Energy-saving operation method of data center machine room air conditioner with cold content matching and load balancing

Abstract

The application discloses a kind of data center computer room air conditioner energy-saving operation methods containing cold content matching and load balancing, comprising: establishing computer room local temperature abnormal position prediction model;Set regional coverage matching, health state matching and air conditioner load balancing matching mechanism, to eliminate local temperature anomaly and air conditioning unit load balancing as the fastest goal, build first layer air conditioner dynamic grouping optimization model, obtain the optimal local temperature anomaly control air conditioner group;According to the optimal local temperature anomaly control air conditioner group, combine cold demand, the refrigeration operating characteristics of each air conditioning unit, local temperature anomaly degree, predict the required air volume and temperature set value at local temperature anomaly position, the total air supply temperature and return air temperature set value of control air conditioner group, and with the collaborative operation energy consumption of air conditioning unit minimum, air conditioning unit load balancing and cold supply-demand matching optimal as goal, build second layer multi-air conditioner optimization operation model, obtain the optimal control parameter of each air conditioner collaborative refrigeration.

Description

Technical Field

[0001] This invention belongs to the field of data center air conditioning operation technology, specifically relating to an energy-saving operation method for data center air conditioning that includes cooling capacity matching and load balancing. Background Technology

[0002] As various IT devices operate dynamically within a data center, the heat dissipation of high-density server racks also changes dynamically. If the heat dissipation is too large and cannot be dissipated, it can cause the equipment temperature to rise rapidly, resulting in system crashes or even damage. Therefore, multiple air conditioning units configured in the server room are used for cooling, and cold air is sent into the cold aisle. The server rack air inlets draw in cold air from the cold aisle, while hot air is discharged into the hot aisle from the server rack air outlets. The hot air is then recovered by the air conditioning return vents, cooled, and sent back into the cold aisle, ultimately ensuring the normal operation of the equipment within the server racks.

[0003] However, traditional data center air conditioning units are controlled by pre-setting cooling capacity. But the actual load on IT equipment fluctuates dynamically. When the pre-set cooling capacity exceeds the cooling load required for equipment heat dissipation, the excess cooling capacity results in energy waste. When the pre-set cooling capacity is less than the cooling load required for equipment heat dissipation, localized temperature anomalies occur in the data center. Furthermore, the air conditioning load in areas with localized temperature anomalies increases due to increased cooling output, while other air conditioning units experience low loads. This leads to localized overload or idle air conditioning systems, failing to fully utilize the equipment's optimal operating efficiency and increasing energy consumption. Therefore, how to quickly eliminate localized temperature anomalies, ensure that air conditioning units can dynamically coordinate to handle the cooling tasks of localized temperature anomalies, avoid overloading individual units, achieve a more balanced air conditioning load, minimize energy consumption, and match cooling supply and demand are urgent problems to be solved.

[0004] Based on the aforementioned technical issues, a new energy-saving operation method for data center air conditioning that includes cooling capacity matching and load balancing needs to be designed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a data center air conditioning energy-saving operation method with cooling capacity matching and load balancing. This method can eliminate local temperature anomalies in advance, ensure stable and controllable temperature in the computer room, and ensure that each air conditioning unit operates with balanced load and high efficiency, resulting in a significant reduction in energy consumption costs. It adopts dynamic grouping and collaborative control to adapt to dynamic changes in the computer room.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: This invention provides a method for energy-saving operation of data center air conditioning, including cooling capacity matching and load balancing, comprising: S1. Establish a digital twin model of the data center cooling system, which includes multiple air conditioning units and multiple server racks; S2. Based on the digital twin model of the data center computer room cooling system, obtain the temperature, airflow distribution data, operating tasks of each cabinet, cooling demand and operating parameters of each air conditioning unit at different locations in the data center computer room, and establish a local temperature anomaly location prediction model to obtain the local temperature anomaly locations in the computer room at different times in the future. S3. Based on the location of local temperature anomalies in the computer room in future time periods, and combined with the operating characteristic parameters of each air conditioning unit in the corresponding time period, set up regional coverage matching, health status matching and air conditioning load balancing matching mechanisms. With the goal of eliminating local temperature anomalies and balancing the load of air conditioning units as quickly as possible, construct the first-level air conditioning dynamic grouping optimization model to obtain the optimal local temperature anomaly control air conditioning group. S4. Based on the optimal local temperature anomaly control of the air conditioning unit, combined with the cooling demand, the cooling operation characteristics of each air conditioning unit, and the degree of local temperature anomaly, predict the required air volume and temperature setpoint at the location of the local temperature anomaly, and adjust the total supply air temperature and return air temperature setpoint of the air conditioning unit. With the goal of minimizing the energy consumption of the coordinated operation of the air conditioning units, balancing the load of the air conditioning units, and achieving optimal matching of cooling supply and demand, construct a second-layer multi-air conditioning optimization operation model to obtain the optimal control parameters for coordinated cooling of each air conditioning unit.

[0007] Furthermore, S1 includes: Based on the distribution and basic parameters of each air conditioning unit, cabinet, cold aisle, and hot aisle, and combined with the relationship between air conditioning unit and cold aisle, cold aisle and cabinet, cabinet and hot aisle, and hot aisle and air conditioning unit, a physical entity mechanism model is established to couple the cooling of the computer room, heat generation of the cabinet, airflow and temperature distribution. The distribution location of the air conditioning units includes the three-dimensional coordinates of each air conditioning unit, the air supply method of the air conditioner, the location of the return air outlet, the direction of the air duct, and the initial direction of the air supply airflow; the basic parameters of the air conditioning units include the external dimensions, rated cooling capacity, rated air supply volume, rated fan frequency, water valve adjustment range, rated supply air temperature, and rated return air temperature. The distribution of the server racks includes the three-dimensional coordinates of each rack, the orientation of the racks, and the spacing between racks; the basic parameters of the racks include their external dimensions, air intake area, air exhaust area, rack heat dissipation efficiency, and IT equipment installed power. The distribution of the cold aisle includes its three-dimensional spatial range, enclosure form, and floor air outlet distribution; the basic parameters of the cold aisle include its length, width, and height, the total number of air outlets, the rated air volume of a single air outlet, and the design air velocity within the aisle. The distribution location of the hot aisle includes the three-dimensional spatial range of the hot aisle, the return air method of the hot aisle, and the isolation boundary between the hot aisle and the cold aisle; the basic parameters of the hot aisle include the length, width, and height of the hot aisle, the area of ​​the return air inlet, and the return air resistance coefficient. Data is collected from sensors deployed at different key locations in each air conditioning unit, cabinet, cold aisle, and hot aisle to obtain data on air conditioning unit supply air temperature, return air temperature, fan frequency, water valve opening, cooling capacity, and COP; cabinet inlet air temperature, exhaust air temperature, cooling capacity requirement, and IT equipment operating power; as well as temperature and wind speed in the cold and hot aisles and temperature and humidity field data of the entire data center. This data drives the corresponding physical entity mechanism model to perform virtual-real mapping, forming a digital twin model of the data center cooling system.

