Data center cooling control method and device based on thermal prediction model

Abstract

The present invention discloses a data center cooling control method and device based on a thermal prediction model, the method comprising the following steps: continuously collecting the server rack power of the data center, the cooling temperature and fan speed of the air conditioner, and the rack inlet temperature to construct a data set; training a thermal prediction model based on a long short-term memory network (LSTM) based on the data set to evaluate the rack inlet temperature distribution change in the computer room; inputting the system state information at the current moment into the thermal prediction model, and outputting the predicted value of the rack inlet temperature at the next moment; and using a genetic algorithm to solve the set point of the air conditioner cooling temperature and fan speed with minimum cooling power consumption under the condition of meeting the thermal constraints of the computer room. The cooling control method of the present invention can quickly and accurately predict the rack inlet temperature distribution according to the system state change based on the thermal prediction model of the LSTM, and can actively adjust the cooling temperature and fan speed of multiple air conditioners under the condition of meeting the thermal constraints of the computer room to minimize the cooling power consumption.

Description

Technical Field

[0001] The present invention belongs to the field of cloud data center cooling system control optimization, and in particular relates to a data center cooling control method and device based on a thermal prediction model. Background Art

[0002] As the physical carrier of information system operation, cloud data centers have become an indispensable key infrastructure for the operation of modern information society. At the same time, massive data centers also bring huge energy consumption and carbon emission problems. According to an article published in Science in early 2020, data centers now consume about 3% of the global electricity supply and about 4% of total greenhouse gas emissions, and are growing rapidly. The cooling system is a core component of the data center, providing a safe thermal environment for IT equipment, but it also accounts for more than 40% of the energy consumption of the entire data center. Therefore, conducting research on cooling management optimization is of great significance to the energy efficiency and safe operation of cloud data centers. A key requirement for cooling optimization management is to quickly and accurately predict the temperature distribution of the data center in order to effectively supply and distribute cold air while meeting the server redline temperature limit.

[0003] However, the complex geometry of data centers and the coupling of airflow and heat transfer within each server pose great challenges to the development of thermal models for air-cooled data centers. Existing data center thermal modeling methods can be roughly divided into simplified physical models, computational fluid dynamics / heat transfer models (CFD / HT) models, and data-driven models. Simplified physical models can predict temperature distribution in a very short computing time and can be used for parameter research and rapid prediction of flow field distribution, but their prediction accuracy is poor. CFD / HT-based thermal modeling methods can provide complete and accurate thermal field prediction results, but due to the huge amount of numerical calculations and the need to input a large number of specific geometric parameters, they are only suitable for scheme evaluation in the design phase and cannot be used for rapid simulation and analysis of temperature distribution in data centers during the operation phase. In order to achieve an acceptable trade-off between the above two modeling methods, data-driven deep neural networks are proposed as a promising thermal modeling method. This modeling method is based on the analysis of system data to explore the connection between system state variables (input, internal and output variables) without building a physical model. This method can achieve a good trade-off between model complexity, accuracy and computational overhead, and is suitable for real-time thermal management of large data centers.

[0004] The cooling system is one of the key parts of data center infrastructure. Multiple servers in the computer room generate a lot of heat, and high temperature may cause the performance of IT equipment to degrade or damage. Although existing work has made significant progress in improving the cooling efficiency of data centers. However, in terms of cooling control, it is still quite challenging to design efficient cooling control strategies due to the complex interactions between multiple nonlinear mechanical and electrical systems. Most existing controllers tend to be quite simple, somewhat conservative, and use manual methods to adjust specific equipment, layouts, and configurations. This leaves potential for improving efficiency and automation using more adaptive data-driven technologies. Recently, DeepMind proposed a power usage effectiveness (PUE) prediction model for large-scale data centers and proved that it can reduce the PUE of data centers by controlling the supply water temperature of the cooling tower and the return chilled water temperature. This work mainly considers the energy-saving optimization of the chilled water system, and does not consider the relationship between the air conditioning terminal system and the thermal environment in the computer room. In addition, Google researchers Lazic et al. constructed a linear thermal model that characterizes the temperature distribution of the data center computer room to guide the control of air conditioning cooling parameters. However, the coarse-grained thermal model used in this work still has a lot of room for improvement in the accuracy of thermal field distribution prediction. In addition, some studies have applied deep reinforcement learning (DRL) methods to the optimization of data center cooling systems, learning effective cooling control strategies by interacting with the environment. Considering the requirements of air free cooling data centers for supply air temperature and relative humidity, Van et al. proposed a constrained DRL scheme to control the air flow set points of the supply and exhaust fans in the computer room. However, the DRL model takes a long time to train and is inefficient before converging to the optimal strategy, which may lead to unexpected service quality violations and functional failures.

