Model training method and apparatus, and readable storage medium

Abstract

A model training method and apparatus, and a readable storage medium. The method comprises: inputting multiple sets of verification parameters of a refrigeration device into a scenario simulation model to obtain scenario simulation data corresponding to each set of verification parameters among the multiple sets of verification parameters, wherein the refrigeration device is used for cooling multiple network devices, the multiple sets of verification parameters comprise multiple sets of refrigeration parameters under different boundary conditions, and the scenario simulation data comprises the energy consumption of refrigeration device and temperature data of a region where each network device among the multiple network devices is located; and training a reinforcement model on the basis of the multiple sets of verification parameters to obtain a target reinforcement model, wherein the target reinforcement model is used for determining a target refrigeration parameter on the basis of the current boundary condition of the refrigeration device, the target refrigeration parameter is the one, among multiple sets of candidate refrigeration parameters, that results in the lowest energy consumption of the refrigeration device, and when the refrigeration device executes the candidate refrigeration parameters, the temperature data of the region where each network device is located is lower than a temperature threshold.

Description

Model training methods, devices, and readable storage media

[0001] This application claims priority to Chinese patent application No. 202410986167.3, filed on July 22, 2024, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This disclosure relates to the field of computer technology, and in particular to a model training method, apparatus, and readable storage medium. Background Technology

[0003] Currently, energy consumption in data centers is increasing. Cooling equipment accounts for a significant portion of the total energy consumption of data centers. To reduce the energy consumption of cooling equipment, it is necessary to optimize the energy consumption of cooling equipment based on the temperature of the areas where network equipment is located in different parts of the data center. Summary of the Invention

[0004] Firstly, a model training method is provided, comprising: inputting multiple sets of verification parameters of a cooling device into a scenario simulation model to obtain scenario simulation data corresponding to each set of verification parameters; the cooling device is used to cool multiple network devices; the multiple sets of verification parameters include multiple sets of cooling parameters under different boundary conditions; the scenario simulation data includes the energy consumption of the cooling device and the temperature data of the area where each of the multiple network devices is located. A reinforcement model is trained based on the multiple sets of verification parameters to obtain a target reinforcement model; the target reinforcement model is used to determine the target cooling parameter based on the current boundary conditions of the cooling device; the target cooling parameter is the cooling parameter with the lowest energy consumption among the multiple sets of candidate cooling parameters; when the cooling device executes the candidate cooling parameter, the temperature data of the area where each network device is located is less than a temperature threshold.

[0005] In some embodiments, training the reinforcement model based on multiple sets of validation parameters to obtain the target reinforcement model includes: determining the reward value for executing each set of validation parameters based on the scenario simulation data corresponding to each set of validation parameters and the reward function; and training the reinforcement model based on multiple sets of validation parameters and the reward value of each set of validation parameters to obtain the target reinforcement model.

[0006] In some embodiments, multiple sets of verification parameters of the refrigeration equipment are input into a scenario simulation model to obtain scenario simulation data corresponding to each set of verification parameters. This includes: inputting multiple sets of verification parameters into a mechanism model to obtain air supply parameters corresponding to each set of verification parameters; the air supply parameters include air volume and air supply temperature; and inputting the air supply parameters corresponding to each set of verification parameters, outdoor temperature, outdoor humidity, and the power of multiple network devices into an artificial intelligence computational fluid dynamics (AI-CFD) model to obtain scenario simulation data corresponding to each set of verification parameters.

[0007] In some embodiments, the method further includes: generating multiple sets of training data based on the target computational fluid dynamics (CFD) model; each set of training data includes air supply parameters, the power of each network device, and the temperature data of the area where each network device is located; and training a preset neural network model based on the multiple sets of training data to obtain an AI-CFD model.

[0008] In some embodiments, the method further includes: acquiring historical temperature data, historical air supply parameters, and three-dimensional structural data of the target area; the three-dimensional structural data includes the location of each device in the target area and the configuration of the flow field; each device includes network devices and refrigeration devices; and constructing a target CFD model based on the historical temperature data, historical air supply parameters, and three-dimensional structural data.

[0009] In some embodiments, constructing a target CFD model based on historical temperature data, historical air supply parameters, and three-dimensional structural data includes: constructing an initial CFD model based on historical temperature data, historical air supply parameters, and three-dimensional structural data; determining the loss value of the initial CFD model, whereby the loss value characterizes the degree of loss between the temperature data output by the initial CFD model and the actual temperature data; and determining the CFD model as the target CFD model if the loss value of the initial CFD model is less than a loss value threshold.

[0010] In some embodiments, the boundary conditions include at least one of the following: outdoor dry-bulb temperature, outdoor wet-bulb temperature, chilled water supply temperature, chilled water flow rate, cooling load power, fan power, or temperature threshold. Cooling parameters include at least one of the following: return air valve opening, fresh air valve opening, return air fan frequency, supply air fan frequency, chilled water valve opening, or supply air temperature setpoint.