[0008] Furthermore, S2 specifically includes: Simulation analysis was conducted based on a digital twin model of the data center cooling system to obtain temperature data, airflow distribution data, rack operation task data, rack cooling capacity demand data, and operating parameters of each air conditioning unit at different locations within the data center rack. After feature extraction, classification, and dimensionality reduction of the acquired data, a feature set including spatial features, temperature-related features, and operating status features is formed. The XGBoost model is then used for training and learning to establish a local temperature prediction model for the computer room, which outputs the predicted temperature values ​​for each temperature measurement point and each cabinet location in the computer room at different times in the future. Temperature predictions and spatial and temperature-related features from the feature set are used as input data. The DBSCAN clustering model is used for training and learning to establish a model for identifying the location of local temperature anomalies in the computer room, and the location of the abnormal temperature area in the spatial cluster is output.

[0009] Furthermore, the temperature data for different locations within the server rack includes: rack inlet temperature, outlet temperature, top, middle, and bottom temperatures, rack spacing aisle temperature, cold aisle temperature, and hot aisle temperature; the airflow distribution data for the data center includes: cold aisle supply air velocity, hot aisle return air velocity, airflow velocity and direction at the air conditioning outlet, airflow velocity between server racks, and coordinates of the cold and hot aisle short-circuit and return flow areas; the server rack operation task data includes: CPU utilization, service deployment volume, memory and GPU computing power usage, server rack operating load rate, and task start / stop time; the server rack cooling demand data includes: cooling demand value per server rack, cooling demand by area in the data center, and cooling supply-demand difference; the operating parameters of each air conditioning unit include: air conditioning unit supply air temperature, return air temperature, supply air volume, cooling power, water pump speed, humidifier opening, fan frequency, and water valve opening.

[0010] Furthermore, in S3, based on the locations of local temperature anomalies in the computer room during future time periods, and combined with the operating characteristic parameters of each air conditioning unit during the corresponding time periods, a regional coverage matching, health status matching, and air conditioning load balancing matching mechanism is set, including: Based on the local temperature location of the data center in future time periods, analyze the physical coordinate range of the abnormal location area, the spatial coverage area of ​​the abnormal location area, the number of server racks, and at the same time, analyze the difference between the abnormal temperature value and the preset normal temperature value, the temperature change rate of the previous time period, the required cooling capacity, the abnormal start time, the duration, and the temperature diffusion trend for the local temperature location, forming a local abnormal space-temperature dataset. The system acquires the physical location coordinates, air supply coverage, cooling radiation area, air supply method, air supply angle, and wind speed adjustment range of each air conditioning unit in future time periods, as well as the current percentage of cooling output to rated cooling capacity, adjustable cooling capacity range, air supply temperature adjustment range, air supply volume adjustment range, cooling response rate, cooling capacity plan for subsequent time periods, and health status score of each air conditioning unit, forming an air-conditioned space-cooling dataset. The system performs spatial and operational characteristic matching between the local anomaly space-temperature dataset and the air conditioning-cooling dataset, and sets up a regional coverage matching mechanism: for local temperature anomaly locations, the air conditioning units in the computer room are evaluated for their effective spatial coverage at that location. Among the candidate air conditioning units at the same anomaly location, those with higher effective spatial coverage are given higher matching priority, forming a candidate air conditioning unit pool. An air conditioning health status matching mechanism is also set up: air conditioning units in the candidate air conditioning unit pool that do not meet the preset health status threshold are excluded. An air conditioning load balancing matching mechanism is also set up: based on the overall load balance of the air conditioning units and the absence of overload risk for individual air conditioners, it avoids situations where some air conditioning units are limited to low loads while others are overloaded. For the air conditioning units in the candidate air conditioning unit pool, the real-time load margin is calculated. The larger the load margin, the more cooling demand the air conditioning unit undertakes. Based on the real-time load margin, the cooling demand required at the temperature anomaly location is preferentially allocated to air conditioning units with low loads, forming a preferred dynamic grouping scheme set for air conditioning units.

[0011] It should be noted that the grouping is dynamic. If the location of the local temperature anomaly changes, such as due to changes in the workload of IT equipment in the cabinet, the group will automatically adjust its members, such as adding or removing an air conditioning unit. At the same time, based on the current load of each air conditioning unit, more air supply tasks will be assigned to air conditioning units with low loads, increasing their fan frequency, while the tasks of air conditioning units with high loads will be reduced, so as to keep the fan frequency of all air conditioning units in a medium range and avoid the phenomenon of one high and many low, so as to make full use of the cooling capacity of each device.

[0012] Furthermore, in S3, with the goal of eliminating local temperature anomalies and balancing the load of air conditioning units as quickly as possible, a first-level dynamic grouping optimization model for air conditioning is constructed, including: The objective function, taking the fastest elimination of local temperature anomalies, is expressed as: ; This represents the total number of locations with localized temperature anomalies within a future time period. Let be the load change in the j-th abnormal location area when the abnormality is eliminated; The set of dynamic grouping schemes for air conditioning units selected for the j-th abnormal location region; The cooling capacity is allocated to the j-th abnormal location area for the i-th air conditioning unit using a dynamic grouping scheme. The coverage effectiveness of the dynamic grouping scheme of the i-th air conditioning unit for the j-th abnormal location area; Let be the cooling efficiency of the dynamic grouping scheme for the i-th air conditioning unit; Let i be the cooling response rate of the dynamic grouping scheme for the i-th air conditioning unit; Taking load balancing of air conditioning units as the objective function, it can be expressed as: ; ; K represents the total number of dynamic grouping schemes for air conditioning units; Let be the variance of the total load rate of all air conditioners in the i-th dynamic grouping scheme of air conditioner units. The smaller the variance, the more balanced the load. m is the total number of air conditioner units in the dynamic grouping scheme of air conditioner units. Let be the original operating load rate of the s-th air conditioning unit in the dynamic grouping scheme of the i-th air conditioning unit; The additional load rate after the s-th air conditioning unit provides cooling to the j-th abnormal location area in the dynamic grouping scheme of the i-th air conditioning unit; Let be the average total load rate of all air conditioning units in the i-th dynamic grouping scheme of air conditioning units.

