Cooling capacity resource cross-machine-room dynamic adjusting and optimizing method and device and medium

Abstract

The invention provides a cooling capacity resource cross-machine-room dynamic adjusting and optimizing method and device and a medium, and relates to the technical field of energy consumption control. The method comprises the following steps: acquiring a source machine room of which the cooling capacity supply quantity is greater than the cooling capacity demand quantity and a target machine room of which the cooling capacity demand quantity is greater than the cooling capacity supply quantity in a plurality of machine rooms of which the cooling capacity can be mutually dispatched; the cooling capacity surplus amount and the first cooling capacity price of each source machine room, the cooling capacity lacking amount and the second cooling capacity price of each target machine room, and the pipeline heat loss and the cooling capacity liability from each source machine room to each target machine room are obtained; according to the cooling capacity surplus amount, the first cooling capacity price, the cooling capacity lacking amount, the second cooling capacity price, the pipeline heat loss and the cooling capacity liability, the cooling capacity dispatching weight from each source machine room to each target machine room is obtained; and obtaining a cooling capacity scheduling routing table from each source machine room to each target machine room according to the cooling capacity scheduling weight, wherein the cooling capacity scheduling routing table is the cooling capacity scheduling weight for completing all the source machine rooms to the target machine rooms and the maximum cooling capacity scheduling scheme.

Description

Technical Field

[0001] This application relates to the field of energy consumption control technology, and in particular to a method, device and medium for dynamic optimization of cooling resources across computer rooms. Background Technology

[0002] As data centers expand in scale, cooling energy consumption accounts for as much as 45% of total energy consumption. Existing technologies mainly focus on optimizing cooling capacity within a single data center, but lack cross-data center collaboration capabilities, resulting in bottlenecks in global energy efficiency optimization. Summary of the Invention

[0003] To address the aforementioned shortcomings, this application provides a method, apparatus, and medium for dynamic optimization of cooling resources across computer rooms, in order to solve the following technical problem: how to achieve dynamic optimization of cooling resources across computer rooms.

[0004] In a first aspect, this application provides a method for dynamic optimization of cooling resources across data centers, the method comprising:

[0005] Among multiple data centers that can mutually schedule cooling capacity, the source data center has a cooling capacity supply greater than the cooling capacity demand, and the target data center has a cooling capacity demand greater than the cooling capacity supply.

[0006] Obtain the surplus cooling capacity and first cooling capacity price of each source computer room, the shortage cooling capacity and second cooling capacity price of each target computer room, the heat loss of the pipeline from each source computer room to each target computer room, and the cooling capacity liability.

[0007] Based on the surplus cooling capacity, the first cooling capacity price, the cooling capacity shortage, the second cooling capacity price, pipeline heat loss, and cooling capacity liability, obtain the cooling capacity scheduling weights from each source computer room to each target computer room.

[0008] Based on the cooling capacity scheduling weights, obtain the cooling capacity scheduling routing table from each source data center to each target data center. The cooling capacity scheduling routing table contains the cooling capacity scheduling weights and the maximum cooling capacity scheduling scheme to complete all source data centers to target data centers.

[0009] Furthermore, among multiple data centers capable of mutually scheduling cooling capacity, the source data center where the cooling capacity supply exceeds the cooling capacity demand, and the target data center where the cooling capacity demand exceeds the cooling capacity supply, specifically include:

[0010] Connect cooling supply pipelines in multiple computer rooms, use software-defined networking (SDN) to establish a shared cooling resource pool for multiple computer rooms, and collect real-time cooling control data, pipeline length and heat loss rate, electricity price of each computer room, and cooling dispatch volume of each computer room in previous cycles.

[0011] Based on the real-time cooling control data of each computer room, obtain the predicted cooling supply and predicted cooling demand of each computer room during the forecast period;

[0012] Identify n source data centers where the predicted cooling supply is greater than the predicted cooling demand, and m target data centers where the predicted cooling demand is greater than the predicted cooling supply.

[0013] Furthermore, obtain the surplus cooling capacity and first cooling capacity price of each source computer room, the shortage cooling capacity and second cooling capacity price of each target computer room, the pipe heat loss from each source computer room to each target computer room, and the cooling capacity liability, specifically including:

[0014] Based on the predicted cooling supply and the predicted cooling demand, obtain the cooling surplus Tsi for each source computer room (si, i=[1-n]) and the cooling shortage Ttj for each target computer room (tj, j=[1-m]);

[0015] Based on the electricity price of each computer room location, obtain the first cooling price Psi for each source computer room (si, i=[1-n]) during the forecast period, and the second cooling price Ptj for each target computer room (tj, j=[1-m]) during the forecast period;

[0016] Based on the pipe length and pipe loss rate, obtain the pipe heat loss Lsitj for each source computer room (si, i=[1-n]) and each target computer room (tj, j=[1-m]);

[0017] Based on the cooling capacity scheduling volume of each computer room in previous cycles, obtain the cooling capacity liability Ysitj of each source computer room (si, i=[1-n]) and each target computer room (tj, j=[1-m]) within the cooling capacity borrowing and repayment period.

[0018] Furthermore, based on the surplus cooling capacity, the price of the first cooling capacity, the shortage of cooling capacity, the price of the second cooling capacity, pipeline heat loss, and cooling capacity liabilities, the cooling capacity scheduling weights from each source computer room to each target computer room are obtained, specifically including:

[0019] The cooling load scheduling weights from each source server room (si, i=[1-n]) to each target server room (tj, j=[1-m]) are obtained according to the following formula:

[0020] Wsitj = α × [1 / (Tsi – Ttj + ε)] + β × (Psi / Ptj) + γ × (1 / Lsitj)+ δ × Ysitj;

[0021] Where Tsi is the surplus cooling capacity of each source room (si, i=[1-n]), Ttj is the shortage of cooling capacity of each target room (tj, j=[1-m]), ε is the set value to prevent the denominator from being 0, Psi is the first cooling capacity price of each source room (si, i=[1-n]), Ptj is the second cooling capacity price of each target room (tj, j=[1-m]), Lsitj is the heat loss of the pipeline from each source room (si, i=[1-n]) to each target room (tj, j=[1-m]), Ysitj is the cooling capacity liability of each source room (si, i=[1-n]) and each target room (tj, j=[1-m]), and α, β, γ, δ are the first set of preset coefficients.

[0022] Furthermore, based on the cooling load scheduling weights, a cooling load scheduling routing table is obtained from each source data center to each target data center. This routing table represents the cooling load scheduling weights and the maximum cooling load scheduling scheme for all source data centers to target data centers, specifically including:

[0023] Obtain the cooling capacity transfer priority of each source computer room (si, i=[1-n]) or the cooling capacity transfer priority of each target computer room (tj, j=[1-m]), and sort each source computer room (si, i=[1-n]) according to the cooling capacity transfer priority or sort each target computer room (tj, j=[1-m]) according to the cooling capacity transfer priority;

[0024] After each matching of the source or target data center with the highest Wsitj, the source or target data center with the highest Wsitj is deleted, and the corresponding matching target or source data center is deleted. This process is repeated until all source or target data centers are matched, so as to obtain the cooling capacity scheduling scheme with the highest ∑Wsitj from all source data centers to the target data center and form the cooling capacity scheduling routing table.

[0025] Furthermore, the priority of cooling capacity transfer from each source data center (si, i=[1-n]) or the priority of cooling capacity transfer from each target data center (tj, j=[1-m]) is obtained, specifically including:

[0026] The priority of cooling capacity allocation for each source computer room (si, i=[1-n]) is obtained according to the following formula:

[0027] Osi=k1 × Tsi + k2 × (1 / Psi);

[0028] Alternatively, the cooling capacity transfer priority for each target computer room (tj, j=[1-m]) can be obtained according to the following formula:

[0029] Isi=k1' × Tti + k2' × Pti;

[0030] Among them, k1, k2, k1', k2' are the second set of preset coefficients.

[0031] Furthermore, the method also includes:

[0032] According to the cooling capacity scheduling routing table, cooling capacity is scheduled from each source computer room (si, i=[1-n]) to each target computer room (tj, j=[1-m]).

[0033] The scheduling effect after obtaining the scheduling cooling capacity constitutes the historical scheduling effect, which includes the actual temperature control accuracy and energy consumption cost of all source and target computer rooms.

[0034] Optimize the first set of preset coefficients α, β, γ, and δ based on historical scheduling results.