[0005] Although the above cooling control methods have improved the cooling efficiency of data centers to a certain extent, there are still some problems that have not been completely solved. For example, the simplified physics and linear thermal models cannot accurately describe the impact of various time-varying factors in the computer room on the thermal field distribution of the computer room. In addition, the use of reinforcement learning control methods based on trial and error mechanisms may lead to risks and operational safety issues in data centers. Finally, since the cooling system itself has a certain cooling lag, that is, it may take several minutes or more than ten minutes from the change in the cooling parameter state to the temperature change that causes a specific target. Therefore, the use of passive reactive cooling control strategies is prone to thermal risks. Summary of the invention

[0006] In view of the deficiencies of the above-mentioned prior art, the present invention provides a data center cooling control method and device based on a thermal prediction model. The present invention adopts a data-driven method to construct a thermal prediction model based on a long short-term memory network (LSTM) that can quickly and accurately evaluate the rack inlet temperature distribution of the future computer room. Secondly, based on the predicted rack inlet temperature distribution, the cooling temperature and fan speed set points of multiple air conditioners are actively adjusted in advance to minimize cooling power consumption while meeting the thermal constraints of the computer room.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] In a first aspect, the present invention provides a data center cooling control method based on a thermal prediction model, comprising the following steps:

[0009] The data set is constructed by continuously collecting the power of the server racks in the real data center, the cooling temperature and fan speed of the air conditioner, and the rack inlet temperature data;

[0010] Based on the constructed data set, a thermal prediction model based on a long short-term memory network is trained to evaluate the rack inlet temperature distribution of the computer room; the thermal prediction model based on the long short-term memory network is composed of at least three layers of networks, including an input layer, a hidden layer, and an output layer; a Bayesian optimization algorithm is used to globally optimize multiple hyperparameters of the long short-term memory network to determine the best hyperparameter combination;

[0011] Input the current system status information into the thermal prediction model based on the long short-term memory network, and output the rack inlet temperature distribution at the next moment; the status information includes the rack power, air conditioning cooling temperature and fan speed and rack inlet temperature at the current moment;

[0012] Under the condition of meeting the thermal constraints of the computer room, a genetic algorithm is used to solve the set points of the air conditioning cooling temperature and fan speed with minimum cooling power consumption.

[0013] As a preferred technical solution, when constructing a data set, the power of all racks in the computer room, the cooling temperature of the air conditioner and the fan speed, and the temperature sensor data at three different heights of the top, middle and bottom of the rack are collected once at a set interval.

[0014] As a preferred technical solution, the thermal prediction model based on the long short-term memory network is specifically:

[0015] The rack temperature distribution T at time t-1 rack_inlet [t-1], air conditioning cooling temperature T sup (t-1), fan speed FS(t-1) and rack power P(t-1) are input into the LSTM-based thermal prediction model f to predict the rack temperature distribution T at time t. rack_inlet[t], its expression is:

[0016] T rack_inlet [t] = f(T rack_inlet [t-1], T sup [t-1], FS[t-1], P[t-1]).

[0017] As a preferred technical solution, the genetic algorithm is used to solve the set point of the air conditioning cooling temperature and fan speed with the minimum cooling power consumption, specifically:

[0018] Construct the power consumption model of air conditioners. Assume that the cooling system of the computer room consists of K air conditioners CRAC. The power consumption of CRAC mainly comes from the compressor and fan. Therefore, CRAC k The power consumption is defined as:

[0019]

[0020] The fan power consumption p fan It shows a cubic exponential growth relationship with the speed FS, expressed as:

[0021] P fan =k 1 ·FS 3 +k 2 ·FS 2 +k 3 ·FS+C

[0022] The model parameter k 1 , k 2 , k 3 , C uses the measured data to fit the power consumption of the compressor P comp Defined as:

