Method and device for dynamically controlling cooling process for intelligent data center

Abstract

The embodiment of the invention discloses a method and device for dynamically controlling the cooling process for an intelligent data center, and the method comprises the steps: determining a control action based on the current data of a to-be-processed data center, and enabling the current data to comprise the current air inlet temperature of a rack, the total power of a system and the IT load, the control actions comprise the cooling water supply temperature, the air supply temperature of the CRAC and the flow speed of the CRAC; inputting the control action into a preset thermodynamic model for steady-state simulation to obtain an optimized control action; and taking the optimized control action as the input of an energy model, and carrying out cooling control on the data center to be processed.

Description

Technical Field

[0001] This specification relates to the fields of computer technology and cooling control technology, and in particular to a method and apparatus for dynamically controlling the cooling process for intelligent data centers. Background Technology

[0002] As a crucial component of modern information technology infrastructure, data centers are experiencing rapid growth in scale and complexity due to the rapid development of new technologies such as cloud computing, big data, and big data models. The servers and equipment in data centers generate significant heat during operation. If this heat is not dissipated promptly, it can create localized hotspots, leading to decreased server performance and shortened equipment lifespan, ultimately impacting the stability and security of the data center. Therefore, data center cooling is of paramount importance.

[0003] In existing technologies, data center cooling control is generally implemented using Model Predictive Control (MPC) and Deep Reinforcement Learning (DRL) methods. MPC is widely used due to its proactive control capabilities; however, it relies on precise cooling control models, which are often difficult to obtain in practical applications, thus hindering its practical application. To overcome this challenge, researchers have attempted to obtain dynamic models through data-driven methods and combine them with MPC. However, the accuracy of these methods is limited by the availability of exploratory data and raises thermal safety compliance issues. DRL methods, by directly learning control strategies that interact with the cooling system, have demonstrated significant energy-saving potential, overcoming the limitations of model dependence. Nevertheless, DRL methods also require exploratory data during training and may violate thermal safety compliance. To improve thermal safety, some studies employ reward shaping techniques, adding penalty terms to the reward function. However, this method cannot guarantee compliance with safety constraints and is difficult to avoid safety violations in the early stages of learning. Recently, some works have used post-rectification techniques to project DRL actions onto a safety set, effectively ensuring thermal safety, but without considering detailed temperature distribution and state-related constraints.

[0004] Advances in security reinforcement learning have provided new approaches to handling Constrained Markov Decision Processes (CMDPs), broadly categorized into security policy optimization and security exploration. In security policy optimization, the goal is to simultaneously consider reward maximization and constraint satisfaction during the policy optimization process. These methods typically employ cumulative discounted cost constraints and usually use primal-dual methods or construct approximate convex programming to solve the optimization problem. However, they suffer from two main problems: the need for an initial feasible policy and the inability to guarantee satisfaction of specific state or time-averaged constraints. In security exploration, existing work only considers state safety constraints, preventing access to insecure states by projecting actions into safe sets, but this approach cannot handle time-averaged constraints. Therefore, how to improve the efficiency and security of data center cooling control while ensuring satisfaction of time-averaged constraints is one of the urgent problems to be solved.

[0005] Based on this, embodiments of this specification provide a method for dynamically controlling the cooling process of intelligent data centers. Summary of the Invention

[0006] This specification provides a method and apparatus for dynamically controlling the cooling process of intelligent data centers, addressing the following technical problems: Existing technologies cannot solve the following two issues: they require an initial feasible strategy and cannot guarantee satisfaction of specific state or time-averaged constraints. Regarding security exploration, existing work only considers state security constraints, preventing access to unsafe states by projecting actions into safe sets; however, this method cannot handle time-averaged constraints.

[0007] To solve the above-mentioned technical problems, the embodiments in this specification are implemented as follows:

[0008] This specification provides an embodiment of a method for dynamically controlling the cooling process in intelligent data centers, including:

[0009] Based on the current data of the data center to be processed, control actions are determined. The current data includes: the current rack inlet temperature, the total system power, and the IT load. The control actions include: the cooling water supply temperature, the CRAC supply air temperature, and the CRAC flow rate.

