Power transmission and distribution collaborative scheduling method and system considering data center computer multi-element flexibility

Abstract

The application provides a power transmission and distribution collaborative scheduling method and system considering data center computing multi-element flexibility, and belongs to the technical field of power dispatching; the method comprises the following steps: constructing a multi-element flexibility resource quantification model based on the characteristics of the computing task at the load flexibility resource level and the physical equipment level; constructing a three-layer collaborative optimization framework of a power transmission network-power distribution network-data center; the upper layer of the power transmission network is used for decision-making of unit operation state and node marginal electricity price, the middle layer of the power distribution network is used for optimization of local resource configuration and transmission of electricity price signals, and the lower layer of the data center is used for optimization and decision-making of the space-time distribution strategy of the computing load; and a target cascade analysis method is used to solve the multi-level coupling problem among the power transmission network layer, the power distribution network layer and the data center layer; the application breaks through the limitation of the existing research which is limited to local optimization of the power distribution network, and realizes global collaborative optimization of power resources and computing resources across layers.

Description

Technical Field

[0001] This invention belongs to the field of power dispatching technology, and in particular relates to a transmission and distribution coordinated dispatching method and system that takes into account the multi-dimensional flexibility of data center computing. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the accelerated development of new power system construction, the deep integration of data centers and power systems has become an inevitable trend in the digital transformation of energy. Data centers are not only core nodes of information networks but also important loads of power systems. Data centers contain abundant flexible resources, including the time and space scheduling flexibility of computing loads, the thermal inertia regulation potential of cooling systems, and the charging and discharging flexibility of energy storage devices. By rationally scheduling computing loads, the spatiotemporal transfer of power loads can be realized, providing high-quality demand response resources for the power system.

[0004] However, existing scheduling methods generally suffer from the following technical shortcomings: (1) At present, the utilization of data center flexibility resources is mostly limited to a single type of resource (such as only considering the migration of computing load or only considering the adjustment of cooling system). There is a lack of systematic collaborative modeling and optimized scheduling of multiple flexibility resources such as computing load, cooling system and energy storage equipment, resulting in the overall demand response potential of data center not being fully explored.

[0005] (2) Existing research on the collaboration between data centers and power systems mostly focuses on the distribution network level, meaning that data centers only interact with the local distribution network. However, the load scheduling of large data center clusters often spans multiple distribution network areas, and their operation strategies not only need to respond to the node marginal electricity price signals at the transmission level, but also need to meet the operational constraints of different distribution networks. Existing technologies lack a cross-level global collaborative optimization mechanism from the transmission network layer to the distribution network layer and then to the data center layer, making it difficult to achieve efficient allocation of power resources and computing resources in multi-level power grids, thus restricting the contribution of data center flexibility resources to the overall economy and security of the power system. Summary of the Invention

[0006] To overcome the shortcomings of the existing technologies, this invention provides a transmission and distribution coordinated scheduling method and system that takes into account the multi-dimensional flexibility of data center computing. It aims to break through the limitations of existing research which is limited to local optimization of distribution networks and realize cross-level global coordinated optimization of power resources and computing resources.

[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides a transmission and distribution coordinated scheduling method that takes into account the multi-dimensional flexibility of data center computing.

[0008] A coordinated scheduling method for power transmission and distribution that takes into account the diverse flexibility of data center computing includes: Based on the characteristics of computing tasks at the load flexibility resource level and physical device level, a multi-dimensional flexibility resource quantification model is constructed. The multi-dimensional flexibility resource quantification model includes a spatiotemporal flexibility model that takes into account the load's electricity consumption, a refrigeration system model that takes into account the building's thermal inertia, and an energy storage device model. Based on the aforementioned multi-dimensional flexible resource quantification model, a three-layer collaborative optimization framework of transmission network, distribution network, and data center is constructed: the upper transmission network layer aims to minimize the system's power generation cost and is used to decide on the unit's operating status and the node's marginal electricity price; the middle distribution network layer is used to optimize local resource allocation and transmit electricity price signals; and the lower data center layer is used to optimize the spatiotemporal allocation strategy for computing load. The target cascade analysis method is used to solve the multi-level coupling problem between the transmission network layer, the distribution network layer and the data center layer, so as to achieve the coordinated optimization of power dispatching and computing power dispatching.

[0009] Furthermore, based on the characteristics of the computing tasks at the load flexibility resource level, the computing load is divided into latency-sensitive tasks and latency-tolerant tasks; based on the characteristics of the computing tasks at the physical device level, the computing load is divided into cooling systems and energy storage devices.

[0010] Furthermore, the three-layer collaborative optimization framework of transmission network-distribution network-data center has bidirectional interaction between its various layers. Specifically, in the forward transmission path, the transmission network generates the node marginal electricity price by solving network constraints, unit combination and load demand, and then transmits it layer by layer to the distribution network and data center.

[0011] Furthermore, the construction of the spatiotemporal flexibility model includes the introduction of flexibility constraints, namely: introducing computational load spatial flexibility constraints and computational load spatiotemporal flexibility constraints for latency-sensitive tasks and latency-tolerant tasks, respectively.

[0012] Furthermore, the construction of the cooling system model includes: establishing an energy efficiency ratio model characterizing the relationship between cooling power and cooling capacity of the cooling system; constructing a thermal balance differential equation and corresponding discrete solution describing the dynamic changes in indoor temperature of the data center based on the principle of building thermal inertia; and introducing safe operation constraints for indoor temperature. The cooling system is configured to achieve time transfer of heat load by releasing pre-cooling capacity during periods of high electricity price and storing cooling capacity during periods of low electricity price.

[0013] Furthermore, the transmission network layer aims to minimize the system's power generation cost and optimizes decisions regarding the start-up and shutdown of conventional generating units and power output scheduling. The distribution network layer optimizes local resource allocation and load distribution based on the tie-line power transmitted from the transmission network layer and in conjunction with local distribution network constraints, while simultaneously transmitting local node electricity price information to the data center layer. The data center layer, from the perspective of the operator, configures computing power load and power consumption strategies for each data center.

[0014] Furthermore, the objective cascade analysis method is used to solve the multi-level coupling problem between the transmission network layer, the distribution network layer and the data center layer. This includes: using tie line power as the inner loop convergence criterion, using the power system operating cost and the computing power system electricity cost as the outer loop convergence condition, and realizing the collaborative optimization of power dispatch and computing power dispatch through the dynamic update of the penalty function multiplier.

[0015] The second aspect of the present invention provides a transmission and distribution coordinated scheduling system that takes into account the diverse flexibility of data center computing.

[0016] A coordinated dispatching system for power distribution that takes into account the diverse flexibility of data center computing includes: The multi-dimensional flexibility modeling module is configured to: construct a multi-dimensional flexibility resource quantification model based on the characteristics of computing power tasks at the load flexibility resource level and the physical device level; the multi-dimensional flexibility resource quantification model includes a spatiotemporal flexibility model that takes into account the load's electricity consumption, a refrigeration system model that takes into account the building's thermal inertia, and an energy storage device model. The three-layer collaborative optimization framework construction module is configured to: construct a three-layer collaborative optimization framework of transmission network, distribution network, and data center based on the aforementioned multi-dimensional flexible resource quantification model; the upper transmission network layer aims to minimize the system's power generation cost and is used to decide on the unit operating status and node marginal electricity price; the middle distribution network layer is used to optimize local resource allocation and transmit electricity price signals; and the lower data center layer is used to optimize the spatiotemporal allocation strategy of computing load. The transmission and distribution coordinated scheduling module is configured to use the target cascade analysis method to solve the multi-level coupling problem between the transmission network layer, the distribution network layer and the data center layer, so as to achieve coordinated optimization of power scheduling and computing power scheduling.

