Adaptive Operation and Maintenance Control Method and Device for Liquid Cooling Systems in Intelligent Computing Data Centers

Abstract

This invention relates to the field of intelligent operation and maintenance technology, and discloses an adaptive operation and maintenance control method and device for a liquid cooling system in an intelligent computing data center. The method includes: collecting multi-dimensional real-time data associated with the liquid cooling system; the multi-dimensional real-time data includes computing load data of the intelligent computing data center, real-time temperature data, and real-time flow data of the liquid cooling system; analyzing the pre-processed real-time data based on an operation and maintenance coupling analysis model to obtain the predicted flow data required by the intelligent computing data center; determining the optimal flow command based on the predicted flow data using an adaptive optimization algorithm; the optimization objectives of the adaptive optimization algorithm include at least temperature control objectives and energy efficiency optimization objectives; and controlling the liquid cooling system to perform cooling operations for the intelligent computing data center according to the optimal flow command. Therefore, implementing this invention can achieve intelligent adaptive adjustment of the liquid cooling system, improve the control accuracy of the liquid cooling system, and thus improve the heat dissipation efficiency of the intelligent computing data center.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology, and in particular to an adaptive operation and maintenance control method and device for liquid cooling systems in intelligent computing data centers. Background Technology

[0002] With the rapid development of artificial intelligence technology, intelligent computing data centers, as core infrastructure supporting high computing power demands such as large-scale model training, autonomous driving, and scientific computing, are experiencing exponential growth in computing density. Compared to traditional data centers, high-density servers generate a large amount of heat during operation. If heat dissipation is not timely, chip temperatures will rapidly rise to the protection threshold, causing not only computing power throttling and performance degradation but also severely impacting hardware lifespan. Therefore, efficient heat dissipation has become a core challenge for ensuring the stable operation of intelligent computing data centers.

[0003] In existing technologies, liquid cooling technology has gradually become the mainstream heat dissipation solution for high-density computing clusters due to its excellent heat dissipation efficiency and low energy consumption. However, the operation and maintenance methods of existing liquid cooling systems still have significant shortcomings, making it difficult to adapt to the dynamic changes in computing load, temperature sensitivity, and high energy efficiency requirements of intelligent computing scenarios. Specifically, current flow regulation strategies are disconnected from the computing load, often using fixed flow rates or control methods based on a single temperature threshold. This fails to respond to load fluctuations in real time, easily leading to problems of "overcooling" or "undercooling," resulting in low heat dissipation efficiency and even creating a vicious cycle of temperature increase and performance degradation.

[0004] Therefore, it is particularly important to propose a technical solution that can realize intelligent adaptive adjustment of the liquid cooling system, improve the control accuracy of the liquid cooling system, thereby improving the heat dissipation efficiency of the intelligent computing center and thus improving the operational stability of the intelligent computing center. Summary of the Invention

[0005] This invention provides an adaptive operation and maintenance control method and device for liquid cooling systems in intelligent computing data centers. It can realize intelligent adaptive adjustment of the liquid cooling system, improve the control accuracy of the liquid cooling system, thereby improving the heat dissipation efficiency of the intelligent computing center and enhancing the operational stability of the intelligent computing center.

[0006] To address the aforementioned technical problems, the first aspect of this invention discloses an adaptive operation and maintenance control method for a liquid cooling system in an intelligent computing data center, the method comprising:

[0007] Collect multi-dimensional real-time data associated with the liquid cooling system; the multi-dimensional real-time data includes computing load data, real-time temperature data, and real-time flow data of the liquid cooling system corresponding to the intelligent computing data center that the liquid cooling system needs to dissipate heat; the real-time temperature data includes first real-time temperature data corresponding to the intelligent computing data center and second real-time temperature data corresponding to the liquid cooling system;

[0008] Preprocessing operations are performed on the multi-dimensional real-time data to obtain preprocessed real-time data; the preprocessing operations include outlier removal, data standardization, and data fusion.

[0009] Based on the determined operation and maintenance coupling analysis model, the preprocessed real-time data is analyzed to obtain the predicted traffic data required by the intelligent computing data center.

