A data center low-carbon control method and system

Abstract

The application discloses a kind of data center low-carbon control method and system, it is related to data center technical field, the method includes: based on multimodal sensor network real-time collection summer hot winter warm region data center's environmental parameter and equipment operation data;Climate adaptive spatiotemporal graph convolution network is constructed, with server rack, cooling tower, switchgear as node, cold and hot flow path is dynamic weighted edge, embedding climate feature vector, generates heat load prediction atlas;Through physical constraint reinforcement learning framework, thermodynamics equation and fluid dynamics model are regarded as strategy optimization boundary condition, combined with heat load prediction atlas, generate dynamic refrigeration strategy and server load migration strategy;Based on dynamic refrigeration strategy, control air conditioning system according to real-time climate type starts mixed refrigeration mode, water-cooling main cycle and indirect evaporative cooling standby loop are synchronously operated, and according to electricity price time period, the logic of filling and discharging cold of water storage refrigeration system is switched.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data centers, in particular to a data center low-carbon control method and system. BACKGROUND

[0002] In recent years, with the rapid development of the Internet, big data, cloud computing and other services have emerged, and the demand for the construction of the foundation supporting their development, large data centers, has also increased. As a basic platform for carrying, transmitting and computing power demand, data centers are the physical base for supporting new generation digital technology applications such as artificial intelligence, cloud computing and blockchain, and are an important part of national strategic development. Data centers have become an important part of new infrastructure construction.

[0003] With the rapid development of digital economy, the number and scale of data centers are continuously expanding, resulting in a large amount of energy consumption and carbon emissions. In particular, in summer-hot and winter-warm regions such as Guangzhou and Shenzhen, the environment is hot and humid for a long time, and the utilization rate of natural cold sources is low. This climate feature poses higher challenges to the energy efficiency improvement and carbon emission control of data centers.

[0004] At present, although research on the design and construction of data centers is actively carried out at home and abroad, certain achievements have been made. However, data center design and construction still face many challenges. For example, traditional data center refrigeration systems rely on fixed refrigeration modes and do not fully consider the dynamic changes of climate conditions and real-time requirements of equipment operation, resulting in low energy utilization efficiency.

[0005] In view of the above problems, no effective solution has been proposed so far. SUMMARY

[0006] The embodiments of the present application provide a data center low-carbon control method and system to solve the above technical problems.

[0007] The present application provides a data center low-carbon control method, comprising:

[0008] Real-time collection of environmental parameters and equipment operation data of data centers in summer-hot and winter-warm regions based on a multi-modal sensor network, wherein the environmental parameters include indoor and outdoor temperature and humidity gradient, heat flow density distribution and cold and hot channel differential pressure change rate;

[0009] Construction of a climate-adaptive spatio-temporal graph convolution network, taking server racks, cooling towers and power distribution cabinets as nodes, cold and hot flow paths as dynamically weighted edges, embedding climate feature vectors, and generating a heat load prediction map;

[0010] Through a physical constraint reinforcement learning framework, the thermodynamic equation and the fluid dynamics model are taken as the boundary conditions of strategy optimization, combined with the heat load prediction map, to generate a dynamic refrigeration strategy and a server load migration strategy.

[0011] Based on the dynamic cooling strategy, the air conditioning system is controlled to start the hybrid cooling mode according to the real-time climate type, synchronously run the water-cooled main loop and the indirect evaporative cooling backup loop, and switch the charging and discharging logic of the water storage cooling system according to the electricity price period.

[0012] Further, the climate-adaptive spatiotemporal graph convolution network comprises:

[0013] The climate feature embedding layer encodes the seasonal humidity fluctuation and typhoon period pressure change in the hot summer and cold winter region into a climate feature vector, and splices the real-time power data of the device node;

[0014] The dynamic edge weight calculation module dynamically adjusts the weight coefficient of the edge according to the temperature rise rate, pressure difference change and path length of the cold and heat flow path, wherein the climate adaptive coefficient is generated by training historical temperature and humidity data;

[0015] The thermodynamic attention mechanism introduces a heat conduction equation constraint in the convolution layer to suppress feature propagation that violates local energy conservation.

[0016] Further, the physical constraint reinforcement learning framework comprises:

[0017] The heat load prediction graph, device energy efficiency ratio and construction stage pipeline pre-embedding deviation data are fused to define the state space;

[0018] The adjustment range of the air conditioning refrigeration capacity is limited to not exceed the critical condensation threshold calculated by the fluid dynamics model, and the server load migration path meets the redundancy power supply safety rules;

[0019] A hierarchical reward function is designed, and the short-term reward is based on the real-time PUE value and the cold and heat channel temperature difference, and the long-term reward is based on the total carbon emissions in the whole life cycle and the construction progress deviation.

[0020] Further, the hybrid cooling mode uses a refrigerant flow dynamic allocation algorithm to preferentially cool high-density computing cluster areas and uses the waste heat of the liquid cooling loop to drive local air circulation;

[0021] The control air conditioning system starts the hybrid cooling mode according to the real-time climate type, comprising:

[0022] In a high-temperature and high-humidity environment, the main water cooling loop adopts a variable flow pump control strategy, and adjusts the branch valve opening degree according to the heat load distribution predicted by the spatiotemporal graph convolution network;

[0023] When the indirect evaporative cooling backup loop is started, meteorological prediction data is introduced to dynamically adjust the spraying frequency of the evaporative water curtain and the fresh air mixing ratio;

[0024] The charging and discharging logic of the water-based cooling system is determined based on the time-of-use electricity price signal and the cooling tower efficiency curve. It prioritizes cooling storage during periods of low electricity prices and triggers cooling release when the cooling tower efficiency falls below a first set threshold.