[0010] The control action is input into a preset thermodynamic model for steady-state simulation to obtain an optimized control action.

[0011] The optimized control actions are used as input to the energy model for cooling control of the data center to be processed.

[0012] This specification also provides an embodiment of a device for dynamically controlling the cooling process in intelligent data centers, comprising:

[0013] The control action determination module determines the control action based on the current data of the data center to be processed. The current data includes: the current rack inlet temperature, the total system power and the IT load. The control action includes: the cooling water supply temperature, the CRAC supply air temperature and the CRAC flow rate.

[0014] The control action optimization module inputs the control action into a preset thermodynamic model for steady-state simulation to obtain optimized control actions;

[0015] The cooling control module uses the optimized control actions as input to the energy model to perform cooling control of the data center to be processed.

[0016] This specification provides a method for dynamically controlling the cooling process of intelligent data centers. It develops a coupled thermal simulation model, introduces a new CMDP formula to evaluate compliance with the ASHRAE thermal guidelines, and introduces a virtual queue with RCI constraints, deriving a virtual queue modulation safety set using Lyapunov stability theory. This significantly reduces data center energy consumption without violating any thermal safety compliance, enabling dynamic control of intelligent data centers. It offers high performance and security, and can reduce energy consumption in specific industrial applications (such as pathological images and defect images), demonstrating high industrial value. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic diagram of a system architecture for a method of dynamically controlling the cooling process in a smart data center, provided as an embodiment of this specification;

[0019] Figure 2 A flowchart illustrating a method for dynamically controlling the cooling process in a smart data center, provided as an embodiment of this specification.

[0020] Figure 3 A framework diagram of a method for dynamically controlling the cooling process in a smart data center, provided as an embodiment of this specification;

[0021] Figure 4 This is a schematic diagram of a device for dynamically controlling the cooling process of a smart data center, provided as an embodiment of this specification. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0023] Figure 1 This is a schematic diagram of a system architecture for a method of dynamically controlling the cooling process in a smart data center, provided as an embodiment of this specification. Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0024] Terminal devices 101, 102, and 103 interact with server 105 via network 104 to receive or send messages, etc. Various client applications can be installed on terminal devices 101, 102, and 103, such as specialized programs for dynamically controlling cooling processes in intelligent data centers.

[0025] Terminal devices 101, 102, and 103 can be either hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various dedicated or general-purpose electronic devices, including but not limited to smartphones, tablets, laptops, and desktop computers. When terminal devices 101, 102, and 103 are software, they can be installed in the electronic devices listed above. They can be implemented as multiple software programs or software modules (e.g., multiple software programs or software modules used to provide distributed services) or as a single software program or software module.

[0026] Server 105 can be a server that provides various services, such as a backend server that provides services to client applications installed on terminal devices 101, 102, and 103. For example, the server can control a dynamic control cooling process for intelligent data centers, so as to display the control results of the dynamic control cooling process for intelligent data centers on terminal devices 101, 102, and 103. The server can also perform dynamic control cooling processes for intelligent data centers, so as to display the control results of the dynamic control cooling process for intelligent data centers on terminal devices 101, 102, and 103.

[0027] Server 105 can be either hardware or software. When server 105 is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When server 105 is software, it can be implemented as multiple software programs or software modules (e.g., multiple software programs or software modules used to provide distributed services), or as a single software program or software module.

[0028] Figure 2 This is a flowchart illustrating a method for dynamically controlling the cooling process in a smart data center, as provided in an embodiment of this specification. From a programming perspective, the execution entity of the process can be a program mounted on an application server or application terminal. It is understood that this method can be executed by any device, equipment, platform, or cluster of devices with computing and processing capabilities. Figure 2 As shown, the method for dynamically controlling the cooling process includes:

[0029] Step S201: Based on the current data of the data center to be processed, determine the control action. The current data includes: the current rack inlet temperature, the total system power, and the IT load. The control action includes: the cooling water supply temperature, the CRAC supply air temperature, and the CRAC flow rate.