[0023]

[0024] Where P IT It represents the total power of IT equipment, COP is the coefficient of performance, and the higher the value, the higher the cooling efficiency of the CRAC;

[0025] Maximize the CRAC supply temperature T while satisfying the thermal constraints sup It can effectively improve cooling efficiency and reduce cooling power consumption. Therefore, the total power consumption of the cooling system E cooling The estimate is:

[0026]

[0027] Formulate a multi-constrained optimization problem. The optimization goal is to minimize the cooling power consumption while satisfying the thermal constraints of the computer room. It can be expressed as:

[0028]

[0029] s.tMax(T rack_inlet_pred )≤T threhold

[0030] T sup_LB ≤T sup ≤T sup_UB

[0031] FS LB ≤FS≤FS UB

[0032] The air conditioning supply temperature T sup and fan speed FS represent the optimizable control variables, T sup_LB , T sup_UB , FS LB , FS UB Represent the adjustable upper and lower thresholds of air conditioning temperature and fan speed respectively; in addition, the thermal constraint of the computer room is expressed as the predicted rack inlet temperature T rack_inlet_pred The temperature threshold T cannot be exceeded threhold ;

[0033] A genetic algorithm is used to solve the supply temperature and fan set point of the air conditioner that minimizes power consumption. The specific steps include: (1) Initializing the population and randomly generating multiple candidate solutions, which refers to the cooling set point of the air conditioner, represented by {T sup,1 ,...,T sup，k , FS 1 , ..., FS k}; (2) Evaluate the feasibility and fitness of individuals in the population according to the optimization objectives and constraints; (3) Select individuals with smaller fitness from the population according to a certain probability; (4) Generate new individuals through crossover operation; (5) Generate new individuals through mutation operation; Repeat the above steps (2)-(5) until the number of iterations reaches the maximum number or the end condition is met, and select the individual solution with the smallest fitness as the optimal cooling set point of the computer room air conditioner.

[0034] As a preferred technical solution, according to the data measured by HP Laboratory, COP is related to the air supply temperature T of CRAC. sup There is a positive correlation, which is expressed as:

[0035] COP(T sup )=0.0068T sup 2 +0.0008T sup +0.458.

[0036] As a preferred technical solution, the temperature threshold T threhold =30℃.

[0037] As a preferred technical solution, the thermal prediction model based on the long short-term memory network has 28, 100, and 60 nodes respectively; the activation function is ReLU, the training batch size is 12, and the learning rate is 0.01.

[0038] In a second aspect, the present invention provides a data center cooling control system based on a thermal prediction model, which is applied to the data center cooling control method based on a thermal prediction model, and includes a data acquisition module, a model training module, a model processing module, and a temperature solving module;

[0039] The data acquisition module is used to continuously collect the power of the server racks in the real data center, the cooling temperature and fan speed of the air conditioner, and the rack inlet temperature data to construct a data set;

[0040] The model training module is used to train a thermal prediction model based on a long short-term memory network based on the constructed data set to evaluate the rack inlet temperature distribution of the computer room; the thermal prediction model based on the long short-term memory network is composed of at least three layers of networks, including an input layer, a hidden layer and an output layer; a Bayesian optimization algorithm is used to globally optimize multiple hyperparameters of the long short-term memory network to determine the best hyperparameter combination;

[0041] The model processing module is used to input the system status information at the current moment into the thermal prediction model based on the long short-term memory network, and output the rack inlet temperature distribution at the next moment; the status information includes the rack power, air conditioning cooling temperature and fan speed and rack inlet temperature at the current moment;

[0042] The temperature solving module is used to use a genetic algorithm to solve the set points of the air-conditioning cooling temperature and the fan speed with the minimum cooling power consumption while satisfying the thermal constraints of the computer room.

[0043] In a third aspect, the present invention provides an electronic device, the electronic device comprising:

[0044] at least one processor; and,

[0045] a memory communicatively connected to the at least one processor; wherein,

[0046] The memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor to enable the at least one processor to perform the data center cooling control method based on the thermal prediction model.

[0047] In a fourth aspect, the present invention provides a computer-readable storage medium storing a program, which, when executed by a processor, implements the data center cooling control method based on a thermal prediction model.