[0025] Furthermore, the method also includes:

[0026] By combining the BIM twin model during the construction phase, the layout of refrigeration equipment and the pre-embedded pipelines are reverse-optimized to reduce cooling loss during the construction period.

[0027] The inverse optimization of the BIM twin model includes:

[0028] By analyzing the discrepancy between laser scanning point cloud and design model, installation errors in refrigeration pipes can be identified.

[0029] The pipeline route was replanned using a topology optimization algorithm to ensure that the refrigerant delivery distance was minimized and the pressure drop met the fluid model constraints.

[0030] The optimized pipeline layout is then updated in reverse to the design model, and modular prefabricated component processing instructions are generated.

[0031] Furthermore, the realization of the waste heat driving airflow circulation in the liquid cooling circuit includes:

[0032] A thermoelectric conversion module is integrated on the surface of the liquid cooling plate to convert waste heat into electrical energy to drive a micro turbofan.

[0033] Adjust the turbine fan speed based on infrared thermal imaging data to match the local airflow velocity with the heat dissipation requirements of the rack.

[0034] When the ambient humidity exceeds the second set threshold, it automatically switches to anti-condensation mode, limits the maximum speed of the turbofan, and starts auxiliary dehumidification.

[0035] Furthermore, the method also includes:

[0036] Carbon metering coding rules are associated in the BIM twin model to record carbon emissions from concrete pouring during the construction phase and refrigerant leakage during the operation and maintenance phase in real time.

[0037] Predict the evolution path of carbon footprint throughout the entire life cycle using hidden Markov models, and dynamically optimize cooling strategies and equipment replacement cycles;

[0038] When carbon emissions are predicted to exceed the threshold, the system will automatically trigger equipment energy efficiency upgrades or renewable energy procurement plans.

[0039] Furthermore, the method also includes an adaptive fault-tolerant mechanism:

[0040] Pre-train equipment failure response strategies under high humidity scenarios during typhoon seasons in a physical field model;

[0041] When the humidity sensor detects a sudden change in value, it triggers the humidity and heat compensation scheme in the reserve strategy library, including increasing the cooling tower fan speed, migrating the edge computing load to the rack in the low humidity area, and starting the standby dehumidifier unit.

[0042] Furthermore, the method also includes:

[0043] Integrating miniature heat sinks and airflow guide channels into lighting fixtures guides LED waste heat to designated heat dissipation areas;

[0044] Based on the rack heat distribution predicted by the spatiotemporal graph convolutional network, the tilt angle and brightness of the lamps are dynamically adjusted to coordinate the heat dissipation airflow with the air conditioning supply path.

[0045] During maintenance work, activate the enhanced lighting and heat dissipation mode in the work area to simultaneously increase the local cooling capacity.

[0046] This application provides a low-carbon control system for a data center, including:

[0047] The environmental and equipment parameter acquisition module is used to collect environmental parameters and equipment operation data of data centers in hot summer and warm winter regions in real time based on a multimodal sensor network. The environmental parameters include indoor and outdoor temperature and humidity gradients, heat flux density distribution, and the rate of change of pressure difference between hot and cold channels.

[0048] The heat load prediction map generation module is used to construct a climate-adaptive spatiotemporal graph convolutional network. It uses server racks, cooling towers, and power distribution cabinets as nodes, cold and heat flow paths as dynamically weighted edges, and embeds climate feature vectors to generate heat load prediction maps.

[0049] The cooling and load migration strategy generation module is used to generate dynamic cooling strategies and server load migration strategies by using a physical constraint reinforcement learning framework, taking thermodynamic equations and fluid dynamics models as boundary conditions for strategy optimization, and combining the heat load prediction map.

[0050] The cooling mode switching module is used to control the air conditioning system to start a hybrid cooling mode according to the real-time climate type based on the dynamic cooling strategy, simultaneously run the water-cooled main circulation and the indirect evaporative cooling backup circuit, and switch the charging and discharging logic of the water storage system according to the electricity price period.

[0051] Based on the embodiments provided in this application, a multimodal sensor network is deployed to collect multidimensional parameters such as indoor and outdoor temperature and humidity gradients, heat flux density distribution, and hot and cold channel pressure difference change rates in real time, overcoming the limitations of traditional fixed sensor layouts. This network can capture the dynamic changes in the microclimate within the data center, providing high-resolution data support for subsequent strategy generation and significantly improving the spatiotemporal accuracy of heat load prediction. The constructed spatiotemporal graph convolutional network uses key equipment as nodes and hot and cold flow paths as dynamically weighted edges, embedding climate feature vectors into the graph structure. This dynamic modeling method can adjust network weights in real time, adapting to the climate characteristics of large diurnal temperature differences and frequent humidity fluctuations in hot-summer and warm-winter regions. Compared with traditional static prediction models, this method can predict the risk of local thermal runaway 2-3 hours in advance, reserving response time for cooling strategy optimization. By using thermodynamic equations and fluid dynamics models as boundary conditions for reinforcement learning, this invention achieves a deep integration of physical laws and artificial intelligence. This framework can automatically generate dynamic cooling and server load migration strategies, dynamically optimizing the PUE value while ensuring equipment thermal safety. Based on real-time climate type assessment, the air conditioning system can seamlessly switch between the water-cooled main loop and the indirect evaporative cooling backup loop. Especially during peak and off-peak electricity periods, the optimized charging and discharging logic of the water-based cooling system can achieve peak shaving and valley filling of cooling energy consumption, reducing operating costs. This hybrid cooling mode demonstrates excellent energy efficiency in high-temperature and high-humidity environments. Through real-time heat load prediction and dynamic strategy generation, this method can significantly reduce the data center's dependence on fossil fuels. Under typical operating conditions in hot-summer and warm-winter regions, it can significantly reduce the annual average carbon emission intensity, providing a practical technical path for data centers to achieve carbon neutrality. Attached Figure Description

[0052] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0053] Figure 1 This is a flowchart of an optional low-carbon control method for data centers according to an embodiment of this application;

[0054] Figure 2 A flowchart illustrating another optional low-carbon control method for data centers according to an embodiment of this application;

[0055] Figure 3 This is a structural diagram of an optional low-carbon control system for a data center according to an embodiment of this application.