[0013] Furthermore, in S4, based on the optimal local temperature anomaly control of the air conditioning unit, and combining the cooling demand, the cooling operation characteristics of each air conditioning unit, and the degree of local temperature anomaly, the required air volume and temperature setpoint at the location of the local temperature anomaly are predicted, and the overall supply air temperature setpoint of the air conditioning unit is adjusted, including: For locations with localized temperature anomalies within the computer room, after matching the optimal localized temperature anomaly control air conditioning unit, the following data are obtained: the degree of temperature anomaly in the anomaly area, the number of server racks covered by the anomaly area, the rack layout density, the location of hot and cold aisles, airflow velocity, total cooling capacity demand, inherent attribute parameters of the air conditioning unit, current operating load rate, cooling capacity, supply air volume, supply air temperature, and cooling efficiency. This data serves as the control dataset. Features strongly correlated with air volume and temperature are extracted, and the LightGBM algorithm is used for training to establish a control temperature-air volume prediction model. The model outputs the required air volume and temperature setpoints for the relevant air vents at the localized temperature anomaly location, as well as the total supply air temperature setpoint of the control air conditioning unit.

[0014] Furthermore, in S4, the objective of minimizing the coordinated operation energy consumption of the air conditioning units is expressed as: ; Let be the energy consumption coefficient of the s-th air conditioning unit; The cooling capacity provided by the s-th air conditioning unit to the j-th abnormal location area; With the goal of load balancing for air conditioning units, it can be represented as: ; The goal of achieving optimal matching between cooling supply and demand is expressed as: ; Let s be the air volume supplied by the s-th air conditioning unit when cooling the j-th abnormal location area; Let be the supply air temperature difference when the s-th air conditioning unit cools the j-th abnormal location area; Let be the cooling efficiency of the s-th air conditioning unit; The effective spatial coverage of the s-th air conditioning unit for the j-th abnormal location area; Let be the attenuation coefficient of cooling capacity delivery when the s-th air conditioning unit is cooling the j-th abnormal location area; Let J be the airflow at the air outlet of the j-th abnormal location area; The temperature of the air outlet in the j-th abnormal location area; Let be the temperature of the air inlet in the j-th abnormal location area.

[0015] Furthermore, in S4, the constraints of the first-layer air conditioning dynamic grouping optimization model are set, including: space unit coverage constraints, air conditioning health status constraints, cooling capacity demand constraints, air conditioning unit load upper and lower limits constraints, and grouping integrity constraints. The constraints for the second-layer multi-air conditioning optimization operation model are set as follows: cooling supply and demand balance constraints, air conditioning unit cooling capacity adjustment constraints, air conditioning unit air volume adjustment constraints, and air conditioning unit load rate safety constraints.

[0016] Furthermore, after S4, the method further includes: iteratively interacting the first-layer air conditioning dynamic grouping optimization model and the second-layer multi-air conditioning optimization operation model to output the optimal control parameters for the local temperature anomaly control air conditioning group and the coordinated cooling of each air conditioner; and executing the optimal control parameters based on the digital twin model of the data center computer room cooling system to determine whether the local temperature anomaly location has disappeared. If it has disappeared, the optimal control parameters are issued; otherwise, the control parameters are corrected until the local temperature anomaly location disappears.

[0017] The beneficial effects of this invention are: (1) This invention restores the physical space and cooling system of the computer room through a digital twin model. It can collect data in all dimensions such as temperature, airflow, cooling demand, and air conditioning parameters in real time. It can also combine a local temperature anomaly prediction model to obtain the location of local temperature anomalies in advance. By making advance predictions, the air conditioning control can be changed from a passive response to an active prevention, which greatly reduces the probability of local temperature anomalies. (2) This invention constructs a dynamic grouping model through regional coverage matching, health status matching and load balancing matching mechanisms. It prioritizes the selection of air conditioning units with high coverage effectiveness in abnormal areas, avoids long-term high-load operation of a single air conditioning unit, takes into account the load distribution of air conditioning units within the group, and prioritizes the selection of air conditioning units with high cooling efficiency and low failure rate. The final output of the optimal control air conditioning unit can eliminate local temperature anomalies with the fastest response speed, while ensuring the stability of air conditioning unit operation. It solves the problem of resource mismatch and load imbalance that is easy to cause in the traditional scheme of using fixed air conditioning units to cover fixed areas. (3) This invention achieves the three objectives of matching cooling capacity, balancing load and minimizing energy consumption through a multi-air conditioning optimization operation model, avoiding energy waste caused by excessive cooling capacity and failure to eliminate hot spots due to insufficient cooling capacity, so that the air conditioning units can achieve balanced load and play their maximum role.

[0018] Other features and advantages will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a data center air conditioning energy-saving operation method that includes cooling capacity matching and load balancing, according to the present invention. Figure 2 This is a schematic diagram of the data center structure of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figure 1 , Figure 2 As shown, this embodiment provides a data center air conditioning energy-saving operation method that includes cooling capacity matching and load balancing, which includes: S1. Establish a digital twin model of the data center cooling system, which includes multiple air conditioning units and multiple server racks; S2. Based on the digital twin model of the data center computer room cooling system, obtain the temperature, airflow distribution data, operating tasks of each cabinet, cooling demand and operating parameters of each air conditioning unit at different locations in the data center computer room, and establish a local temperature anomaly location prediction model to obtain the local temperature anomaly locations in the computer room at different times in the future. S3. Based on the location of local temperature anomalies in the computer room in future time periods, and combined with the operating characteristic parameters of each air conditioning unit in the corresponding time period, set up regional coverage matching, health status matching and air conditioning load balancing matching mechanisms. With the goal of eliminating local temperature anomalies and balancing the load of air conditioning units as quickly as possible, construct the first-level air conditioning dynamic grouping optimization model to obtain the optimal local temperature anomaly control air conditioning group. S4. Based on the optimal local temperature anomaly control of the air conditioning unit, combined with the cooling demand, the cooling operation characteristics of each air conditioning unit, and the degree of local temperature anomaly, predict the required air volume and temperature setpoint at the location of the local temperature anomaly, and adjust the total supply air temperature and return air temperature setpoint of the air conditioning unit. With the goal of minimizing the energy consumption of the coordinated operation of the air conditioning units, balancing the load of the air conditioning units, and achieving optimal matching of cooling supply and demand, construct a second-layer multi-air conditioning optimization operation model to obtain the optimal control parameters for coordinated cooling of each air conditioning unit.