[0017] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the transmission and distribution coordinated scheduling method considering the multi-functional flexibility of data center computing as described in the first aspect of the present invention.

[0018] The fourth aspect of the present invention provides an electronic device including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the transmission and distribution coordinated scheduling method considering the multi-functional flexibility of data center computing as described in the first aspect of the present invention.

[0019] The above one or more technical solutions have the following beneficial effects: (1) This invention constructs a multi-dimensional flexibility resource quantification model that includes a computing load spatiotemporal flexibility model, a cooling system model taking into account building thermal inertia, and an energy storage device model. This model achieves unified representation and collaborative optimization scheduling of multi-dimensional flexibility resources such as computing load, cooling system, and energy storage device. The model fully explores the spatial migration capability of latency-sensitive tasks, the spatiotemporal dual flexibility of latency-tolerant tasks, the "virtual energy storage" adjustment capability of the cooling system based on building thermal inertia, and the potential of energy storage devices to transform from backup power to active adjustment resources. This significantly improves the overall demand response capability of data centers and reduces the electricity cost and total operating cost of data centers.

[0020] (2) This invention constructs a three-layer collaborative optimization framework based on nodal marginal electricity prices for transmission networks, distribution networks, and data centers, and employs an improved target cascade analysis method for hierarchical iterative solution. This framework generates electricity price signals at the transmission network layer and transmits them layer by layer to the distribution network layer and the data center layer. Simultaneously, the optimized electricity demand at the data center layer is fed back to the transmission network layer via the distribution network layer, forming a two-way dynamic matching closed loop of power dispatch and computing power dispatch. This mechanism overcomes the limitations of existing research, which is confined to local optimization of the distribution network, and realizes global collaborative allocation of power resources and computing power resources at the three levels of transmission networks, distribution networks, and data centers, effectively improving the economy and security of large power grid operation.

[0021] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0022] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0023] Figure 1 This is a flowchart of the power-computing power collaboration mechanism in Embodiment 1 of the present invention.

[0024] Figure 2 This is a schematic diagram illustrating the principle of building thermal inertia in Embodiment 1 of the present invention; wherein, Figure 2 (a) in the diagram represents a schematic of heat conduction between indoors and outdoors. Figure 2(b) in the diagram represents the amplitude attenuation effect of the temperature response. Figure 2 (c) in the diagram represents the hysteresis effect of the temperature response.

[0025] Figure 3 This is a logical flowchart of the three-layer collaborative optimization framework of power transmission network-distribution network-data center in Embodiment 1 of the present invention.

[0026] Figure 4 This is a schematic diagram of the delay-sensitive and delay-tolerant load scheduling results of IDC1 in Embodiment 1 of the present invention.

[0027] Figure 5 This is a schematic diagram of the load scheduling results for IDC3 latency-sensitive and latency-tolerant types in Embodiment 1 of the present invention.

[0028] Figure 6 This is a schematic diagram of the IDC delay-sensitive load scheduling results for different power distribution networks in Embodiment 1 of the present invention.

[0029] Figure 7 This is a schematic diagram of the IDC delay-tolerant load scheduling results for different power distribution networks in Embodiment 1 of the present invention.

[0030] Figure 8 This is a schematic diagram of the power variation of the IDC1 cooling system in Embodiment 1 of the present invention.

[0031] Figure 9 This is a schematic diagram of the power variation of the IDC3 cooling system in Embodiment 1 of the present invention.

[0032] Figure 10 This is a schematic diagram comparing the charging and discharging power of different IDC energy storage devices in Embodiment 1 of the present invention.

[0033] Figure 11 This is a schematic diagram of the power of the tie lines between the transmission network and each distribution network in Embodiment 1 of the present invention; wherein, Figure 11 (a) in the diagram represents the power diagram of the tie line between the transmission network and distribution network No. 1. Figure 11 (b) in the diagram represents the power diagram of the tie line between the transmission network and distribution network No. 2. Figure 11 (c) in the diagram represents the power diagram of the tie line between the transmission network and the No. 3 distribution network.

[0034] Figure 12 This is a schematic diagram of the LMP nodes connecting the distribution network to the transmission network in Embodiment 1 of the present invention; wherein... Figure 12 (a) in the diagram represents a schematic diagram of the connection between distribution network 1 and transmission network node LMP. Figure 12 (b) in the diagram represents a schematic diagram of the connection between distribution network 2 and transmission network node LMP. Figure 12 (c) in the diagram represents a schematic diagram of the connection between distribution network No. 3 and transmission network node LMP. Detailed Implementation

[0035] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0036] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0037] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0038] The overall approach of this invention is based on the concepts of distributed autonomy and global coordination. It proposes a transmission and distribution coordinated scheduling method that considers the multi-dimensional flexibility of data center computing. This method presents a three-layer coordinated optimization framework of transmission network, distribution network, and data center. This framework fully considers the heterogeneous operating mechanisms and bidirectional interaction between power and computing systems. Specifically, the power system provides reliable power supply to computing power and guides load for flexible spatiotemporal scheduling through local marginal price (LMP). The computing system, in turn, provides demand response support to the power system by flexibly adjusting the spatiotemporal distribution of load and the operating strategies of auxiliary equipment. Based on this, the transmission network layer generates nodal price signals through unit combination optimization, the distribution network layer optimizes local resource allocation and transmits price signals downwards, and the data center layer performs multi-dimensional flexible resource scheduling based on price signals. This framework overcomes the limitations of existing research, which is confined to local optimization of the distribution network, and achieves cross-level global coordinated optimization of power and computing resources.

[0039] Example 1 This embodiment discloses a transmission and distribution coordinated scheduling method that takes into account the multi-dimensional flexibility of data center computing.

[0040] A coordinated scheduling method for power transmission and distribution that takes into account the diverse flexibility of data center computing includes: Based on the characteristics of computing tasks at the load flexibility resource level and physical device level, a multi-dimensional flexibility resource quantification model is constructed. The multi-dimensional flexibility resource quantification model includes a spatiotemporal flexibility model that takes into account the load's electricity consumption, a refrigeration system model that takes into account the building's thermal inertia, and an energy storage device model. Based on the aforementioned multi-dimensional flexible resource quantification model, a three-layer collaborative optimization framework of transmission network, distribution network, and data center is constructed: the upper transmission network layer aims to minimize the system's power generation cost and is used to decide on the unit's operating status and the node's marginal electricity price; the middle distribution network layer is used to optimize local resource allocation and transmit electricity price signals; and the lower data center layer is used to optimize the spatiotemporal allocation strategy for computing load. The target cascade analysis method is used to solve the multi-level coupling problem between the transmission network layer, the distribution network layer and the data center layer, so as to achieve the coordinated optimization of power dispatching and computing power dispatching.

[0041] Based on the above process, this invention overcomes the shortcomings of existing research which is limited to local optimization of power distribution networks, and realizes cross-level global collaborative optimization of power resources and computing resources. To facilitate understanding of the technical solution of this invention, the specific implementation methods of this invention will be further explained and described below.

[0042] like Figure 1 As shown, this invention constructs a three-layer interactive framework of "transmission network-distribution network-data center" to form a power-computing power collaborative scheduling mechanism, achieving synergistic optimization of power resources and computing power flexibility resources. This mechanism relies on the flexibility resources of the data center and the bidirectional interaction between the power system and the data center, enabling the data center to participate in power system scheduling decisions, forming a new scheduling mode with deep integration of power and computing power. Specifically, this can be achieved through the following methods: Step S1: Based on the characteristics of computing tasks at the load flexibility resource level and the physical device level, construct a multi-dimensional flexibility resource quantification model.