[0035] Furthermore, the method also includes:

[0036] Establish cooling capacity lending and return agreements for multiple data centers. The agreements stipulate the lending and return period and the rules for returning the cooling capacity. If any two data centers have cooling capacity liabilities that exceed the lending and return period, an economic compensation plan between the two data centers shall be formulated according to the cooling capacity return rules.

[0037] Secondly, this application provides a device for dynamic optimization of cooling resources across computer rooms, the device comprising:

[0038] The supply and demand assessment module is used to identify the source data center where the cooling supply exceeds the cooling demand, and the target data center where the cooling demand exceeds the cooling supply, among multiple data centers that can mutually schedule cooling capacity.

[0039] The parameter selection module is connected to the supply and demand assessment module to obtain the surplus cooling capacity and first cooling capacity price of each source computer room, the shortage cooling capacity and second cooling capacity price of each target computer room, the pipeline heat loss and cooling capacity liability from each source computer room to each target computer room.

[0040] The weight calculation module, connected to the parameter selection module, is used to obtain the cooling capacity scheduling weight from each source computer room to each target computer room based on the surplus cooling capacity, the first cooling capacity price, the cooling capacity shortage, the second cooling capacity price, pipeline heat loss, and cooling capacity liability.

[0041] The scheduling and routing module, connected to the weight calculation module, is used to obtain the cooling capacity scheduling routing table from each source data center to each target data center based on the cooling capacity scheduling weight. The cooling capacity scheduling routing table contains the cooling capacity scheduling weights and the maximum cooling capacity scheduling scheme for all source data centers to target data centers.

[0042] Thirdly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for dynamic optimization of cooling resources across data centers as described above.

[0043] This application provides a method, device, and medium for dynamic optimization of cooling resources across data centers. By reasonably selecting parameters between the source data center and the target data center to calculate the cooling scheduling weight, a routing table is formulated to complete the cooling scheduling from all source data centers to the target data center based on the cooling scheduling weight. The selected parameters comprehensively consider four dimensions: cooling supply and demand satisfaction, economy, transmission efficiency, and balance. The resulting cooling scheduling scheme enables multiple data centers that mutually schedule cooling to work together for cooling, thereby achieving global energy efficiency optimization. Attached Figure Description

[0044] Figure 1 This is a flowchart of a method for dynamic optimization of cooling resources across computer rooms according to an embodiment of this application;

[0045] Figure 2 This is a schematic diagram of the structure of a dynamic optimization device for cooling resources across computer rooms according to an embodiment of this application;

[0046] Figure 3 This is an architecture diagram of a cross-data center dynamic optimization system for cooling resources according to an embodiment of this application;

[0047] Figure 4 This is a flowchart of another method for dynamic optimization of cooling resources across computer rooms according to an embodiment of this application;

[0048] Figure 5 This is a schematic diagram of the structure of a computer-readable storage medium according to an embodiment of this application;

[0049] Figure 6 This is a schematic diagram of the structure of a computer device according to an embodiment of this application. Detailed Implementation

[0050] To enable those skilled in the art to better understand the technical solution of this application, the embodiments of this application will be further described in detail below with reference to the accompanying drawings.

[0051] It is understood that the specific embodiments and accompanying drawings described herein are merely for explaining this application and are not intended to limit this application.

[0052] It is understood that, without conflict, the various embodiments and features in the embodiments of this application can be combined with each other.

[0053] It is understood that, for ease of description, only the parts relevant to this application are shown in the accompanying drawings, while parts unrelated to this application are not shown in the drawings.

[0054] It is understood that each module or unit involved in the embodiments of this application may correspond to only one entity structure, or may be composed of multiple entity structures, or multiple modules or units may be integrated into one entity structure.

[0055] It is understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of this application may occur in a different order than that marked in the accompanying drawings.

[0056] It is understood that the flowcharts and block diagrams of this application illustrate the possible architecture, functions, and operations of systems, apparatuses, devices, and methods according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, unit, program segment, or code, containing executable instructions for implementing the specified function. Furthermore, each block or combination of blocks in the block diagrams and flowcharts may be implemented using a hardware-based device to implement the specified function, or using a combination of hardware and computer instructions.

[0057] It is understood that the modules and units involved in the embodiments of this application can be implemented by software or by hardware. For example, the modules and units can be located in the processor.

[0058] Example 1:

[0059] like Figure 1 As shown, this application provides a method for dynamic optimization of cooling resources across data centers, the method comprising:

[0060] S1. Among multiple computer rooms that can mutually schedule cooling capacity, the source computer room whose cooling capacity supply is greater than its cooling capacity demand, and the target computer room whose cooling capacity demand is greater than its cooling capacity supply.

[0061] S2. Obtain the surplus cooling capacity and first cooling capacity price of each source computer room, the shortage cooling capacity and second cooling capacity price of each target computer room, the heat loss of the pipeline from each source computer room to each target computer room, and the cooling capacity liability.

[0062] S3. Based on the surplus cooling capacity, the first cooling capacity price, the cooling capacity shortage, the second cooling capacity price, pipeline heat loss, and cooling capacity liability, obtain the cooling capacity scheduling weights from each source computer room to each target computer room.

[0063] S4. Obtain the cooling capacity scheduling routing table from each source computer room to each target computer room based on the cooling capacity scheduling weight. The cooling capacity scheduling routing table contains the cooling capacity scheduling weights and the maximum cooling capacity scheduling scheme to complete all source computer rooms to target computer rooms.

[0064] In this embodiment, the provided method calculates cooling capacity scheduling weights by reasonably selecting parameters between the source and target data centers. Based on these weights, a routing table is created to schedule cooling capacity from all source data centers to the target data center. The selected parameters comprehensively consider four dimensions: cooling capacity supply and demand satisfaction, economy, transmission efficiency, and balance. This results in a cooling capacity scheduling scheme that enables multiple data centers that mutually schedule cooling capacity to work together for coordinated cooling, achieving global energy efficiency optimization. Figure 1 The method shown is applicable to, for example, Figure 2 The apparatus shown.

[0065] Specifically, the prior art most similar to this embodiment includes: a single-room cooling capacity dynamic scheduling system, which monitors the rack environment parameters in real time through a sensor network and optimizes local cooling efficiency by combining a dynamic load balancing algorithm and an adaptive cooling strategy; a computational cooling co-optimization method, which jointly models the IT load and cooling system energy consumption and uses a physical information neural network to predict the optimal operating temperature; and the use of natural cold sources, which involves modifying the refrigeration room and using cooling towers as a natural cold source to replace the refrigeration unit in winter.

[0066] The main drawbacks of the existing technologies mentioned above include: lack of cross-data center collaboration capabilities, with existing technologies focusing on cooling optimization within a single data center (such as dynamic load balancing algorithms and adaptive cooling strategies), but unable to achieve cross-data center cooling resource sharing and load migration; insufficient dynamic response accuracy, with existing control strategies struggling to adapt to sudden load changes and complex environmental fluctuations; and limited failover mechanisms, with existing redundancy designs limited to local backups and lacking cross-data center emergency scheduling capabilities.

[0067] To address the aforementioned shortcomings, this embodiment provides a cross-data center dynamic optimization system and method based on cooling resource pooling, aiming to solve the problems of insufficient cross-data center cooling capacity coordination, low dynamic response accuracy, and limited failover mechanisms in existing technologies. Specifically, it includes:

[0068] 1. Enable pooled scheduling of cooling resources across computer rooms, construct a virtualized shared pool of cooling sources (chillers, cold storage tanks, liquid cooling backplanes, etc.) across computer rooms, and break the physical boundaries of a single computer room through dynamic routing protocols to solve the problem of limited global energy efficiency optimization caused by regional cooling supply and demand imbalance.

[0069] 2. Improve the real-time matching accuracy of dynamic load and cooling supply by designing a multi-dimensional weight allocation model (integrating thermal urgency, economic weight, transmission loss factor, etc.) and adjusting the cooling allocation strategy in real time through an adaptive algorithm to eliminate local overheating or cooling redundancy caused by sudden load changes or environmental fluctuations.

[0070] 3. Establish a global fault emergency dispatch mechanism, develop a cross-data center cooling capacity borrowing and returning protocol and priority response mechanism, and realize second-level cross-domain dispatch of cooling resources in the event of a single data center cooling failure, so as to ensure the continuity of temperature control in high-priority data centers and avoid the risk of system downtime due to insufficient local redundancy.

