the power balance of the gas boiler and other heating equipment. The DCP, as a distributed production and consumption unit, integrates renewable energy such as wind power and photovoltaic, IT equipment, cold and heat load facilities (such as electric / absorption refrigerators, gas boilers) and residential buildings, forms an integrated energy system of "power-storage-use", and its electric energy and heat energy are preferentially obtained through the P2P market coordinated by the DCCO, thereby reducing the direct interaction cost with the power grid (PG) and the heat network (HN). The framework innovatively integrates the dispersed producers and consumers into a collaborative network, buffers through SES, couples multi-energy flow, and uses the P2P transaction mechanism to build a high-elasticity, low-cost decentralized energy sharing ecosystem, and provides a new path for efficient operation of multi-energy systems.

[0026] Figure 3 The specific structure of the DCP is shown. In Figure 3 The external energy supply mainly comes from the electric energy and heat energy of the electric and heat P2P transaction market coordinated by the DCCO, and the natural gas purchased from the natural gas network (GN). The data center cluster (DCC) proposed in this embodiment is composed of a data center cluster operator (DCCO) and multiple data center producers and consumers (DCPs), and its multi-level architecture includes the following three levels: (1) Physical layer. Each DCP integrates distributed renewable energy (wind power, photovoltaic), IT server cluster, hybrid cooling system (air cooling + liquid cooling), gas boiler, energy storage device and attached residential building (including electricity / heat load), forming an energy flow of electricity, heat, cold and data quadruple coupling. Among them, the IT load is affected by the space-time distribution of data flow, the heat load is related to the building thermal inertia, and the cold load is driven by the server heat dissipation demand.

[0027] (2) Information layer. Each DCP deploys edge computing nodes to realize local renewable energy output prediction, load sensing and equipment monitoring; DCCO builds a blockchain platform to record P2P energy trading, SES charging and discharging and demand response (IDR) data, ensuring transparency and traceability. Through the two-stage optimization interface (based on OPC-UA protocol), the vertical connection of day-ahead optimization plan and real-time adjustment instruction is realized, supporting efficient communication across systems.

[0028] (3) Market layer. DCC as an aggregation entity participates in external electricity wholesale market (day-ahead / real-time market) and heat market transactions; internally, it establishes a shared energy storage rental market (DCCO pricing) and a P2P energy sharing market (dynamic supply and demand ratio pricing) to promote optimal allocation of resources within the cluster. This architecture realizes the deep integration of energy flow, information flow and value flow through the multi-level coordination of physical-information-market.

[0029] In terms of energy flow coordination, electricity flow realizes clean supply by preferentially consuming local renewable energy, and the insufficient part is purchased through the grid or calls for shared energy storage system (SES) reserves, and the excess electricity is traded through point-to-point (P2P) or stored in SES, forming a dynamic balance of energy supply and demand network. Heat flow adopts gas boiler and server waste heat recovery system to supply energy, which not only meets the building heating demand, but also improves system energy efficiency through heat storage or energy form conversion, reflecting the characteristics of multi-energy complementation. Cold flow is based on a hybrid cooling system that dynamically adjusts the air cooling and liquid cooling ratio according to the server heat dissipation demand and environmental temperature, achieving precise matching of cold supply and equipment load.

[0030] In terms of time scale coordination, the day-ahead stage relies on prediction data to formulate the SES charging and discharging plan, external market transaction volume, and baseline operation strategy of data center power supply equipment (DCPs), to conduct multi-energy coupling optimization with the goal of global economic optimization; the real-time stage combines ultra-short-term prediction data and actual operation status to enhance the flexibility of the system in response to load fluctuations and energy price changes through dynamic adjustment of P2P transaction prices, demand side response (IDR) strategies, and SES charging and discharging power. This mechanism realizes coordinated regulation of multiple time scales through hierarchical optimization, ensuring the economic efficiency and reliability of the energy system.