[0043] The flexibility of data center computing loads and physical equipment can exhibit multi-dimensional characteristics. Fully exploring and integrating these resources is the foundation for achieving power-computing synergy. Specifically, this invention analyzes the issue from two levels: the flexibility of computing loads and the physical equipment.

[0044] In terms of computing load flexibility resources, computing loads can be divided into two categories based on task characteristics: 1) Latency-sensitive tasks (such as online transactions and real-time communication), which, while not allowing for latency, can migrate between different data centers, enabling flexible scheduling in the spatial dimension; 2) Latency-tolerant tasks (such as batch processing and data backup), which can be flexibly scheduled for execution within a time window and can migrate across data centers, possessing both spatiotemporal flexibility. This classification allows computing tasks to be scheduled differently based on the spatiotemporal distribution characteristics of electricity price signals.

[0045] At the level of physical equipment (refrigeration system and energy storage equipment), the refrigeration system utilizes the building's thermal inertia to form a "virtual energy storage" resource. By precisely controlling the refrigeration power, it pre-cools and stores cold energy during periods of low electricity prices and reduces the release of cold energy during periods of high electricity prices, thereby achieving a time-based transfer of heat load. The energy storage equipment, on the other hand, transforms from a traditional emergency backup role to an active resource regulator, optimizing charging and discharging strategies based on electricity price signals and participating in peak shaving and valley filling of the power grid.

[0046] This invention establishes an integrated optimization model (i.e., a multi-dimensional flexibility resource quantification model) that covers computing load, cooling system and energy storage equipment, to achieve coordinated scheduling of multi-dimensional flexibility resources, thereby avoiding conflicts and efficiency losses that may occur from independent optimization of each resource.

[0047] In practical integrated optimization modeling, data center power consumption mainly consists of three parts: computing load power consumption, cooling system power consumption, and energy storage device charging and discharging. The spatiotemporal adjustability of the computing load needs to be quantitatively described through task classification and migration constraints; the dynamic characteristics of the cooling system are affected by building thermal inertia, requiring the establishment of a thermodynamic model considering temperature-power coupling; and the operating strategy of energy storage devices needs to reduce charging and discharging costs while ensuring emergency power supply functionality. The construction of a multi-dimensional flexible resource quantification model can be achieved through the following methods: S1-1, Computational load spatiotemporal flexibility modeling.

[0048] The computing load migration described in this invention is an initial allocation decision for newly arriving tasks, applicable to general batch processing computing tasks that do not involve sensitive data compliance constraints and have relatively relaxed data locality requirements. Based on cloud service service level agreements (SLAs), computing loads are divided into latency-sensitive and latency-tolerant types. Latency-sensitive tasks (such as online transactions, real-time communication, and live video streaming) have strict latency requirements (typically <100ms) and cannot tolerate time delays; they can only be migrated spatially between data centers. Latency-tolerant tasks (such as data backup and synchronization, offline batch processing, and log analysis—tasks with execution times in the minutes range and flexible scheduling within the SLA time window) can tolerate delays of several hours or even across days. They can flexibly schedule execution times within the SLA-allowed time window and have cross-data center migration capabilities, offering both temporal and spatial flexibility.

[0049] 1) Calculate the number of loads.

[0050] The total computing load in a data center remains constant before and after flexible scheduling, but its distribution in time and space can be optimized based on electricity price signals. The balance of computing load is as follows: (1) In the formula, For data centers i During the period t The amount of computational load processed internally. For data centers i During the period t The number of latency-sensitive computational loads processed internally. For data centers i During the periodt The number of latency-tolerant computational loads processed internally. Among them, the number of computational loads... Indicates data center i During the period t The aggregate computational workload is processed internally, rather than the entire execution process of a single job from submission to completion. In real-world data center scheduling systems (such as Kubernetes and YARN), large jobs are automatically broken down into parallel-processable task instances after submission, and the server cluster continuously processes the stream of task instances from different jobs in parallel during each time period.

[0051] 2) Calculate the power consumption of the load.

[0052] When a data center handles a computing load, the power consumption of the servers is proportional to the amount of computing load they are processing. Considering the processing speed and number of servers, the power consumption of the computing load is: (2) In the formula, For data centers i During the period t Internally calculates the power consumption of the load. For data centers i The full-load power consumption of a single server For data centers i The idle power consumption of a single server; The rate at which a single server can handle computational load; This refers to the number of servers.

[0053] 3) Limitations on computational resource utilization.

[0054] To ensure the reliable operation of the data center, a certain amount of redundant computing resources needs to be reserved to cope with sudden demand, specifically as follows: (3) In the formula, The proportion of redundant computing resources in the data center. For data centers i Total computing resources. Among them, data centers... i Total computing resources Due to the number of its servers and the speed at which a single server can handle computing load Joint decision, that is This formula shows that the total computing power of a data center is equal to the product of the capacity of a single server and the number of servers.

[0055] 4) Space flexibility constraints for latency-sensitive computing loads.

[0056] Latency-sensitive computing loads, due to their stringent response time requirements, can only track low-electricity-cost areas through spatial migration between different data centers, and cannot tolerate time delays. Therefore, the total amount of latency-sensitive computing load in the system remains unchanged before and after scheduling. (4) In the formula, and Data centers before and after flexible scheduling, respectively. i During the period t The number of latency-sensitive computing loads within the scope, For data center collections.

[0057] 5) Delay-tolerant computing load spatiotemporal flexibility constraints.

[0058] Latency-tolerant computing workloads exhibit unique time elasticity characteristics in data center load scheduling. Unlike latency-sensitive computing workloads, these tasks have relatively relaxed requirements for response time, allowing for flexible time-dimensional scheduling while meeting service level agreement (SLA) constraints. In other words, the time flexibility of latency-tolerant computing workloads lies in their ability to delay execution time from high-electricity-cost periods to low-electricity-cost periods, thereby optimizing costs while maintaining service quality. Based on latency elasticity, latency-tolerant computing workloads can be scheduled from time-to-time... Delayed to time implement.

[0059] (5) (6) In the formula, For flexible scheduling of the former data center i The number of latency-tolerant computing loads, For flexible scheduling of the data center i The number of latency-tolerant computing loads; For data centers i Maximum delay time for internal delay-tolerant computing loads The scheduling period is [number].

[0060] Equation (5) is used to ensure that during the time period Arrival latency-tolerant computing load It must be within its maximum tolerable delay time Processing is completed within a certain time period within the scheduling cycle; Equation (6) is used to handle the special case of the end of the scheduling cycle when the task is within the time period. The theoretical maximum delay time window upon arrival. It will exceed the current scheduling cycle. The boundary, therefore this constraint modifies the time window to... Ensure that all tasks are completed within the current scheduling cycle.

[0061] In addition to flexible scheduling over time, latency-tolerant computing tasks can also achieve load redistribution across data centers through spatial migration: (7) Equations (5) to (7) together constitute the spatiotemporal flexibility constraint system for delay-tolerant computing loads. Equation (5) is the sliding time window constraint, Equation (6) is the periodic boundary correction constraint, and Equation (7) is the spatial migration conservation constraint, used to jointly ensure that all delay-tolerant tasks are completed within the maximum delayable time. The process is completed within the time limit.

[0062] This spatiotemporal coupling scheduling mechanism enables latency-tolerant tasks to fully utilize the spatiotemporal complementarity of data center networks.

[0063] S1-2, Modeling of a refrigeration system that takes into account the building's thermal inertia.