[0024] In this embodiment, S1 includes: Based on the distribution and basic parameters of each air conditioning unit, cabinet, cold aisle, and hot aisle, and combined with the relationship between air conditioning unit and cold aisle, cold aisle and cabinet, cabinet and hot aisle, and hot aisle and air conditioning unit, a physical entity mechanism model is established to couple the cooling of the computer room, heat generation of the cabinet, airflow and temperature distribution. The distribution location of the air conditioning units includes the three-dimensional coordinates of each air conditioning unit, the air supply method of the air conditioner, the location of the return air outlet, the direction of the air duct, and the initial direction of the air supply airflow; the basic parameters of the air conditioning units include the external dimensions, rated cooling capacity, rated air supply volume, rated fan frequency, water valve adjustment range, rated supply air temperature, and rated return air temperature. The distribution of the server racks includes the three-dimensional coordinates of each rack, the orientation of the racks, and the spacing between racks; the basic parameters of the racks include their external dimensions, air intake area, air exhaust area, rack heat dissipation efficiency, and IT equipment installed power. The distribution of the cold aisle includes its three-dimensional spatial range, enclosure form, and floor air outlet distribution; the basic parameters of the cold aisle include its length, width, and height, the total number of air outlets, the rated air volume of a single air outlet, and the design air velocity within the aisle. The distribution location of the hot aisle includes the three-dimensional spatial range of the hot aisle, the return air method of the hot aisle, and the isolation boundary between the hot aisle and the cold aisle; the basic parameters of the hot aisle include the length, width, and height of the hot aisle, the area of ​​the return air inlet, and the return air resistance coefficient. Data is collected from sensors deployed at different key locations in each air conditioning unit, cabinet, cold aisle, and hot aisle to obtain data on air conditioning unit supply air temperature, return air temperature, fan frequency, water valve opening, cooling capacity, and COP; cabinet inlet air temperature, exhaust air temperature, cooling capacity requirement, and IT equipment operating power; as well as temperature and wind speed in the cold and hot aisles and temperature and humidity field data of the entire data center. This data drives the corresponding physical entity mechanism model to perform virtual-real mapping, forming a digital twin model of the data center cooling system.

[0025] It should be noted that the digital twin model of the data center cooling system is a 1:1 digital mirror of the physical cooling system. Through precise virtual modeling, physical data acquisition, bidirectional data mapping, and simulation, the operating status of the air conditioning units, the heat load of the racks, the airflow temperature, and the temperature and humidity field in the physical world are all replicated and mapped into the virtual digital space. At the same time, it supports virtual-physical interaction, dynamic simulation, and decision reasoning. Ultimately, it enables the virtual side to deduce and control the operation strategy for the cooling problem on the physical side, and sends the optimized control parameters to the physical side for execution. This solves the problems of local hot spots, wasted cooling capacity, and uneven air conditioning load when multiple air conditioning units work together to cool the data center.

[0026] A cold aisle is defined as the passageway formed by the front (air intake) of the server rack, where cool air from the air conditioning unit is delivered. The IT equipment inside the rack draws in cool air from the cold aisle for cooling. A hot aisle is defined as the passageway formed by the back (exhaust) of the server rack, where hot air from the IT equipment gathers and is then drawn in by the air conditioning unit for cooling.

[0027] Sensors are deployed at key locations in every air conditioner, every server rack, and hot / cold aisle within the physical server room to form a comprehensive sensing network. Furthermore, the cooling-heat dissipation mechanism is a core modeling element. Based on the heat and mass transfer mechanisms and fluid dynamics mechanisms of the data center server room, a coupled mechanism model of server room cooling, server rack heat generation, airflow, and temperature distribution can be constructed.

[0028] In this embodiment, S2 specifically includes: Simulation analysis was conducted based on a digital twin model of the data center cooling system to obtain temperature data, airflow distribution data, rack operation task data, rack cooling capacity demand data, and operating parameters of each air conditioning unit at different locations within the data center rack. After feature extraction, classification, and dimensionality reduction of the acquired data, a feature set including spatial features, temperature-related features, and operating status features is formed. The XGBoost model is then used for training and learning to establish a local temperature prediction model for the computer room, which outputs the predicted temperature values ​​for each temperature measurement point and each cabinet location in the computer room at different times in the future. Temperature predictions and spatial and temperature-related features from the feature set are used as input data. The DBSCAN clustering model is used for training and learning to establish a model for identifying the location of local temperature anomalies in the computer room, and the location of the abnormal temperature area in the spatial cluster is output.

[0029] It should be noted that feature dimensionality reduction can be achieved using Pearson correlation coefficient analysis: calculating the correlation between features and the temperature of the measurement point, eliminating features with a correlation less than a preset value, and then using principal component analysis to reduce the dimensionality of the remaining features, retaining principal components with a large cumulative variance contribution rate, thereby reducing feature dimensionality and model computation. The feature set, including spatial features, temperature-related features, and operational status features, is divided into training, validation, and test sets. The training set is input into the XGBoost model, and multiple decision trees are constructed through iterative training. Each tree fits the residual of the previous tree, and the parameters are adjusted using the validation set. Grid search and 5-fold cross-validation are used to find the optimal parameter combination. Finally, the model accuracy is evaluated using the test set, with the core evaluation metric being the mean squared error, which should be as small as possible. Once trained, the XGBoost model can output temperature predictions for the next 5, 10, and 30 minutes in the time dimension, and temperature predictions for each temperature measurement point, each rack air inlet / outlet, and the central area in the spatial dimension, by inputting feature set data at any time. Here, the temperature predictions are only the temperature predictions for the relevant locations within the server room. The next step is to combine the temperature predictions to identify and judge temperature anomalies and abnormal locations. Subsequently, the predicted future temperatures, spatial features, and temperature-related features of each point output by the XGBoost model are used as input features. The DBSCAN model is then used for clustering training. The model iterates through all input points, calculates the number of points within the neighborhood radius of each point, marks points with a neighborhood number greater than the minimum number of points as core points, connects adjacent core points to form clusters, creating continuous temperature regions, and sets a temperature anomaly threshold to filter the clustering results: if the temperature of all points within a cluster is higher than normal, the cluster is determined to be a local temperature anomaly region; if some points within a cluster are abnormal and some are normal, it is determined to be a suspected anomaly region. This is further verified by combining historical temperature data. Finally, the spatial location of the anomaly region, the number of racks covered, and the spatial area are output. For example, if the top temperatures of multiple consecutive racks exceed normal values, DBSCAN will cluster the top locations of these three racks into an anomaly region cluster, rather than an isolated single-point anomaly.