[0048] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0049] (1) The long short-term memory network (LSTM) model adopted in the present invention adopts the Bayesian (BO) optimization algorithm to determine the optimal hyperparameter combination, which has better prediction performance.

[0050] (2) This paper proposes a thermal prediction model based on long short-term memory network (LSTM), which uses the powerful nonlinear fitting ability of deep neural network to characterize the complex thermodynamic evolution characteristics in the computer room and accurately evaluate the rack inlet temperature distribution of the future computer room. In addition, compared with the thermal model based on CFD and the thermal model based on simplified physics, the proposed thermal prediction model based on LSTM can achieve a better trade-off between model complexity, accuracy and computational overhead.

[0051] (3) The present invention proposes a data center cooling control method based on a thermal prediction model. The method can sense the temperature distribution changes in the computer room and use a genetic algorithm to solve the supply temperature and fan speed set points of multiple air conditioners in the computer room. Under the premise of meeting the cooling demand, the total power consumption of the air conditioner is minimized, thereby improving the overall energy efficiency of the data center. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0053] Figure 1 This is a flow chart of a data center cooling control method based on a thermal prediction model according to an embodiment of the present invention;

[0054] Figure 2 Provide a schematic diagram of the data center infrastructure layout and rack sensor deployment locations;

[0055] Figure 3 This is a schematic diagram of the cold aisle in the middle of the rack row;

[0056] Figure 4 Schematic diagram of a heat prediction model based on a long short-term memory network (LSTM);

[0057] Figure 5 Schematic diagram of cooling setting optimization based on genetic algorithm;

[0058] Figure 6 It is a structural schematic diagram of a data center cooling control system based on a thermal prediction model according to an embodiment of the present invention;

[0059] Figure 7 4 is a structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application.

[0061] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments.

[0062] like Figure 1 As shown, a data center cooling control method based on a thermal prediction model proposed in this embodiment includes the following steps:

[0063] S1. Continuously collect the power of the server racks in the real data center, the cooling temperature and fan speed of the air conditioner, and the rack inlet temperature data to build a data set.

[0064] Furthermore, in step S1, the power of all racks in the computer room, the cooling temperature and fan speed of the air conditioner, and the temperature sensor data at three different heights of the upper, middle and lower racks are collected every 15 minutes, and divided into a training set and a test set in a ratio of 8:2; the test set is used to verify the model performance.

[0065] S2. Based on the dataset constructed in S1, a thermal prediction model based on the long short-term memory network (LSTM) is trained to evaluate the rack inlet temperature changes in the computer room.

[0066] Furthermore, the LSTM-based thermal prediction model is composed of at least three layers of networks, including an input layer, a hidden layer, and an output layer. Specifically, the rack temperature distribution T at time t-1 is rack_inlet [t-1], air conditioning cooling temperature T sup(t-1), fan speed FS(t-1) and rack power P(t-1) are input into the LSTM-based thermal prediction model f to predict the rack temperature distribution T at time t. rack_inlet [t], its expression is,

[0067] T rack_inlet [t] = f(T rack_inlet [t-1], T sup [t-1], FS[t-1], P[t-1]), (1)

[0068] S3. Input the current system status information (the current rack power, air conditioning cooling temperature and fan speed, and rack inlet temperature) into the LSTM-based thermal prediction model, and output the rack inlet temperature distribution at the next moment.

[0069] S4. Under the condition that the thermal constraints of the computer room are met, a genetic algorithm (GA) is used to solve the set points of the air conditioning cooling temperature and fan speed with the minimum cooling power consumption.