[0064] Modeling data center cooling systems requires full consideration of building thermal inertia. Building thermal inertia refers to a building's ability to resist temperature changes due to its heat capacity and thermal resistance. This characteristic means that indoor temperature does not respond instantaneously to changes in ambient temperature, but rather manifests as a decrease in the amplitude of temperature fluctuations and a response delay. For example... Figure 2 The diagram shown illustrates the principle of building thermal inertia; where, Figure 2 (a) in the diagram represents a schematic of heat conduction between indoors and outdoors. Figure 2 (b) in the diagram represents the temperature-dependent amplitude attenuation effect. Figure 2 (c) in the diagram represents the hysteresis effect of the temperature response. Specifically, as shown in the diagram... Figure 2 As shown in (a), outdoor ambient temperature transfers heat to the interior through the building envelope (walls) via radiation and conduction. Changes in indoor temperature lag behind and are less pronounced than changes in outdoor ambient temperature. Figure 2 As shown in (b), the outdoor temperature exhibits periodic fluctuations with large amplitude (solid line). After passing through a wall with a certain thermal inertia, the amplitude of the indoor temperature fluctuations (dashed line) significantly decreases. That is, the greater the thermal inertia, the lower the peak value and the higher the trough value of the indoor temperature fluctuations. This phenomenon is called the amplitude attenuation effect. Figure 2 As shown in (c), the waveform of the indoor temperature response (dashed line) exhibits a significant time delay compared to the outdoor temperature waveform (solid line). That is, the peak or trough of the indoor temperature occurs later than the outdoor temperature. The greater the thermal inertia, the longer this delay. This phenomenon is called the hysteresis effect. The amplitude attenuation effect and the hysteresis effect together demonstrate the moderating effect of building thermal inertia on the indoor thermal environment.

[0065] In data centers, the heat generated by server equipment, the cooling capacity provided by the cooling system, and the heat transfer from the building envelope collectively determine the dynamic balance of the indoor thermal environment. The thermal time constant of this invention is on the same order of magnitude as the scheduling period, indicating that the room temperature has not yet fully reached thermal equilibrium within a single scheduling period. The coupling relationship between thermal states across time periods is significant, and the "virtual energy storage" effect of building thermal inertia physically exists within the scheduling time scale of this invention, satisfying the physical prerequisite for flexible cross-time period adjustment using the thermal inertia of the cooling system.

[0066] 1) Relationship between cooling power and cooling capacity.

[0067] There is a non-linear relationship between the power consumption of a refrigeration system and the cooling capacity it provides, and this relationship is affected by the equipment's energy efficiency ratio. Furthermore, even when not cooling, the refrigeration system still requires a certain amount of base power to maintain normal operation. (8) In the formula, For data centers i In the medium refrigeration system Power consumption during the time period For data centers i In the medium refrigeration system The energy efficiency ratio (COP) of the refrigeration equipment during the specified time period; For data centers i In the medium refrigeration system t Cooling capacity during the period This refers to the minimum electrical power required by the refrigeration system, also known as the lower limit of refrigeration power; where, the lower limit of refrigeration power... The settings ensure that even during periods of extremely low heat load, the cooling system fans and basic heat exchange equipment continue to operate at minimum power, and also ensure the continuity of airflow organization in the computer room, preventing the accumulation of local hot spots.

[0068] 2) Calculation of energy efficiency ratio.

[0069] The COP (Coefficient of Performance) of refrigeration equipment is significantly affected by ambient temperature, typically decreasing as outdoor temperature rises. Based on performance test data fitting of refrigeration equipment under different outdoor temperature conditions, the relationship between the energy efficiency ratio and outdoor temperature is as follows: (9) In the formula, For data centers i exist t Energy efficiency ratio of refrigeration equipment during different time periods For data centers i exist t Outdoor temperature (°C) during the time period; , and This is an empirical constant for energy efficiency ratio related to the region where the data center is located.

[0070] 3) Cooling power constraints.

[0071] The operating power of a refrigeration system is limited by the physical characteristics of the equipment, satisfying the following relationship: (10) In the formula, This represents the maximum power consumption of the refrigeration system.

[0072] 4) Calculation of indoor cooling capacity.

[0073] The actual cooling capacity of a data center room equals the cooling capacity provided by the cooling system minus the heat generated by computing equipment and the basic heat generated by lighting and other electrical equipment. Since almost all the electrical energy of computing equipment is converted into heat energy, the indoor cooling balance is: (11) In the formula, For data centers i During the period t The indoor cooling capacity; This refers to the basic heat generated in a data center due to personnel, lighting, and power supply equipment.

[0074] 5) Indoor thermal balance differential equation.

[0075] Considering the building's thermal inertia, the temperature change inside the data center follows the first law of thermodynamics. Abstracting the data center as a lumped-parameter thermodynamic system with indoor air as the heat transfer medium, the differential equation for indoor heat balance is established: (12) In the formula, For data centers i During the period t The indoor temperature (°C); The coefficient of thermal inertia; It is a constant related to factors such as indoor heat transfer coefficient and indoor area; It is the equivalent coefficient for temperature change, and is usually related to indoor air volume, heat capacity, and room heat storage capacity.

[0076] 6) Discrete solutions to the thermal equilibrium differential equation.

[0077] To facilitate calculation, the continuous differential equations are discretized: (13) In the formula, For data centers i During the periodt- 1. Indoor temperature (°C); For time intervals.

[0078] 7) Indoor temperature constraints.

[0079] (14) In the formula, The minimum permissible indoor temperature (°C), The maximum permissible indoor temperature (°C), where The values ​​are based on the ASHRAE Class A1 data center recommended intake air temperature standard (15℃ to 32℃), with sufficient safety margin.

[0080] It should be noted that the present invention adopts a lumped parameter first-order thermodynamic model with indoor air as the equivalent heat transfer medium, and regards the entire computer room as a single thermal node with uniform temperature. Therefore, it is particularly suitable for dynamic modeling of thermal state on an hourly time scale.

[0081] S1-3, Modeling of energy storage equipment.

[0082] 1) Energy storage capacity constraints.

[0083] (15) In the formula, For data centers i Energy storage devices during time periods t The amount of electricity; For data centers i Energy storage devices ensure the minimum power capacity required by the data center; For data centers i The rated capacity of the energy storage device.

[0084] 2) Constraints on the operating status of energy storage.

[0085] (16) In the formula, , For data centers i Energy storage devices during time periods t The charging and discharging state within.

[0086] 3) Energy storage charging and discharging power constraints.

[0087] (17) In the formula, , Data Center i Energy storage devices during time periods t Internal charging and discharging power; , Data Center i Energy storage devices during time periods t Maximum charging and discharging power within the device.

[0088] 4) Energy storage energy balance constraints.

[0089] (18) In the formula, For data centers i Energy storage devices during time periods t- 1 unit of battery power; , These are the charging and discharging efficiencies of the energy storage device, respectively.

[0090] 5) Constraints on the periodicity of energy storage.

[0091] (19) In the formula, and Data Center i The amount of electricity stored in the energy storage device during the initial period and the initial period of the next cycle.

[0092] Step S2: Based on the aforementioned multi-dimensional flexible resource quantification model, construct a three-layer collaborative optimization framework for the transmission network, distribution network, and data center.