[0056] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0058] Currently, both domestic and international researchers are actively engaged in research on key technologies for data center construction, such as intelligence, automation, and energy conservation, achieving certain results. However, there is still significant room for improvement in areas such as energy efficiency, site selection (under different climate and terrain conditions), space utilization, security, reliability, and rapid construction. Therefore, this application primarily focuses on the impact of complex operating environment conditions on design, research on reducing data PUE (Power Usage Effectiveness) and rapid construction. With the rapid development of the digital economy, the demand for data storage and data center computing power from various industries continues to increase, leading to the continuous expansion of data center scale. Data centers primarily rely on server racks to house IT equipment, and the number of racks is a crucial indicator of data center scale. High computing power demands result in high energy consumption and high heat dissipation, causing the energy consumption of data center equipment and its associated cooling and heat dissipation to rise accordingly.

[0059] The applicant's objective is to improve the efficiency and reduce carbon emissions of ultra-large data centers in hot-summer and warm-winter regions during their design and construction. By researching key technologies for efficiency enhancement and carbon reduction in ultra-large data centers in these regions and their applications, the aim is to reduce energy consumption during the design and construction process, improve space utilization, construction efficiency, and construction quality, continuously improve data center construction technology, and accelerate the development of the national data center industry. Through data collection on different environmental conditions in hot-summer and warm-winter regions, energy conservation, emission reduction, and lower operation and maintenance costs will be fully considered from the planning and design stage. The goal is to achieve a leading overall PUE value for similar projects at the same latitude globally, while simultaneously improving the automation level of the data center's low-carbon control system and reducing operation and maintenance costs. Specific research content includes:

[0060] Research on low-carbon design for ultra-large data centers under complex environmental conditions mainly includes the following aspects:

[0061] (1) Research on the design of air conditioning and refrigeration technology for ultra-large data centers under different climatic and environmental conditions.

[0062] Northern climates are dry and water-scarce, with low average annual temperatures and limited water resources. Therefore, data center projects typically use air-cooled systems to take advantage of the low temperatures and avoid the disadvantages of water scarcity. Southern climates are hot, with higher average temperatures than the north, and abundant water resources. Therefore, data center projects in the south typically use water-cooled systems. Air-cooled systems require a semi-open, unenclosed building envelope, with the cold air acting as the medium for the air conditioning units. Water-cooled systems do not require an enclosed building envelope and have no special requirements for it. It is necessary to study what climatic characteristics and average temperatures are best suited for air-cooling, what water resource conditions are best suited for water-cooling, and the impact of different air conditioning technologies on the overall functional design of the building.

[0063] (2) Research on the planning and design of the functional layout of ultra-large data centers under different terrain conditions.

[0064] The location and functional layout of a data center affect the building's functional layout, as well as the cost of outdoor roads and pipelines.

[0065] (3) Research on key design technologies for reducing PUE in ultra-large data centers under complex environmental conditions.

[0066] PUE (Power Usage Effectiveness) is an internationally recognized metric for measuring the power efficiency of data centers. A PUE value closer to 1 indicates a higher level of greenness in a data center. According to relevant Chinese regulations, the Shaoguan cluster data center must reduce its PUE value to below 1.25, striving to create a green and low-carbon data center cluster. This study establishes an energy consumption calculation model for a super-large data center in a hot-summer, warm-winter region, analyzes energy consumption indicators, and summarizes the factors affecting these indicators. Based on these factors, different energy-saving technologies are implemented and compared to identify the optimal carbon reduction and energy-saving technologies (equipment energy saving, water cooling, and energy storage). During the operation phase, computing power is a dynamic process; the number of racks is not a one-time investment at peak times. Therefore, the dynamic PUE variation pattern under different rack numbers (power) is worth exploring and researching.

[0067] Research on the carbon reduction application of BIM technology in ultra-large data center projects includes:

[0068] Against the backdrop of global climate change, reducing carbon emissions has become a crucial task facing all industries. As core infrastructure of the information age, ultra-large data centers present particularly significant carbon emission challenges during their construction and operation. Therefore, research on the application of BIM technology for carbon reduction in ultra-large data center projects is of great importance to promoting the green and sustainable development of the data center industry.

[0069] This study aims to explore how to effectively reduce carbon emissions throughout the entire lifecycle of a hyperscale data center project, including design, detailed design, and construction, by utilizing BIM technology. Specifically, we will focus on the application of BIM technology in optimized design during the design phase and refined management during the construction phase, with the goal of achieving carbon reduction and emission reduction.

[0070] (1) Research on the application of BIM technology in the design phase of ultra-large data center projects to reduce carbon emissions.