[0031] Figure 4A two-stage collaborative optimization mechanism is adopted for the decision-making process of the DCC framework. The decision-making process of the DCC framework adopts a two-stage collaborative optimization mechanism - day-ahead optimization and real-time optimization, which deals with uncertainty at different time scales through hierarchical modeling. In the day-ahead optimization stage, the time coupling problem of the state of charge (SOC) of the shared energy storage system (SES) is solved. Due to the continuity of the SOC (i.e. the current charging and discharging decision affects the availability of future capacity), if it is completely placed in the real-time stage optimization, it may lead to short-sighted strategy and capacity mismatch. Therefore, this stage formulates the globally optimal charging and discharging power curve for the SES based on a stochastic optimization model: first, a scenario tree is constructed with the objective of minimizing the operating cost of the SES, including the probability distribution of renewable energy output (wind power / photovoltaic) and load fluctuation; second, the weighted conditional value at risk (WCVaR) is introduced as a robustness indicator to quantify the expected loss under extreme scenarios, and the risk aversion intensity is adjusted through the weight coefficient, so as to generate a day-ahead plan that takes into account both economic and robustness. The output of this stage not only provides a benchmark charging and discharging strategy for the SES, but also lays the boundary conditions for subsequent real-time optimization. In the real-time optimization stage, a Stackelberg game model is constructed with DCCO as the leader and DCP as the follower, realizing multi-agent collaborative decision-making in a dynamic market environment. In the upper model, DCCO as the market leader, adjusts the SES selling / buying price range according to the day-ahead charging and discharging plan of the SES and the real-time price signal of the external energy market (power grid PG, heat network HN, natural gas GN): on the one hand, through the electric-thermal coupled pricing mechanism (such as ESDR-based electric power pricing, HSDR-based heat pricing), the capacity constraints of the SES and market fluctuations are converted into price incentive signals; on the other hand, combined with the heat price information of the heat network, the electric-thermal P2P transaction market rules are constructed to guide the energy sharing behavior among DCPs. In the lower model, each DCP as a price accepter optimizes two-dimensional decisions - energy consumption plan (such as IT equipment operation period, refrigeration / heating equipment output) and integrated demand response (IDR) plan (such as load shifting, interruptible load calling) with the objective of minimizing its own energy consumption cost under the given P2P internal electricity and heat prices. Through the iterative solution of the upper and lower models (such as dual transformation based on KKT conditions or heuristic algorithm), the Stackelberg equilibrium state is finally converged: DCP obtains the optimal scheduling scheme that meets individual economic efficiency, and DCCO determines the real-time electricity price of the SES, while ensuring that the SOC state of the SES dynamically matches the day-ahead plan. This two-level mechanism not only avoids the information overload problem of centralized optimization, but also realizes the autonomous collaboration of distributed agents through price signals, significantly improving the system's adaptability to real-time fluctuations.

[0032] Next, the day-ahead optimization model is described.

[0033] During the day-ahead optimization phase, the goal is to develop a global charging and discharging plan for the shared energy storage system (SES) by coordinating the multi-energy coupling relationship between electricity, thermal energy, and data loads within the data center cluster (DCC). This ensures system robustness while reserving flexibility for real-time operation. The core approach includes three aspects: First, addressing the uncertainty in the probabilistic distribution of wind and solar power output and load demand, a worst-case value-at-risk (WCVaR) model is used to quantify extreme scenario risks, and risk aversion constraints are constructed to enhance the robustness of the scheduling strategy. Second, based on the energy storage capacity limitations and charging and discharging efficiency characteristics of the SES, temporal coupling constraints are designed to ensure continuity between day-ahead planning and real-time energy scheduling, avoiding capacity shortages or overcharging and discharging. Finally, by coordinating energy declaration information from multiple energy providers (DCPs) through the data center operator (DCCO), and with the goal of optimizing the overall economic efficiency of the DCC, the SES charging and discharging periods and external market trading strategies are optimized to prioritize the consumption of local renewable energy and reduce the cost of purchased electricity. The key design feature of this stage is the triple coordination of "risk-time-agent": suppressing the impact of wind and solar fluctuations and load mutations through WCVaR risk constraints, ensuring the physical feasibility of the SES scheduling plan through time coupling constraints, and achieving a balance between global optimization and flexibility reservation through a multi-agent coordination mechanism, thus laying the foundation for real-time dynamic adjustment.