[0071] In one embodiment, S1, among multiple data centers capable of mutually scheduling cooling capacity, the source data center whose cooling capacity supply is greater than its cooling capacity demand, and the target data center whose cooling capacity demand is greater than its cooling capacity supply, specifically includes:

[0072] Connect cooling supply pipelines in multiple computer rooms, use software-defined networking (SDN) to establish a shared cooling resource pool for multiple computer rooms, and collect real-time cooling control data, pipeline length and heat loss rate, electricity price of each computer room, and cooling dispatch volume of each computer room in previous cycles.

[0073] Based on the real-time cooling control data of each computer room, obtain the predicted cooling supply and predicted cooling demand of each computer room during the forecast period;

[0074] Identify n source data centers where the predicted cooling supply is greater than the predicted cooling demand, and m target data centers where the predicted cooling demand is greater than the predicted cooling supply.

[0075] In this embodiment, the physical implementation guarantee of the system includes: the physical implementation of cooling capacity scheduling, the core of which is a closed circulation pipeline system, which mainly consists of insulated pipelines, circulation pump sets, temperature and pressure sensors, and control valves.

[0076] Piping Materials and Insulation: Pipelines typically use seamless steel pipes or high-density polyethylene (HDPE) pipes, with an outer layer of polyurethane foam insulation. The insulation layer thickness needs to be calculated based on the design temperature difference, usually 40-100mm, to ensure its thermal conductivity is below 0.035 W / (m·K). Pipe connections are made using flanges or welding, with additional insulation treatment to prevent the formation of cold bridges.

[0077] Power System: The system is equipped with a variable frequency centrifugal water pump as the power source for transportation. The pump's head and flow rate need to be precisely selected based on the resistance loss of the most unfavorable loop in the system and the required cooling capacity to ensure sufficient power to overcome the frictional resistance of pipelines for long-distance transportation.

[0078] Real-time monitoring and closed-loop control: High-precision temperature sensors (accuracy ±0.1℃) and pressure sensors are installed at key nodes in the pipeline (such as outlets, inlets, and intermediate points every 500-1000 meters). Data is transmitted to the central control system in real time. The system sets a target temperature range, and once the temperature at a certain point deviates from the set value by more than ±0.5℃, it will automatically adjust the operating frequency of the variable frequency pump or adjust the opening of the corresponding valve, forming a dynamic temperature control closed loop.

[0079] Cross-domain distance and temperature rise control:

[0080] Typical distance: Under good engineering design and insulation conditions, economical and reliable cooling capacity delivery distances can typically reach 5 to 20 kilometers. For example, in district cooling systems, there are successful cases of operation exceeding 15 kilometers.

[0081] Measures to address temperature rise over long distances:

[0082] 1. Reduce the temperature difference during delivery: In system design, avoid excessively low delivery temperatures. Appropriately increasing the supply and return water temperature difference (e.g., using 6℃ supply water and 12℃ return water instead of the traditional 5℃ / 13℃) can reduce the flow rate while delivering the same amount of cooling capacity, thereby reducing pump consumption and pipeline temperature rise.

[0083] 2. Intermediate pressurization and cooling stations: For ultra-long distances (e.g., exceeding 15 kilometers), intermediate pressurization pump stations and auxiliary cooling stations can be set up at pre-designed locations along the route. Auxiliary cooling stations typically employ small, high-efficiency refrigeration units to perform "secondary cooling" of the circulating refrigerant to compensate for the cooling loss in the pipeline.

[0084] 3. Optimize hydraulic balance: Through hydraulic balancing valves and precise pipeline design, ensure uniform flow distribution in each branch and avoid excessively low local flow velocity leading to additional temperature rise.

[0085] In summary: Physically, it is protected by efficient insulation, reliable pumping, and real-time monitoring; the cross-domain distance can reach tens of kilometers; to cope with long-distance temperature rise, it mainly relies on system optimization design (such as temperature difference), the addition of intermediate cooling stations, and precise hydraulic balancing.

[0086] The cooling capacity scheduling system is not a single-function system; its core role is multi-dimensional and systematic, mainly reflected in the following three aspects:

[0087] 1. Fault Recovery and Reliability Enhancement: When a data center (cooling station) experiences a sudden cooling system failure (such as compressor shutdown or cooling tower malfunction), the system can quickly allocate cooling capacity from other normally operating sites to ensure continuous cooling of IT equipment and prevent server downtime or data loss due to overheating. This reduces potential business interruption time from tens of hours to minutes, meeting the stringent SLA (Service Level Agreement) requirements of high-level data centers and ensuring business continuity at the cooling level.

[0088] 2. Energy Efficiency Optimization and Cost Savings: Based on the principle of marginal efficiency, cooling capacity is diverted from sites operating inefficiently (e.g., low COP at partial load) or in unfavorable natural environments (e.g., high outdoor temperatures) to high-efficiency sites. Simultaneously, by leveraging time-of-use electricity pricing in different regions, data centers are encouraged to utilize more diverted cooling capacity during peak electricity price periods, reducing the start-up and shutdown of local high-energy-consuming cooling equipment. This significantly reduces the overall PUE (Power Usage Effectiveness) of the entire data center cluster, resulting in an average annual energy saving of 15%-30% in cooling system electricity costs.

[0089] 3. Load Balancing and Equipment Life Extension: Preventing a single data center's cooling system from operating at or near full load for extended periods. Through scheduling, the load rate of cooling equipment across sites is balanced, maintaining it within an efficient range (e.g., 40%-80%). This reduces the frequency of start-ups and shutdowns of cooling units and minimizes fatigue wear, significantly extending the lifespan of critical equipment such as compressors and cooling towers (by more than 15%), while simultaneously reducing maintenance frequency and costs.

[0090] System architecture and composition as follows Figure 3 As shown, this system adopts a layered architecture design, mainly including three layers: Edge acquisition layer: responsible for real-time acquisition of cooling-related physical parameters in each computer room; Intelligent analysis layer: analyzes, predicts, and makes decisions on the uploaded data to realize the virtualization of cooling resources, demand prediction, and scheduling decisions; Global scheduling layer: receives decision instructions from the intelligent analysis layer and drives physical devices to perform cross-computer room cooling scheduling and fault transfer.

[0091] The edge acquisition layer includes the following key components:

[0092] Sensor Network: Deployed at key nodes in the air inlets, hot aisles, and refrigerant (e.g., chilled water) supply / return pipelines of each server rack in the computer room. Specifically, this includes: Temperature sensors: Real-time monitoring of ambient and refrigerant temperature with an accuracy better than ±0.1℃ (e.g., using PT100 platinum resistance thermometers or equivalent sensors); Humidity sensors: Real-time monitoring of ambient humidity with an accuracy better than ±2%; Flow meters: Installed on the supply / return pipelines of chiller units and key branches to monitor the accumulated refrigerant flow rate in real time (range, e.g., 0-100 m³ / h). 3 / h, accuracy for example ±0.5%); Pressure transmitter: installed at key pipeline nodes to monitor pipeline differential pressure in real time (range for example 0-1MPa, accuracy for example ±0.25%).

[0093] Edge computing nodes: Deployed in various data centers, they receive and preprocess raw data from sensor networks. They have built-in data filtering algorithms (such as moving average filtering and Kalman filtering) to remove sensor noise and outliers. The processed data is aggregated into data packets with a fixed time granularity (e.g., 5 seconds) and uploaded to the intelligent analysis layer through a reliable communication network (such as 5G private network or industrial Ethernet).

[0094] In one embodiment, S2, obtaining the surplus cooling capacity and first cooling capacity price of each source computer room, the shortage cooling capacity and second cooling capacity price of each target computer room, the heat loss of the pipeline from each source computer room to each target computer room, and the cooling capacity liability, specifically including:

[0095] Based on the predicted cooling supply and the predicted cooling demand, obtain the cooling surplus Tsi for each source computer room (si, i=[1-n]) and the cooling shortage Ttj for each target computer room (tj, j=[1-m]);

[0096] Based on the electricity price of each computer room location, obtain the first cooling price Psi for each source computer room (si, i=[1-n]) during the forecast period, and the second cooling price Ptj for each target computer room (tj, j=[1-m]) during the forecast period;

[0097] Based on the pipe length and pipe loss rate, obtain the pipe heat loss Lsitj for each source computer room (si, i=[1-n]) and each target computer room (tj, j=[1-m]);

[0098] Based on the cooling capacity scheduling volume of each computer room in previous cycles, obtain the cooling capacity liability Ysitj of each source computer room (si, i=[1-n]) and each target computer room (tj, j=[1-m]) within the cooling capacity borrowing and repayment period.