[0093] This invention constructs a three-layer collaborative optimization model based on LMP (Light Principles and Modulation) for the transmission network, distribution network, and data center. This model decomposes the complex multi-agent collaborative problem into multiple optimization problems at the transmission network, distribution network, and data center levels. Thus, the bidirectional interaction between the power system and the data center transforms from a one-way price response to a two-way collaborative optimization. By establishing a complete closed loop of information interaction and power demand across the three layers of the transmission network, distribution network, and data center, bidirectional interaction between each level is achieved: In the forward transmission path, the transmission network generates nodal marginal prices by solving network constraints, unit combination, and load demand. This LMP signal is transmitted layer by layer to the distribution network and data center, providing them with accurate economic incentive signals. In the reverse transmission path, the data center, based on the received LMP signal, comprehensively considers multiple factors such as computing power task characteristics, equipment operating status, and service quality requirements to optimize and formulate electricity consumption strategies. Its decision results change the spatiotemporal distribution of the system's load. This load change is aggregated by the distribution network and fed back to the transmission network, affecting the system's supply-demand balance and power flow distribution, triggering a new round of unit scheduling optimization and price updates. Through iterative optimization, Pareto optimality is achieved. This two-way interactive mechanism is achieved by constructing a three-layer operating framework of transmission network-distribution network-data center. It not only improves the accuracy of price signals, but more importantly, it transforms data centers from traditional "price takers" to "price formers," truly realizing the coordinated optimization of power dispatch and computing power dispatch.

[0094] 1) Transmission network layer model.

[0095] The transmission network layer aims to minimize the system's power generation cost by optimizing the start-up and shutdown decisions and output scheduling of conventional generating units, thereby minimizing system operating costs while ensuring a balance between power supply and demand.

[0096] The optimization objective function for the power transmission network layer is: (20) In the formula, For the operating costs of the power transmission network; For the set of scheduling periods; A collection of thermal power units; For the unit During the period t contribution; , and For the unit Fuel cost coefficient; For the unit Start-stop status; Costs for unit start-up and shutdown; and These are respectively the upward and downward adjustments to the unit's standby costs; and The two measures are to increase and decrease the reserve capacity of the generating units, respectively.

[0097] 2) Distribution network layer model.

[0098] The distribution network layer, as an intermediate layer, receives tie-line power from the transmission network layer, optimizes local resource allocation and load distribution in conjunction with local distribution network constraints, and transmits local node electricity price information to the data center layer.

[0099] Distribution network m The optimization objective is to minimize the operating cost within the scheduling cycle, including network loss cost, load shedding penalty cost, and light and wind curtailment penalty cost, expressed as: (twenty one) In the formula, For the first m The operating cost of a distribution network; For distribution network m The set of branches; , and Distribution network m A collection of wind turbine generators, photovoltaic power sources, and general loads of the distribution network; , and Distribution network m Network loss cost coefficient, load d The load shedding cost coefficient and the wind / solar curtailment cost coefficient; For distribution network m Middle Branch Road During the period The square of the current; For distribution network m Middle Branch Road The resistance; , and Distribution network m medium load d Photovoltaics v and wind power w During the period t The amount of load shedding, solar curtailment, and wind curtailment.

[0100] 3) Data center layer model.

[0101] From the perspective of cloud service providers' operations, the data center layer has constructed an operating cost optimization model that comprehensively considers various cost factors. The core objective of this model is to minimize the overall operating cost of a multi-data center cluster by rationally configuring the computing load and power consumption strategies of each data center while ensuring service quality.

[0102] (twenty two) In the formula, For data center operating costs; , and These are the data center electricity costs, data transmission costs, and temperature deviation penalties, respectively.

[0103] Electricity costs are a major component of data center operating costs, and these costs are calculated based on LMP signals transmitted from the upper-level power system. (twenty three) In the formula, For the time period Data Center i The marginal electricity price transmitted at the node in the distribution network; This represents the basic load power of the data center.

[0104] Data transfer costs reflect the bandwidth costs incurred in performing load migration and data synchronization between data centers: (twenty four) In the formula, Calculate the network bandwidth cost required to migrate the basic network switching equipment unit to handle the load.

[0105] To further optimize temperature control, a temperature deviation penalty mechanism is introduced, incurring additional penalty costs when the temperature deviates from the optimal operating range. (25) (26) In the formula, For data centers i exist t Temperature deviation penalty during a specific time period; The lowest temperature (°C) at which no penalty will be imposed; The highest indoor temperature (°C) at which no penalty will be imposed; and These are the penalty coefficients for downward and upward deviations in indoor temperature, respectively.

[0106] Other constraints in the data center layer include: spatiotemporal flexibility constraints for computing load (1)-(7), operating constraints for cooling system (8)-(14), and operating constraints for energy storage equipment (15)-(19).

[0107] Step S3: Use the target cascade analysis method to solve the multi-level coupling problem between the transmission network layer, distribution network layer and data center layer, so as to achieve coordinated optimization of power dispatching and computing power dispatching.

[0108] To address the multi-level coupling problem among transmission networks, distribution networks, and data centers in power-computing collaborative optimization, this invention employs a target cascading analysis method to construct a hierarchical coordination solution framework. Considering the characteristics of data centers participating in power system dispatch, the traditional analytical target cascading (ATC) method is adaptively improved: the diverse flexibility resources of the data center are incorporated as an independent decision-making layer into the optimization framework, establishing the aforementioned three-layer coordination mechanism of "transmission network-distribution network-data center," and designing a hierarchical coordination solution method based on the improved target cascading analysis method. This algorithm achieves ordered coordination optimization among the three systems by constructing an inner and outer iterative structure. Specifically, the algorithm framework has the following characteristics: 1) System architecture design: The system adopts an inner and outer dual-loop mechanism. The outer loop coordinates the coupling constraints between layers, while the inner loop achieves fast convergence within the layer. It supports multiple power distribution networks and data centers to solve their respective sub-problems in parallel, which greatly improves computational efficiency.

[0109] 2) Coupling Mechanism Implementation: A distributed solution approach is adopted, with each layer optimizing its own subproblems in parallel. By introducing a virtual variable mechanism, the transmission and distribution networks are coupled through "virtual generators" and "virtual loads" variables, decomposing the originally tightly coupled three-layer optimization problem into loosely coupled subproblems. Global coordination can be achieved simply by exchanging boundary information.

[0110] 3) Information Interaction Mechanism: Considering that transmission network, distribution network, and data center operators often belong to different stakeholders, information interaction between levels is strictly limited to boundary variables: the transmission network only provides the marginal electricity price of nodes and the upper limit of exchange power; the distribution network transmits electricity price signals; and the data center provides feedback on aggregated electricity demand. Multiple data centers belonging to the same operator can coordinate the allocation of computing power internally and only provide aggregated demand externally, achieving a balance between privacy protection and coordinated optimization.

[0111] Compared to existing methods such as alternating direction method of multipliers (ADMM), heterogeneous decomposition (HGD), optimal condition decomposition (OCD), and proximal message passing (PMP), the objective cascade analysis method supports multiple forms of penalty functions. The penalty function multipliers can be flexibly updated, the parameters are easy to select, global convergence can be guaranteed, and it can directly handle nonlinear multilevel optimization problems with discrete variables. Therefore, it is more suitable for solving the transportation and distribution cooperative optimization model established in this invention.

[0112] In actual solution implementation, the algorithm uses tie-line power as the inner loop convergence criterion and power system operating cost and computing power system electricity cost as the outer loop convergence conditions. It achieves collaborative optimization of power dispatch and computing power dispatch through dynamic updates of the penalty function multiplier. The specific algorithm flow is as follows: Figure 3 As shown: Step 1: Parameter initialization.

[0113] Initialize algorithm parameters and set inter-layer multipliers between transmission and distribution networks. and Virtual load and virtual generator Initialize values ​​and set the number of iterations. . and These represent the new variables after optimization.

[0114] Step 2: Optimization decision-making at the transmission network level.