[0034] (1) Day-ahead optimization model Preferably, in the day-ahead optimization model, an objective function is constructed with hourly time intervals and the goal of minimizing the total day-ahead operating cost of the data center cluster. This objective function aims to minimize the total day-ahead cost of DCC, including the cost of purchased electricity, gas cost, and energy storage loss cost, while reducing the financial risk brought by the uncertainty of wind and solar output through the WCVaR constraint. The objective function of the day-ahead optimization model is expressed as: (1) in, express The electricity purchase cost for the period, Indicates that all gas boilers are Total fuel cost for the period, Indicates that the shared energy storage system The loss cost of charging and discharging during the period (yuan); is the worst-case risk value, used to quantify the risk cost in extreme scenarios (yuan); is the risk aversion coefficient (dimensionless), the larger the value, the more conservative the decision; Indicates the scheduling period of the day-ahead optimization model. In the day-ahead optimization model constructed in this embodiment, the scheduling period of the day-ahead optimization model is 24 hours a day and the time interval is one hour.

[0035] (2) in, express The grid electricity price for the time period, express The purchased electricity power during the period.

[0036] (3) in, Indicates the The gas boiler in the DCP Gas consumption during the period, express Gas and electricity prices for the time period, Indicates the total number of DCPs.

[0037] (4) in, represents the loss cost coefficient of the shared energy storage system (yuan / kWh), 、 Represents the shared energy storage system in Charging power and discharging power (kW) during the time period.

[0038] In the day-ahead optimization model, the decision variables include 、 、 、 .

[0039] In the specific implementation process, the decision variables in the day-ahead optimization model are 、 It should be noted that since the day-ahead optimization model uses hourly intervals, the output of the day-ahead optimization model is 、 It is also the parameter of each hour.

[0040] The constraints of the day-ahead optimization model include power balance constraints, thermal balance constraints, SES operation constraints, and WCVaR risk constraints, which are described in detail below. 1) Power balance constraints (5) in, Indicates that the shared energy storage system Net discharge power during the period (kW), ; Indicates the The RES in the DCP is Wind and solar power output (kW) during the period, i.e., the power generated by renewable energy; Indicates the The IT servers in the DCP Power consumption during the time period (kW); Indicates the The cooling system in a DCP is Power for the time period (kW); Indicates the The application of DCP The load power (kW) of the time period.

[0041] 2) Thermal energy balance constraints (6) in, Indicates the The gas boiler in the DCP Heat generated during the period (kWh); Indicates the The cooling system in a DCP is Waste heat recovery amount during the period (kWh); Indicates the The corresponding buildings in the DCP are Heat demand during the period.

[0042] (7) in, Indicates the thermal efficiency of the gas boiler.

[0043] 3) Operational constraints of shared energy storage systems (8) in, 、 Shared energy storage system State variables of charging and discharging during the time period; is the rated capacity of SES, 、 are charging efficiency and discharging efficiency respectively; For shared energy storage systems The state of charge during the time period.

[0044] The energy storage state of charge must meet the upper and lower capacity limits and the minimum capacity requirements at the end of the cycle, as shown below: (9) in, 、 Respectively represent the minimum and maximum values ​​of the charge of the shared energy storage system; SOC shared storage system at the end of the cycle, which usually refers to the state of charge of the battery at the end of a specific time period (such as a day, a week, etc.); Minimum capacity requirement of shared storage system at the end of the cycle.

[0045] 4) WCVaR risk constraint Risk cost is defined by quantile optimization, with the constraint: (10) Where F represents the set of probabilistic distribution uncertainties of wind and solar power output and load, including all possible probabilistic distribution scenarios of wind and solar power output and load; is an element in set F, i.e. a specific probabilistic distribution in the set, used to consider uncertainty in the calculation of risk constraints; Risk value (yuan) under confidence level ; Loss function is represented by L; Loss greater than or equal to ; The maximum value of the expected value of loss .

[0046] WCVaR risk constraint defines risk cost through quantile optimization. Specifically, it considers the average loss in extreme cases where loss exceeds under a given confidence level . This method can more comprehensively assess risk, as it not only considers the maximum loss that may occur (VaR), but also considers the average loss in extreme cases. By incorporating the WCVaR risk constraint into the objective function, the system's performance is optimized while considering risk cost. It serves as an additional constraint, requiring the system to meet certain risk level requirements during operation, ensuring stable operation under different scenarios. This constraint works together with other constraints such as power balance, thermal balance, and storage operation to optimize the system.