[0070] Furthermore, step S4 is specifically as follows:

[0071] S401. Construct a power consumption model for air conditioners. Assume that the cooling system of the computer room consists of K air conditioners (CRACs). The power consumption of CRACs mainly comes from compressors and fans. Therefore, k The power consumption can be defined as,

[0072]

[0073] The fan power consumption P fan It has a cubic exponential growth relationship with the speed FS, which can be expressed as:

[0074] P fan =k 1 ·FS 3 +k 2 ·FS 2 +k 3 FS+C, (3)

[0075] The model parameter k 1 , k 2 , k 3 , C needs to be fitted using measured data. In addition, the power consumption of the compressor P comp Defined as,

[0076]

[0077] Where P ITIt represents the total power of IT equipment. In addition, COP is the coefficient of performance. The higher the value, the higher the cooling efficiency of CRAC. According to the data measured by HP Laboratory, COP is related to the air supply temperature T of CRAC. sup There is a positive correlation, which is expressed as,

[0078] COP(T sup )=0.0068T sup 2 +0.0008T sup +0.458, (5)

[0079] From formula (5), we can see that under the condition of satisfying the thermal constraints, the supply temperature T of CRAC is maximized. sup It can effectively improve cooling efficiency and reduce cooling power consumption. Therefore, the total power consumption of the cooling system E cooling It can be estimated as,

[0080]

[0081] S402. Formulate a multi-constrained optimization problem. The optimization goal of the present invention is to minimize cooling power consumption while satisfying the thermal constraints of the computer room, which can be expressed as:

[0082]

[0083] The air conditioning supply temperature T sup and fan speed FS represent the optimizable control variables, T sup_LB , T sup_UB , FS LB ,FS UB Represent the adjustable upper and lower thresholds of air conditioning temperature and fan speed respectively. In addition, the thermal constraint of the computer room is expressed as the predicted rack inlet temperature T rack_inlet_pred The temperature threshold T cannot be exceeded threhold =30℃.

[0084] S403. Using a genetic algorithm (GA) to solve the supply temperature and fan set point of the air conditioner that minimizes power consumption, the specific steps include:

[0085] (1) Initialize the population and randomly generate multiple candidate solutions, which refers to the cooling set point of the air conditioner, which can be expressed as {T sup,1 ,...,T sup,k ,FS 1 , ..., FS k};

[0086] (2) Evaluate the feasibility and fitness of individuals in the population according to the optimization objectives and constraints (Formula 6);

[0087] (3) Select individuals with smaller fitness from the population according to a certain probability;

[0088] (4) Generate new individuals through crossover operation;

[0089] (5) Generate new individuals through mutation operations.

[0090] Repeat steps 2-5 above until the number of iterations reaches the maximum number or the end condition is met, and select the individual solution with the smallest fitness as the optimal cooling set point for the computer room air conditioner.

[0091] In another specific embodiment of the present application, the present invention is applied to a data center using a raised floor cooling method. The data center infrastructure layout and rack sensor deployment locations of this embodiment are as follows: Figure 2 The parameter configuration of the data center is shown in Table 1. The cooling system of the computer room has 4 precision air conditioners, and 20 racks are divided into 4 rows, arranged face to face and back to back to form hot and cold channels. Figure 3 As shown in the figure, the front of the cabinet is a cold channel and a ventilation floor is laid. The cold air of the precision air conditioner enters from the front of the cabinet, takes away the heat of the server, and then naturally returns to the precision air conditioner from the top of the back of the cabinet. In addition, in order to collect the server inlet temperature data, three temperature sensors at different heights are deployed on the front of each rack.

[0092] Table 1

[0093]

[0094] In this example, real data of the data center (cooling temperature and fan speed of 4 CRACs, operating power of 20 racks, and 60 temperature sensor data) were collected every 15 minutes for 10 days, with a total of 960 sets of sample data, which were divided into training set and test set in a ratio of 8:2. Figure 4 As shown in the figure, the LSTM model has three network layers, with 28, 100, and 60 nodes respectively. The activation function of the network model is ReLU, the training batch size is 12, and the learning rate is 0.01. The prediction results show that the prediction error RMSE value of the LSTM-based thermal prediction model is about 1.1℃, the MAPE is 3.1%, and the calculation time is 3.5ms, which can meet the needs of prediction accuracy and real-time prediction.