[0011] Secondly, a model training device is provided, comprising an input unit and a processing unit. The input unit is used to input multiple sets of verification parameters from a cooling device into a scene simulation model, obtaining scene simulation data corresponding to each set of verification parameters. The cooling device is used to cool multiple network devices. The multiple sets of verification parameters include multiple sets of cooling parameters under different boundary conditions. The scene simulation data includes the energy consumption of the cooling device and the temperature data of the area where each of the multiple network devices is located. The processing unit is used to train a reinforcement model based on the multiple sets of verification parameters to obtain a target reinforcement model. The target reinforcement model is used to determine target cooling parameters based on the current boundary conditions of the cooling device. The target cooling parameter is the cooling parameter with the lowest energy consumption among the multiple sets of candidate cooling parameters. When the cooling device executes the candidate cooling parameters, the temperature data of the area where each network device is located is less than a temperature threshold.

[0012] In some embodiments, the processing unit is configured to: determine the reward value for executing each set of verification parameters based on the scene simulation data corresponding to each set of verification parameters and the reward function; and train the reinforcement model based on multiple sets of verification parameters and the reward value of each set of verification parameters to obtain the target reinforcement model.

[0013] In some embodiments, the processing unit is configured to: input multiple sets of verification parameters into a mechanistic model to obtain air supply parameters corresponding to each set of verification parameters. The air supply parameters include air volume and air supply temperature. The processing unit is also configured to input the air supply parameters corresponding to each set of verification parameters, outdoor temperature, outdoor humidity, and the power of multiple network devices into an artificial intelligence computational fluid dynamics (AI-CFD) model to obtain scenario simulation data corresponding to each set of verification parameters.

[0014] In some embodiments, the processing unit is further configured to generate multiple sets of training data based on the target computational fluid dynamics (CFD) model. Each set of training data includes air supply parameters, the power of each network device, and the temperature data of the area where each network device is located. The processing unit is further configured to train a preset neural network model based on the multiple sets of training data to obtain an AI-CFD model.

[0015] In some embodiments, the model training apparatus further includes an acquisition unit. The acquisition unit is used to acquire historical temperature data, historical air supply parameters, and three-dimensional structural data of the target area. Here, the three-dimensional structural data includes the location of each device in the target area and the configuration of the flow field; each device includes network devices and refrigeration devices; the processing unit is further used to construct a target CFD model based on the historical temperature data, historical air supply parameters, and three-dimensional structural data.

[0016] In some embodiments, the processing unit is configured to: construct an initial CFD model based on historical temperature data, historical air supply parameters, and three-dimensional structural data; determine the loss value of the initial CFD model, wherein the loss value is used to characterize the degree of loss between the temperature data output by the initial CFD model and the actual temperature data; and determine the CFD model as the target CFD model if the loss value of the initial CFD model is less than a loss value threshold.

[0017] In some embodiments, the boundary conditions include at least one of the following: outdoor dry-bulb temperature, outdoor wet-bulb temperature, chilled water supply temperature, chilled water flow rate, cooling load power, fan power, or temperature threshold. Cooling parameters include at least one of the following: return air valve opening, fresh air valve opening, return air fan frequency, supply air fan frequency, chilled water valve opening, or supply air temperature setpoint.

[0018] Thirdly, a model training device is provided, which can realize the functions performed by the model training device in the above aspects or possible designs. The functions can be implemented by hardware. For example, in one embodiment, the model training device may include a processor and a communication interface. The processor can be used to support the model training device in realizing the methods and functions involved in the first aspect or any embodiment of the first aspect.

[0019] In some embodiments, the model training apparatus may further include a memory for storing necessary computer execution instructions and data. When the model training apparatus is running, the processor executes the computer execution instructions stored in the memory to cause the model training apparatus to perform the model training method of the first aspect or any embodiment thereof.

[0020] Fourthly, a computer-readable storage medium is provided, which may be a readable non-volatile storage medium storing computer instructions or programs that, when run on a computer, enable the computer to execute the model training method described in the first aspect or any of the methods described in the first aspect.

[0021] Fifthly, a computer program product containing instructions is provided, which, when run on a computer, enables the computer to execute the first aspect or any of the model training methods described in the first aspect.

[0022] A sixth aspect provides an electronic device comprising one or more processors and one or more memories. The one or more memories are coupled to the one or more processors, and the one or more memories are used to store computer program code, including computer instructions, which, when executed by the one or more processors, cause the electronic device to perform a model training method as described in the first aspect or any of the methods described in the first aspect.

[0023] In a seventh aspect, a chip system is provided, the chip system including a processor and a communication interface, the chip system being used to implement the functions performed by the first aspect or any of the model training devices of the first aspect.

[0024] In some embodiments, the chip system further includes a memory for storing at least one of program instructions or data. The chip system comprises a chip, or the chip system comprises a chip and other discrete devices. Attached Figure Description

[0025] Figure 1 is a schematic diagram of a model training system according to some embodiments;

[0026] Figure 2 is a schematic diagram of another model training system according to some embodiments;

[0027] Figure 3 is a schematic diagram of another model training device according to some embodiments;

[0028] Figure 4 is a flowchart of a model training method according to some embodiments;

[0029] Figure 5 is a flowchart of another model training method according to some embodiments;

[0030] Figure 6 is a flowchart of yet another model training method according to some embodiments;

[0031] Figure 7 is a block diagram of a model training apparatus according to some embodiments. Detailed Implementation

[0032] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in some embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0033] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of some embodiments of this disclosure as detailed in the appended claims.