[0030] In the embodiments of this specification, the data center to be processed is an intelligent data center, which is an intelligent data center that requires cooling process control.

[0031] In the embodiments of this specification, the control action is determined based on the current data of the data center to be processed, that is, based on the data of the current state of the data center to be processed.

[0032] Specifically, based on the current data of the data center to be processed, the control action is determined, which includes: the intelligent control body determines the control action based on the monitored current data of the data center to be processed, and uses the acquired control action as the input of the preset thermodynamic model.

[0033] In the embodiments of this specification, the total system power is the total power of the data center to be processed, and the IT load is the workload, which is an external random variable that changes with the different working devices in the data center to be processed and the different working stages.

[0034] CRAC (Computer Room Air Conditioning) units, in this embodiment of the specification, refer to the data center being processed using CRAC units. When determining control actions, it is necessary to determine the supply air temperature and flow rate of the CRAC unit, referred to as the CRAC supply air temperature and CRAC flow rate, respectively.

[0035] Step S203: Input the control action into a preset thermodynamic model for steady-state simulation to obtain optimized control action.

[0036] Since the control action corresponding to the current data is not the optimal control action, it is necessary to optimize the control action corresponding to the current data. In the embodiments of this specification, the control action corresponding to the current data is input into a preset thermodynamic model for optimization in order to obtain an optimized control action.

[0037] To further explain the solutions provided in the embodiments of this specification, the thermodynamic model will be described in detail below.

[0038] The preset thermodynamic model described in the embodiments of this specification is a differentiable model based on POD technology;

[0039] The preset thermodynamic model is as follows:

[0040] T[k+1]=f(P IT [k],T sup[k],V sup[k]) T ·φ

[0041] in,

[0042] T[k+1] represents the (k+1)th control action;

[0043] P IT [k] represents the IT load power consumption during the kth control action;

[0044] Tsup[k] is the air supply temperature of the CRAC during the kth control action;

[0045] Vsup[k] is the flow rate of the CRAC during the k-th control action;

[0046] f(·) represents a continuously differentiable function that maps the k-th control action to the POD;

[0047] φ represents the vector of the first k PODs.

[0048] POD technology is a surrogate model that can obtain the mature temperature field under given boundary conditions in real time, approximating the temperature field through a linear combination of orthogonal POD modes. Therefore, the embodiments in this specification use POD technology to perform real-time temperature simulation. POD technology is prior art and does not constitute a limitation of this application, and will not be described in detail here.

[0049] In the embodiments described in this specification, the preset thermodynamic model is a constrained Markov decision process;

[0050] The preset thermodynamic model's state s[k] includes: the rack inlet temperature Tin[k] at the k-th control action and the IT load power consumption P at the k-th control action. IT [k];

[0051] The preset thermodynamic model action a[k] includes: the CRAC supply air temperature Tsup[k] and the CRAC flow rate Vsup[k] at the kth control action, and the cooling water supply temperature Tchw[k] at the kth control action;

[0052] The reward r[k] of the preset thermodynamic model includes: the grid carbon emission factor e[k] at the kth control action and the total energy consumption Etol[k] of the data center to be processed at the kth control action;

[0053] The constraints of the preset thermodynamic model include RCI constraints and hard constraints.

[0054] In the embodiments described in this specification, the reward of the preset thermodynamic model is the reward for reducing carbon emissions.

[0055] In the embodiments of this specification, the preset thermodynamic model is a constrained Markov decision process (CDMP). Based on the state, action, reward and constraint conditions of the CMDP, the objective function of the CMDP can be established, that is, the objective function is established.