[0095] In order to test the performance of the cooling control method under real working conditions, this embodiment deploys the trained LSTM-based thermal prediction model and GA-based cooling control algorithm to the cooling control system of the above data center, collects the power, air conditioning cooling set point and current temperature distribution of the racks in the computer room in real time every 15 minutes, and uses the thermal prediction model to predict the temperature distribution of the computer room at the next moment. Figure 5 As shown in the figure, a genetic algorithm (GA) is used to solve the cooling temperature and fan speed set points of the air conditioner with the minimum power consumption. The key steps include: (1) Evaluate the feasibility and fitness of the individuals in the population by optimizing the objectives and constraints, and determine whether the algorithm iteration stop condition is met (the maximum number of iterations is set to 30). (2) New individual solutions are continuously generated through the selection, crossover and mutation operations of the genetic algorithm. (3) Until the stopping criterion is met, the individual with the smallest fitness value is output as the optimal solution to determine the cooling temperature and fan speed set points of multiple air conditioners in the computer room.

[0096] After 10 consecutive days of observation and data statistics, compared with the original PID-based cooling control method, this method can increase the average cooling temperature of the air conditioner by 1°C without any thermal risk, thereby effectively reducing the cooling energy consumption of the data center.

[0097] It should be noted that, for the sake of convenience, the aforementioned method embodiments are all expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited to the described order of actions, because according to the present invention, certain steps can be performed in other orders or simultaneously.

[0098] Based on the same idea as the data center cooling control method based on the thermal prediction model in the above embodiment, the present invention also provides a data center cooling control system based on the thermal prediction model, which can be used to execute the above data center cooling control method based on the thermal prediction model. For ease of explanation, the structural schematic diagram of the data center cooling control system embodiment based on the thermal prediction model only shows the parts related to the embodiment of the present invention. Those skilled in the art can understand that the illustrated structure does not constitute a limitation on the device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0099] See also Figure 6 In another embodiment of the present application, a data center cooling control system 100 based on a thermal prediction model is provided, the system comprising a data acquisition module 101, a model training module 102, a model processing module 103 and a temperature solving module 104;

[0100] The data acquisition module 101 is used to continuously collect the power of the server racks in the real data center, the cooling temperature and fan speed of the air conditioner, and the rack inlet temperature data to construct a data set;

[0101] The model training module 102 is used to train a thermal prediction model based on a long short-term memory network based on the constructed data set to evaluate the rack inlet temperature distribution of the computer room; the thermal prediction model based on the long short-term memory network is composed of at least three layers of networks, including an input layer, a hidden layer and an output layer; a Bayesian (BO) optimization algorithm is used to perform global optimization on multiple hyperparameters of the long short-term memory network to determine the best hyperparameter combination;

[0102] The model processing module 103 is used to input the current system status information into the thermal prediction model based on the long short-term memory network, and output the rack inlet temperature distribution at the next moment; the status information includes the rack power, air conditioning cooling temperature and fan speed and rack inlet temperature at the current moment;

[0103] The temperature solving module 104 is used to use a genetic algorithm to solve the set points of the air-conditioning cooling temperature and the fan speed with the minimum cooling power consumption while satisfying the thermal constraints of the computer room.

[0104] It should be noted that the data center cooling control system based on the thermal prediction model of the present invention corresponds one-to-one to the data center cooling control method based on the thermal prediction model of the present invention. The technical features and beneficial effects described in the above-mentioned embodiment of the data center cooling control method based on the thermal prediction model are applicable to the embodiment of the data center cooling control method based on the thermal prediction model. For specific contents, please refer to the description in the embodiment of the method of the present invention, which will not be repeated here. This is hereby declared.

[0105] In addition, in the implementation of the data center cooling control system based on the thermal prediction model in the above-mentioned embodiment, the logical division of each program module is only an example. In actual applications, the above-mentioned functions can be assigned to different program modules as needed, for example, for the configuration requirements of the corresponding hardware or the convenience of software implementation. That is, the internal structure of the data center cooling control system based on the thermal prediction model is divided into different program modules to complete all or part of the functions described above.

[0106] See also Figure 7 In one embodiment, an electronic device for implementing a data center cooling control method based on a thermal prediction model is provided. The electronic device 200 may include a first processor 201, a first memory 202 and a bus, and may also include a computer program stored in the first memory 202 and executable on the first processor 201, such as a data center cooling control program 203 based on a thermal prediction model.

[0107] Among them, the first memory 202 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the first memory 202 can be an internal storage unit of the electronic device 200, such as a mobile hard disk of the electronic device 200. In other embodiments, the first memory 202 can also be an external storage device of the electronic device 200, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 200. Further, the first memory 202 can also include both an internal storage unit of the electronic device 200 and an external storage device. The first memory 202 can not only be used to store application software and various types of data installed in the electronic device 200, such as the code of the data center cooling control program 203 based on the thermal prediction model, but also can be used to temporarily store data that has been output or is to be output.