[0047] For the energy efficiency model of the hybrid cooling system, the cooling power consumption of the hybrid cooling system in time period is represented as: (11) Where , represent the dynamic energy efficiency ratio and IT equipment heat dissipation in time period, respectively; , represent the chilled water temperature and ambient temperature in time period, respectively, , COPf and COPl represent the efficiency coefficient of air-cooling and liquid-cooling subsystems, respectively; Tref represents the reference temperature.

[0048] (2) Real-time optimization model In the real-time optimization model, a collaborative decision-making framework based on Stackelberg game is constructed: the leader, data center cluster operator (DCCO), adjusts the charging and discharging price strategy of the shared energy storage system (SES) dynamically to maximize its own revenue while encouraging prosumers (DCPs) to participate in energy sharing; and the DCPs as followers optimize the interactive demand response (IDR) strategy and peer-to-peer (P2P) energy transaction plan based on the SES price signal to minimize the real-time operation cost.

[0049] On this basis, the embodiment designs a dynamic pricing mechanism for supply and demand linkage - relying on the dynamic adjustment of P2P transaction price based on the electricity supply and demand ratio (ESDR) and the heat supply and demand ratio (HSDR). When there is excess energy supply, the price is reduced to stimulate demand; otherwise, the price is increased to suppress excessive transactions, and the joint bidding and matching of electricity and waste heat are supported to promote the cross-category collaborative use of electricity and heat energy.

[0050] The model excavates load flexibility through a three-level refined IDR mechanism: on the data side, relying on the spatiotemporal flexibility of server workloads, non-real-time computing tasks are migrated to low-price valleys or adjacent data centers; on the thermal side, using building thermal inertia and user thermal comfort tolerance intervals, the heating system operation period is intelligently adjusted; on the power side, based on the hierarchical response strategy of rigid load, transferable load, and adjustable load (air conditioner), differentiated load form optimization is achieved, thereby systematically improving the energy supply and demand balance capability.

[0051] The real-time optimization model in the embodiment is realized based on the Stackelberg game framework, consisting of an upper model for DCCO decision and a lower model for each DCP decision.

[0052] 1) Upper model (DCCO decision) Exemplarily, the upper model constructs an objective function with a 15-minute time interval and a goal of maximizing DCCO revenue, represented as: (12) wherein, COPf and COPl represent the efficiency coefficient of air-cooling and liquid-cooling subsystems, respectively; COPf and COPl represent the efficiency coefficient of air-cooling and liquid-cooling subsystems, respectively; COPf and COPl represent the efficiency coefficient of air-cooling and liquid-cooling subsystems, respectively; COPf and COPl represent the efficiency coefficient of air-cooling and liquid-cooling subsystems, respectively; The scheduling period of the real-time optimization model, the scheduling period is an integer multiple of 15 minutes, for example, 4 hours can be set as a scheduling period, one scheduling period includes 16 time periods.

[0053] In the implementation process, the shared energy storage system charging and discharging plan with an hour as the time interval 、 is transmitted to the upper model in the real-time optimization model to solve the hourly 、 ; on this basis, the upper model is solved to obtain the hourly with 15 minutes as the time interval.

[0054] The upper model satisfies the IDR flexibility constraint. The IDR flexibility constraint is an important limiting condition in the optimization process of the objective function. It ensures the flexibility and stability of the power dispatching scheme, thereby helping to achieve the optimization of the objective function. In the objective function, whether it is to maximize the DCCO income or to minimize the real-time cost of the DCP, the influence of the IDR flexibility constraint needs to be considered. This helps to balance the stability and economy of the power system in the optimization process. The IDR flexibility constraint limits the selection range of the power dispatching scheme, ensuring that the scheme can meet the stability and reliability requirements of the power system, thereby helping to maximize the overall benefit.

[0055] 2) Lower model (DCP decision) Each DCP makes individual decisions to construct an objective function with the goal of minimizing the real-time cost of the current DCP; the objective function of the first DCP is represented as: (13) Where, represents the transaction power (kW) of the first DCP and the shared energy storage system in the time period; represents the transaction power (kW) of the first DCP in the time period in the P2P market; represents the transaction price of the P2P market in the time period; represents the adjustment cost (yuan) of the first DCP in the time period, such as task delay penalty or thermal comfort compensation.

[0056] The upper model and the lower model satisfy: (14) The dynamic P2P pricing mechanism satisfies: (15) wherein, denotes the price elasticity coefficient (dimensionless), denotes the power supply-demand ratio of the time period, , denotes the load power of the th DCP in the time period; denotes the thermal energy supply-demand ratio of the time period, .