[0034] It should also be understood that the term "comprising" indicates the presence of at least one of the described features, wholes, steps, operations, elements, or components, but does not exclude the presence or addition of at least one of one or more other features, wholes, steps, operations, elements, or components.

[0035] The following is an explanation of the terms used in this disclosure.

[0036] Reinforcement learning consists of four elements: agent, environment, state, action, and reward. Here, the agent's input is the state, and its output is the action.

[0037] The training process of reinforcement learning is as follows: the agent interacts with the environment multiple times to obtain the action, state, and reward for each interaction; these multiple sets of (action, state, reward) are used as training data to train the agent once. This process is repeated for the next round of training until the convergence condition is met.

[0038] Intelligent agents are entities capable of thought and interaction with their environment. For example, an intelligent agent can be a computer system or part of a computer system within a specific environment. Based on its own perception of the environment, following existing instructions or through autonomous learning, and by communicating and collaborating with other intelligent agents, an intelligent agent can autonomously achieve predetermined goals within its environment. Intelligent agents can be software or a combination of software and hardware.

[0039] With the rapid development of cloud computing, new communication services, and internet data services, computing power is becoming a new form of productivity, and data centers, as the physical carriers of computing power, are rapidly expanding in scale. Air conditioning systems account for 40% to 60% of the total energy consumption of a data center. Central air conditioning systems consist of two main components: terminal units and the cooling source system. The energy consumption of terminal cabinets can account for approximately 40% of the total energy consumption of the entire air conditioning system. Therefore, energy-saving optimization of data center terminal systems can effectively reduce the energy consumption of the data center.

[0040] In related technologies, the operation mode of the terminal system of the data center still mainly adopts the traditional proportional integral derivative (PID) control, and sets a relatively conservative control strategy by increasing the air volume and reducing the air temperature, which leads to energy waste.

[0041] When adapting to the needs of modeling, simulation, and optimization control of data center terminal systems, the preset methods that can be used include: artificial intelligence technology and computational fluid dynamics (AI-CFD) methods based on AI learning and simulation of temperature and velocity fields, terminal mechanism modeling methods based on Modelica, terminal system optimization control methods based on reinforcement learning, deep learning modeling algorithms, and model-free reinforcement learning optimization algorithms, etc.

[0042] Because there are currently few measuring points in the terminal systems of data centers, information such as water flow and temperature in each branch and airflow and temperature in the ducts is lacking. Therefore, it is difficult to use model-based optimization methods, such as heuristic search algorithms based on genetic algorithms or model predictive control (MPC) algorithms based on the dynamic characteristics of the controlled system.

[0043] Model-free reinforcement learning algorithms require a large amount of historical data for training. However, many data center edge systems lack sufficient training data due to missing measurement points.

[0044] Traditional data center edge optimization control methods only monitor the return air temperature of the edge system and use the return air temperature for strategy optimization control, but the return air temperature of the edge system cannot accurately characterize the temperature of the rack.

[0045] Optimizing data center edge systems cannot simply consider the temperature of a single rack; it requires considering the temperature field cloud map of the entire data center, and this temperature field cloud map must be linked with the optimization control algorithm. However, traditional Transformer models are not good at handling long-range dependencies, and multi-layer Transformer models consume a lot of memory.

[0046] In view of this, some embodiments of this disclosure provide a model training method, which includes: inputting multiple sets of verification parameters of a cooling device into a scene simulation model to obtain scene simulation data corresponding to each set of verification parameters. Here, the cooling device is used to cool multiple network devices; the multiple sets of verification parameters include multiple sets of cooling parameters under different boundary conditions; the scene simulation data includes the energy consumption of the cooling device and the temperature data of the area where each of the multiple network devices is located.

[0047] The method further includes training the reinforcement model based on multiple sets of validation parameters to obtain a target reinforcement model. Here, the target reinforcement model is used to determine the target refrigeration parameters based on the current boundary conditions of the refrigeration equipment; the target refrigeration parameter is the refrigeration parameter that minimizes the energy consumption of the refrigeration equipment from multiple sets of candidate refrigeration parameters; when the refrigeration equipment executes the candidate refrigeration parameters, the temperature data of the area where each network device is located is less than the temperature threshold.

[0048] The methods provided by some embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0049] It should be noted that the network systems described in some embodiments of this disclosure are for the purpose of more clearly illustrating the technical solutions of some embodiments of this disclosure, and do not constitute a limitation on the technical solutions provided by some embodiments of this disclosure. As those skilled in the art will know, with the evolution of network systems and the emergence of other network systems, the technical solutions provided by some embodiments of this disclosure are also applicable to similar technical problems.

[0050] Figure 1 is a schematic diagram of a model training system according to some embodiments. As shown in Figure 1, the model training system 10 may include a terminal device 11 and a model training apparatus 12.

[0051] The terminal device 11 is connected to the model training device 12. For example, the connection can be made wirelessly or via a wired connection.

[0052] In some embodiments of this disclosure, the terminal device 11 is used to send training data to the model training device 12. For example, the terminal device 11 can be any computer device or server. Here, the computer device includes, but is not limited to, mobile phones, tablets, desktop computers, laptops, in-vehicle terminals, handheld terminals, augmented reality (AR) devices, or virtual reality (VR) devices, etc. This disclosure does not limit the form of the terminal device 11. The terminal device 11 can interact with the user through one or more methods such as a keyboard, touchpad, touch screen, remote control, voice interaction, or handwriting device. This disclosure does not limit the technology and device form adopted by the terminal device 11.