[0056] In a specific embodiment, the recommended maximum rack inlet temperature T is 27°C; the recommended maximum rack inlet temperature T is 32°C. If the rack inlet temperature exceeds the recommended maximum rack inlet temperature... This will trigger costs, so it is necessary to limit the cost over an average time period to within a threshold range.

[0057] In the embodiments described in this specification, the RCI constraint is:

[0058]

[0059] in,

[0060] ψ is the cost threshold;

[0061] Maximum permissible rack inlet temperature;

[0062] The recommended maximum rack inlet temperature;

[0063] K is the number of time steps;

[0064] Let be the intake air temperature of the i-th server at time step k;

[0065] This means that for all i = 1, 2, 3, ..., the above condition must be true.

[0066] The existence of the RCI constraint ensures that the RCI index is less than ψ under all circumstances.

[0067] The hard constraint is:

[0068]

[0069] in,

[0070] Tin[k] is the rack inlet temperature at the kth control action;

[0071] Maximum permissible rack inlet temperature;

[0072] This means that for all i = 1, 2, 3, ..., the above condition must be true.

[0073] In the embodiments described in this specification, the purpose of the hard constraint is to ensure that the rack inlet temperature remains at the recommended maximum rack inlet temperature at all times. The following is an explanation of how, due to hard constraints, the rack inlet temperature is consistently maintained at the recommended maximum rack inlet temperature. the following.

[0074] The objective function of the preset thermodynamic model is:

[0075]

[0076] in,

[0077] π θ Represents strategy θ;

[0078] k is the time step;

[0079] γ k Let be the discount factor for the cost function at the k-th time step;

[0080] r k The reward for the preset thermodynamic model.

[0081] By using the objective function of this pre-defined thermodynamic model, we can obtain the strategy π that maximizes the expected reward. θ It also meets thermal safety constraints, meaning that the RCI is less than ψ under all conditions, while ensuring that the rack inlet temperature is always maintained at the recommended maximum rack inlet temperature. the following.

[0082] To provide a strong guarantee for the satisfaction of the time averaging constraint, this specification further transforms the RCI constraint into a control problem.

[0083] In the embodiments described in this specification, the RCI constraint is obtained based on instantaneous cost and virtual queue, wherein,

[0084] The instantaneous cost c i (k) is:

[0085]

[0086] The virtual queue is

[0087] q i [k+1]=max{q i [k]+c i [k],0}≥0

[0088] in,

[0089] q i [k] represents the virtual queue at the kth control action;

[0090] q i [k+1] represents the virtual queue during the (k+1)th control action;

[0091] The stable strategy π of the virtual queue θ Satisfying RCI constraints:

[0092]

[0093] in,

[0094] q i (k) represents the virtual queue at the kth control action;

[0095] ψ is the cost threshold.

[0096] The stable strategy based on virtual queues satisfies the RCI constraint and can guarantee that the constraint requirements are automatically met within the average time constraint.

[0097] To control the RCI-constrained virtual queue, in the embodiments of this specification, the strategy for stabilizing the virtual queue is achieved by minimizing the upper bound of the single-step Lyapunov drift, specifically expressed as:

[0098]

[0099] in,

[0100] ΔV[k] represents the single-step Lyapunov drift during the k-th control action;

[0101] qi (k) represents the virtual queue at the kth control action;

[0102] c i (k) represents instantaneous cost;

[0103] J represents the number of temperature sensors.

[0104] Based on minimizing the upper bound of single-step Lyapunov drift It can assess Lyapunov drift without knowing the future state of the system, thereby pushing the virtual queue to a stable state.

[0105] In Lyapunov drift, the Lyapunov function is used to characterize the degree to which the virtual queue deviates from the stable state. The higher the Lyapunov function, the greater the degree of deviation. Minimizing the Lyapunov function is equivalent to minimizing the single-step Lyapunov drift.

[0106] In the embodiments described in this specification, the expression for the Lyapunov function is:

[0107]

[0108] Where V[k] represents the Lyapunov function for the k-th control action;

[0109] M represents the number of CRAC units;

[0110] q i (k) represents the virtual queue at the kth control action.