[0057] The decision variables of the lower-layer model include , and , which are used to schedule the corresponding DCPs in real time.

[0058] The lower-layer model satisfies the P2P transaction balance constraint, the data load delay constraint, and the thermal comfort constraint. 1) P2P transaction balance constraint (16) The total amount of electricity trading within the cluster needs to be balanced to avoid external intervention.

[0059] 2) Data load delay constraint The spatiotemporal transferability of data load is reflected through the task delay constraint, which is: (17) wherein, denotes the execution state of task in the time period, taking 1 to indicate running and 0 to indicate not running; denotes the task start time, before which the task will not start execution; is the maximum allowed delay time, within which the task can be delayed for execution without violating the constraint condition; denotes the set of tasks.

[0060] The purpose of this constraint is to ensure that the task is completed within the specified time, avoiding performance degradation and service quality reduction due to delay.

[0061] 3) Thermal comfort constraint (18) wherein: is the predicted average vote value (quantifying human thermal sensation) in the time period; is the expected unsatisfactory percentage of the time period.

[0062] (19) wherein, 、 data load delay penalty coefficient of the task ; 、 thermal comfort violation penalty coefficient actual execution time of the task.

[0063] The flowchart of the optimization algorithm is shown in Figure 5 , and the specific description is as follows: the above optimization process is developed around dynamic game: first, the data center operator (DCCO) calculates and publishes the global price signal containing the P2P transaction price and the demand side response (IDR) incentive based on the real-time state of the shared energy storage system (SES) and the energy declaration demand of the energy supply device (DCP); then, each DCP optimizes the local strategy according to the price signal, adjusts the server load, starts and stops the gas boiler, and other IDR means, and decides to generate a scheduling scheme that takes into account the cost and constraints combined with the P2P transaction volume; finally, through multiple rounds of price-demand interaction iteration, the game parties (DCCO and DCPs) reach an equilibrium state - at this time, the DCCO cannot increase its income by unilateral price adjustment, and the DCPs cannot further reduce costs by strategy adjustment, forming a stable Pareto optimal solution.

[0064] The core advantages of the above model are reflected in three aspects:‌two-stage complementarity‌, the day-ahead robust optimization and real-time dynamic game form a "risk pre-control-flexible adjustment" collaborative framework, the former avoids extreme scenario risk through WCVaR constraint, and the latter excavates the flexibility potential of distributed resources through Stackelberg game;‌multi-dimensional flexibility integration‌, the IDR mechanism realizes the cross-time and space coordination of electric energy, thermal energy and computing power through data load space-time migration, thermal comfort constraint optimization and multi-type load hierarchical management, while the P2P transaction mechanism opens up the cross-agent flow channel of surplus electricity / heat, and builds a multi-energy complementary flexible network;‌incentive compatible design‌, based on the interest coordination mechanism of principal-agent game, it not only guarantees the global optimization authority of the DCCO as the system coordinator, but also gives the DCPs the right to make autonomous decisions based on the price signal, avoiding efficiency loss through strategy equilibrium.

[0065] This scheme, through the "risk-game-market" triple synergy mechanism, not only improves the economy of the data center cluster, but also effectively alleviates the double impact of renewable energy volatility and computing power load space-time unevenness, providing a closed-loop solution for low-carbon operation of data centers under high proportion of renewable energy penetration.

[0066] (1) Call CPLEX / Gurobi to solve the day-ahead optimization model That is, in the day-ahead stage, robust optimization is realized and mixed with MILP solution. Specifically, the day-ahead optimization takes minimizing the total operation cost (purchasing power cost, fuel cost, wind curtailment penalty, etc.) as the optimization objective, and introduces WCVaR constraints to control the risk in extreme scenarios. Among them, the uncertainty set mainly models the wind and light output prediction error and load fluctuation as an interval uncertainty set: (20) ‌In the formula, respectively represent the predicted value of wind power in the uncertainty set, the predicted value of photovoltaic power in the uncertainty set, the predicted value of power system load in the uncertainty set, and the predicted value of heat supply system load in the uncertainty set; 、 respectively represent the upper and lower bounds of the power system load; 、 respectively represent the upper and lower bounds of the heat supply system load.