[0053] The model training apparatus 12 described in some embodiments of this disclosure can be used to receive training data from a terminal device and perform model training based on the training data. The model training apparatus 12 can be an electronic device with processing capabilities, such as a computer or a server. For example, the model training apparatus 12 can be a computer or a server. Here, the server can be a single server, or it can be a server cluster consisting of multiple servers.

[0054] In some embodiments, the server cluster may also be a distributed cluster. This disclosure does not limit the technology, quantity, or form of the terminal device 11, nor the technology or form of the model training device 12.

[0055] Figure 1 is an exemplary framework diagram. The names of the various devices included in Figure 1 are not limited, and other nodes may be included in addition to the functional nodes shown in Figure 1. This disclosure does not limit this.

[0056] Figure 2 is a schematic diagram of another model training system according to some embodiments. As shown in Figure 2, the model training system 10 may include a computational fluid dynamics (CFD) model, an AI-CFD model, a mechanistic model, and a reinforcement learning model.

[0057] Here, the AI-CFD model is trained based on the CFD model. The model training system 10 is used for joint simulation based on the AI-CFD model, the mechanistic model, and the reinforcement learning model to optimize the energy consumption of the cooling equipment.

[0058] In actual implementation, the devices in Figure 1 can adopt the composition structure shown in Figure 3, or include the components shown in Figure 3. Figure 3 is a schematic diagram of a model training device according to some embodiments. The model training device 200 can be a network device, or the model training device 200 can be a chip or system on a chip (SoC) in a network device. As shown in Figure 3, the model training device 200 includes a processor 201, a communication interface 202, and a communication line 203.

[0059] In some embodiments, the model training apparatus 200 may further include a memory 204. Here, the processor 201, the memory 204, and the communication interface 202 can be connected via a communication line 203.

[0060] Processor 201 is a central processing unit (CPU), a general-purpose processor, a network processor (NP), a digital signal processor (DSP), a microprocessor, a microcontroller, a programmable logic device (PLD), or any combination thereof. Processor 201 can also be other devices with processing capabilities, such as circuits, devices, or software modules, which are not limited herein.

[0061] Communication interface 202 is used to communicate with other devices or other communication networks. Communication interface 202 can be a module, circuit, communication interface, or any device capable of enabling communication.

[0062] Communication line 203 is used to transmit information between the components included in the model training device 200.

[0063] Memory 204 is used to store instructions. Here, the instructions can be computer program instructions.

[0064] For example, memory 204 may be read-only memory (ROM) or other types of static storage devices that can store static information and / or instructions; it may also be random access memory (RAM) or other types of dynamic storage devices that can store information and / or instructions; it may also be electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs or Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, etc., and this disclosure does not limit it in this way.

[0065] It should be noted that the memory 204 can exist independently of the processor 201 or can be integrated with the processor 201. The memory 204 can be used to store instructions, program code, or some data. The memory 204 can be located inside or outside the model training device 200, and this disclosure does not limit this. The processor 201 is used to execute the instructions stored in the memory 204 to implement the model training methods provided in some embodiments of this disclosure.

[0066] In some embodiments, processor 201 may include one or more CPUs, such as CPU0 and CPU1 in FIG3.

[0067] In one implementation, the model training device 200 includes multiple processors. For example, in addition to the processor 201 in FIG3, the model training device 200 may also include a processor 205.

[0068] It should be noted that the composition shown in Figure 3 does not constitute a limitation on the various devices in Figure 1. In addition to the components shown in Figure 3, the various devices in Figure 1 may include more or fewer components than those shown in Figure 3, or combine certain components, or have different component arrangements.

[0069] In some embodiments of this disclosure, the chip system may consist of chips or may include chips and other discrete devices.

[0070] Furthermore, the actions, terms, etc., involved in the various embodiments of this disclosure can be referenced interchangeably, and this disclosure does not impose any limitations on them. The message names or parameter names in the messages exchanged between the various devices in the embodiments of this disclosure are merely examples, and other names may be used in the implementation, and this disclosure does not impose any limitations on them.

[0071] To facilitate a clear description of the technical solutions of some embodiments of this disclosure, the terms "first" and "second" are used in the embodiments of this disclosure to distinguish identical or similar items with substantially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0072] It should be noted that in this disclosure, the words "exemplarily" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplarily" or "for example" in this disclosure should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of words such as "exemplarily" or "for example" is intended to present the relevant concepts by way of example.

[0073] In this disclosure, "at least one" means one or more, "more than one" means two or more, and "at least two (items)" means two or three or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: only A, only B, and A and B, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, a and b, a and c, b and c, or a, b, and c, where a, b, and c can be single or multiple.

[0074] The model training methods provided in some embodiments of this disclosure are described below with reference to the model training system shown in Figure 1.

[0075] Figure 4 is a flowchart of a model training process according to some embodiments, which is applied to a model training device and can also be applied to devices in the model training device, such as chips.

[0076] Some embodiments of this disclosure are illustrated using an application to a model training device as an example, as shown in FIG4. The method includes the following steps S301-S302.