[0010] Based on the determined adaptive optimization algorithm, the optimal flow command is determined according to the predicted flow data; the adaptive optimization algorithm is constructed based on the MPC algorithm, and the adaptive optimization algorithm is a multi-objective optimization algorithm, and the objectives to be optimized by the adaptive optimization algorithm include at least temperature control objectives and energy efficiency optimization objectives;

[0011] According to the optimal flow command, the liquid cooling system is controlled to perform cooling operations on the intelligent computing data center.

[0012] A second aspect of this invention discloses an adaptive operation and maintenance control device for a liquid cooling system in an intelligent computing data center, the device comprising:

[0013] The acquisition module is used to acquire multi-dimensional real-time data associated with the liquid cooling system; the multi-dimensional real-time data includes computing load data, real-time temperature data, and real-time flow data corresponding to the intelligent computing data center that the liquid cooling system needs to dissipate heat from; the real-time temperature data includes first real-time temperature data corresponding to the intelligent computing data center and second real-time temperature data corresponding to the liquid cooling system;

[0014] The preprocessing module is used to perform preprocessing operations on the multi-dimensional real-time data to obtain preprocessed real-time data; the preprocessing operations include outlier removal, data standardization, and data fusion.

[0015] The analysis module is used to analyze the preprocessed real-time data based on the determined operation and maintenance coupling analysis model to obtain the predicted traffic data required by the intelligent computing data center.

[0016] The determination module is used to determine the optimal flow instruction based on the predicted flow data and the determined adaptive optimization algorithm; the adaptive optimization algorithm is constructed based on the MPC algorithm and is a multi-objective optimization algorithm, and the objectives to be optimized by the adaptive optimization algorithm include at least temperature control objectives and energy efficiency optimization objectives;

[0017] The cooling control module is used to control the liquid cooling system to perform cooling operations on the intelligent computing data center according to the optimal flow command.

[0018] A third aspect of this invention discloses another adaptive operation and maintenance control device for liquid cooling systems in intelligent computing data centers, the device comprising:

[0019] Memory containing executable program code;

[0020] A processor coupled to the memory;

[0021] The processor calls the executable program code stored in the memory to execute some or all of the steps of the adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers disclosed in the first aspect of the present invention.

[0022] The fourth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute some or all of the steps of the adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers disclosed in the first aspect of the present invention.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] In this invention, multi-dimensional real-time data associated with the liquid cooling system is collected. This multi-dimensional real-time data includes computing load data, real-time temperature data, and real-time flow data corresponding to the intelligent computing data center requiring heat dissipation from the liquid cooling system. The real-time temperature data includes first real-time temperature data corresponding to the intelligent computing data center and second real-time temperature data corresponding to the liquid cooling system. Preprocessing operations are performed on the multi-dimensional real-time data to obtain preprocessed real-time data. These preprocessing operations include outlier removal, data standardization, and data fusion. Based on a determined operation and maintenance coupling analysis model, the preprocessed real-time data is analyzed to obtain predicted flow data required by the intelligent computing data center. Based on a determined adaptive optimization algorithm, the optimal flow command is determined according to the predicted flow data. The adaptive optimization algorithm is built based on the MPC algorithm and is a multi-objective optimization algorithm, with the optimization objectives including at least temperature control and energy efficiency optimization objectives. According to the optimal flow command, the liquid cooling system is controlled to perform cooling operations on the intelligent computing data center. As can be seen, implementing this invention can obtain the predicted flow data required by the intelligent computing data center by collecting multi-dimensional real-time data, analyzing the pre-processed real-time data based on the operation and maintenance coupling analysis model, and then determining the optimal flow command based on the adaptive optimization algorithm and the predicted flow data. The liquid cooling system is then controlled to cool the intelligent computing data center based on the optimal flow command, which enables intelligent adaptive adjustment of the liquid cooling system. Through multi-objective optimization and optimal flow calculation, the operation and maintenance adjustment scheme of the liquid cooling system can be made more adaptable to the high volatility of the intelligent computing load, while reducing the energy consumption of the liquid cooling system. This improves the control accuracy of the liquid cooling system, thereby improving the cooling precision and efficiency of the liquid cooling system for the intelligent computing data center, and thus improving the heat dissipation efficiency of the intelligent computing data center. It effectively avoids GPU frequency reduction in the intelligent computing data center due to excessive temperature, which in turn helps the intelligent computing data center to execute intelligent computing tasks more efficiently and improve the operational stability of the intelligent computing data center. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart illustrating an adaptive operation and maintenance control method for a liquid cooling system in an intelligent computing data center, as disclosed in an embodiment of the present invention.