[0071] During the design phase, BIM technology's precise modeling and simulation analysis capabilities are used to study and evaluate the building energy consumption and carbon emissions of the ultra-large data center project, enabling optimization from the early design stages. The study utilizes BIM technology to optimize design parameters such as building orientation, window-to-wall ratio, and building shape to reduce solar radiation and heat transfer, thereby lowering building energy consumption and carbon emissions. Furthermore, BIM technology is used to dynamically simulate and predict the building energy consumption of the ultra-large data center project, providing real-time energy consumption data feedback and allowing for timely adjustments to the design scheme, ensuring that the project meets functional requirements while achieving the lowest possible energy consumption and carbon emissions.

[0072] (2) Research on the application of BIM technology in the detailed design of ultra-large data center projects to reduce carbon emissions.

[0073] In the detailed design phase of ultra-large data center projects, BIM technology is applied to create detailed models for integrated pipeline layout and clash detection. This enables equipment layout and management, professional coordination, pipeline integration, clearance control, parameter verification, support and hanger design, and electromechanical terminal and pre-embedded positioning. The output includes integrated electromechanical pipeline drawings, detailed construction design drawings for electromechanical disciplines, and coordination condition drawings for related disciplines. This effectively optimizes equipment layout, pipeline routing, and material usage, reduces design changes and material waste, and thus lowers carbon emissions.

[0074] (3) Research on the application of BIM technology in the construction phase of ultra-large data center projects to reduce carbon emissions.

[0075] By establishing a BIM model and applying BIM technology to carbon emission statistics during the construction phase of a large-scale data center project, carbon emission data can be accurately tracked and recorded during construction, including energy consumption of construction equipment and carbon emissions from material transportation and processing. The BIM model provides powerful data integration and analysis capabilities, making carbon emission data management and reporting more accurate and efficient. Through real-time monitoring and analysis of carbon emission data, carbon emission hotspots and potential improvement points can be identified in a timely manner, enabling corresponding measures to reduce emissions.

[0076] Optionally, such as Figure 1 As shown, this application provides a low-carbon control method for data centers, including:

[0077] S101 is based on a multimodal sensor network to collect environmental parameters and equipment operation data of data centers in hot summer and warm winter regions in real time. The environmental parameters include indoor and outdoor temperature and humidity gradients, heat flux density distribution and cold and hot aisle pressure difference change rate.

[0078] S102, construct a climate-adaptive spatiotemporal graph convolutional network, with server racks, cooling towers, and power distribution cabinets as nodes, cold and heat flow paths as dynamically weighted edges, embedding climate feature vectors, and generating heat load prediction maps;

[0079] S103 uses a physical constraint reinforcement learning framework to take thermodynamic equations and fluid dynamics models as boundary conditions for policy optimization, and combines them with heat load prediction maps to generate dynamic cooling strategies and server load migration strategies.

[0080] S104, based on a dynamic cooling strategy, controls the air conditioning system to start a hybrid cooling mode according to the real-time climate type, simultaneously running the water-cooled main circulation and the indirect evaporative cooling backup circuit, and switches the charging and discharging logic of the water storage system according to the electricity price period.

[0081] Based on the embodiments provided in this application, a multimodal sensor network is deployed to collect multidimensional parameters such as indoor and outdoor temperature and humidity gradients, heat flux density distribution, and hot and cold channel pressure difference change rates in real time, overcoming the limitations of traditional fixed sensor layouts. This network can capture the dynamic changes in the microclimate within the data center, providing high-resolution data support for subsequent strategy generation and significantly improving the spatiotemporal accuracy of heat load prediction. The constructed spatiotemporal graph convolutional network uses key equipment as nodes and hot and cold flow paths as dynamically weighted edges, embedding climate feature vectors into the graph structure. This dynamic modeling method can adjust network weights in real time, adapting to the climate characteristics of large diurnal temperature differences and frequent humidity fluctuations in hot-summer and warm-winter regions. Compared with traditional static prediction models, this method can predict the risk of local thermal runaway 2-3 hours in advance, reserving response time for cooling strategy optimization. By using thermodynamic equations and fluid dynamics models as boundary conditions for reinforcement learning, this invention achieves a deep integration of physical laws and artificial intelligence. This framework can automatically generate dynamic cooling and server load migration strategies, dynamically optimizing the PUE value while ensuring equipment thermal safety. Based on real-time climate type assessment, the air conditioning system can seamlessly switch between the water-cooled main loop and the indirect evaporative cooling backup loop. Especially during peak and off-peak electricity periods, the optimized charging and discharging logic of the water-based cooling system can achieve peak shaving and valley filling of cooling energy consumption, reducing operating costs. This hybrid cooling mode demonstrates excellent energy efficiency in high-temperature and high-humidity environments. Through real-time heat load prediction and dynamic strategy generation, this method can significantly reduce the data center's dependence on fossil fuels. Under typical operating conditions in hot-summer and warm-winter regions, it can significantly reduce the annual average carbon emission intensity, providing a practical technical path for data centers to achieve carbon neutrality.

[0082] Furthermore, the climate-adaptive spatiotemporal graph convolutional network includes:

[0083] Climate feature embedding layer: The seasonal humidity fluctuations and typhoon-season air pressure changes in hot-summer and warm-winter regions are encoded into climate feature vectors and spliced ​​with the real-time power data of the device nodes.

[0084] Dynamic edge weight calculation module: dynamically adjusts the edge weight coefficients based on the temperature rise rate, pressure difference change, and path length of the hot and cold flow paths, where the climate adaptive coefficient is generated through training with historical temperature and humidity data;

[0085] Thermodynamic attention mechanism: Introducing the heat conduction equation constraint into the convolutional layer to suppress the propagation of features that violate local energy conservation.