[0077] S301. Input multiple sets of verification parameters of the refrigeration equipment into the scenario simulation model to obtain the scenario simulation data corresponding to each set of verification parameters.

[0078] Here, the cooling equipment is used to cool multiple network devices; for example, the cooling equipment could be an air conditioner. Multiple sets of verification parameters include cooling parameters under different boundary conditions; the scenario simulation data includes the energy consumption of the cooling equipment and the temperature data of the area where each network device is located.

[0079] In some embodiments, the boundary conditions include at least one of the following: outdoor dry-bulb temperature, outdoor wet-bulb temperature, chilled water supply temperature, chilled water flow rate, cooling load power, fan power, or temperature threshold.

[0080] The refrigeration parameters include at least one of the following: return air valve opening, fresh air valve opening, return air fan frequency, supply air fan frequency, chilled water valve opening, or supply air temperature setpoint.

[0081] As one implementation method, the model training device can input multiple sets of verification parameters into the mechanism model to obtain the air supply parameters corresponding to each set of verification parameters; input the air supply parameters corresponding to each set of verification parameters, outdoor temperature, outdoor humidity, and the power of multiple network devices into the artificial intelligence computational fluid dynamics (AI-CFD) model to obtain the scene simulation data corresponding to each set of verification parameters.

[0082] Here, the air supply parameters include air volume and air temperature.

[0083] It should be noted that the mechanistic model was obtained by modeling the data center air conditioning terminal system using Modelica, based on the conditions of the cooling equipment terminals inside the data center computer room. The modeling content includes the all-air handling unit (AHU), the make-up air unit (MAU), the fan coil unit (FCU), valve equipment, fan equipment, water and air ducts, etc.

[0084] It should be noted that the model training device can determine the maximum and minimum values ​​of each boundary condition and different cooling parameters, as well as the data deviation, and combine different boundary conditions and different cooling parameters to obtain multiple sets of verification parameters.

[0085] In some embodiments, when the boundary conditions include a first boundary condition and a second boundary condition, and the refrigeration parameters include a first refrigeration parameter and a second refrigeration parameter, if the parameter values ​​of the first boundary condition are a1 and a2, the parameter values ​​of the second boundary condition are b1 and b2, the parameter values ​​of the first refrigeration parameter are c1 and c2, and the parameter values ​​of the second refrigeration parameter are d1 and d2, then multiple sets of verification parameters can be (a1, b1, c1, d1), (a1, b1, c1, d2), (a1, b1, c2, d1), (a1, b1, c2, d2), (a1, b2, c1, d1), (a1, b2, c1, d2), (a1, b2, c1, d1), (a1, b2, c1, d2), (a1, b2, c1, d2), (a1, b2, c1, d2), (a1, b2, c1, d2), (a1, b2, c1, d2), (a1, b2, c1, d2), (a1, b2, c1, d2), (a1, b1 ... c1, d2), (a1, b1, c1, c1, d2), (a1, b1, c1, c1, d2), (a1, b1, c1, c1, d2), (a1, b1, c1, c1, d2), (a1, b1, c1, c 1, d2), (a1, b2, c2, d1), (a1, b2, c2, d2), (a2, b1, c1, d1), (a2, b1, c1, d2), (a2, b1, c2, d1), (a2, b1, c2, d2), (a2, b2, c1, d1), (a2, b2, c1, d1), (a2, b2, c1, d2), (a2, b2, c2, d1), (a2, b2, c2, d2).

[0086] S302. The reinforcement model is trained based on multiple sets of validation parameters to obtain the target reinforcement model.

[0087] Here, the target reinforcement model is used to determine the target cooling parameters based on the current boundary conditions of the cooling equipment. The target cooling parameters are the cooling parameters that minimize the energy consumption of the cooling equipment from a set of candidate cooling parameters. When the cooling equipment executes the candidate cooling parameters, the temperature data of the area where each network device is located is lower than the temperature threshold.

[0088] Here, the temperature threshold can be set as needed. For example, the temperature threshold can be 40 degrees or 50 degrees, etc.

[0089] As one implementation method, the model training device can divide the validation parameters into training and test sets; construct a reinforcement model using the Deep Q Network (DQN) algorithm; input the validation parameters (used as the training set) into the DQN algorithm for training; adjust the model parameters and functions to obtain the trained reinforcement model; input the validation parameters (used as the test set) into the trained reinforcement model to verify the accuracy of the trained reinforcement model; and if the accuracy is greater than the accuracy threshold, determine the trained reinforcement model as the target reinforcement model.

[0090] As another implementation method, the model training device can determine the reward value for executing each set of verification parameters based on the scene simulation data and reward function corresponding to each set of verification parameters; and train the reinforcement model based on multiple sets of verification parameters and the reward value of each set of verification parameters to obtain the target reinforcement model.

[0091] It should be noted that the termination condition for reinforcement model training can be that the temperature data of the area where the network device is located is greater than or equal to the temperature threshold.

[0092] It should be noted that the reward function is used to reward the reinforcement learning model (such as increasing the reward value of the validation parameters) when a set of cooling parameters makes the temperature data of the area where the network device is located less than the temperature threshold and the energy consumption of the cooling device decreases.

[0093] In some embodiments, the reward function is also used to penalize the reinforcement learning model (e.g., reduce the reward value of the verification parameters) when a set of cooling parameters causes the temperature data of the area where the network device is located to be greater than or equal to a temperature threshold, or when the energy consumption of the cooling device does not decrease.