[0108] In some embodiments, the first processor 201 may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The first processor 201 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes various functions and processes data of the electronic device 200 by running or executing programs or modules stored in the first memory 202, and calling data stored in the first memory 202.

[0109] Figure 7 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 7 The structure shown does not constitute a limitation on the electronic device 200 , and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0110] The data center cooling control program 203 based on the thermal prediction model stored in the first memory 202 of the electronic device 200 is a combination of multiple instructions. When running in the first processor 201, it can achieve:

[0111] The data set is constructed by continuously collecting the power of the server racks in the real data center, the cooling temperature and fan speed of the air conditioner, and the rack inlet temperature data;

[0112] Based on the constructed data set, a thermal prediction model based on a long short-term memory network is trained to evaluate the rack inlet temperature distribution of the computer room; the thermal prediction model based on the long short-term memory network is composed of at least three layers of networks, including an input layer, a hidden layer and an output layer; a Bayesian (BO) optimization algorithm is used to globally optimize multiple hyperparameters of the long short-term memory network to determine the best hyperparameter combination;

[0113] Input the current system status information into the thermal prediction model based on the long short-term memory network, and output the rack inlet temperature distribution at the next moment; the status information includes the rack power, air conditioning cooling temperature and fan speed and rack inlet temperature at the current moment;

[0114] Under the condition of meeting the thermal constraints of the computer room, a genetic algorithm is used to solve the set points of the air conditioning cooling temperature and fan speed with minimum cooling power consumption.

[0115] Furthermore, if the module / unit integrated in the electronic device 200 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile computer-readable storage medium. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0116] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0117] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0118] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. Data center cooling control method based on thermal prediction model, It is characterized in that The steps include: The data set is constructed by continuously collecting the power of the server racks in the real data center, the cooling temperature and fan speed of the air conditioner, and the rack inlet temperature data; Based on the constructed data set, a thermal prediction model based on a long short-term memory network is trained to evaluate the rack inlet temperature distribution of the computer room; the thermal prediction model based on the long short-term memory network is composed of at least three layers of networks, including an input layer, a hidden layer, and an output layer; a Bayesian optimization algorithm is used to globally optimize multiple hyperparameters of the long short-term memory network to determine the best hyperparameter combination; Input the current system status information into the thermal prediction model based on the long short-term memory network, and output the rack inlet temperature distribution at the next moment; the status information includes the rack power, air conditioning cooling temperature and fan speed and rack inlet temperature at the current moment; Under the condition of meeting the thermal constraints of the computer room, a genetic algorithm is used to solve the set points of the air conditioning cooling temperature and fan speed with the minimum cooling power consumption; The genetic algorithm is used to solve the set point of the air conditioning cooling temperature and fan speed with the minimum cooling power consumption, specifically: Construct the power consumption model of air conditioners. Assume that the cooling system of the computer room consists of K air conditioners CRAC. The power consumption of CRAC mainly comes from the compressor and fan. Therefore, CRAC k The power consumption is defined as: The fan power consumption P fan It shows a cubic exponential growth relationship with the speed FS, expressed as: P fan =k 1 ·FS 3 +k 2 ·FS 2 +k 3 ·FS+C The model parameter k 1 ,k 2 ,k 3 ,C uses the measured data to fit the power consumption P of the compressor comp Defined as: Where P IT It represents the total power of IT equipment, COP is the coefficient of performance, and the higher the value, the higher the cooling efficiency of the CRAC; Maximize the CRAC supply temperature T while satisfying the thermal constraints sup It can effectively improve cooling efficiency and reduce cooling power consumption. Therefore, the total power consumption of the cooling system E cooling The estimate is: Formulate a multi-constrained optimization problem. The optimization goal is to minimize the cooling power consumption while satisfying the thermal constraints of the computer room. It can be expressed as: s.tMax(T rack_inletpred )≤T threhold T sup_LB ≤T sup ≤T sup_UB FS LB ≤FS≤FS UB The air conditioning supply temperature T sup and fan speed FS represent the optimizable control variables, T sup_LB , T sup_UB , FS LB , FS UB Represent the adjustable upper and lower thresholds of air conditioning temperature and fan speed respectively; in addition, the thermal constraint of the computer room is expressed as the predicted rack inlet temperature T rack_inlet_pred The temperature threshold T cannot be exceeded threhold ; A genetic algorithm is used to solve the supply temperature and fan set point of the air conditioner that minimizes power consumption. The specific steps include: (1) Initializing the population and randomly generating multiple candidate solutions, which refers to the cooling set point of the air conditioner, represented by {T sup,1 ,...,T sup，k , FS 1 , ..., FS k }; (2) Evaluate the feasibility and fitness of individuals in the population according to the optimization objectives and constraints; (3) Select individuals with smaller fitness from the population according to a certain probability; (4) Generate new individuals through crossover operation; (5) Generate new individuals through mutation operation; Repeat the above steps (2)-(5) until the number of iterations reaches the maximum number or the end condition is met, and select the individual solution with the smallest fitness as the optimal cooling set point of the computer room air conditioner.