[0099] In this embodiment, the key modules of the intelligent analysis layer include:

[0100] The cooling resource pooling management module abstracts and virtualizes physical cooling resources (including but not limited to centrifugal / screw chillers, liquid-cooled backplane heat dissipation systems, cold storage tanks, dry coolers, and natural cold source utilization facilities) scattered across different computer rooms, constructing a global cooling resource pool for unified management and scheduling, and maintaining a dynamically updated resource status matrix. This matrix records the real-time status information of each virtualized resource, for example:

[0101] Centrifugal chiller unit (Resource_A1) in computer room A: Current available cooling capacity (e.g., 650 kW, rated 800 kW), health status, and computer room it belongs to.

[0102] Liquid cooling backplane system of computer room B (Resource_B2): Current available heat dissipation power (e.g., total capacity 500kW, 300kW used, 200kW available), and the computer room it belongs to.

[0103] The cold storage tank (Resource_C3) in computer room C: the current available cooling capacity (e.g., 150 kW·h), the charging / discharging rate, and the computer room it belongs to.

[0104] Dry cooler in computer room D (Resource_D4): Currently available natural cooling source power (e.g., 300 kW, affected by ambient temperature and humidity), and the computer room it belongs to.

[0105] At the same time, key parameters of cooling capacity transfer between the machine rooms are recorded, such as the transfer path, pipe heat loss rate (e.g., the pipe heat loss rate from machine room A to machine room B = 3.2 kW / km), and maximum transfer capacity.

[0106] In one implementation, S3, based on the surplus cooling capacity, the first cooling capacity price, the cooling capacity shortage, the second cooling capacity price, pipeline heat loss, and cooling capacity liability, obtain the cooling capacity scheduling weights from each source computer room to each target computer room, specifically including:

[0107] The cooling load scheduling weights from each source server room (si, i=[1-n]) to each target server room (tj, j=[1-m]) are obtained according to the following formula:

[0108] Wsitj = α × [1 / (Tsi – Ttj + ε)] + β × (Psi / Ptj) + γ × (1 / Lsitj)+ δ × Ysitj;

[0109] Where Tsi is the surplus cooling capacity of each source room (si, i=[1-n]), Ttj is the shortage of cooling capacity of each target room (tj, j=[1-m]), ε is the set value to prevent the denominator from being 0 (a minimum value set based on experience, such as 1), Psi is the first cooling capacity price of each source room (si, i=[1-n]), Ptj is the second cooling capacity price of each target room (tj, j=[1-m]), Lsitj is the heat loss of the pipeline from each source room (si, i=[1-n]) to each target room (tj, j=[1-m]), Ysitj is the cooling capacity liability of each source room (si, i=[1-n]) and each target room (tj, j=[1-m]), and α, β, γ, δ are the first set of preset coefficients.

[0110] In this embodiment, the key modules of the intelligent analysis layer include:

[0111] Spatiotemporal load prediction module: Uses machine learning models (specifically, a long short-term memory network model that incorporates attention mechanisms, i.e., the LSTM-Attention model) to predict cooling demand.

[0112] Model inputs include historical cooling load data (e.g., high-frequency data from the past 72 hours), data center IT equipment load forecasts, external environmental parameters (e.g., outdoor dry and wet bulb temperatures, weather forecasts), and future electricity price curves (obtained via the power grid API).

[0113] Model output: Predicted cooling demand for each data center over a future period of time (e.g., the next 30 minutes).

[0114] Dynamic weight allocation engine: Based on a multi-dimensional weight allocation model, the cooling capacity scheduling weight Wsitj of each pair of source data centers to the target data center is calculated in real time. The weight calculation formula comprehensively considers cooling demand, cooling price, cooling transmission loss and the original cooling borrowing and returning relationship between data centers.

[0115] Cooling requirements can be obtained based on the temperature control threshold of computer rooms i and j and the current measured values ​​of key temperatures. For example, the temperature control threshold is set to 25°C for core financial business computer rooms and 28°C for ordinary office computer rooms. This value is preset according to the type of equipment and business importance in the computer room. The measured values ​​of key temperatures are such as the temperature of hotspot cabinets or the temperature of key return air.

[0116] The price of cooling capacity can be obtained from the real-time electricity price of the power system, and a reference low electricity price threshold can be set (such as off-peak electricity price), or the expected value of low electricity price can be dynamically calculated.

[0117] The cold load transmission loss is the estimated comprehensive loss factor for the cold load scheduling of computer room i and j (mainly considering factors such as transmission distance and pipe insulation efficiency; the larger this value is, the greater the scheduling loss).

[0118] The engine periodically (e.g., every minute) calculates the W of all data centers and determines the source data center (providing cooling), target data center (receiving cooling), and their scheduling priorities for cooling capacity scheduling.

[0119] The direction of cooling flow is dynamic and determined by the real-time calculation of cooling supply and demand by the system. The determination is based on the difference between cooling supply and demand, and the logic is based on the calculation and comparison of the following two core parameters:

[0120] Cooling supply: refers to the total cooling capacity that a data center can currently provide. It is determined by the cooling capacity of the operating cooling equipment (chillers, natural cooling systems, etc.).

[0121] Cooling demand (Demand): refers to the total heat load currently generated by IT equipment and other facilities within a data center.

[0122] Judgment rules:

[0123] Cooling capacity allocation criteria (source data center): When the cooling capacity supply of a data center exceeds the cooling capacity demand, "redundant cooling capacity" is generated. This data center is marked as a potential source data center. The scheduling system will decide whether and how much cooling capacity to allocate based on factors such as the size of the redundant cooling capacity and external demand.

[0124] Cooling demand conditions (target data center): When the cooling demand of a data center exceeds the cooling supply, a "cooling gap" is generated. The data center is marked as the target data center (Destination). The system will assess the size and urgency of the gap and initiate a cooling demand request.

[0125] Dynamic decision-making process: The scheduling system (central controller) collects real-time supply and demand data from all data centers at fixed intervals (e.g., every 5 minutes). It acts like a "cooling exchange," continuously matching source data centers with "surplus" capacity with target data centers with "shortage," and dynamically generating scheduling instructions based on preset optimization goals (e.g., minimum total energy consumption, minimum cost) to determine the direction and volume of cooling flow. Therefore, a data center that might be a supplier at one moment could become a supplier at the next.

[0126] Forecast values ​​(such as day-ahead cold load forecasts based on machine learning) play the role of "strategic commander" in the scheduling system. Although they do not directly participate in the fine-tuning formulas of certain real-time control loops, they provide foresight and optimization guidance for the entire scheduling strategy. Their function is as follows:

[0127] 1. Setting Scheduling Baselines: Forecast values ​​are used to formulate basic scheduling plans for a future period (e.g., the next 24 hours). For example, if the system predicts that data center A will experience a surge in cooling demand at 2 PM tomorrow due to peak business hours, while data center B will have ample cooling capacity due to cooler local weather, the system will plan ahead to start allocating cooling capacity from B to A at 1:30 PM.

[0128] 2. Adjusting Real-Time Control Parameters: In the real-time scheduling formula, the key parameter, target flow (Q_target), is often a weighted fusion of predicted and real-time values. An adjustable weight (e.g., 0.7) can be set to indicate the degree of trust in the real-time data. When the prediction model is highly accurate, this weight can be reduced, allowing the system to favor smoother, prediction-based scheduling and avoid overreacting to real-time noise.

[0129] 3. Economic optimization: Forecast values ​​are crucial for utilizing time-of-use pricing. The system can predict the future electricity price curves of each data center, allowing lower-cost sites to generate more cooling during off-peak periods and then using the reserves to meet demand during peak periods, thus achieving energy arbitrage across time and space.

[0130] Conclusion: Predicted values ​​function by influencing high-level scheduling strategies and critical setpoints, and are key to transitioning from "passive response" to "proactive optimization." If predictions are completely inaccurate, the system can still operate based on real-time data, but it will lose its optimization capabilities and degenerate into a basic assurance mode.

[0131] Electricity price is an important spatial and temporal variable in cooling load dispatching. The differences lie in:

[0132] Regional differences: Industrial electricity pricing policies vary across provinces and cities, influenced by factors such as power generation costs, grid structure, and local government subsidies. For example, City A relies on hydropower with an average price of 0.6 yuan / kWh; City B relies on imported electricity with an average price of 0.9 yuan / kWh. While meeting latency requirements, the system will prioritize scheduling computing loads to data centers in areas with lower electricity prices, and accordingly, will relocate cooling capacity from those areas or reduce their cooling demand, thereby lowering total energy costs.