[0094] Based on the technical solutions provided in some embodiments of this disclosure, multiple sets of verification parameters of the cooling equipment are input into a scene simulation model to obtain scene simulation data corresponding to each set of verification parameters; the enhancement model is trained based on multiple sets of verification parameters to obtain a target enhancement model. In this way, by constructing different simulation data in the scene simulation model, the number of parameters in the optimization model can be effectively increased. Training the enhancement model based on more simulation data ensures the training effect of the enhancement model, effectively reduces the energy consumption of the cooling equipment, and avoids overheating in areas where local network devices are located, thus ensuring the effectiveness of optimizing the energy consumption of the cooling equipment.

[0095] In some embodiments, as shown in FIG5, in order to determine the AI-CFD model, the model training method may further include the following S401-S402.

[0096] S401. Based on the target computational fluid dynamics (CFD) model, generate multiple sets of training data.

[0097] Here, each set of training data includes air supply parameters, the power of each network device, and the temperature data of the area where each network device is located.

[0098] As one implementation method, the model training device can input air supply parameters and network device power into the CFD model, so that the target CFD model outputs temperature data for the area where each network device is located, corresponding to the air supply parameters and network device power. Furthermore, the air supply parameters, network device power, and the corresponding temperature data for the area where each network device is located are used as training data.

[0099] As one implementation method, the model training device can input air supply parameters and network device power into the CFD model, so that the target CFD model outputs temperature data for the area where each network device is located, corresponding to the air supply parameters and network device power. Furthermore, feature processing is performed on the temperature data of the area where each network device is located to obtain a temperature feature map of that area. The air supply parameters, network device power, and the corresponding temperature feature map of each network device's area are then used as training data.

[0100] It should be noted that the temperature feature map is data that can be recognized by the neural network model.

[0101] In some embodiments, the training data may further include defining a symbolic distance feature map, a velocity feature map, and a pressure feature map.

[0102] S402. Based on the training data, train the preset neural network model to obtain the AI-CFD model.

[0103] Here, the default neural network model can be a multi-feature Reformer neural network.

[0104] As one implementation method, the model training device can train an AI-CFD model by adding constraints of thermophysical equations to the intermediate and output layers of a pre-defined neural network model based on training data.

[0105] In some embodiments, as shown in FIG6, in order to construct the target CFD model, the model training method may further include the following S501-S502.

[0106] S501: Acquire historical temperature data, historical air supply parameters, and three-dimensional structural data of the target area.

[0107] Here, the target area can be the internal area of ​​a data center. The three-dimensional structural data includes the location of each device within the target area and the configuration of the airflow; each device includes network equipment and cooling equipment. Historical temperature data and historical airflow parameters can refer to temperature data and airflow parameters for a preset historical time period, respectively. For example, the preset historical time period could be the month preceding the current time, or the week preceding the current time, etc.

[0108] As one implementation method, the refrigeration equipment is connected to a cloud server, which can store the operating data of the refrigeration equipment. The model training device can send request messages to the cloud server to obtain historical temperature data and historical air supply parameters for the target area. The model training device obtains 3D structural data based on the design drawings of the target area, or it can perform a 3D scan of the target area to obtain 3D structural data.

[0109] S502. Construct a target CFD model based on historical temperature data, historical air supply parameters, and three-dimensional structural data.

[0110] Here, the process of constructing the target CFD model may include geometric modeling, mesh generation, or boundary condition setting.

[0111] As one implementation method, the model training device can simulate the temperature field distribution of the target region based on three-dimensional structural data, historical temperature data, and CFD simulation software for the target region to determine the initial CFD model. Furthermore, it can determine the loss value of the initial CFD model, and if the loss value of the initial CFD model is less than a loss value threshold, the CFD model is selected as the target CFD model.

[0112] It should be noted that the loss value is used to characterize the degree of loss between the temperature data output by the initial CFD model and the actual temperature data.

[0113] In some embodiments, the model training device can, based on the CAD drawings of the computer room layout, utilize the rich component library of CFD simulation software (such as air conditioning components, network equipment components, chillers, cooling towers, etc.) to perform detailed modeling and simulation of the spatial structure of the refrigeration equipment (including the spatial layout of server information technology (IT) equipment, hot and cold aisle spatial layout, refrigeration equipment spatial layout, temperature sensor spatial layout, refrigeration equipment system structure, etc.) and the refrigeration system (such as the models of refrigeration equipment and network equipment, and various parameter indicators of chillers and cooling towers). This establishes a computer room temperature field distribution model and a cooling system model, accurately depicting the changes in temperature at each measuring point in the target area over time and space, as well as the relationships between various variables of the cooling system.

[0114] In some embodiments, if the loss value of the initial CFD model is determined to be greater than or equal to a loss value threshold, the model training device may update the model parameters until the loss value of the CFD model is less than the loss value threshold.

[0115] Understandably, if the loss value of the initial CFD model is greater than or equal to the loss value threshold, it indicates that the accuracy of the initial CFD model is low. At this time, the model training device can update the model parameters until the loss value is less than the loss value threshold, and a target CFD model with higher accuracy can be obtained.