[0027] Figure 2 This is a flowchart illustrating another adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers disclosed in an embodiment of the present invention.

[0028] Figure 3 This is a schematic diagram of the architecture of an adaptive operation and maintenance control system for a liquid cooling system in an intelligent computing data center, as disclosed in an embodiment of the present invention.

[0029] Figure 4 This is a flowchart illustrating another adaptive operation and maintenance control method for a liquid cooling system in an intelligent computing data center, as disclosed in an embodiment of the present invention.

[0030] Figure 5 This is a flowchart illustrating another adaptive operation and maintenance control method for a liquid cooling system in an intelligent computing data center, as disclosed in an embodiment of the present invention.

[0031] Figure 6 This is a schematic diagram of the structure of an adaptive operation and maintenance control device for a liquid cooling system in an intelligent computing data center, as disclosed in an embodiment of the present invention.

[0032] Figure 7 This is a schematic diagram of another adaptive operation and maintenance control device for a liquid cooling system in an intelligent computing data center, as disclosed in an embodiment of the present invention.

[0033] Figure 8 This is a schematic diagram of the structure of another adaptive operation and maintenance control device for liquid cooling systems in intelligent computing data centers, as disclosed in an embodiment of the present invention. Detailed Implementation

[0034] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0036] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0037] This invention discloses an adaptive operation and maintenance control method and device for liquid cooling systems in intelligent computing data centers. It can collect multi-dimensional real-time data, analyze the pre-processed real-time data based on an operation and maintenance coupling analysis model to obtain the predicted flow data required by the intelligent computing data center, and then determine the optimal flow command based on an adaptive optimization algorithm and the predicted flow data. This optimal flow command controls the liquid cooling system to cool the intelligent computing data center, enabling intelligent adaptive adjustment of the liquid cooling system. Through multi-objective optimization and optimal flow calculation, it can make the operation and maintenance adjustment scheme of the liquid cooling system more adaptable to the high volatility of intelligent computing loads while reducing the energy consumption of the liquid cooling system. This improves the accuracy of liquid cooling system control, thereby enhancing the cooling precision and efficiency of the liquid cooling system for the intelligent computing data center, improving the heat dissipation efficiency of the intelligent computing data center, effectively avoiding GPU throttling caused by excessive temperature, and ultimately enabling the intelligent computing data center to execute intelligent computing tasks more efficiently and improve its operational stability. Detailed descriptions follow.

[0038] Example 1

[0039] Please see Figure 1 , Figure 1 This is a flowchart illustrating an adaptive operation and maintenance control method for a liquid cooling system in an intelligent computing data center, as disclosed in an embodiment of the present invention. Figure 1 The described adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers can be applied to adaptive operation and maintenance control devices for liquid cooling systems in intelligent computing data centers. These devices may include one of the following: intelligent devices, intelligent terminals, intelligent systems, and servers. The servers may include local servers or cloud servers; this embodiment of the invention does not limit the specific type of server. Figure 1 As shown, the adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers can include the following operations:

[0040] 101. Collect multi-dimensional real-time data related to the liquid cooling system.

[0041] In this embodiment of the invention, the multi-dimensional real-time data associated with the liquid cooling system may include computing load data, real-time temperature data, and real-time flow data corresponding to the intelligent computing data center that the liquid cooling system needs to dissipate heat; wherein, the real-time temperature data may include first real-time temperature data corresponding to the intelligent computing data center and second real-time temperature data corresponding to the liquid cooling system.

[0042] In this embodiment of the invention, optionally, the computing load data corresponding to the intelligent computing data center can be collected in the following ways:

[0043] Using GPU (Graphics Processing Unit) management tools, such as NVIDIA DCGM (Data Center GPU Manager) or GPU-related commands in the operating system, the real-time computing power utilization, memory usage, and computing core frequency of each GPU are collected based on a pre-set sampling frequency (e.g., once per second). Additionally, based on a pre-set sampling frequency (e.g., once every 5 seconds), intelligent computing task information is collected through the intelligent computing platform API interface. This intelligent computing task information may include one or more of the following: task type (e.g., training / inference), task priority (e.g., high / medium / low), and task parallelism (e.g., data parallelism / tensor parallelism). This embodiment of the invention does not impose any limitations on these aspects.