[0115] Inner loop iteration begins, set The transmission network optimizes unit combination decisions based on initial information to obtain the nodal marginal electricity price. Assign these values ​​to the lower-level distribution network and data center layers as known quantities for parallel optimization of the distribution network layer and optimization of the data center layer.

[0116] Step 3: Parallel optimization of the distribution network layer.

[0117] Parallel optimization of distribution network It is assigned to the upper-level power grid and transmitted to the lower-level data center.

[0118] Step 4: Optimize data center power demand.

[0119] Optimize global load distribution across multiple data centers, based on the current... Signal and outdoor temperature of each data center Minimize data center electricity costs. Optimize load through load migration within and across data centers within the distribution network. and The allocation.

[0120] Step 5: Determine the convergence condition of the inner loop.

[0121] Determine whether the convergence condition (27) is satisfied. If both conditions in (27) are satisfied, proceed to step 6; otherwise, return to step 2 and continue the inner loop.

[0122] (27) Step 6: Determine the convergence condition of the outer loop.

[0123] If both the switching power and cost satisfy the convergence condition of the following formula, the optimization ends and the optimal value is output. Otherwise, proceed to step 7.

[0124] (28) Step 7: Update the penalty function multiplier.

[0125] Place Update the multipliers according to equation (28): (29) Then proceed to step 2 and restart the inner loop. To accelerate convergence, The general range of values ​​for is [2,3]. In this model, we take . Initial value of penalty function multiplier , .

[0126] Furthermore, to verify the effectiveness of the proposed transmission and distribution coordinated scheduling method for data center computing with multi-functional flexibility, this embodiment uses an improved 6-node transmission network plus three 7-node distribution network systems as a test case for analysis and verification. IDC1 and 2 are connected to distribution network 1, IDC3 and 4 are connected to distribution network 2, and IDC5 is connected to distribution network 3. It should be noted that the ambient temperature varies in different regions where data centers are located. Specifically, the ambient temperature in the area where IDC5 is located is relatively high, which affects the energy efficiency ratio and load-bearing capacity of its cooling system; while the ambient temperature in the areas where IDC3 and 4 are located is relatively low, resulting in higher operating efficiency of their cooling systems. The model is solved using MATLAB 2022b software with the Gurobi solver. The computer configuration is Windows 11, with an Intel Core i9-13900H CPU (2.6GHz) and 16GB of memory. The load time-series data of the computing tasks is referenced from the time structure characteristics of the Alibaba cluster operation tracking dataset and proportionally mapped to match the scale of each data center.

[0127] 1) Data center multi-dimensional flexibility analysis.

[0128] To verify the effectiveness of the data center multi-flexibility model proposed in this invention, the following three scenarios were constructed for comparative analysis: Scenario 1: The data center lacks flexibility and has no ability to schedule or migrate latency-sensitive or latency-tolerant loads. All electrical loads are treated as fixed loads, there are no energy storage devices, and the cooling system operates at a base temperature of 22°C to maintain the indoor temperature. Scenario 2: The data center only has the ability to schedule and migrate latency-sensitive and latency-tolerant loads, has no energy storage devices, and the cooling system cools at a reference temperature of 22°C to maintain the indoor temperature. Scenario 3: Data centers not only have the ability to schedule and migrate latency-sensitive and latency-tolerant loads, but also the ability to regulate energy storage devices and the ability to flexibly regulate cooling systems.

[0129] The operating costs for each scenario are shown in Table 1.

[0130] Table 1. Changes in system costs across different scenarios

[0131] As can be seen from the cost data in Table 1, the total system cost shows a step-down trend with the gradual development of data center flexibility resources. From Scenario 1 to Scenario 2, and then to Scenario 3, the cumulative reduction in total system cost reached 9.81%, fully validating the economic value of diverse flexibility resources in data centers. Compared to Scenario 1 and Scenario 2, after introducing computing load scheduling, the electricity cost of data centers decreased significantly, by 30.57%, fully demonstrating the economic value of computing load scheduling to the data center itself. After Scenario 3 further introduced flexible adjustment of the cooling system and energy storage equipment, the data center cost decreased again, by 6.31%. It is evident that the data center diversified flexibility model proposed in this invention, through the collaborative optimization of computing load, cooling system, and energy storage equipment, not only significantly reduces the operating cost of the data center itself but also reduces the system's power generation cost, achieving a win-win situation for both grid operators and data center operators, with all parties gaining substantial benefits from collaborative optimization.

[0132] 1.1) Spatiotemporal flexibility scheduling analysis of computing power load.

[0133] 1.1.1) Analysis of typical data center load scheduling characteristics.

[0134] To verify the effectiveness of latency-sensitive and latency-tolerant load scheduling optimization for data centers, the latency-sensitive and latency-tolerant load scheduling results for IDC1 and IDC3, and the corresponding LMP changes, are as follows: Figure 4 and Figure 5 As shown.

[0135] Depend on Figure 4 As can be seen, from the perspective of delay-sensitive loads, the LMP of the distribution network where IDC1 is located is higher than that of other distribution networks in most periods, resulting in a significant relocation of loads. During the peak electricity price periods of 8:00-10:00 and 19:00-20:00, the load reduction of delay-sensitive loads reached 20%-47%, effectively avoiding high electricity price periods. From the perspective of delay-tolerant loads, IDC1 exhibits stronger spatiotemporal dual optimization characteristics. The load from 1:00 to 11:00 is almost zero because the model not only delays tasks to low electricity price periods but also migrates most tasks to other distribution networks for execution. After 12:00, the load increases significantly, concentrating on processing previously delayed tasks. The number of loads reaches a high of 6787 at 21:00. Although the electricity price is still high during this period, considering the delay constraints and other IDC states, migrating some loads to IDC1 for processing is actually the most economically efficient solution overall, fully demonstrating the model's comprehensive synergistic ability to handle multiple factors.

[0136] Depend on Figure 5It is evident that the overall LMP (Lower Minimum Load) of the No. 2 distribution network where IDC3 is located is relatively low, making it the primary area for load transfer. Looking at delay-sensitive loads, the number of loads surged from 4600 to 15021 at 3:00 AM, an increase of 226.5%, fully leveraging the electricity price advantage to achieve large-scale load migration. For delay-tolerant loads, the number of loads reached 8531 at 8:00 AM. Although the LMP was no longer the lowest throughout the day, considering the cooling efficiency advantage brought by the ambient temperature, IDC3 remained the optimal load transfer point. This spatiotemporal coupled scheduling mode fully demonstrates the spatiotemporal flexibility of delay-tolerant loads, achieving dual optimization in handling "low-price periods + low-price areas".

[0137] It is evident that while delay-sensitive loads cannot be delayed in time, they possess significant spatial portability; delay-tolerant loads, on the other hand, offer both temporal and spatial flexibility.

[0138] 1.1.2) Collaborative scheduling analysis of multi-data center clusters across power distribution networks.

[0139] To verify the collaborative optimization effect of data center clusters across distribution network regions, the scheduling results of latency-sensitive and latency-tolerant loads for IDCs in different distribution networks are as follows: Figure 6 and Figure 7 As shown.

[0140] Depend on Figure 6 As can be seen, there are significant differences in electricity prices among the three distribution networks, with delay-sensitive loads exhibiting a clear tendency to seek lower prices. Although distribution network 3 maintained the lowest electricity price throughout the dispatch cycle, IDC3 and IDC4 in distribution network 2 absorbed more of the migrated load. This is because the ambient temperature in the area where IDC5 is located is high; large-scale load migration there would lead to a decrease in the cooling system's energy efficiency ratio, a significant increase in power consumption, and could easily push the computer room temperature close to the constraint limit, affecting the reliable operation of the equipment. Therefore, the model proposed in this invention comprehensively balances the advantages of electricity prices and cooling energy consumption, prioritizing the migration of most loads to distribution network 2. This achieves economic optimization while ensuring the safe operation of the system, demonstrating the collaborative dispatching effect of multi-dimensional flexible resources.