[0116] The various solutions in the above embodiments of this disclosure can be combined without contradiction.

[0117] Some embodiments of this disclosure can divide the model training device into functional modules or functional units according to the above method examples. For example, each function can be divided into a separate functional module or functional unit, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or in software as a functional module or functional unit. Here, the division of modules or units in some embodiments of this disclosure is illustrative and is only a logical functional division; in actual implementation, there may be other division methods.

[0118] Figure 7 shows a block diagram of a model training device 700, which can be a model training device or a chip applied in a model training device. The model training device 700 can be used to perform the functions of the model training device involved in the above embodiments.

[0119] The model training device 700 shown in Figure 7 may include an input unit 701 and a processing unit 702. The input unit 701 is used to input multiple sets of verification parameters of the cooling device into the scene simulation model to obtain scene simulation data corresponding to each set of verification parameters. Here, the cooling device is used to cool multiple network devices; the multiple sets of verification parameters include multiple sets of cooling parameters under different boundary conditions; the scene simulation data includes the energy consumption of the cooling device and the temperature data of the area where each network device is located.

[0120] Processing unit 702 is used to train the reinforcement model based on multiple sets of validation parameters to obtain the target reinforcement model. Here, the target reinforcement model is used to determine the target cooling parameters based on the current boundary conditions of the cooling equipment; the target cooling parameters are the cooling parameters that minimize the energy consumption of the cooling equipment from multiple sets of candidate cooling parameters; when the cooling equipment executes the candidate cooling parameters, the temperature data of the area where each network device is located is less than the temperature threshold.

[0121] In some embodiments, the processing unit 702 is configured to: determine the reward value for executing each set of verification parameters based on the scene simulation data and reward function corresponding to each set of verification parameters; and train the reinforcement model based on multiple sets of verification parameters and the reward value of each set of verification parameters to obtain the target reinforcement model.

[0122] In some embodiments, the processing unit 702 is configured to: input multiple sets of verification parameters into a mechanistic model to obtain air supply parameters corresponding to each set of verification parameters. Here, the air supply parameters include air volume and air supply temperature. The processing unit 702 is also configured to input the air supply parameters corresponding to each set of verification parameters, outdoor temperature, outdoor humidity, and the power of multiple network devices into an artificial intelligence computational fluid dynamics (AI-CFD) model to obtain scenario simulation data corresponding to each set of verification parameters.

[0123] In some embodiments, the processing unit 702 is further configured to generate multiple sets of training data based on the target computational fluid dynamics (CFD) model. Here, each set of training data includes air supply parameters, the power of each network device, and the temperature data of the area where each network device is located. The processing unit 702 is further configured to train a preset neural network model based on the multiple sets of training data to obtain an AI-CFD model.

[0124] In some embodiments, the model training apparatus 700 further includes an acquisition unit 703. The acquisition unit 703 is used to acquire historical temperature data, historical air supply parameters, and three-dimensional structural data of the target area. Here, the three-dimensional structural data includes the location of each device in the target area and the configuration of the flow field; each device includes network devices and cooling devices.

[0125] The processing unit 702 is also used to construct a target CFD model based on historical temperature data, historical air supply parameters and three-dimensional structural data.

[0126] In some embodiments, the processing unit 702 is configured to: construct an initial CFD model based on historical temperature data, historical air supply parameters, and three-dimensional structural data; determine the loss value of the initial CFD model, wherein the loss value is used to characterize the degree of loss between the temperature data output by the initial CFD model and the actual temperature data; and determine the CFD model as the target CFD model if the loss value of the initial CFD model is less than a loss value threshold.

[0127] In some embodiments, the boundary conditions include at least one of the following: outdoor dry-bulb temperature, outdoor wet-bulb temperature, chilled water supply temperature, chilled water flow rate, cooling load power, fan power, or temperature threshold. Cooling parameters include at least one of the following: return air valve opening, fresh air valve opening, return air fan frequency, supply air fan frequency, chilled water valve opening, or supply air temperature setpoint.

[0128] This disclosure also provides a computer-readable storage medium in some embodiments. All or part of the processes in the above method embodiments can be implemented by a computer program instructing related hardware. This program can be stored in the computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. The computer-readable storage medium can be an internal storage unit of the model training apparatus (including at least one of a data transmitter or data receiver) in any of the foregoing embodiments, such as the hard disk or memory of the model training apparatus. The computer-readable storage medium can also be an external storage device of the terminal device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the terminal device. In some embodiments, the computer-readable storage medium can include both the internal storage unit of the model training apparatus and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by the model training apparatus. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0129] It should be noted that the terms "comprising" and "having," and any variations thereof, in the specification, claims, and drawings of this disclosure are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may also include steps or units not listed, or may include other steps or units inherent to such processes, methods, products, or apparatus.

[0130] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0131] In the several embodiments provided in this disclosure, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection between the devices or units shown or discussed may be through some interfaces, and the indirect coupling or communication connection between the devices or units may be electrical, mechanical, or other forms.

[0132] The units described as separate components may or may not be physically separate. A component shown as a unit can be one or more physical units; that is, it can be located in one place or distributed in multiple different locations. Some or all of the units can be selected to achieve the objectives of some embodiments of this disclosure, depending on actual needs.