[0133] Time-of-use (TOU) pricing: This is the most common price difference. To balance grid load, power companies divide the day into peak, off-peak, and valley periods, resulting in significant price variations. For example, peak hours (2:00 PM - 5:00 PM) can cost up to 1.2 yuan / kWh, while off-peak hours (12:00 AM - 8:00 AM) can be as low as 0.3 yuan / kWh. During off-peak hours, the dispatch system encourages stations to fully utilize local cooling and cold storage; during peak hours, it minimizes the operation of high-energy-consuming local cooling equipment, relying more on cooling stored from low-price periods or from off-peak stations in other areas where prices are still in off-peak conditions.

[0134] Contractual Differences: Large data centers may enter into special Power Purchase Agreements (PPAs) with the power grid, such as directly purchasing green electricity. Their electricity pricing models may differ from standard catalog prices. Data centers with contracts for inexpensive green electricity have lower marginal costs for cooling, making them more competitive in dispatch systems and more likely to become the primary source of cooling capacity.

[0135] Therefore, the electricity price factor in the formula is a different and dynamically changing value for each data center, and it is one of the core inputs for the scheduling algorithm to optimize costs.

[0136] In one embodiment, S4, the cooling capacity scheduling routing table from each source data center to each target data center is obtained according to the cooling capacity scheduling weight. The cooling capacity scheduling routing table is the cooling capacity scheduling weight and the maximum cooling capacity scheduling scheme that completes the cooling capacity scheduling from all source data centers to the target data centers, specifically including:

[0137] Obtain the cooling capacity transfer priority of each source computer room (si, i=[1-n]) or the cooling capacity transfer priority of each target computer room (tj, j=[1-m]), and sort each source computer room (si, i=[1-n]) according to the cooling capacity transfer priority or sort each target computer room (tj, j=[1-m]) according to the cooling capacity transfer priority;

[0138] After each matching of the source or target data center with the highest Wsitj, the source or target data center with the highest Wsitj is deleted, and the corresponding matching target or source data center is deleted. This process is repeated until all source or target data centers are matched, so as to obtain the cooling capacity scheduling scheme with the highest ∑Wsitj from all source data centers to the target data center and form the cooling capacity scheduling routing table.

[0139] In this embodiment, the scheduling priority value of a single data center is the "bid" or "request" submitted by that data center to the central controller. Detailed explanation of the decision-making mechanism:

[0140] 1. Local calculation: Each data center (or its local controller) calculates its own "priority score" as a source or target data center based on its own temperature, load, electricity price and other parameters.

[0141] 2. Global summary: The priority scores of all data centers are reported to the central scheduler (global optimization controller).

[0142] 3. Path Optimization: The central dispatcher has a complete understanding of the network topology of the entire system, including the physical connections between data centers, pipe distances, pipe diameters, valve status, etc. It acts like a "traffic controller" and performs the following operations: Matching: Pairing high-priority _source data centers with high-priority _target data centers; Routing: Calculating the optimal delivery path for each matched "source-target" pair, taking into account pipe losses, congestion on the current path, and transit costs; Flow Allocation: Determining the amount of cooling capacity delivered along the path.

[0143] 4. Command Issuance: Finally, the central dispatcher generates specific control commands (such as: opening valves V1, V2, and V3 on the path from A to B, and setting the pump speed to X), and issues them to the relevant pumps, valves, and actuators.

[0144] Conclusion: Scheduling decision is a centralized optimization problem. The weight of a single data center is the input, while "where to schedule to where and how many to schedule" is the output, which is determined by the central controller based on global information.

[0145] In one embodiment, obtaining the cooling capacity transfer priority of each source computer room (si, i=[1-n]) or the cooling capacity transfer priority of each target computer room (tj, j=[1-m]) specifically includes:

[0146] The priority of cooling capacity allocation for each source computer room (si, i=[1-n]) is obtained according to the following formula:

[0147] Osi=k1 × Tsi + k2 × (1 / Psi);

[0148] Alternatively, the cooling capacity transfer priority for each target computer room (tj, j=[1-m]) can be obtained according to the following formula:

[0149] Isi=k1' × Tti + k2' × Pti;

[0150] Among them, k1, k2, k1', k2' are the second set of preset coefficients.

[0151] In this embodiment, the priority calculation formulas for the source and target data centers should logically be opposite or different to accurately reflect their respective status requirements. The formulas are designed as follows: Source data center (sender) priority_source = k1 × (T_actual - T_target) + k2 × (1 / P_price) + ..., the priority is positively correlated with redundant cooling capacity (supply-demand) and negatively correlated with its own electricity price (the higher the electricity price, the less willing it is to use local cooling for external supply); Target data center (receiver) priority_target = k1' × (T_actual - T_target) + k2' × P_price + ..., the priority is positively correlated with cooling capacity gap (demand-supply) and its own electricity price. Among them, k1, k2, k1', and k2' are weighting coefficients used to balance the importance of different factors.

[0152] Local logic of the formula: In the source data center's priority _source ∝ (T_target / T_actual): When T_actual (current temperature of the data center) is much lower than T_target (set target temperature), it means that the data center is very cold inside, its cooling capacity is much higher than the current heat load, and it has a large amount of redundant cooling capacity that can be dispatched. Therefore, this value is high, which improves the data center's qualification and priority as a "supplier".

[0153] The system's global logic (bidirectional matching) states that a high priority for a single source data center is meaningless. The system also calculates the inbound priority of the target data center, using the opposite formula: priority_target ∝ (T_actual / T_target). When the target data center's T_actual approaches or exceeds T_target, it indicates internal overheating and an urgent need for cooling, thus requiring a high inbound priority.

[0154] Final Decision: Scheduling decisions are a two-way matching process. The system searches for data centers with high priority_source and high priority_target, and pairs them together. A high priority data center only indicates that it is eligible to participate in scheduling, but whether the scheduling is ultimately executed depends on whether there is a suitable "partner". Therefore, the system is logically complete and supports bidirectional flow of cooling capacity.

[0155] In one embodiment, the method further includes:

[0156] According to the cooling capacity scheduling routing table, cooling capacity is scheduled from each source computer room (si, i=[1-n]) to each target computer room (tj, j=[1-m]).

[0157] The scheduling effect after obtaining the scheduling cooling capacity constitutes the historical scheduling effect, which includes the actual temperature control accuracy and energy consumption cost of all source and target computer rooms.

[0158] Optimize the first set of preset coefficients α, β, γ, and δ based on historical scheduling results.

[0159] In this embodiment, the key components of the global scheduling layer include:

[0160] Refrigerant routing controller: Generates specific cooling scheduling instructions based on the decision results of the intelligent analysis layer (especially the cooling resource pooling management module and the dynamic weight allocation engine).

[0161] This command uses Software-Defined Networking (SDN) technology to control electrically operated regulating valves deployed on refrigerant pipelines (such as chilled water pipelines) interconnected across data centers. In this system, "Software-Defined Networking (SDN)" does not refer to SDN in pure communication networks, but rather borrows its concepts to construct a "software-defined refrigerant delivery network." Its core is the separation of the refrigerant delivery control logic from the physical execution devices, defined as follows:

[0162] 1. Separation of control plane and data plane:

[0163] Control plane: This is handled by a central SDN controller (software platform). It does not directly drive the valves, but is responsible for formulating the "traffic rules" and scheduling strategies for the entire network.

[0164] The data plane consists of all the physical devices in the network, such as smart electric valves, variable frequency pumps, and sensors. These devices receive and execute instructions from the control plane.

[0165] 2. Programmable Flow Table:

[0166] This is the core of SDN. In a cooling network, "flow" doesn't refer to data packets, but rather the flow rate of the refrigerant. The controller, through software programming, issues "flow table" entries to each valve.

[0167] Example: The controller can be programmed to define a rule: "When the return air temperature of machine room A is >26℃, automatically create a transport path from cold source station B to machine room A, and the valves V1, V2, and V3 on the path will be opened to 50% in sequence, and the auxiliary pump P1 will be started."

[0168] 3. Open Northbound API:

[0169] The controller provides a standardized programming interface (API), allowing upper-level energy management systems or operations and maintenance personnel to write custom scheduling strategies and applications. For example, a "cost optimization app" can be developed and installed on the controller. This app will read real-time electricity prices and automatically generate the most economical scheduling strategy.

[0170] In summary, SDN defines the intelligent behavior of networks, making cold-air networks flexible, programmable, and easy to manage, capable of responding quickly to changing needs and failures, rather than being a rigid hardware system.