[0141] Depend on Figure 7It is evident that delay-tolerant loads across multiple data centers exhibit a significant spatiotemporal coupling and collaborative scheduling effect, fully demonstrating the dual optimization characteristics of delay-tolerant loads. The allocation of delay-tolerant loads after 13:00 is the most typical: at 13:00, the total delay-tolerant load of the IDC corresponding to distribution network 1 reached 9098, compared to only 6300 before optimization. This is because a large number of tasks originally planned for high-price periods were delayed until this time, achieving load transfer in the time dimension. Furthermore, the load volume decreased significantly at 20:00 due to the high LMP (Local Performance Management) of all distribution networks at that time, resulting in a large amount of load being postponed to later periods. The delay-tolerant load of the IDCs corresponding to distribution networks 2 and 3 was almost zero from 1:00 to 11:00, while the IDC corresponding to distribution network 1 bore a large load during the same period. This fully demonstrates the spatiotemporal dual optimization characteristics of delay-tolerant loads, not only delaying tasks from high-price periods to low-price periods but also simultaneously performing spatial migration across data centers. It is evident that the collaborative scheduling of cross-distribution network data center clusters fully utilizes the price differences between different distribution networks, achieving both spatial optimization for delay-sensitive loads and spatiotemporal optimization for delay-tolerant loads.

[0142] 1.2) Flexibility analysis of the refrigeration system.

[0143] To verify the effectiveness of the proposed virtual energy storage model based on building thermal inertia in unlocking the regulation potential of the cooling system, IDC1 and IDC3 were selected as typical cases for analysis. The operating conditions of the cooling systems of IDC1 and IDC3 are as follows: Figure 8 and Figure 9 As shown.

[0144] Depend on Figure 8 As can be seen, the temperature of IDC1's server room is strictly controlled within the safe range of 23.67~24.71℃, meeting the temperature constraint requirements. It is worth noting that the server room temperature is not constantly maintained at a baseline value of 22℃, but rather dynamically adjusted within the allowable range. This embodies the virtual energy storage mechanism based on building thermal inertia proposed in this invention. (Comparison) Figure 4 and Figure 8 The cooling system power is highly correlated with load migration and electricity price signals, reflecting the synergistic optimization of multi-dimensional flexibility resources. The periods when cooling power experiences significant lows are related to... Figure 4 The large-scale load relocation period shown perfectly matches the actual situation. This large-scale load relocation directly reduces server heat generation, allowing the cooling system to significantly reduce its operating power. At this time, as the data center temperature rises, the system chooses to release previously stored cooling capacity instead of relying entirely on active cooling, effectively reducing electricity costs during periods of high electricity prices. This verifies the synergistic effect between the virtual energy storage model of the cooling system and the computing load scheduling.

[0145] Depend on Figure 9 As can be seen, the operating curves of the cooling system in IDC3 differ significantly from those in IDC1. Due to the lower electricity prices and generally lower ambient temperatures in the area where IDC3 is located, its load-bearing characteristics are significantly different from those of IDC1, resulting in differentiated operating strategies for its cooling system. The advantage of lower ambient temperatures means that IDC3's cooling power is extremely low or zero during several periods from 2:00 to 11:00, with natural heat dissipation basically meeting the room's temperature requirements. Only after 12:00, when the ambient temperature rises, does the cooling system's power need to be significantly increased to maintain the room's temperature. The cooling system dynamically adjusts its operating strategy based on load migration and electricity price signals using a virtual energy storage model, achieving heat load time transfer within the allowable temperature range and effectively reducing cooling costs.

[0146] 1.3) Energy storage equipment regulation analysis.

[0147] To verify the effectiveness of IDC energy storage devices in transitioning from traditional backup power to proactive resource regulation, this study analyzes the flexible charging and discharging strategies of each IDC energy storage device while ensuring emergency power supply. The charging and discharging power of each IDC energy storage device is as follows: Figure 10 As shown.

[0148] Depend on Figure 10 It is evident that the charging and discharging decisions of energy storage devices comprehensively consider LMP signals and load migration strategies, exhibiting a coordinated operation characteristic of "charging at low prices and discharging at high prices" and "charging when loads migrate out and discharging when loads migrate in." During the low-price period from 1:00 to 6:00, each IDC energy storage device charges; during the high-price period, such as 8:00, the IDC in distribution network 2 discharges at full power to cope with the surge in electricity demand due to the large number of migrated loads; the IDC in distribution network 1 reduces its discharge demand due to load migration, demonstrating the coordinated cooperation between energy storage devices and load migration; from 18:00 to 20:00, the LMP of all three distribution networks is at a high level, and each IDC energy storage device discharges at full power to reduce costs; the charging behavior at 24:00 is to meet the periodic constraints of energy storage. The invention verifies that the optimized operation model of the energy storage device proposed in this invention, under the premise of ensuring emergency power supply, realizes the transformation of the energy storage device from a traditional backup role to an active adjustment resource by tracking electricity price signals and coordinating with load spatiotemporal allocation for flexible charging and discharging, and verifies the synergistic optimization effect of multiple flexible resources.

[0149] 2) Analysis of the synergistic effect of transportation and distribution.

[0150] To verify the effectiveness of the proposed LMP-based three-layer collaborative optimization framework for transmission network-distribution network-data center, the changes in tie-line power and LMP before and after transmission and distribution collaborative optimization were analyzed. The changes in tie-line power between the transmission network and each distribution network, and the LMP before and after optimization, are shown below. Figure 11 and Figure 12 As shown.

[0151] Depend on Figure 11 As shown in (a), the tie-line power of all three distribution networks exhibits significant peak-shaving and valley-filling characteristics. In distribution network No. 1, the peak power during the optimized period from 8:00 to 10:00 decreased to 49.00MW, 51.85MW, and 49.79MW respectively, representing a peak reduction of 30.0%. This is mainly attributed to the load relocation and energy storage discharge of IDC1 and IDC2 during periods of high electricity prices. Figure 11 As shown in (b), the tie-line power change in distribution network 2 is more pronounced, surging from 34.62MW before optimization to 59.75MW at 8:00 AM, an increase of 72.6%. This is because the three-layer collaborative optimization framework proposed in this invention guides distribution network 2 to take on a large amount of load from distribution network 1 during this period through the layer-by-layer transmission of LMP. Although the tie-line power increases, the overall system operating cost is actually reduced because the LMP of distribution network 2 is significantly lower than that of distribution network 1. Figure 11 As shown in (c), the tie-line power of distribution network No. 3 shows an overall downward trend after optimization, especially during the period from 1:00 to 11:00, when the tie-line power is significantly lower than before optimization. This is because although distribution network No. 3 has the lowest electricity price, the three-layer collaborative optimization framework proposed in this invention comprehensively considers the environmental temperature and cooling capacity limitations of IDC5. If IDC5 takes on too much load, it will significantly increase the power of the cooling system, thus increasing costs.