[0111] In the embodiments described in this specification, the expression for single-step Lyapunov drift is:

[0112] ΔV[k]=V[k+1]-V[k]

[0113] in,

[0114] V[k] represents the Lyapunov function for the kth control action;

[0115] V[k+1] represents the Lyapunov function for the (k+1)th control action.

[0116] Based on the aforementioned definition of a virtual queue, the objective function is further redefined. In the embodiments of this specification, the objective function of the preset thermodynamic model is the objective function of the virtual queue-modulated security set, and the objective function of the preset thermodynamic model is:

[0117]

[0118]

[0119]

[0120] in,

[0121] ε represents the equilibrium index set, and the expression for ε is ε={iq i (k) = 0, 1, 2, ..., M;

[0122] The virtual queue modulation security set δ is:

[0123]

[0124] In the embodiments described in this specification, the set of equilibrium indices includes indices for virtual queue zero backlog. Therefore, the virtual queue modulation security set can be derived from the set of equilibrium indices.

[0125] The virtual queue modulation security set indicates that if the backlog of the virtual queue is zero, then only the T value of the next time step is restricted. in Less than the maximum allowable rack inlet temperature That's fine; if the backlog in the virtual queue is not zero, then limit the time step T for the next time step. in Less than the recommended maximum rack inlet temperature That's all.

[0126] To further understand how the pre-defined thermodynamic model is used, further explanation will follow.

[0127] In the embodiments of this specification, the step of inputting the control action into a preset thermodynamic model for steady-state simulation to obtain an optimized control action specifically includes:

[0128] The control action is input into a preset thermodynamic model for steady-state simulation to obtain candidate control actions;

[0129] If the candidate control action belongs to the virtual queue modulation security set δ, then the candidate control action is taken as the optimized control action.

[0130] If the candidate control action does not belong to the virtual queue modulation security set δ, then an adjustment is made based on convex quadratic programming to obtain an optimized control action;

[0131] in,

[0132] The convex quadratic programming is as follows:

[0133]

[0134]

[0135]

[0136] in,

[0137] express At that time, the gradient of the i-th constraint in the equilibrium state set;

[0138] This represents the gradient of the i-th constraint in the equilibrium state set when i ∈ ε.

[0139] yes At that time, the distance between the control action and the boundary of the virtual queue's safety set δ is adjusted;

[0140] When i∈ε, the distance of the control action from the boundary of the virtual queue adjustment safety set δ;

[0141] i represents the index of the constraint, referring to the i-th constraint in a series of constraints;

[0142] T represents temperature.

[0143] It is important to note that the preset thermodynamic model should be initialized before use. Specifically, the initialization of the preset thermodynamic model includes: setting the parameter to be optimized θ, setting the first commentator network φ. i and the second commentator network φ i Temperature threshold: the maximum permissible rack inlet temperature and recommended maximum rack intake temperature Cost threshold ψ, parameters α and learning rate λ, weights η; initialize all J virtual queues q i [0] is 0; initialize the replay buffer D.

[0144] In the embodiments of this specification, after inputting the control action into a preset thermodynamic model for steady-state simulation to obtain the optimized control action, the following steps are also included:

[0145] Update the backlog of the virtual queue to obtain the updated virtual queue backlog;

[0146] The updated virtual queue backlog q i [k+1]' is:

[0147]

[0148] in,

[0149] q i (k) represents the virtual queue at the kth control action;

[0150] ε represents the set of equilibrium exponents;

[0151] Let be the intake air temperature of the i-th server at time step k;

[0152] The recommended maximum rack inlet temperature;

[0153] Maximum permissible rack inlet temperature;

[0154] ε represents the set of equilibrium exponents;

[0155] J represents the number of temperature sensors.

[0156] In the embodiments of this specification, when the time step k is a preset multiple of the predetermined time interval I, the strategy network and evaluation network of the preset thermodynamic model are updated to obtain an updated strategy network and an updated evaluation network.