[0133] Furthermore, the functional units in the various embodiments of this disclosure can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0134] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of some embodiments of this disclosure, in essence, or the part that contributes to related technologies, or all or part of the technical solution, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0135] The above are merely specific embodiments of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any changes or substitutions within the technical scope disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A model training method comprising: inputting a plurality of sets of verification parameters of a refrigeration device into a scene simulation model to obtain scene simulation data corresponding to each set of verification parameters in the plurality of sets of verification parameters; the refrigeration device is used to cool a plurality of network devices; the plurality of sets of verification parameters comprises a plurality of sets of refrigeration parameters under different boundary conditions; the scene simulation data comprises energy consumption of the refrigeration device and temperature data of a region where each network device in the plurality of network devices is located; and training a reinforcement model based on the plurality of sets of verification parameters to obtain a target reinforcement model; the target reinforcement model is used to determine a target refrigeration parameter based on a current boundary condition of the refrigeration device; the target refrigeration parameter is a refrigeration parameter with the minimum energy consumption of the refrigeration device in a plurality of sets of candidate refrigeration parameters; temperature data of the region where each network device is located when the refrigeration device executes the candidate refrigeration parameter is less than a temperature threshold.

2. The method of claim 1, wherein, The training of the reinforcement model based on the plurality of sets of verification parameters to obtain the target reinforcement model comprises: determining a reward value of executing each set of verification parameters based on the scene simulation data corresponding to each set of verification parameters and a reward function; training the reinforcement model based on the plurality of sets of verification parameters and the reward value of each set of verification parameters to obtain the target reinforcement model.

3. The method of claim 1 or 2, wherein, The inputting of the plurality of sets of verification parameters of the refrigeration device into the scene simulation model to obtain the scene simulation data corresponding to each set of verification parameters in the plurality of sets of verification parameters comprises: inputting the plurality of sets of verification parameters into a mechanism model to obtain air supply parameters corresponding to each set of verification parameters; the air supply parameters comprise air supply amount and air supply temperature; inputting the air supply parameters corresponding to each set of verification parameters, outdoor temperature, outdoor humidity, and power of the plurality of network devices into an artificial intelligence computational fluid dynamics (AI-CFD) model to obtain scene simulation data corresponding to each set of verification parameters.

4. The method of claim 3, further comprising: generating a plurality of sets of training data based on a target computational fluid dynamics (CFD) model; each set of training data in the plurality of sets of training data comprises air supply parameters, power of each network device, and temperature data of a region where the corresponding each network device is located; training a preset neural network model based on the plurality of sets of training data to obtain the AI-CFD model.

5. The method of claim 4, further comprising: obtaining historical temperature data, historical air supply parameters, and three-dimensional structure data of a target region; the three-dimensional structure data comprises positions of devices in the target region and configurations of flow fields; the devices comprise network devices and refrigeration devices; constructing the target CFD model based on the historical temperature data, the historical air supply parameters, and the three-dimensional structure data.

6. The method of claim 5, wherein, The construction of the target CFD model based on the historical temperature data, the historical air supply parameters, and the three-dimensional structure data comprises: constructing an initial CFD model based on the historical temperature data, the historical air supply parameters, and the three-dimensional structure data; determine a loss value of the initial CFD model, the loss value being used to represent a degree of loss between temperature data output by the initial CFD model and real temperature data; in a case where the loss value of the initial CFD model is less than a loss value threshold, determine the CFD model as the target CFD model.

7. The method of any one of claims 1 to 6, wherein, The boundary condition comprises at least one of the following: outdoor air dry-bulb temperature, outdoor air wet-bulb temperature, chilled water supply temperature, chilled water flow, cooling load power, fan power, or temperature threshold. The refrigeration parameter comprises at least one of the following: return air damper opening, fresh air damper opening, return air fan frequency, supply air fan frequency, chilled water valve opening, or supply air temperature set value.

8. A model training apparatus, comprising: an input unit configured to input a plurality of sets of verification parameters of a refrigeration device into a scenario simulation model, to obtain scenario simulation data corresponding to each set of verification parameters in the plurality of sets of verification parameters; the refrigeration device being configured to cool a plurality of network devices; the plurality of sets of verification parameters comprising a plurality of sets of refrigeration parameters under different boundary conditions; the scenario simulation data comprising energy consumption of the refrigeration device and temperature data of an area in which each network device in the plurality of network devices is located; and a processing unit configured to train a reinforcement model based on the plurality of sets of verification parameters, to obtain a target reinforcement model; the target reinforcement model being configured to determine a target refrigeration parameter based on a current boundary condition of the refrigeration device; the target refrigeration parameter being a refrigeration parameter in a plurality of sets of candidate refrigeration parameters, which has the minimum energy consumption of the refrigeration device; when the refrigeration device executes the candidate refrigeration parameter, the temperature data of the area in which each network device is located is less than a temperature threshold.

9. A computer readable storage medium, wherein, The readable storage medium has instructions stored therein, which, when executed, implement the method according to any one of claims 1 to 7.

10. A model training apparatus comprising: a processor, a memory, and a communication interface; wherein the communication interface is configured to communicate with other devices or networks; the memory is configured to store one or more programs, the one or more programs comprising computer execution instructions; and the processor is configured to execute the computer execution instructions stored in the memory, so that the model training apparatus executes the method according to any one of claims 1 to 7.

Images