[0030] In practical applications, heat maps can be used to visually display the temperature distribution in a computer room. Blue represents low-temperature areas and red represents high-temperature areas. This can be understood as the temperature decreasing step by step from red, orange, yellow, green, light blue, to blue. At the same time, the location of sensors or cabinets can be marked on the heat map with small dots, which can accurately correspond to which area or cabinet has a high temperature.

[0031] It's important to note that localized temperature anomalies aren't limited to hot spots; cold spots also exist, indicating localized areas where the temperature is significantly below the normal threshold. These cold spots, in contrast to hot spots, also affect the stability and cooling efficiency of the data center. When the cooling supply to a cold spot far exceeds actual demand, it constitutes overcooling. Similar to hot spots, cold spots require dynamic grouping and optimized operation of multiple air conditioners. Cooling capacity is adjusted according to demand, reducing the airflow and increasing the air temperature in cold spot areas to decrease the operating power of the corresponding air conditioners.

[0032] In this embodiment, the temperature data at different locations of the rack includes: rack inlet temperature, rack outlet temperature, rack top, middle and bottom temperatures, rack spacing aisle temperature, cold aisle temperature, and hot aisle temperature; the airflow distribution data of the data center includes: cold aisle supply air velocity, hot aisle return air velocity, airflow velocity and direction at the air conditioning outlet, airflow velocity between racks, and coordinates of the cold and hot aisle short-circuit and return flow areas; the rack operation task data includes: CPU utilization, service deployment volume, memory, GPU computing power usage, rack operating load rate, and task start and stop time of the servers in the rack; the rack cooling demand data includes: cooling demand value per rack, cooling demand of different areas of the data center, and cooling supply and demand difference; the operating parameters of each air conditioning unit include: air conditioning unit supply air temperature, return air temperature, supply air volume, cooling power, water pump speed, humidifier opening, fan frequency, and water valve opening.

[0033] In this embodiment, in step S3, based on the location of local temperature anomalies in the computer room for future time periods, and combined with the operating characteristic parameters of each air conditioning unit for the corresponding time period, a regional coverage matching, health status matching, and air conditioning load balancing matching mechanism is set, including: Based on the local temperature location of the data center in future time periods, analyze the physical coordinate range of the abnormal location area, the spatial coverage area of ​​the abnormal location area, the number of server racks, and at the same time, analyze the difference between the abnormal temperature value and the preset normal temperature value, the temperature change rate of the previous time period, the required cooling capacity, the abnormal start time, the duration, and the temperature diffusion trend for the local temperature location, forming a local abnormal space-temperature dataset. The system acquires the physical location coordinates, air supply coverage, cooling radiation area, air supply method, air supply angle, and wind speed adjustment range of each air conditioning unit in future time periods, as well as the current percentage of cooling output to rated cooling capacity, adjustable cooling capacity range, air supply temperature adjustment range, air supply volume adjustment range, cooling response rate, cooling capacity plan for subsequent time periods, and health status score of each air conditioning unit, forming an air-conditioned space-cooling dataset. The system performs spatial and operational characteristic matching between the local anomaly space-temperature dataset and the air conditioning-cooling dataset, and sets up a regional coverage matching mechanism: for local temperature anomaly locations, the air conditioning units in the computer room are evaluated for their effective spatial coverage at that location. Among the candidate air conditioning units at the same anomaly location, those with higher effective spatial coverage are given higher matching priority, forming a candidate air conditioning unit pool. An air conditioning health status matching mechanism is also set up: air conditioning units in the candidate air conditioning unit pool that do not meet the preset health status threshold are excluded. An air conditioning load balancing matching mechanism is also set up: based on the overall load balance of the air conditioning units and the absence of overload risk for individual air conditioners, it avoids situations where some air conditioning units are limited to low loads while others are overloaded. For the air conditioning units in the candidate air conditioning unit pool, the real-time load margin is calculated. The larger the load margin, the more cooling demand the air conditioning unit undertakes. Based on the real-time load margin, the cooling demand required at the temperature anomaly location is preferentially allocated to air conditioning units with low loads, forming a preferred dynamic grouping scheme set for air conditioning units.

[0034] In practical applications, the effectiveness of the spatial coverage of each air conditioning unit in the computer room at that location is expressed as: ; For spatial coverage effectiveness, ; This is a correction factor for the air supply method. Let i be the rated cooling capacity of the i-th air conditioning unit; Let be the physical center distance from the i-th air conditioning unit to the abnormal location area j; This is the air supply angle matching coefficient.

[0035] Calculate the real-time load margin, expressed as: ; The load margin indicates the additional cooling load capacity that the i-th air conditioning unit can handle. The larger the value, the more the air conditioner can meet new cooling demands. Let be the operating load rate of the i-th air conditioning unit, indicating the percentage of its current cooling output relative to its rated cooling capacity.