[0086] In this embodiment, the climate-adaptive dynamic edge weights are calculated based on the following formula:

[0087]

[0088] Among them, W {ij} Let α(T) be the edge weight from node i to j, reflecting the priority of the hot and cold flow paths; {season} ) is the seasonal humidity adaptive coefficient, generated based on historical temperature and humidity data, used to amplify / suppress the effects of temperature rise; ΔT {ij} β(H) represents the temperature rise rate (°C / s) from node i to j, and the real-time monitored temperature difference change of the cold and hot flows. typhoon ) represents the typhoon-period pressure adaptive coefficient, encoding the pressure fluctuation characteristics under high humidity during typhoon seasons; ΔP {ij} The change in pressure difference (Pa) along the hot and cold flow paths reflects the dynamics of airflow resistance; L {ij} The path length (m) is used to optimize refrigerant delivery efficiency; Δt is the time interval, which in data center scenarios can represent the time step of sensor data acquisition, such as data acquisition every minute or every 5 minutes. The value of Δt depends on the system's real-time requirements and is usually in the range of 1-10 minutes; α is a coefficient related to seasonal humidity, used to adjust the impact of temperature rise rate on edge weights. Its value is usually between 0.5 and 1.5, and specific values ​​can be generated through training with historical temperature and humidity data. For example, in seasons with low humidity (such as winter), α can be set to 0.8 to suppress the impact of temperature rise; in seasons with high humidity (such as summer), α can be set to 1.2 to amplify the impact of temperature rise; β is a coefficient related to air pressure during typhoon seasons, used to adjust the impact of pressure difference changes on edge weights. Its value range is usually between 0.5 and 1.5. The specific value can be generated by training with historical air pressure data. For example, during typhoon season, when air pressure fluctuations are large, β can be set to 1.5 to amplify the impact of pressure difference changes; during non-typhoon season, when air pressure fluctuations are small, β can be set to 0.8 to suppress the impact of pressure difference changes.

[0089] Based on the embodiments provided in this application, the climate feature embedding layer and dynamic edge weight calculation module, by encoding seasonal humidity fluctuations and typhoon-season air pressure changes, enable the climate-adaptive spatiotemporal graph convolutional network to accurately predict the heat load distribution under abrupt climate changes, thereby reducing the heat load prediction error compared to traditional static graph models.

[0090] Furthermore, the physical constraint reinforcement learning framework includes:

[0091] The state space is defined by integrating heat load prediction maps, equipment energy efficiency ratios, and pipeline pre-embedding deviation data during the construction phase.

[0092] The adjustment range of air conditioning cooling capacity is limited to not exceeding the critical condensation threshold calculated by the fluid dynamics model, and the server load migration path meets the redundancy power supply safety rules;

[0093] The design incorporates a tiered reward function, with short-term rewards based on real-time PUE values ​​and the temperature difference between hot and cold aisles, and long-term rewards based on total carbon emissions over the entire lifecycle and deviations from construction schedule.

[0094] In this embodiment of the application, the hierarchical reward function for physical constraint reinforcement learning is:

[0095] R = γ1 × (1 - PUE) t )+γ2×CDR t -γ3×Delay construction

[0096] Where R is the overall reward value, used for multi-objective balancing in policy optimization; PUE t Real-time energy efficiency is calculated via a sensor network; CDR t Real-time carbon emission intensity (kgCO2 / kWh), correlated with refrigerant leakage and equipment energy consumption; Delay construction The penalty for construction schedule deviation is dynamically calculated based on the BIM model's process conflict detection results; γ1, γ2, and γ3 are weighting coefficients that are adaptively adjusted using historical operation and maintenance data.

[0097] For example, in a high-energy-consumption data center scenario, the objectives are: reduce energy consumption, reduce carbon emissions, and ensure construction progress. Weighting coefficients are assigned as follows: γ1 = 0.5: Power Usage Effectiveness (PUE) has a significant impact on energy consumption, therefore it receives a high weight. γ2 = 0.3: Carbon Emission Intensity (CDR) has a significant environmental impact, but in this scenario, it is secondary to energy consumption optimization. γ3 = 0.2: Delay construction has a relatively small impact on the overall project, hence its low weight. In high-energy-consumption data centers, reducing energy consumption is the primary task, therefore PUE has a high weight. Carbon emission intensity also needs attention, but it is relatively less important. The impact of delay construction is relatively small, therefore it has the lowest weight.

[0098] In key carbon emission monitoring areas, the objective is to prioritize carbon emission reduction while also considering energy consumption and construction progress. Weighting coefficients are assigned as follows: γ1 = 0.3: The impact of Power Usage Effectiveness (PUE) on energy consumption, with a moderate weight. γ2 = 0.5: Carbon Emission Intensity (CDR) is the primary optimization objective, with the highest weight. γ3 = 0.2: Delay construction deviation has a relatively low weight. In key carbon emission monitoring areas, reducing carbon emissions is the primary task, hence the highest weight for CDR. While PUE and delay construction deviation have relatively low weights, they still require attention.

[0099] In data centers with tight construction schedules, the goal is to ensure on-time completion while considering energy consumption and carbon emissions. Weighting coefficients are assigned as follows: γ1 = 0.3: Moderate weight for Power Usage Effectiveness (PUE). γ2 = 0.3: Moderate weight for Carbon Intensity Ratio (CDR). γ3 = 0.4: Highest weight for construction schedule deviation. In data centers with tight construction schedules, ensuring on-time completion is the primary task; therefore, Delay_construction has the highest weight. PUE and CDR have relatively lower weights but still require attention.