[0067] Robust counterpart model transformation, using column and constraint generation (C&CG) algorithm, decompose the problem into: Master Problem, determine the scheduling baseline of energy storage (SES) and hydrogen energy system (H2); Subproblem, search for the worst uncertainty scenario and verify the robustness of the solution. Call CPLEX / Gurobi to solve the mixed integer linear programming (MILP) model, the specific solving steps are as follows: 1) Initialize the master problem and set the initial feasible solution (such as SES charging and discharging baseline, hydrogen production plan).

[0068] 2) Solve the subproblem to verify the feasibility of the current solution in the uncertainty set. If there is a scenario that violates the constraints, generate Benders cut (Benders Cut) and add it to the master problem.

[0069] 3) Iteratively update the master problem and the subproblem until the difference between the objective functions of the master problem and the subproblem is less than a threshold.

[0070] (2) Use distributed ADMM to solve the real-time optimization model. ‌Upper layer (DCCO): maximize revenue as the target, update the transaction price of shared energy storage system through gradient descent method and broadcast to all DCPs.

[0071] (21) Among them, represents the transaction price of the shared energy storage system in the Transaction price of sharing energy storage system in sub-iteration; Learning rate, Profit function of DCCO.

[0072] Lower layer (DCPs): Adjust the load distribution and energy demand to minimize energy cost, and solve it through distributed ADMM algorithm.

[0073] The advantages of the above model constructed in this embodiment are embodied in: (1) Two-stage complementary optimization The day-ahead robust optimization stage effectively avoids risks in extreme scenarios by introducing WCVaR constraints, ensuring the stability of the system under uncertain conditions. This optimization strategy provides a solid foundation for subsequent real-time dynamic games, achieving "risk pre-control".

[0074] The real-time dynamic game stage uses the Stackelberg game model to tap the flexibility potential of distributed resources, achieving flexible adjustment of the system. Through multiple rounds of price-demand interaction iteration, both parties reach an equilibrium state, forming a stable Pareto optimal solution.

[0075] (2) Multi-dimensional flexibility integration: The IDR mechanism realizes the cross-time and space coordination of electricity, heat, and computing power through data load space-time migration, thermal comfort constraint optimization, and multi-type load hierarchical management. This multi-dimensional flexibility integration improves the overall efficiency and response speed of the system.

[0076] The P2P transaction mechanism opens up the cross-agent flow channel for surplus electricity / heat, building a multi-energy complementary flexible network. This mechanism promotes the optimal allocation and efficient use of resources, further enhancing the economic efficiency of the system.

[0077] (3) Incentive compatible design The interest coordination mechanism based on principal-agent game ensures the global optimization authority of DCCO as the system coordinator and gives DCPs the right to make autonomous decisions based on price signals. This design avoids efficiency loss and achieves a win-win situation for both parties.

[0078] Through strategic equilibrium, it is ensured that when the system reaches an equilibrium state, DCCO cannot increase its revenue by unilateral price adjustment, and DCPs cannot further reduce their costs by strategic adjustment, thus forming a stable Pareto optimal solution.

[0079] In summary, the embodiment proposes a two-stage energy sharing model for data center clusters. By integrating multi-load integrated demand response (IDR), shared energy storage (SES), and P2P transaction mechanisms, the model achieves efficient energy sharing and collaborative optimization. In terms of technical feasibility, the combined scheduling of hydrogen-thermal devices and SES effectively smooths the intermittency of wind and solar power output and improves system flexibility. In terms of economic advantage, the dynamic pricing mechanism combined with multi-energy flow optimization significantly reduces operating costs and promotes mutual benefits between operators and producers and consumers. In terms of expansion value, the model can be extended to comprehensive energy scenarios such as industrial parks and transportation hubs, supporting cross-domain low-carbon transformation. Simulation results show that the model reduces data center operating costs while considering user satisfaction and renewable energy consumption rate. Future research will further deepen multi-energy flow coupling dynamics modeling, server-level real-time control algorithms, and data privacy protection technologies to provide universal solutions for high-energy consumption scenarios.

[0080] Those skilled in the art can understand that all or part of the processes of the above-mentioned embodiment methods can be completed by instructing relevant hardware through a computer program, and the program can be stored in a computer readable storage medium. Among them, the computer readable storage medium is a disk, an optical disk, a read-only memory, or a random access memory, etc.