[0171] Valve type: Electric control valves or electric butterfly valves are commonly used in the system. They have a mechanical valve body, valve core, valve seat, and an electric actuator (motor).

[0172] Control method: These real physical valves receive electrical signals (such as 4-20mA analog signals or Modbus TCP / IP digital commands) from the central SDN controller through field controllers (such as PLC or DDC), thereby driving the motor to rotate and changing the valve opening (from fully closed to fully open, or any intermediate position), thus achieving precise control of refrigerant flow.

[0173] The valve response time is required to be ≤1 second in order to achieve rapid and precise control of the cooling flow.

[0174] The controller sets the valve opening based on the optimal transmission path (calculated by the cooling resource pooling management module using graph theory algorithms such as Dijkstra's algorithm, with the optimization objective being to minimize transmission loss or time).

[0175] Failover actuator: Real-time monitoring of the cooling system status of each data center (receiving fault alarm signals from the edge acquisition layer, such as compressor current over-limit and water pump failure). Once a data center (denoted as faulty data center X) is detected to have severely insufficient or failed cooling capacity, the cross-data center fault emergency dispatch mechanism is immediately triggered.

[0176] Cooling load scheduling is not limited to fault scenarios; it is a routine, multi-scenario optimization tool. Predicted values ​​play a crucial role in non-fault scenarios. Scheduling trigger scenarios include:

[0177] 1. Emergency Dispatch for Faults (Passive): Such as when the refrigeration unit malfunctions, this is the most direct triggering condition.

[0178] 2. Preventive Scheduling (Proactive, Relying on Forecasting): Based on load forecasting: If the system predicts that a data center will experience a surge in load due to a major task going online in 3 hours, it will schedule cooling capacity from other sites in advance; Based on weather forecasting: If extreme high temperatures are predicted for the next day, the system will increase cooling and cold storage during the off-peak electricity price period at night and activate the reserve cooling capacity during the high temperature period during the day.

[0179] 3. Economic Dispatch (Proactive, Forecast-dependent): Primarily based on electricity price forecasts. Cooling distribution is adjusted in advance before electricity prices surge to maximize the use of low-cost energy.

[0180] α, β, γ, and δ represent the cold / heat urgency coefficient, economic weighting coefficient, transmission efficiency coefficient, and data center balance coefficient, respectively. These coefficients are not fixed but are dynamically adjusted and optimized based on historical scheduling results (such as actual temperature control accuracy and energy consumption costs) through reinforcement learning algorithms (such as Q-learning and Deep Q-Network).

[0181] In one embodiment, the method further includes:

[0182] Establish cooling capacity lending and return agreements for multiple data centers. The agreements stipulate the lending and return period and the rules for returning the cooling capacity. If any two data centers have cooling capacity liabilities that exceed the lending and return period, an economic compensation plan between the two data centers shall be formulated according to the cooling capacity return rules.

[0183] In this embodiment, the cold energy borrowing and returning protocol is automatically invoked based on the preset business priority strategy and the status of the cold energy resource pool:

[0184] Initiate a "cooling capacity borrowing" request to the data center with sufficient available cooling capacity in the resource pool (denoted as support data center Y).

[0185] The agreement includes: borrowed cooling capacity (e.g., 150kW), borrowing period, and return rules (e.g., double the cooling capacity to be returned within 24 hours after the fault is resolved, or payment of the corresponding cost).

[0186] "Repayment" occurs after the cooling system of the faulty computer room has been repaired and returned to normal operation. There are two ways to repay:

[0187] 1. Physical repayment of cooling capacity: This is the most direct method. After the faulty data center recovers, it can choose a time when its IT load is low (such as at night) to send the excess cooling capacity back to the data center that previously assisted it through the existing pipeline network. The total amount of cooling capacity repaid will be equal to (or as agreed in the agreement, such as 1.5 times) the cooling capacity previously borrowed.

[0188] 2. Financial Settlement and Repayment: If the system design includes internal cost accounting, virtual or real financial settlement can also be carried out. That is, the business unit belonging to the faulty data center pays the business unit of the assisting data center a fee equivalent to the cost of the borrowed cooling energy. The fee can be calculated based on the average energy cost at the time.

[0189] The core principle is that the "repayment" mechanism is the credit foundation for ensuring the sustainable mutual assistance of the system, and its premise is that the faulty data center restores its production capacity.

[0190] The refrigerant routing controller synchronously executes valve operations to quickly dispatch the cooling capacity from the supporting computer room Y to the faulty computer room X.

[0191] Once the faulty computer room X regains its cooling capacity, the actuator automatically initiates the return process, delivering the agreed-upon cooling capacity back to the supporting computer room Y according to the agreement (for example, after computer room X starts its standby unit, it delivers 300kW of cooling capacity to Y).

[0192] like Figure 4 As shown, the core workflow of this embodiment includes three main stages: cooling resource pooling, dynamic weight allocation and scheduling, and cross-domain fault transfer.

[0193] Cold energy resource pooling stage:

[0194] Resource virtualization: The cooling resource pooling management module abstracts the physical attributes (type, location, capacity) and real-time status (available power / cooling capacity, health status) of all available cooling equipment (such as Resource_A1 chiller, Resource_B2 liquid cooling backplane, Resource_C3 cold storage tank, Resource_D4 dry chiller) in each computer room, and registers them into a unified global cooling resource pool. The resource pool exists in the form of a matrix or database.

[0195] Status synchronization: The system periodically (e.g., every 5 minutes) retrieves the latest data from the edge acquisition layer and updates the resource status matrix. The updates include the current available cooling capacity / power of each resource, the data center environment status, and key cross-data center transmission parameters (e.g., the heat loss rate of the pipes from data center A to data center B = 3.2 kW / km).

[0196] Cold load routing table generation: Based on the latest resource state matrix and data center network topology (pipeline connection relationships), using path planning algorithms (such as Dijkstra's algorithm, with the optimization objective being to minimize total transmission loss or time), calculate the optimal path for cold load transmission between any two data centers and its estimated loss. Generate or update the global cold load routing table.

[0197] Dynamic weight allocation and scheduling phase:

[0198] Weight coefficient calculation: The dynamic weight allocation engine periodically (e.g., every minute) or when a triggering event (such as a sharp rise in the temperature of a data center) occurs, calculates the current scheduling priority weight coefficient for each data center, where the value of the coefficient is continuously optimized and adjusted by the reinforcement learning module based on historical performance.

[0199] Cooling load scheduling decision: The engine sorts and analyzes the scheduling priority weight coefficients of all data centers. When a significant sudden increase in the weight of a data center is detected (indicating that it urgently needs cooling load, such as data center B starting a high computing load causing T_current_B to approach T_critical_B), and its local resources are insufficient:

[0200] Based on the resource status matrix and the cooling capacity routing table, select a suitable cooling capacity provider from the global resource pool (such as the available cooling capacity of data center A).

[0201] Calculate the required cooling capacity (e.g., 200kW) and the optimal transmission path (e.g., A→B).

[0202] Send a scheduling instruction to the refrigerant routing controller of the global scheduling layer (target: computer room B, source: computer room A resources, cooling capacity: 200kW, path: A→B).

[0203] Real-time feedback control: The refrigerant routing controller, based on scheduling instructions, precisely controls the opening of electric valves on relevant pipelines via the SDN network (e.g., increasing the opening of the main pipeline valve from A to B to 85%). Simultaneously, it monitors the actual refrigerant flow rate delivered to server room B in real time (via flow meter data). A closed-loop control algorithm (such as PID control) is used to dynamically adjust the valve opening, ensuring that the error between the actual refrigerant flow rate and the target flow rate is ≤5%. Sensors at the edge acquisition layer continuously provide feedback on temperature changes in server room B.

[0204] Fault transfer phase:

[0205] Fault Diagnosis: The system monitors the status of the refrigeration equipment in real time (through edge acquisition layer or device alarms). When a serious fault is identified (such as the chiller compressor current exceeding the threshold in computer room C), the computer room (C) is automatically marked as a "refrigeration fault" state, and its cooling capacity shortfall is assessed.

[0206] Cooling capacity borrowing and returning agreement execution: Failover actuator intervenes immediately: Based on business priority (e.g., financial data center priority) and resource pool status, select a suitable supporting data center (e.g., data center D) and determine the available cooling capacity (e.g., 150kW).

[0207] Automatically initiate and sign an electronic cooling capacity borrowing and returning agreement (core terms of the agreement: Computer room C immediately borrows 150kW of cooling capacity from computer room D; computer room C promises to return double the cooling capacity (300kW) or equivalent compensation to computer room D within 24 hours after the fault is resolved).