[0100] Based on the embodiments provided in this application, the hierarchical reward function integrates real-time PUE, carbon emission intensity and construction schedule deviation to ensure that the strategy achieves a balance between short-term energy efficiency and long-term carbon footprint, thus overcoming the limitations of single-objective optimization in the prior art.

[0101] Furthermore, the hybrid cooling mode prioritizes cooling the high-density computing cluster area through a dynamic refrigerant flow allocation algorithm, and utilizes the waste heat of the liquid cooling circuit to drive local airflow circulation.

[0102] like Figure 2 As shown, the control system for the air conditioning system to activate a hybrid cooling mode based on the real-time climate type includes:

[0103] S201, in a high temperature and high humidity environment, the main water cooling circuit adopts a variable flow pump control strategy, and adjusts the opening of the branch valves according to the heat load distribution predicted by the spatiotemporal graph convolutional network.

[0104] S202, when the indirect evaporative cooling backup circuit is started, meteorological forecast data is introduced to dynamically adjust the spraying frequency of the evaporative water curtain and the mixing ratio of fresh air.

[0105] S203, the charging and discharging logic of the water storage cooling system is determined based on the time-of-use electricity price signal and the cooling tower efficiency curve. It prioritizes cooling storage during periods of low electricity prices and triggers cooling release when the cooling tower efficiency is lower than the first set threshold.

[0106] The first set threshold may include 0.7, 0.6, etc.

[0107] Based on the embodiments provided in this application, the water-cooled branch valve opening adjustment model incorporates a humidity correction factor and meteorological forecast data to dynamically optimize the fresh air mixing ratio of the indirect evaporative cooling circuit, thereby reducing cooling tower energy consumption in high-temperature and high-humidity environments.

[0108] Furthermore, the method also includes:

[0109] By combining the BIM twin model during the construction phase, the layout of refrigeration equipment and the pre-embedded pipelines are reverse-optimized to reduce cooling loss during the construction period.

[0110] The inverse optimization of the BIM twin model includes:

[0111] By analyzing the discrepancy between laser scanning point cloud and design model, installation errors in refrigeration pipes can be identified.

[0112] The pipeline route was replanned using a topology optimization algorithm to ensure that the refrigerant delivery distance was minimized and the pressure drop met the fluid model constraints.

[0113] The optimized pipeline layout is then updated in reverse to the design model, and modular prefabricated component processing instructions are generated.

[0114] Based on the embodiments provided in this application, pipeline reverse optimization based on laser scanning, combined with topology algorithms, replans the refrigerant path, reduces cooling loss during construction, and improves pipeline installation accuracy to the millimeter level.

[0115] The generation of modular prefabricated component processing instructions includes:

[0116] The piping system is broken down into prefabricated sections with standardized interfaces, and temperature and pressure sensors are pre-installed in the prefabricated sections.

[0117] During construction, drones are used to inspect and match the prefabricated sections with the on-site positioning coordinates. Combined with the BIM twin model, the assembly accuracy is verified in real time, achieving rapid assembly with errors less than a set threshold.

[0118] Furthermore, the realization of waste heat driving airflow circulation in the liquid cooling circuit includes:

[0119] A thermoelectric conversion module is integrated on the surface of the liquid cooling plate to convert waste heat into electrical energy to drive a micro turbofan.

[0120] Adjust the turbine fan speed based on infrared thermal imaging data to match the local airflow velocity with the heat dissipation requirements of the rack.

[0121] When the ambient humidity exceeds the second set threshold, it automatically switches to anti-condensation mode, limits the maximum speed of the turbofan, and starts auxiliary dehumidification.

[0122] The second threshold can be set to a humidity level close to the condensation point, such as 90% RH. If the equipment has special humidity requirements, such as some servers needing lower humidity to avoid condensation, the threshold can be set to 85%-90% RH to provide a certain safety margin. The threshold can be dynamically adjusted according to the actual operating environment and the equipment's heat dissipation needs. For example, in a high humidity environment, the threshold can be appropriately lowered to enter anti-condensation mode earlier, ensuring equipment safety.

[0123] Based on the embodiments provided in this application, the micro turbofan system driven by waste heat of liquid cooling circuit converts waste heat into airflow power through thermoelectric conversion. When the humidity exceeds the threshold, it automatically switches to anti-condensation mode to avoid equipment failure caused by sudden changes in humidity in traditional heat dissipation solutions.

[0124] Furthermore, the method also includes:

[0125] In the BIM twin model, carbon metering coding rules are associated to record carbon emissions from concrete pouring during the construction phase and refrigerant leakage during the operation and maintenance phase in real time.

[0126] Predict the evolution path of carbon footprint throughout the entire life cycle using hidden Markov models, and dynamically optimize cooling strategies and equipment replacement cycles;

[0127] When carbon emissions are predicted to exceed the threshold, the system will automatically trigger equipment energy efficiency upgrades or renewable energy procurement plans.

[0128] In this embodiment of the application, the hidden Markov carbon footprint state transition is calculated based on the following formula:

[0129] S {t+1} =S t ×A s +E construction ×B operate

[0130] Among them, S t Let A be the carbon footprint state vector at time t, containing carbon emission components from the construction and operation phases; sThis is a state transition matrix for the construction phase, linking the carbon emission correlations of processes such as concrete pouring and pipeline welding; E construction B is a construction error correction factor calculated based on the deviation of laser scanning point cloud; operate The operation and maintenance phase impact matrix maps the coupling effect of cooling strategies and equipment aging on carbon emissions; S {t+1} Let be the carbon footprint state vector at time t+1;

[0131] Based on the embodiments provided in this application, the Hidden Markov Carbon Footprint Model achieves dynamic tracking and prediction of carbon emissions throughout the entire life cycle by associating construction errors with operation and maintenance strategies, providing accurate decision support for carbon quota management.