[0081] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A low-carbon economic robust optimization method for data center clusters considering multiple types of loads, characterized by: The method comprises: Construct a day-ahead optimization model and a real-time optimization model for a data center cluster that considers multiple load types. The objective function of the day-ahead optimization model is constructed with a first time interval and the goal of minimizing the total day-ahead operating cost of the data center cluster. The real-time optimization model is constructed with a second time interval and includes an upper-level model for DCCO decision-making and a lower-level model for each DCP decision-making. Solve the day-ahead optimization model to calculate the charge and discharge schedule of the shared energy storage system with the first time as the time interval; The shared energy storage system charging and discharging plan with the first time interval is transmitted to the real-time optimization model, and the real-time optimization model is solved to perform real-time scheduling of each DCP; Among them, the first time is greater than the second time.

2. The low-carbon economy robust optimization method for data center clusters considering multiple types of loads according to claim 1 is characterized in that: The objective function of the day-ahead optimization model is expressed as: (1) in, express The electricity purchase cost for the period, Indicates that all gas boilers are Total fuel cost for the period, Indicates that the shared energy storage system The loss cost of charging and discharging during the period; is the worst-case risk value, is the risk aversion coefficient; Indicates the scheduling period of the day-ahead optimization model.

3. The low-carbon economy robust optimization method for data center clusters considering multiple types of loads according to claim 2 is characterized in that: (2) in, express The grid electricity price for the time period, express Power purchased during the time period; (3) in, Indicates the The gas boiler in the DCP Gas consumption during the period, express Gas and electricity prices for the time period, Indicates the total number of DCPs; (4) in, represents the loss cost coefficient of the shared energy storage system, 、 Represents the shared energy storage system in Charging power and discharging power during the time period.

4. According to the low-carbon economy robust optimization method for data center clusters considering multiple types of loads according to claim 3, the decision variables of the day-ahead optimization model include 、 、 、 .

5. According to the low-carbon economic robust optimization method for a data center cluster considering multiple types of loads described in claim 4, the constraints of the day-ahead optimization model include power balance constraints, thermal energy balance constraints, SES operation constraints, and WCVaR risk constraints.

6. According to the low-carbon economy robust optimization method for data center clusters considering multiple load types according to claim 5, the power balance constraint is expressed as: (5) in, Indicates that the shared energy storage system The net discharge power of the time period, ; Indicates the The RES in the DCP is The wind and solar power output during the period, that is, the power generation of renewable energy; Indicates the The IT servers in the DCP Power consumption during the time period; Indicates the The cooling system in a DCP is Power during the time period; Indicates the The application of DCP Load power during the period.

7. The method for low-carbon economic robust optimization of a data center cluster considering multiple load types according to claim 6, wherein the second time is 15 minutes and the first time is 1 hour; The objective function is constructed with the goal of maximizing DCCO benefits, which is expressed as: (6) in, Indicates that the shared energy storage system The transaction price of the period, Indicates that the shared energy storage system Net discharge power during the time period; represents the scheduling period of the real-time optimization model; The shared energy storage system charging and discharging plan will be based on hourly intervals 、 Transmitted to the upper model in the real-time optimization model to solve the hourly 、 ; is the decision variable of the upper model.

8. The method for low-carbon economic robust optimization of a data center cluster considering multiple types of loads according to claim 7 is characterized in that: The upper model satisfies the IDR flexibility constraint.

9. The method for low-carbon economic robust optimization of a data center cluster considering multiple types of loads according to claim 8, characterized in that: Each DCP makes its own decision and builds an objective function with the goal of minimizing the real-time cost of the current DCP. The objective function of a DCP is expressed as: (7) (8) in, Indicates the DCP and shared energy storage system Trading power of the session; Indicates the DCPs in The trading power of the P2P market during the period; Indicates that the P2P market is The transaction price of the session; Indicates the DCP in Adjustment costs for time periods, such as task delay penalties or thermal comfort compensation; The decision variables of the lower model include 、 and , used to perform real-time scheduling of the corresponding DCP.

10. The method for low-carbon economic robust optimization of a data center cluster considering multiple types of loads according to claim 9, characterized in that: The lower-level model satisfies P2P transaction balance constraints, data load delay constraints, and thermal comfort constraints.

Images