[0157] The updated policy network is:

[0158] θ←θ+λ A ▽ θ L(θ, α, B)

[0159] Where θ represents the parameter to be optimized;

[0160] α represents the parameters of the Q function.

[0161] λ represents the learning rate;

[0162] L(·) represents the loss function;

[0163] B represents a sampling batch;

[0164] The updated evaluation network is

[0165]

[0166] φ i '←ηφ i '+(1-η)φ i i = 1, 2

[0167] Where B represents a sampling batch;

[0168] η is the weight used to update the objective Q-function;

[0169] L(·) represents the loss function;

[0170] φ i Indicates the first commentator network;

[0171] φ i 'This refers to the second network of commentators.'

[0172] In the embodiments described in this specification, based on the updated policy network, the parameters θ can be optimized in deep reinforcement learning to minimize the loss function L(θ, α, B).

[0173] B, as a sampling batch, is a batch containing transformed tuples.

[0174] In the embodiments described in this specification, the preset time interval is the policy update interval. In specific embodiments, the policy update time is set according to the specific business scenario, and no specific limitation is made here.

[0175] When the time step k is a preset multiple of the predetermined time interval I, a new sampling batch can be released for updating the policy network and the evaluation network.

[0176] Furthermore, in the embodiments of this specification, the evaluation network employs a critic network, including a first critic network and a second critic network. Using two critic networks helps reduce overestimation bias in value estimation—a problem known as the double-Q learning problem. Each critic network independently approximates the Q-value, which represents the expected reward (sum of rewards) of starting from a state, taking an action, and subsequently following the policy.

[0177] Step S205: Use the optimized control action as input to the energy model to perform cooling control of the data center to be processed.

[0178] Based on the aforementioned steps, the optimal control action, i.e., the optimized control action, is then used as input to the energy model to control the data center to be processed.

[0179] To further illustrate the method for dynamically controlling the cooling process provided in the embodiments of this specification, the following will provide a framework diagram of the method for dynamically controlling the cooling process for intelligent data centers.

[0180] Figure 3 This is a framework diagram of a method for dynamically controlling the cooling process in a smart data center, provided as an embodiment of this specification. Figure 3As shown, the intelligent control center monitors the current data of the data center to be processed and provides control actions based on the monitored data. The intelligent control center then sends the control actions to a preset thermodynamic model for steady-state simulation and optimization of the control actions. The optimized control actions are then sent to an energy model for cooling control of the data center to be processed. The energy model and the preset thermodynamic model are coupled simulation models to jointly realize the dynamic control of the intelligent data center. In the embodiments of this specification, the preset thermodynamic model (POD) is a thermodynamic model based on fluid dynamics (CFD), hence it is called the CFD / POD model.

[0181] The method for dynamically controlling the cooling process of intelligent data centers provided in this specification's embodiments, through the development of a coupled thermal energy simulation model, the introduction of a new CMDP formula to evaluate compliance with the ASHRAE thermal guidelines, and the introduction of a virtual queue with RCI constraints, derives a virtual queue modulation safety set through Lyapunov stability theory. This achieves a significant reduction in data center energy consumption without violating any thermal safety compliance, enabling dynamic control of intelligent data centers. It possesses high performance and security, and can reduce energy consumption in specific industrial applications (such as pathological images and defect images), demonstrating high industrial value.

[0182] The above describes in detail a method for dynamically controlling the cooling process of an intelligent data center. Correspondingly, this specification also provides an apparatus for dynamically controlling the cooling process of an intelligent data center, such as... Figure 4 As shown. Figure 4 This is a schematic diagram of an apparatus for dynamically controlling the cooling process in a smart data center, provided as an embodiment of this specification. The apparatus includes:

[0183] The control action determination module 401 determines the control action based on the current data of the data center to be processed. The current data includes: the current rack inlet temperature, the total system power and the IT load. The control action includes: the cooling water supply temperature, the CRAC supply air temperature and the CRAC flow rate.