[0132] Furthermore, the method also includes an adaptive fault-tolerance mechanism:

[0133] Pre-train equipment failure response strategies under high humidity scenarios during typhoon seasons in a physical field model;

[0134] When the humidity sensor detects a sudden change in value, it triggers the humidity and heat compensation scheme in the reserve strategy library, including increasing the cooling tower fan speed, migrating the edge computing load to the rack in the low humidity area, and starting the standby dehumidifier unit.

[0135] Based on the embodiments provided in this application, the typhoon season humidity and heat compensation mechanism pre-trains response strategies for multiple fault scenarios, quickly migrates the computing load and starts backup dehumidification when humidity changes abruptly, ensuring the continuous and stable operation of the system under extreme climate conditions.

[0136] Furthermore, the method also includes:

[0137] Integrating miniature heat sinks and airflow guide channels into lighting fixtures guides LED waste heat to designated heat dissipation areas;

[0138] Based on the rack heat distribution predicted by the spatiotemporal graph convolutional network, the tilt angle and brightness of the lamps are dynamically adjusted to coordinate the heat dissipation airflow with the air conditioning supply path.

[0139] During maintenance work, activate the enhanced lighting and heat dissipation mode in the work area to simultaneously increase the local cooling capacity.

[0140] Based on the embodiments provided in this application, the lighting-heating coupling control converts LED waste heat into auxiliary heat dissipation resources through lamp structure optimization and heat distribution linkage adjustment, thereby improving local cooling efficiency while reducing lighting energy consumption.

[0141] Optionally, such as Figure 3 As shown, this application provides a low-carbon control system for a data center, comprising:

[0142] The environmental and equipment parameter acquisition module 301 is used to collect environmental parameters and equipment operation data of data centers in hot summer and warm winter regions in real time based on a multimodal sensor network. The environmental parameters include indoor and outdoor temperature and humidity gradients, heat flux density distribution, and cold and hot aisle pressure difference change rate.

[0143] The heat load prediction map generation module 302 is used to construct a climate-adaptive spatiotemporal graph convolutional network, with server racks, cooling towers, and power distribution cabinets as nodes, cold and heat flow paths as dynamically weighted edges, and embedding climate feature vectors to generate a heat load prediction map.

[0144] The cooling and load migration strategy generation module 303 is used to generate dynamic cooling strategies and server load migration strategies by using a physical constraint reinforcement learning framework, taking thermodynamic equations and fluid dynamics models as boundary conditions for strategy optimization, and combining them with heat load prediction maps.

[0145] The cooling mode switching module 304 is used to control the air conditioning system to start the hybrid cooling mode according to the real-time climate type based on the dynamic cooling strategy, simultaneously run the water-cooled main circulation and the indirect evaporative cooling backup circuit, and switch the charging and discharging logic of the water storage system according to the electricity price period.

[0146] It should be noted that the embodiments implemented on the data center low-carbon control system side in this application can be referenced with the embodiments implemented on the data center low-carbon control method side, and will not be described in detail here.

[0147] The above are merely preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A low-carbon control method for data centers, characterized in that, include: The data center in the hot summer and warm winter region is collected in real time based on a multimodal sensor network. The environmental parameters include indoor and outdoor temperature and humidity gradients, heat flux density distribution and the rate of change of pressure difference between hot and cold channels. A climate-adaptive spatiotemporal graph convolutional network is constructed, with server racks, cooling towers, and power distribution cabinets as nodes, cold and heat flow paths as dynamically weighted edges, and climate feature vectors are embedded to generate a heat load prediction map. Using a physical constraint reinforcement learning framework, thermodynamic equations and fluid dynamics models are used as boundary conditions for strategy optimization. Combined with the heat load prediction map, dynamic cooling strategies and server load migration strategies are generated. Based on the dynamic cooling strategy, the air conditioning system is controlled to start a hybrid cooling mode according to the real-time climate type, and the water-cooled main circulation and the indirect evaporative cooling backup circuit are run simultaneously. The charging and discharging logic of the water storage system is switched according to the electricity price period. The climate-adaptive spatiotemporal graph convolutional network include: Climate feature embedding layer: The seasonal humidity fluctuations and typhoon-season air pressure changes in hot-summer and warm-winter regions are encoded into climate feature vectors and spliced ​​with the real-time power data of the device nodes. Dynamic edge weight calculation module: dynamically adjusts the edge weight coefficients based on the temperature rise rate, pressure difference change, and path length of the hot and cold flow paths, wherein the climate adaptive coefficient is generated through training with historical temperature and humidity data; Thermodynamic attention mechanism: Introducing the heat conduction equation constraint into the convolutional layer to suppress the propagation of features that violate local energy conservation.

2. The data center low-carbon control method according to claim 1, characterized in that, The physical constraint reinforcement learning framework includes: The state space is defined by integrating the heat load prediction map, equipment energy efficiency ratio, and pipeline pre-embedding deviation data during the construction phase. The adjustment range of the air conditioning cooling capacity is limited to not exceeding the critical condensation threshold calculated by the fluid dynamics model, and the server load migration path meets the redundancy power supply safety rules; The design incorporates a tiered reward function, with short-term rewards based on real-time PUE values ​​and the temperature difference between hot and cold aisles, and long-term rewards based on total carbon emissions over the entire lifecycle and deviations from construction schedule.