[0036] In this embodiment, step S3 aims to eliminate local temperature anomalies and balance the load of air conditioning units as quickly as possible, and constructs a first-layer dynamic grouping optimization model for air conditioning, including: The objective function, taking the fastest elimination of local temperature anomalies, is expressed as: ; This represents the total number of locations with localized temperature anomalies within a future time period. Let be the load change in the j-th abnormal location area when the abnormality is eliminated; The set of dynamic grouping schemes for air conditioning units selected for the j-th abnormal location region; The cooling capacity is allocated to the j-th abnormal location area for the i-th air conditioning unit using a dynamic grouping scheme. The coverage effectiveness of the dynamic grouping scheme of the i-th air conditioning unit for the j-th abnormal location area; Let be the cooling efficiency of the dynamic grouping scheme for the i-th air conditioning unit; Let i be the cooling response rate of the dynamic grouping scheme for the i-th air conditioning unit; Taking load balancing of air conditioning units as the objective function, it can be expressed as: ; ; K represents the total number of dynamic grouping schemes for air conditioning units; Let be the variance of the total load rate of all air conditioners in the i-th dynamic grouping scheme of air conditioner units. The smaller the variance, the more balanced the load. m is the total number of air conditioner units in the dynamic grouping scheme of air conditioner units. Let be the original operating load rate of the s-th air conditioning unit in the dynamic grouping scheme of the i-th air conditioning unit; The additional load rate after the s-th air conditioning unit provides cooling to the j-th abnormal location area in the dynamic grouping scheme of the i-th air conditioning unit; Let be the average total load rate of all air conditioning units in the i-th dynamic grouping scheme of air conditioning units.

[0037] In this embodiment, in step S4, based on the optimal local temperature anomaly control of the air conditioning unit, and combining the cooling demand, the cooling operation characteristics of each air conditioning unit, and the degree of local temperature anomaly, the required air volume and temperature setpoint at the location of the local temperature anomaly are predicted, and the overall supply air temperature setpoint of the air conditioning unit is adjusted, including: For locations with localized temperature anomalies within the computer room, after matching the optimal localized temperature anomaly control air conditioning unit, the following data are obtained: the degree of temperature anomaly in the anomaly area, the number of server racks covered by the anomaly area, the rack layout density, the location of hot and cold aisles, airflow velocity, total cooling capacity demand, inherent attribute parameters of the air conditioning unit, current operating load rate, cooling capacity, supply air volume, supply air temperature, and cooling efficiency. This data serves as the control dataset. Features strongly correlated with air volume and temperature are extracted, and the LightGBM algorithm is used for training to establish a control temperature-air volume prediction model. The model outputs the required air volume and temperature setpoints for the relevant air vents at the localized temperature anomaly location, as well as the total supply air temperature setpoint of the control air conditioning unit.

[0038] In practical applications, the Pearson correlation coefficient is used to calculate the correlation between features and three labels (target air volume, target temperature, and total supply air temperature), retaining features with high correlation. The LightGBM algorithm model supports multi-output regression and can simultaneously predict the three labels: target air volume, target temperature, and total supply air temperature. It leverages the correlation between features to improve overall accuracy. The training process includes: Initialize the model: Initialize the LightGBM model and determine the model's training framework and learning rules; Model Fitting Training: The pre-divided training set feature data and label data are input into the initialized model, and the model begins iterative training. The core training logic of LightGBM is as follows: starting from the first decision tree, the training of each new tree aims to fit the prediction residual of the previous tree, gradually correcting prediction bias; simultaneously, a histogram algorithm is used to discretize the features, significantly improving training speed and reducing memory usage. During training, the model continuously learns the inherent patterns of data center characteristics, airflow, and temperature, gradually converging to the optimal fitting state. Real-time optimization on the validation set: During training, after each iteration, validation set data is input into the model for prediction. The mean squared error (MSE) of the validation set is used to judge the model's fit. If the MSE of the validation set continues to decrease, it indicates that the model is learning normally; if the MSE of the validation set starts to increase, it is determined that the model is overfitting, training is stopped immediately, and the optimal model state is preserved.

[0039] In this embodiment, step S4 aims to minimize the energy consumption of the air conditioning units during coordinated operation, which is expressed as: ; Let be the energy consumption coefficient of the s-th air conditioning unit; The cooling capacity provided by the s-th air conditioning unit to the j-th abnormal location area; With the goal of load balancing for air conditioning units, it can be represented as: ; The goal of achieving optimal matching between cooling supply and demand is expressed as: ; Let s be the air volume supplied by the s-th air conditioning unit when cooling the j-th abnormal location area; Let be the supply air temperature difference when the s-th air conditioning unit cools the j-th abnormal location area; Let be the cooling efficiency of the s-th air conditioning unit; The effective spatial coverage of the s-th air conditioning unit for the j-th abnormal location area; Let be the attenuation coefficient of cooling capacity delivery when the s-th air conditioning unit is cooling the j-th abnormal location area; Let J be the airflow at the air outlet of the j-th abnormal location area; The temperature of the air outlet in the j-th abnormal location area; Let be the temperature of the air inlet in the j-th abnormal location area.

[0040] In this embodiment, in step S4, the constraints of the first-layer air conditioning dynamic grouping optimization model are set, including: space unit coverage constraints, air conditioning health status constraints, cooling capacity demand constraints, air conditioning unit load upper and lower limits constraints, and grouping integrity constraints. The constraints for the second-layer multi-air conditioning optimization operation model are set as follows: cooling supply and demand balance constraints, air conditioning unit cooling capacity adjustment constraints, air conditioning unit air volume adjustment constraints, and air conditioning unit load rate safety constraints.

[0041] In this embodiment, after step S4, the method further includes: iteratively interacting the first-layer air conditioning dynamic grouping optimization model and the second-layer multi-air conditioning optimization operation model to output the optimal control parameters for the local temperature anomaly control air conditioning group and the coordinated cooling of each air conditioner; and executing the optimal control parameters based on the digital twin model of the data center computer room cooling system to determine whether the local temperature anomaly location has disappeared. If it has disappeared, the optimal control parameters are issued; otherwise, the control parameters are corrected until the local temperature anomaly location disappears.