[0208] Send an emergency dispatch command to the refrigerant routing controller (target: faulty machine room C, source: support machine room D, cooling capacity: 150kW, path: select the optimal path according to the routing table, such as D→C).

[0209] Redundancy Recovery: The refrigerant routing controller performs emergency dispatch to ensure that cooling capacity is delivered to the faulty computer room C within seconds. Once the backup refrigeration equipment in computer room C starts up or the fault is repaired and cooling capacity is restored, the failover actuator automatically triggers the return process. According to the agreement (a specified multiple of cooling capacity compensation), a return dispatch instruction is sent to the refrigerant routing controller (target: support computer room D, source: the restored computer room C, cooling capacity: 300kW, path: C→D).

[0210] The refrigerant routing controller performs a return operation. After the return is complete, the protocol is closed and the status of the relevant resources is updated.

[0211] Example 2:

[0212] like Figure 2 As shown, this application provides a device for dynamic optimization of cooling resources across computer rooms, the device comprising:

[0213] Supply and demand assessment module 1 is used to identify the source data center where the cooling supply is greater than the cooling demand, and the target data center where the cooling demand is greater than the cooling supply, among multiple data centers that can mutually schedule cooling capacity.

[0214] The parameter selection module 2 is connected to the supply and demand assessment module 1 and is used to obtain the surplus cooling capacity and the first cooling capacity price of each source computer room, the shortage cooling capacity and the second cooling capacity price of each target computer room, the pipeline heat loss from each source computer room to each target computer room, and the cooling capacity liability.

[0215] The weight calculation module 3, connected to the parameter selection module 2, is used to obtain the cooling capacity scheduling weight from each source computer room to each target computer room based on the surplus cooling capacity, the first cooling capacity price, the cooling capacity shortage, the second cooling capacity price, pipeline heat loss, and cooling capacity liability.

[0216] The scheduling and routing module 4 is connected to the weight calculation module 3. It is used to obtain the cooling scheduling routing table from each source computer room to each target computer room according to the cooling scheduling weight. The cooling scheduling routing table is the cooling scheduling weight and the maximum cooling scheduling scheme to complete all source computer rooms to target computer rooms.

[0217] In one embodiment, the supply and demand assessment module 1 specifically includes:

[0218] The resource pool unit is used to connect cooling supply pipelines in multiple computer rooms. It uses software-defined networking (SDN) to establish a shared cooling resource pool for multiple computer rooms, and collects real-time cooling control data, pipeline length and heat loss rate, electricity price of each computer room, and cooling dispatch volume of each computer room in previous cycles.

[0219] The cooling capacity forecasting unit, connected to the resource pool unit, is used to obtain the forecasted cooling capacity supply and forecasted cooling capacity demand of each computer room during the forecast period based on the real-time cooling capacity control data of each computer room.

[0220] The data center partitioning unit, connected to the cooling capacity prediction unit, is used to obtain n source data centers where the predicted cooling capacity supply is greater than the predicted cooling capacity demand, and m target data centers where the predicted cooling capacity demand is greater than the predicted cooling capacity supply.

[0221] In one embodiment, the parameter selection module 2 specifically includes:

[0222] The cooling capacity parameter unit is used to obtain the cooling capacity surplus Tsi of each source computer room (si, i=[1-n]) and the cooling capacity shortage Ttj of each target computer room (tj, j=[1-m]) based on the predicted cooling capacity supply and the predicted cooling capacity demand.

[0223] The price parameter unit is used to obtain the first cooling price Psi of each source computer room (si, i=[1-n]) during the prediction period and the second cooling price Ptj of each target computer room (tj, j=[1-m]) during the prediction period based on the electricity price of each computer room's location;

[0224] The heat loss parameter unit is used to obtain the pipe heat loss Lsitj of each source room (si, i=[1-n]) and each target room (tj, j=[1-m]) based on the pipe length and pipe loss rate;

[0225] The borrowing and returning parameter unit is used to obtain the cooling load liability Ysitj of each source computer room (si, i=[1-n]) and each target computer room (tj, j=[1-m]) within the cooling load borrowing and returning period, based on the cooling load scheduling volume of each computer room in previous cycles.

[0226] In one embodiment, the weight calculation module 3 is specifically used for:

[0227] The cooling load scheduling weights from each source server room (si, i=[1-n]) to each target server room (tj, j=[1-m]) are obtained according to the following formula:

[0228] Wsitj = α × [1 / (Tsi – Ttj + ε)] + β × (Psi / Ptj) + γ × (1 / Lsitj)+ δ × Ysitj;

[0229] Where Tsi is the surplus cooling capacity of each source room (si, i=[1-n]), Ttj is the shortage of cooling capacity of each target room (tj, j=[1-m]), ε is the set value to prevent the denominator from being 0, Psi is the first cooling capacity price of each source room (si, i=[1-n]), Ptj is the second cooling capacity price of each target room (tj, j=[1-m]), Lsitj is the heat loss of the pipeline from each source room (si, i=[1-n]) to each target room (tj, j=[1-m]), Ysitj is the cooling capacity liability of each source room (si, i=[1-n]) and each target room (tj, j=[1-m]), and α, β, γ, δ are the first set of preset coefficients.

[0230] In one embodiment, the scheduling and routing module 4 specifically includes:

[0231] The sorting unit is used to obtain the cooling capacity transfer priority of each source computer room (si, i=[1-n]) or the cooling capacity transfer priority of each target computer room (tj, j=[1-m]), and sort each source computer room (si, i=[1-n]) according to the cooling capacity transfer priority or sort each target computer room (tj, j=[1-m]) according to the cooling capacity transfer priority;

[0232] The sequential matching unit, connected to the sorting unit, is used to delete the source or target data center with the largest Wsitj corresponding to the source or target data center that is currently sorted first, and the corresponding matched target or source data center after each matching process. This process is repeated until all source or target data centers are matched, so as to obtain the cooling capacity scheduling scheme with the largest ∑Wsitj from all source data centers to the target data center and form a cooling capacity scheduling routing table.

[0233] In one embodiment, the sorting unit specifically includes:

[0234] The priority unit is used to obtain the priority of cooling capacity retrieval for each source computer room (si, i=[1-n]) according to the following formula:

[0235] Osi=k1 × Tsi + k2 × (1 / Psi);

[0236] Alternatively, a priority unit can be added to determine the cooling capacity allocation priority for each target computer room (tj, j=[1-m]) according to the following formula:

[0237] Isi=k1' × Tti + k2' × Pti;

[0238] Among them, k1, k2, k1', k2' are the second set of preset coefficients.

[0239] In one embodiment, the apparatus further includes:

[0240] The scheduling unit is used to schedule cooling capacity from each source computer room (si, i=[1-n]) to each target computer room (tj, j=[1-m]) according to the cooling capacity scheduling routing table.

[0241] The effect unit, connected to the scheduling unit, is used to obtain the scheduling effect after scheduling cooling capacity to form the historical scheduling effect. The scheduling effect includes the actual temperature control accuracy and energy consumption cost of all source and target computer rooms.

[0242] The optimization unit, connected to the effect unit, is used to optimize the first set of preset coefficients α, β, γ, and δ based on historical scheduling effects.

[0243] In one embodiment, the apparatus further includes:

[0244] The borrowing and returning module is used to formulate cooling capacity borrowing and returning agreements for multiple data centers. The cooling capacity borrowing and returning agreement stipulates the cooling capacity borrowing and returning period and the cooling capacity return rules. If there is a cooling capacity liability between any two data centers that exceeds the cooling capacity borrowing and returning period, an economic compensation plan between the two data centers is formulated according to the cooling capacity return rules.

[0245] Example 3:

[0246] like Figure 5 As shown, Embodiment 3 of this application provides a computer-readable storage medium storing a computer program. When the computer program is run by a processor, it implements the method for dynamic optimization of cooling resources across computer rooms as described in Embodiment 1.

[0247] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program units, or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory or other memory technologies, CD-ROM (Compact Disc Read-Only Memory), DVD or other optical disc storage, cartridges, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer.

[0248] like Figure 6As shown, this application can also provide a computer device, including a memory and a processor. The memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the cross-data center dynamic optimization method for cooling resources as described in Embodiment 1. This computer device can be the cross-data center dynamic optimization device for cooling resources as described in Embodiment 2.

[0249] The memory is connected to the processor. The memory can be flash memory, read-only memory or other types of memory. The processor can be a central processing unit or a microcontroller.