2. According to the data center cooling control method based on the thermal prediction model according to claim 1, It is characterized in that When constructing the data set, the power of all racks in the computer room, the cooling temperature of the air conditioner and the fan speed, and the temperature sensor data at three different heights of the top, middle and bottom of the rack are collected once at a set interval.

3. According to the data center cooling control method based on the thermal prediction model of claim 1, It is characterized in that The thermal prediction model based on the long short-term memory network is specifically as follows: The rack temperature distribution T at time t-1 rack_inlet [t-1], air conditioning cooling temperature T sup (t-1), fan speed FS(t-1) and rack power P(t-1) are input into the LSTM-based thermal prediction model f to predict the rack temperature distribution T at time t. rack_inlet [t], its expression is: T rack_inlet [t]=f(T rack_inlet [t-1],T sup [t-1],FS[t-1],P[t-1])。 4. According to the data center cooling control method based on the thermal prediction model of claim 1, It is characterized in that According to the data measured by HP Labs, COP is related to the air supply temperature T of CRAC. sup There is a positive correlation, which is expressed as: COP(T sup )=0.0068T sup 2 +0.0008T sup +0.458。 5. According to the data center cooling control method based on the thermal prediction model of claim 1, It is characterized in that Temperature threshold T threhold =30℃.

6. According to the data center cooling control method based on the thermal prediction model of claim 1, It is characterized in that The thermal prediction model based on the long short-term memory network has 28, 100, and 60 nodes respectively; the activation function is ReLU, the training batch size is 12, and the learning rate is 0.

01.

7. Data center cooling control system based on thermal prediction model, It is characterized in that A data center cooling control method based on a thermal prediction model applied to any one of claims 1-6, comprising a data acquisition module, a model training module, a model processing module and a temperature solution module; The data acquisition module is used to continuously collect the power of the server racks in the real data center, the cooling temperature and fan speed of the air conditioner, and the rack inlet temperature data to construct a data set; The model training module is used to train a thermal prediction model based on a long short-term memory network based on the constructed data set to evaluate the rack inlet temperature distribution of the computer room; the thermal prediction model based on the long short-term memory network is composed of at least three layers of networks, including an input layer, a hidden layer and an output layer; a Bayesian optimization algorithm is used to globally optimize multiple hyperparameters of the long short-term memory network to determine the best hyperparameter combination; The model processing module is used to input the system status information at the current moment into the thermal prediction model based on the long short-term memory network, and output the rack inlet temperature distribution at the next moment; the status information includes the rack power, air conditioning cooling temperature and fan speed and rack inlet temperature at the current moment; The temperature solving module is used to use a genetic algorithm to solve the set points of the air-conditioning cooling temperature and the fan speed with the minimum cooling power consumption while satisfying the thermal constraints of the computer room.

8. An electronic device, It is characterized in that The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor so that the at least one processor can execute the data center cooling control method based on the thermal prediction model as described in any one of claims 1-6.

9. A computer-readable storage medium storing a program, It is characterized in that When the program is executed by a processor, the data center cooling control method based on a thermal prediction model described in any one of claims 1 to 6 is implemented.

Images