[0184] The control action optimization module 403 inputs the control action into a preset thermodynamic model for steady-state simulation to obtain the optimized control action;

[0185] The cooling control module 405 uses the optimized control actions as input to the energy model to perform cooling control of the data center to be processed.

[0186] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0187] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for apparatus, electronic devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0188] The apparatus, electronic device, and non-volatile computer storage medium and method provided in the embodiments of this specification are corresponding. Therefore, the apparatus, electronic device, and non-volatile computer storage medium also have similar beneficial technical effects as the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the corresponding apparatus, electronic device, and non-volatile computer storage medium will not be repeated here.

[0189] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0190] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0191] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0192] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in one or more software and / or hardware.

[0193] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0194] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0195] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0196] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0197] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0198] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0199] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0200] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0201] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside on local and remote computer storage media, including storage devices.

[0202] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0203] The above description is merely an embodiment of this specification and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this application should be included within the scope of the claims of this application.

Claims

1. A method for dynamically controlling the cooling process of an intelligent data center, characterized in that, The method for dynamically controlling the cooling process includes: Based on the current data of the data center to be processed, control actions are determined. The current data includes: the current rack inlet temperature, the total system power, and the IT load. The control actions include: the cooling water supply temperature, the CRAC supply air temperature, and the CRAC flow rate. The control action is input into a preset thermodynamic model for steady-state simulation to obtain an optimized control action. The optimized control actions are used as input to the energy model for cooling control of the data center to be processed.

2. The method for dynamically controlling the cooling process as described in claim 1, characterized in that, The preset thermodynamic model is a differentiable model based on POD technology; The preset thermodynamic model is: T[k+1]=f(P IT [k],Tsup[k],Vsup[k]) T ·φ in, T[k+1] represents the (k+1)th control action; P IT [k] represents the IT load power consumption during the kth control action; Tsup[k] is the air supply temperature of the CRAC during the kth control action; Vsup[k] is the flow rate of the CRAC during the k-th control action; f(·) represents a continuously differentiable function that maps the k-th control action to the POD; φ represents the vector of the first k PODs.

3. The method for dynamically controlling the cooling process as described in claim 1, characterized in that, The preset thermodynamic model is a constrained Markov decision process. The preset thermodynamic model's state s[k] includes: the rack inlet temperature Tin[k] at the k-th control action and the IT load power consumption P at the k-th control action. IT [k]; The preset thermodynamic model action a[k] includes: the CRAC supply air temperature Tsup[k] and the CRAC flow rate Vsup[k] at the kth control action, and the cooling water supply temperature Tchw[k] at the kth control action; The reward r[k] of the preset thermodynamic model includes: the grid carbon emission factor e[k] at the kth control action and the total energy consumption Etol[k] of the data center to be processed at the kth control action; The constraints of the preset thermodynamic model include RCI constraints and hard constraints.

4. The method for dynamically controlling the cooling process as described in claim 3, characterized in that, The RCI constraint is: in, ψ is the cost threshold; Maximum permissible rack inlet temperature; The recommended maximum rack inlet temperature; K is the number of time steps; Let be the intake air temperature of the i-th server at time step k; This means that the above condition must be true for all i = 1, 2, 3, ...; The hard constraint is: in, Tin[k] is the rack inlet temperature at the kth control action; Maximum permissible rack inlet temperature; This means that the above condition must be true for all i = 1, 2, 3, ...; The objective function of the preset thermodynamic model is: in, π θ Represents strategy θ; k is the time step; γ k Let be the discount factor for the cost function at the k-th time step; r k The reward for the preset thermodynamic model.

5. The method for dynamically controlling the cooling process as described in claim 4, characterized in that, The RCI constraint is obtained based on instantaneous cost and virtual queue, wherein, The instantaneous cost c i (k) is: The virtual queue is q i [k+1]=max{q i [k]+c i [k],0}≥0 in, q i [k] represents the virtual queue at the kth control action; q i [k+1] represents the virtual queue during the (k+1)th control action; The stable strategy π of the virtual queue θ Satisfying RCI constraints: in, q i (k) represents the virtual queue at the kth control action; ψ is the cost threshold.