[0250] Embodiments 1-3 of this application provide a method, device, and medium for dynamic optimization of cooling resources across data centers. By reasonably selecting parameters between the source data center and the target data center to calculate the cooling scheduling weight, a routing table is formulated to complete the cooling scheduling from all source data centers to the target data center based on the cooling scheduling weight. The selected parameters comprehensively consider four dimensions: cooling supply and demand satisfaction, economy, transmission efficiency, and balance. The resulting cooling scheduling scheme enables multiple data centers that mutually schedule cooling to work together for cooling, thereby achieving global energy efficiency optimization.

[0251] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of this application, and this application is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and substance of this application, and these modifications and improvements are also considered to be within the scope of protection of this application.

Claims

1. A method for dynamic optimization of cooling resources across computer rooms, characterized in that, The method includes: Among multiple data centers that can mutually schedule cooling capacity, the source data center has a cooling capacity supply greater than the cooling capacity demand, and the target data center has a cooling capacity demand greater than the cooling capacity supply. Obtain the surplus cooling capacity and first cooling capacity price of each source computer room, the shortage cooling capacity and second cooling capacity price of each target computer room, the heat loss of the pipeline from each source computer room to each target computer room, and the cooling capacity liability. Based on the surplus cooling capacity, the first cooling capacity price, the cooling capacity shortage, the second cooling capacity price, pipeline heat loss, and cooling capacity liability, obtain the cooling capacity scheduling weights from each source computer room to each target computer room. Based on the cooling capacity scheduling weights, obtain the cooling capacity scheduling routing table from each source data center to each target data center. The cooling capacity scheduling routing table contains the cooling capacity scheduling weights and the maximum cooling capacity scheduling scheme to complete all source data centers to target data centers.

2. The method according to claim 1, characterized in that, This involves identifying the source data center (where cooling supply exceeds cooling demand) and the target data center (where cooling demand exceeds cooling supply) among multiple data centers capable of mutually scheduling cooling capacity. Specifically, this includes: Connect cooling supply pipelines in multiple computer rooms, use software-defined networking (SDN) to establish a shared cooling resource pool for multiple computer rooms, and collect real-time cooling control data, pipeline length and heat loss rate, electricity price of each computer room, and cooling dispatch volume of each computer room in previous cycles. Based on the real-time cooling control data of each computer room, obtain the predicted cooling supply and predicted cooling demand of each computer room during the forecast period; Identify n source data centers where the predicted cooling supply is greater than the predicted cooling demand, and m target data centers where the predicted cooling demand is greater than the predicted cooling supply.

3. The method according to claim 2, characterized in that, Obtain the surplus cooling capacity and primary cooling capacity price of each source data center, the shortage cooling capacity and secondary cooling capacity price of each target data center, the pipe heat loss from each source data center to each target data center, and the cooling capacity liability, specifically including: Based on the predicted cooling supply and the predicted cooling demand, obtain the cooling surplus Tsi for each source computer room (si, i=[1-n]) and the cooling shortage Ttj for each target computer room (tj, j=[1-m]); Based on the electricity price of each computer room location, obtain the first cooling price Psi for each source computer room (si, i=[1-n]) during the forecast period, and the second cooling price Ptj for each target computer room (tj, j=[1-m]) during the forecast period; Based on the pipeline length and pipeline loss rate, obtain the pipeline heat loss Lsitj from each source computer room (si, i=[1-n]) to each target computer room (tj, j=[1-m]); Based on the cooling capacity scheduling volume of each computer room in previous cycles, obtain the cooling capacity liability Ysitj of each source computer room (si, i=[1-n]) and each target computer room (tj, j=[1-m]) within the cooling capacity borrowing and repayment period.

4. The method according to any one of claims 1-3, characterized in that, Based on the surplus cooling capacity, the price of the first cooling capacity, the shortage of cooling capacity, the price of the second cooling capacity, pipeline heat loss, and cooling capacity liabilities, the cooling capacity scheduling weights from each source data center to each target data center are obtained, specifically including: The cooling load scheduling weights from each source server room (si, i=[1-n]) to each target server room (tj, j=[1-m]) are obtained according to the following formula: Wsitj = α × [1 / (Tsi – Ttj + ε)] + β × (Psi / Ptj) + γ × (1 / Lsitj)+δ × Ysitj; Where Tsi is the surplus cooling capacity of each source room (si, i=[1-n]), Ttj is the shortage of cooling capacity of each target room (tj, j=[1-m]), ε is the set value to prevent the denominator from being 0, Psi is the first cooling capacity price of each source room (si, i=[1-n]), Ptj is the second cooling capacity price of each target room (tj, j=[1-m]), Lsitj is the heat loss of the pipeline from each source room (si, i=[1-n]) to each target room (tj, j=[1-m]), Ysitj is the cooling capacity liability of each source room (si, i=[1-n]) and each target room (tj, j=[1-m]), and α, β, γ, δ are the first set of preset coefficients.

5. The method according to claim 4, characterized in that, Based on the cooling capacity scheduling weights, a cooling capacity scheduling routing table is obtained from each source data center to each target data center. The cooling capacity scheduling routing table contains the cooling capacity scheduling weights for completing all source data centers to target data centers and the maximum cooling capacity scheduling scheme, specifically including: Obtain the cooling capacity transfer priority of each source computer room (si, i=[1-n]) or the cooling capacity transfer priority of each target computer room (tj, j=[1-m]), and sort each source computer room (si, i=[1-n]) according to the cooling capacity transfer priority or sort each target computer room (tj, j=[1-m]) according to the cooling capacity transfer priority; After each matching of the source or target data center with the highest Wsitj, the source or target data center with the highest Wsitj is deleted, and the corresponding matching target or source data center is deleted. This process is repeated until all source or target data centers are matched, so as to obtain the cooling capacity scheduling scheme with the highest ∑Wsitj from all source data centers to the target data center and form the cooling capacity scheduling routing table.

6. The method according to claim 5, characterized in that, Obtain the priority of cooling capacity transfer from each source data center (si, i=[1-n]) or the priority of cooling capacity transfer from each target data center (tj, j=[1-m]), specifically including: The priority of cooling capacity allocation for each source computer room (si, i=[1-n]) is obtained according to the following formula: Osi=k1 × Tsi + k2 × (1 / Psi); Alternatively, the cooling capacity transfer priority for each target computer room (tj, j=[1-m]) can be obtained according to the following formula: Isi=k1' × Tti + k2' × Pti; Among them, k1, k2, k1', k2' are the second set of preset coefficients.

7. The method according to claim 4, characterized in that, The method further includes: According to the cooling capacity scheduling routing table, cooling capacity is scheduled from each source computer room (si, i=[1-n]) to each target computer room (tj, j=[1-m]). The scheduling effect after obtaining the scheduling cooling capacity constitutes the historical scheduling effect, which includes the actual temperature control accuracy and energy consumption cost of all source and target computer rooms. Optimize the first set of preset coefficients α, β, γ, and δ based on historical scheduling results.

8. The method according to claim 3, characterized in that, The method further includes: Establish cooling capacity lending and return agreements for multiple data centers. The agreements stipulate the lending and return period and the rules for returning the cooling capacity. If any two data centers have cooling capacity liabilities that exceed the lending and return period, an economic compensation plan between the two data centers shall be formulated according to the cooling capacity return rules.

9. A device for dynamic optimization of cooling resources across computer rooms, characterized in that, The device includes: The supply and demand assessment module is used to identify the source data center where the cooling supply exceeds the cooling demand, and the target data center where the cooling demand exceeds the cooling supply, among multiple data centers that can mutually schedule cooling capacity. The parameter selection module is connected to the supply and demand assessment module to obtain the surplus cooling capacity and first cooling capacity price of each source computer room, the shortage cooling capacity and second cooling capacity price of each target computer room, the pipeline heat loss and cooling capacity liability from each source computer room to each target computer room. The weight calculation module, connected to the parameter selection module, is used to obtain the cooling capacity scheduling weight from each source computer room to each target computer room based on the surplus cooling capacity, the first cooling capacity price, the cooling capacity shortage, the second cooling capacity price, pipeline heat loss, and cooling capacity liability. The scheduling and routing module, connected to the weight calculation module, is used to obtain the cooling capacity scheduling routing table from each source data center to each target data center based on the cooling capacity scheduling weight. The cooling capacity scheduling routing table contains the cooling capacity scheduling weights and the maximum cooling capacity scheduling scheme for all source data centers to target data centers.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for dynamic optimization of cooling resources across computer rooms as described in any one of claims 1-8.

Images