3. The data center low-carbon control method according to claim 2, characterized in that, The hybrid cooling mode prioritizes cooling the high-density computing cluster area through a dynamic refrigerant flow allocation algorithm, and utilizes the waste heat of the liquid cooling circuit to drive local airflow circulation. The air conditioning system is controlled to activate a hybrid cooling mode based on real-time climate type, including: In a high-temperature and high-humidity environment, the main water cooling circuit adopts a variable flow pump control strategy to adjust the opening of the branch valves according to the heat load distribution predicted by the spatiotemporal graph convolutional network. When the indirect evaporative cooling backup circuit is started, meteorological forecast data is used to dynamically adjust the spraying frequency of the evaporative water curtain and the mixing ratio of fresh air. The charging and discharging logic of the water-based cooling system is determined based on the time-of-use electricity price signal and the cooling tower efficiency curve. It prioritizes cooling storage during periods of low electricity prices and triggers cooling release when the cooling tower efficiency falls below a first set threshold.

4. The data center low-carbon control method according to claim 1, characterized in that, The method further includes: By combining the BIM twin model during the construction phase, the layout of refrigeration equipment and the pre-embedded pipelines are reverse-optimized to reduce cooling loss during the construction period. The inverse optimization of the BIM twin model includes: By analyzing the discrepancy between laser scanning point cloud and design model, installation errors in refrigeration pipes can be identified. The pipeline route was replanned using a topology optimization algorithm to ensure that the refrigerant delivery distance was minimized and the pressure drop met the fluid model constraints. The optimized pipeline layout is then updated in reverse to the design model, and modular prefabricated component processing instructions are generated.

5. The data center low-carbon control method according to claim 3, characterized in that, The implementation of the waste heat driving airflow circulation in the liquid cooling circuit includes: A thermoelectric conversion module is integrated on the surface of the liquid cooling plate to convert waste heat into electrical energy to drive a micro turbofan. Adjust the turbine fan speed based on infrared thermal imaging data to match the local airflow velocity with the heat dissipation requirements of the rack. When the ambient humidity exceeds the second set threshold, it automatically switches to anti-condensation mode, limits the maximum speed of the turbofan, and starts auxiliary dehumidification.

6. The data center low-carbon control method according to claim 4, characterized in that, The method further includes: Carbon metering coding rules are associated in the BIM twin model to record carbon emissions from concrete pouring during the construction phase and refrigerant leakage during the operation and maintenance phase in real time. Predict the evolution path of carbon footprint throughout the entire life cycle using hidden Markov models, and dynamically optimize cooling strategies and equipment replacement cycles; When carbon emissions are predicted to exceed the threshold, the system will automatically trigger equipment energy efficiency upgrades or renewable energy procurement plans.

7. The data center low-carbon control method according to claim 1, characterized in that, The method also includes an adaptive fault-tolerant mechanism: Pre-train equipment failure response strategies under high humidity scenarios during typhoon seasons in a physical field model; When the humidity sensor detects a sudden change in value, it triggers the humidity and heat compensation scheme in the reserve strategy library, including increasing the cooling tower fan speed, migrating the edge computing load to the rack in the low humidity area, and starting the standby dehumidifier unit.

8. The data center low-carbon control method according to claim 1, characterized in that, The method further includes: Integrating miniature heat sinks and airflow guide channels into lighting fixtures guides LED waste heat to designated heat dissipation areas; Based on the rack heat distribution predicted by the spatiotemporal graph convolutional network, the tilt angle and brightness of the lamps are dynamically adjusted to coordinate the heat dissipation airflow with the air conditioning supply path. During maintenance work, activate the enhanced lighting and heat dissipation mode in the work area to simultaneously increase the local cooling capacity.

9. A low-carbon control system for a data center, characterized in that, include: The environmental and equipment parameter acquisition module is used to collect environmental parameters and equipment operation data of data centers in hot summer and warm winter regions in real time based on a multimodal sensor network. The environmental parameters include indoor and outdoor temperature and humidity gradients, heat flux density distribution, and the rate of change of pressure difference between hot and cold channels. The heat load prediction map generation module is used to construct a climate-adaptive spatiotemporal graph convolutional network. It uses server racks, cooling towers, and power distribution cabinets as nodes, cold and heat flow paths as dynamically weighted edges, and embeds climate feature vectors to generate heat load prediction maps. The cooling and load migration strategy generation module is used to generate dynamic cooling strategies and server load migration strategies by using a physical constraint reinforcement learning framework, taking thermodynamic equations and fluid dynamics models as boundary conditions for strategy optimization, and combining the heat load prediction map. The cooling mode switching module is used to control the air conditioning system to start the hybrid cooling mode according to the real-time climate type based on the dynamic cooling strategy, simultaneously run the water-cooled main circulation and the indirect evaporative cooling backup circuit, and switch the charging and discharging logic of the water storage system according to the electricity price period. The climate-adaptive spatiotemporal graph convolutional network include: Climate feature embedding layer: The seasonal humidity fluctuations and typhoon-season air pressure changes in hot-summer and warm-winter regions are encoded into climate feature vectors and spliced ​​with the real-time power data of the device nodes. Dynamic edge weight calculation module: dynamically adjusts the edge weight coefficients based on the temperature rise rate, pressure difference change, and path length of the hot and cold flow paths, wherein the climate adaptive coefficient is generated through training with historical temperature and humidity data; Thermodynamic attention mechanism: Introducing the heat conduction equation constraint into the convolutional layer to suppress the propagation of features that violate local energy conservation.

Images