6. The method for dynamically controlling the cooling process as described in claim 5, characterized in that, The strategy for stabilizing the virtual queue is achieved by minimizing the upper bound of single-step Lyapunov drift, specifically expressed as: in, ΔV[k] represents the single-step Lyapunov drift during the k-th control action; q i (k) represents the virtual queue at the kth control action; c i (k) represents instantaneous cost; J represents the number of temperature sensors.

7. The method for dynamically controlling the cooling process as described in claim 6, characterized in that, The objective function of the preset thermodynamic model is the objective function based on the virtual queue modulation security set. The objective function of the preset thermodynamic model is: in, ε represents the equilibrium index set, and the expression for ε is ε={iq i (k) = 0, 1, 2, ..., M; γ k Let be the discount factor for the cost function at the k-th time step; The virtual queue modulation security set δ is:

8. The method for dynamically controlling the cooling process as described in claim 7, characterized in that, The step of inputting the control action into a preset thermodynamic model for steady-state simulation to obtain an optimized control action specifically includes: The control action is input into a preset thermodynamic model for steady-state simulation to obtain candidate control actions; If the candidate control action belongs to the virtual queue modulation security set δ, then the candidate control action is taken as the optimized control action. If the candidate control action does not belong to the virtual queue modulation security set δ, then an adjustment is made based on convex quadratic programming to obtain an optimized control action; in, The convex quadratic programming is as follows: in, express At that time, the gradient of the i-th constraint in the equilibrium state set; Let represent the gradient of the i-th constraint in the equilibrium state set when i∈ε; yes At that time, the distance between the control action and the boundary of the virtual queue's safety set δ is adjusted; When i∈ε, the distance of the control action from the boundary of the virtual queue adjustment safety set δ; i represents the index of the constraint, referring to the i-th constraint in a series of constraints; T represents temperature.

9. The method for dynamically controlling the cooling process as described in claim 8, characterized in that, After inputting the control action into a preset thermodynamic model for steady-state simulation to obtain the optimized control action, the method further includes the following steps: Update the backlog of the virtual queue to obtain the updated virtual queue backlog; The updated virtual queue backlog q i [k+1]' is: in, q i (k) represents the virtual queue at the kth control action; ε represents the set of equilibrium exponents; Let be the intake air temperature of the i-th server at time step k; The recommended maximum rack inlet temperature; Maximum permissible rack inlet temperature; ε represents the set of equilibrium exponents; J represents the number of temperature sensors.

10. The method for dynamically controlling the cooling process as described in claim 9, characterized in that, When the time step k is a preset multiple of the predetermined time interval I, the strategy network and evaluation network of the preset thermodynamic model are updated to obtain the updated strategy network and the updated evaluation network. The updated policy network is: Where θ represents the parameter to be optimized; α represents the parameters of the Q function; λ represents the learning rate; L(·) represents the loss function; B represents a sampling batch; The updated evaluation network is f i '←hφ i '+(1-η)φ i ,i=1.2 Where B represents a sampling batch; η is the weight used to update the objective Q-function; L(·) represents the loss function; φ i Indicates the first commentator network; φ i 'This refers to the second network of commentators.' 11. A device for dynamically controlling the cooling process in intelligent data centers, characterized in that, The device for dynamically controlling the cooling process includes: The control action determination module determines the control action based on the current data of the data center to be processed. The current data includes: the current rack inlet temperature, the total system power and the IT load. The control action includes: the cooling water supply temperature, the CRAC supply air temperature and the CRAC flow rate. The control action optimization module inputs the control action into a preset thermodynamic model for steady-state simulation to obtain optimized control actions; The cooling control module uses the optimized control actions as input to the energy model to perform cooling control of the data center to be processed.

Images