[0152] Depend on Figure 12 As shown in (a) above, the power demand of distribution network No. 1 decreased due to the relocation of IDC loads, especially during the peak period from 8:00 to 10:00, and its LMP was significantly reduced after optimization. Figure 12 As shown in (b), the LMP of distribution network No. 2 exhibits differentiated variation characteristics: during the period of 8:00, the increased power demand due to the large influx of relocated loads leads to an increase in LMP; while from 19:00 to 21:00, power demand is reduced through energy storage discharge and optimized load configuration of data centers, resulting in a corresponding decrease in LMP. Figure 12 As shown in (c), although the electricity price of distribution network No. 3 is the lowest, the high ambient temperature in the area where IDC5 is located limits the load-bearing capacity of the cooling system, resulting in a large number of loads being relocated. Therefore, the overall LMP decreases after optimization. The significant change in LMP before and after optimization reveals the guiding role and feedback mechanism of the electricity price signal in transmission and distribution coordination. This further verifies that the three-layer collaborative optimization framework proposed in this invention guides data center load scheduling through the layer-by-layer transmission of LMP signals, achieving peak shaving and valley filling of tie-line power. The dynamic change of LMP reflects the two-way feedback mechanism between power scheduling and computing power scheduling.

[0153] Example 2 This embodiment discloses a transmission and distribution coordinated scheduling system that takes into account the diverse flexibility of data center computing.

[0154] A coordinated dispatching system for power distribution that takes into account the diverse flexibility of data center computing includes: The multi-dimensional flexibility modeling module is configured to: construct a multi-dimensional flexibility resource quantification model based on the characteristics of computing power tasks at the load flexibility resource level and the physical device level; the multi-dimensional flexibility resource quantification model includes a spatiotemporal flexibility model that takes into account the load's electricity consumption, a refrigeration system model that takes into account the building's thermal inertia, and an energy storage device model. The three-layer collaborative optimization framework construction module is configured to: construct a three-layer collaborative optimization framework of transmission network, distribution network, and data center based on the aforementioned multi-dimensional flexible resource quantification model; the upper transmission network layer aims to minimize the system's power generation cost and is used to decide on the unit operating status and node marginal electricity price; the middle distribution network layer is used to optimize local resource allocation and transmit electricity price signals; and the lower data center layer is used to optimize the spatiotemporal allocation strategy of computing load. The transmission and distribution coordinated scheduling module is configured to use the target cascade analysis method to solve the multi-level coupling problem between the transmission network layer, the distribution network layer and the data center layer, so as to achieve coordinated optimization of power scheduling and computing power scheduling.

[0155] Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0156] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the transmission and distribution coordinated scheduling method considering the multi-functional flexibility of data center computing as described in Embodiment 1 of this disclosure.

[0157] Example 4 The purpose of this embodiment is to provide an electronic device.

[0158] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the data center computing multi-functional flexible transmission and distribution coordinated scheduling method as described in Embodiment 1 of this disclosure.

[0159] The steps and methods involved in the apparatuses of Embodiments 2, 3, and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0160] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0161] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A transmission and distribution coordinated scheduling method that takes into account the diverse flexibility of data center computing, characterized in that, include: Based on the characteristics of computing tasks at both the load flexibility resource level and the physical device level, a multi-dimensional flexibility resource quantification model is constructed. The multi-dimensional flexibility resource quantification model includes a spatiotemporal flexibility model that takes into account load electricity consumption, a refrigeration system model that takes into account building thermal inertia, and an energy storage device model. Based on the aforementioned multi-dimensional flexible resource quantification model, a three-layer collaborative optimization framework of transmission network, distribution network, and data center is constructed: the upper transmission network layer aims to minimize the system's power generation cost and is used to decide on the unit's operating status and the node's marginal electricity price; the middle distribution network layer is used to optimize local resource allocation and transmit electricity price signals; and the lower data center layer is used to optimize the spatiotemporal allocation strategy for computing load. The target cascade analysis method is used to solve the multi-level coupling problem between the transmission network layer, the distribution network layer and the data center layer, so as to achieve the coordinated optimization of power dispatching and computing power dispatching. 2.The method of claim 1, wherein, Based on the characteristics of the computing tasks at the load flexibility resource level, the computing load is divided into latency-sensitive tasks and latency-tolerant tasks; based on the characteristics of the computing tasks at the physical device level, the computing load is divided into cooling systems and energy storage devices. 3.The method of claim 1, wherein, The three-layer collaborative optimization framework of transmission network-distribution network-data center has bidirectional interaction between its various layers. Specifically, in the forward transmission path, the transmission network generates the nodal marginal electricity price by solving network constraints, unit combination and load demand, and then transmits it layer by layer to the distribution network and data center. 4.The method of claim 1, wherein, The construction of the spatiotemporal flexibility model includes the introduction of flexibility constraints, namely: introducing computational load spatial flexibility constraints and computational load spatiotemporal flexibility constraints for latency-sensitive tasks and latency-tolerant tasks, respectively. 5.The method of claim 1, wherein, The construction of the cooling system model includes: establishing an energy efficiency ratio model that characterizes the relationship between cooling power and cooling capacity of the cooling system; constructing a thermal balance differential equation and corresponding discrete solution describing the dynamic changes in indoor temperature of the data center based on the principle of building thermal inertia; and introducing safe operation constraints for indoor temperature. The cooling system is configured to achieve time transfer of heat load by releasing pre-cooling capacity during periods of high electricity price and storing cooling capacity during periods of low electricity price. 6.The method of claim 1, wherein, include: The transmission network layer aims to minimize the system's power generation cost and optimizes decisions regarding the start-up and shutdown of conventional generating units and power output scheduling. The distribution network layer optimizes local resource allocation and load distribution based on the tie-line power transmitted from the transmission network layer and in combination with local distribution network constraints, while transmitting local node electricity price information to the data center layer. The data center layer, from the perspective of the operator, configures computing load and power consumption strategies for each data center.

7. The method of claim 1, wherein, The objective cascade analysis method is used to solve the multi-level coupling problem between the transmission network layer, the distribution network layer and the data center layer. This includes: using tie line power as the inner loop convergence criterion, using the power system operating cost and the computing power system electricity cost as the outer loop convergence condition, and realizing the collaborative optimization of power dispatch and computing power dispatch through the dynamic update of the penalty function multiplier.

8. A power transmission and distribution collaborative scheduling system taking into account the computational multiplicity of flexibility of data centers, characterized in that, include: The multi-dimensional flexibility modeling module is configured to: construct a multi-dimensional flexibility resource quantification model based on the characteristics of computing power tasks at the load flexibility resource level and the physical device level; the multi-dimensional flexibility resource quantification model includes a spatiotemporal flexibility model that takes into account the load's electricity consumption, a refrigeration system model that takes into account the building's thermal inertia, and an energy storage device model. The three-layer collaborative optimization framework construction module is configured to: construct a three-layer collaborative optimization framework of transmission network, distribution network, and data center based on the aforementioned multi-dimensional flexible resource quantification model; the upper transmission network layer aims to minimize the system's power generation cost and is used to decide on the unit operating status and node marginal electricity price; the middle distribution network layer is used to optimize local resource allocation and transmit electricity price signals; and the lower data center layer is used to optimize the spatiotemporal allocation strategy of computing load. The transmission and distribution coordinated scheduling module is configured to use the target cascade analysis method to solve the multi-level coupling problem between the transmission network layer, the distribution network layer and the data center layer, so as to achieve coordinated optimization of power scheduling and computing power scheduling.

9. A computer-readable storage medium having stored thereon a program, characterized in that, When executed by the processor, the program implements the steps of the transmission and distribution coordinated scheduling method that takes into account the multi-faceted flexibility of data center computing as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and capable of running on the processor, characterized by When the processor executes the program, it implements the steps in the data center computer-based transmission and distribution coordinated scheduling method that takes into account the multi-functional flexibility of data centers as described in any one of claims 1-7.

Images