[0042] Furthermore, the functional modules in the various embodiments of this invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part. If the function is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0043] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for energy-saving operation of data center air conditioning, incorporating cooling capacity matching and load balancing, characterized in that, It includes: S1. Establish a digital twin model of the data center cooling system, which includes multiple air conditioning units and multiple server racks; S2. Based on the digital twin model of the data center computer room cooling system, obtain the temperature, airflow distribution data, operating tasks of each cabinet, cooling demand and operating parameters of each air conditioning unit at different locations in the data center computer room, and establish a local temperature anomaly location prediction model to obtain the local temperature anomaly locations in the computer room at different times in the future. S3. Based on the location of local temperature anomalies in the computer room in future time periods, and combined with the operating characteristic parameters of each air conditioning unit in the corresponding time period, set up regional coverage matching, health status matching and air conditioning load balancing matching mechanisms. With the goal of eliminating local temperature anomalies and balancing the load of air conditioning units as quickly as possible, construct the first-level air conditioning dynamic grouping optimization model to obtain the optimal local temperature anomaly control air conditioning group. S4. Based on the optimal local temperature anomaly control of the air conditioning unit, combined with the cooling demand, the cooling operation characteristics of each air conditioning unit, and the degree of local temperature anomaly, predict the required air volume and temperature setpoint at the location of the local temperature anomaly, and adjust the total supply air temperature and return air temperature setpoint of the air conditioning unit. With the goal of minimizing the energy consumption of the coordinated operation of the air conditioning units, balancing the load of the air conditioning units, and achieving optimal matching of cooling supply and demand, construct a second-layer multi-air conditioning optimization operation model to obtain the optimal control parameters for coordinated cooling of each air conditioning unit.

2. The energy-saving operation method for data center computer room air conditioning according to claim 1, characterized in that, S1 includes: Based on the distribution and basic parameters of each air conditioning unit, cabinet, cold aisle, and hot aisle, and combined with the relationship between air conditioning unit and cold aisle, cold aisle and cabinet, cabinet and hot aisle, and hot aisle and air conditioning unit, a physical entity mechanism model is established to couple the cooling of the computer room, heat generation of the cabinet, airflow and temperature distribution. The distribution location of the air conditioning units includes the three-dimensional coordinates of each air conditioning unit, the air supply method of the air conditioner, the location of the return air outlet, the direction of the air duct, and the initial direction of the air supply airflow; the basic parameters of the air conditioning units include the external dimensions, rated cooling capacity, rated air supply volume, rated fan frequency, water valve adjustment range, rated supply air temperature, and rated return air temperature. The distribution of the server racks includes the three-dimensional coordinates of each rack, the orientation of the racks, and the spacing between racks; the basic parameters of the racks include their external dimensions, air intake area, air exhaust area, rack heat dissipation efficiency, and IT equipment installed power. The distribution of the cold aisle includes its three-dimensional spatial range, enclosure form, and floor air outlet distribution; the basic parameters of the cold aisle include its length, width, and height, the total number of air outlets, the rated air volume of a single air outlet, and the design air velocity within the aisle. The distribution location of the hot aisle includes the three-dimensional spatial range of the hot aisle, the return air method of the hot aisle, and the isolation boundary between the hot aisle and the cold aisle; the basic parameters of the hot aisle include the length, width, and height of the hot aisle, the area of ​​the return air inlet, and the return air resistance coefficient. Data is collected from sensors deployed at different key locations in each air conditioning unit, cabinet, cold aisle, and hot aisle to obtain data on air conditioning unit supply air temperature, return air temperature, fan frequency, water valve opening, cooling capacity, and COP; cabinet inlet air temperature, exhaust air temperature, cooling capacity requirement, and IT equipment operating power; as well as temperature and wind speed in the cold and hot aisles and temperature and humidity field data of the entire data center. This data drives the corresponding physical entity mechanism model to perform virtual-real mapping, forming a digital twin model of the data center cooling system.

3. The energy-saving operation method for data center computer room air conditioning according to claim 1, characterized in that, S2 specifically includes: Simulation analysis was conducted based on a digital twin model of the data center cooling system to obtain temperature data, airflow distribution data, rack operation task data, rack cooling capacity demand data, and operating parameters of each air conditioning unit at different locations within the data center rack. After feature extraction, classification, and dimensionality reduction of the acquired data, a feature set including spatial features, temperature-related features, and operating status features is formed. The XGBoost model is then used for training and learning to establish a local temperature prediction model for the computer room, which outputs the predicted temperature values ​​for each temperature measurement point and each cabinet location in the computer room at different times in the future. Temperature predictions and spatial and temperature-related features from the feature set are used as input data. The DBSCAN clustering model is used for training and learning to establish a model for identifying the location of local temperature anomalies in the computer room, and the location of the abnormal temperature area in the spatial cluster is output.

4. The energy-saving operation method for data center computer room air conditioning according to claim 3, characterized in that, The temperature data for different locations within the server rack includes: rack inlet temperature, outlet temperature, top, middle, and bottom temperatures, rack spacing aisle temperature, cold aisle temperature, and hot aisle temperature; the airflow distribution data for the data center includes: cold aisle supply air velocity, hot aisle return air velocity, airflow velocity and direction at air conditioning outlets, airflow velocity between racks, and coordinates of cold and hot aisle short-circuit and return flow areas; the server rack operation task data includes: CPU utilization, service deployment volume, memory and GPU computing power usage, rack operating load rate, and task start / stop time; the server rack cooling demand data includes: cooling demand value per rack, cooling demand by area in the data center, and cooling supply-demand difference; the operating parameters of each air conditioning unit include: air conditioning unit supply air temperature, return air temperature, supply air volume, cooling power, water pump speed, humidifier opening, fan frequency, and water valve opening.

5. The energy-saving operation method for data center computer room air conditioning according to claim 1, characterized in that, In S3, based on the locations of local temperature anomalies in the computer room during future time periods, and combined with the operating characteristic parameters of each air conditioning unit during the corresponding time periods, a regional coverage matching, health status matching, and air conditioning load balancing matching mechanism is set, including: Based on the local temperature location of the data center in future time periods, analyze the physical coordinate range of the abnormal location area, the spatial coverage area of ​​the abnormal location area, the number of server racks, and at the same time, analyze the difference between the abnormal temperature value and the preset normal temperature value, the temperature change rate of the previous time period, the required cooling capacity, the abnormal start time, the duration, and the temperature diffusion trend for the local temperature location, forming a local abnormal space-temperature dataset. The system acquires the physical location coordinates, air supply coverage, cooling radiation area, air supply method, air supply angle, and wind speed adjustment range of each air conditioning unit in future time periods, as well as the current percentage of cooling output to rated cooling capacity, adjustable cooling capacity range, air supply temperature adjustment range, air supply volume adjustment range, cooling response rate, cooling capacity plan for subsequent time periods, and health status score of each air conditioning unit, forming an air-conditioned space-cooling dataset. The system performs spatial and operational characteristic matching between the local anomaly space-temperature dataset and the air conditioning-cooling dataset, and sets up a regional coverage matching mechanism: for local temperature anomaly locations, the air conditioning units in the computer room are evaluated for their effective spatial coverage at that location. Among the candidate air conditioning units at the same anomaly location, those with higher effective spatial coverage are given higher matching priority, forming a candidate air conditioning unit pool. An air conditioning health status matching mechanism is also set up: air conditioning units in the candidate air conditioning unit pool that do not meet the preset health status threshold are excluded. An air conditioning load balancing matching mechanism is also set up: based on the overall load balance of the air conditioning units and the absence of overload risk for individual air conditioners, it avoids situations where some air conditioning units are limited to low loads while others are overloaded. For the air conditioning units in the candidate air conditioning unit pool, the real-time load margin is calculated. The larger the load margin, the more cooling demand the air conditioning unit undertakes. Based on the real-time load margin, the cooling demand required at the temperature anomaly location is preferentially allocated to air conditioning units with low loads, forming a preferred dynamic grouping scheme set for air conditioning units.

6. The energy-saving operation method for air conditioning in a data center computer room according to claim 5, characterized in that, In S3, with the goal of eliminating local temperature anomalies and balancing the load of air conditioning units as quickly as possible, a first-level dynamic grouping optimization model for air conditioning is constructed, including: The objective function, taking the fastest elimination of local temperature anomalies, is expressed as: ； This represents the total number of locations with localized temperature anomalies within a future time period. Let be the load change in the j-th abnormal location area when the abnormality is eliminated; The set of dynamic grouping schemes for air conditioning units selected for the j-th abnormal location region; The cooling capacity is allocated to the j-th abnormal location area for the i-th air conditioning unit using a dynamic grouping scheme. The coverage effectiveness of the dynamic grouping scheme of the i-th air conditioning unit for the j-th abnormal location area; Let be the cooling efficiency of the dynamic grouping scheme for the i-th air conditioning unit; Let i be the cooling response rate of the dynamic grouping scheme for the i-th air conditioning unit; Taking load balancing of air conditioning units as the objective function, it can be expressed as: ； ； K represents the total number of dynamic grouping schemes for air conditioning units; Let be the variance of the total load rate of all air conditioners in the i-th dynamic grouping scheme of air conditioner units. The smaller the variance, the more balanced the load. m is the total number of air conditioner units in the dynamic grouping scheme of air conditioner units. Let be the original operating load rate of the s-th air conditioning unit in the dynamic grouping scheme of the i-th air conditioning unit; The additional load rate after the s-th air conditioning unit provides cooling to the j-th abnormal location area in the dynamic grouping scheme of the i-th air conditioning unit; Let be the average total load rate of all air conditioning units in the i-th dynamic grouping scheme of air conditioning units.

7. The energy-saving operation method for air conditioning in a data center computer room according to claim 1, characterized in that, In step S4, based on the optimal local temperature anomaly control of the air conditioning unit, and considering the cooling demand, the cooling operation characteristics of each air conditioning unit, and the degree of local temperature anomaly, the required air volume and temperature setpoint at the location of the local temperature anomaly are predicted, and the overall supply air temperature setpoint of the air conditioning unit is adjusted, including: For locations with localized temperature anomalies within the computer room, after matching the optimal localized temperature anomaly control air conditioning unit, the following data are obtained: the degree of temperature anomaly in the anomaly area, the number of server racks covered by the anomaly area, the rack layout density, the location of hot and cold aisles, airflow velocity, total cooling capacity demand, inherent attribute parameters of the air conditioning unit, current operating load rate, cooling capacity, supply air volume, supply air temperature, and cooling efficiency. This data serves as the control dataset. Features strongly correlated with air volume and temperature are extracted, and the LightGBM algorithm is used for training to establish a control temperature-air volume prediction model. The model outputs the required air volume and temperature setpoints for the relevant air vents at the localized temperature anomaly location, as well as the total supply air temperature setpoint of the control air conditioning unit.

8. The energy-saving operation method for air conditioning in a data center computer room according to claim 1, characterized in that, In S4, the objective is to minimize the energy consumption of the air conditioning units during coordinated operation, which is expressed as: ； Let be the energy consumption coefficient of the s-th air conditioning unit; The cooling capacity provided by the s-th air conditioning unit to the j-th abnormal location area; With the goal of load balancing for air conditioning units, it can be represented as: ; The goal of achieving optimal matching between cooling supply and demand is expressed as: ； Let s be the air volume supplied by the s-th air conditioning unit when cooling the j-th abnormal location area; Let be the supply air temperature difference when the s-th air conditioning unit cools the j-th abnormal location area; Let be the cooling efficiency of the s-th air conditioning unit; The effective spatial coverage of the s-th air conditioning unit for the j-th abnormal location area; Let be the attenuation coefficient of cooling capacity delivery when the s-th air conditioning unit is cooling the j-th abnormal location area; Let J be the airflow at the air outlet of the j-th abnormal location area; The temperature of the air outlet in the j-th abnormal location area; Let be the temperature of the air inlet in the j-th abnormal location area.

9. The energy-saving operation method for air conditioning in a data center computer room according to claim 1, characterized in that, In S4, the constraints of the first-layer air conditioning dynamic grouping optimization model are set, including: space unit coverage constraints, air conditioning health status constraints, cooling capacity demand constraints, air conditioning unit load upper and lower limits constraints, and grouping integrity constraints. The constraints for the second-layer multi-air conditioning optimization operation model are set as follows: cooling supply and demand balance constraints, air conditioning unit cooling capacity adjustment constraints, air conditioning unit air volume adjustment constraints, and air conditioning unit load rate safety constraints.

10. The energy-saving operation method for data center computer room air conditioning according to claim 1, characterized in that, Following S4, the process further includes: iteratively interacting the first-layer air conditioning dynamic grouping optimization model and the second-layer multi-air conditioning optimization operation model to output the optimal control parameters for the local temperature anomaly control air conditioning group and the coordinated cooling of each air conditioner; and executing the optimal control parameters based on the digital twin model of the data center computer room cooling system to determine whether the local temperature anomaly location has disappeared. If it has disappeared, the optimal control parameters are issued; otherwise, the control parameters are corrected until the local temperature anomaly location disappears.

Images