[0044] In this embodiment of the invention, real-time temperature data can optionally be collected in the following ways:

[0045] Based on a pre-set acquisition frequency (e.g., once / 0.5 seconds), one or more of the following are collected using distributed fiber Bragg grating sensors: GPU core temperature, server surface temperature, coolant inlet temperature, and coolant outlet temperature. The first real-time temperature data may include GPU core temperature and / or server surface temperature, and the second real-time temperature data may include coolant inlet temperature and / or coolant outlet temperature. This embodiment of the invention does not limit the specific data collection frequency. Further optionally, the deployment density of the distributed fiber Bragg grating sensors can be set to one per GPU core and one per coolant inlet / outlet. This embodiment of the invention does not limit the specific data collection frequency. Further optionally, based on a pre-set acquisition frequency (e.g., once / 10 seconds), an infrared thermal imaging camera can be used to collect the overall temperature field distribution of the GPU cluster to verify the accuracy of single-point temperature data. This embodiment of the invention does not limit the specific temperature data collection frequency.

[0046] In this embodiment of the invention, the real-time flow data corresponding to the liquid cooling system can optionally be collected in the following ways:

[0047] Based on a pre-set acquisition frequency (e.g., once per second), real-time flow rate (L / min) and / or flow pressure (MPa) are collected by electromagnetic flow meters installed in the inlet pipes of each coolant; and, based on a pre-set acquisition frequency (e.g., once per 2 seconds), the operating parameters of the liquid cooling pump group (e.g., speed, power, current) are obtained through the Modbus protocol of the pump group controller. This embodiment of the invention is not limited.

[0048] 102. Perform preprocessing operations on multi-dimensional real-time data to obtain preprocessed real-time data.

[0049] In this embodiment of the invention, the preprocessing operations include outlier removal, data standardization, and data fusion.

[0050] In this embodiment of the invention, the outlier removal operation may optionally include: removing outlier data (such as sudden temperature rise / fall or flow rate of 0 caused by sensor failure) using the 3σ criterion, and filling in the missing data after removal using linear interpolation to ensure data continuity.

[0051] In this embodiment of the invention, optionally, the data standardization operation may specifically include: standardizing data of different dimensions such as computing load (0-100%), temperature (20-100℃), and flow rate (0-10L / min) to the [0,1] interval, wherein the standardization formula may be:

[0052]

[0053] Where x is the original data, , These are the historical minimum and maximum values ​​of the parameter, respectively, to avoid the impact of dimensional differences on subsequent model calculations.

[0054] In this embodiment of the invention, optionally, the data fusion operation may specifically include: using a Kalman filter algorithm to fuse the single-point temperature data collected by the fiber optic grating sensor and the temperature field data collected by the infrared thermal imaging, so as to correct the single-point temperature measurement error and reduce the error from ±0.5℃ to ±0.2℃; and, by weighted summation, fusing the GPU computing power utilization rate and the computing core frequency (e.g., the weights can be 0.6 and 0.4 respectively) to obtain the Comprehensive Computing Load Index (CLI) to quantify the actual load level of the GPU;

[0055] Alternatively, the formula for calculating the comprehensive computing power load index can be:

[0056] CLI = 0.6 × CU + 0.4 × CF / CFmax

[0057] Where CU is the computing power utilization rate (%), CF is the actual core frequency (MHz), and CFmax is the GPU maximum core frequency (MHz).

[0058] 103. Based on the established operation and maintenance coupling analysis model, analyze and preprocess real-time data to obtain the predicted traffic data required by the intelligent computing data center.

[0059] In this embodiment of the invention, the operation and maintenance coupling analysis model includes a heat generation analysis model, a temperature distribution analysis model, and a flow rate analysis model; wherein, optionally, the heat generation analysis model is a model used to reflect the mapping relationship between computing load and heat generation, the temperature distribution analysis model is a heat conduction model used to describe the heat generation-temperature distribution, and the flow rate analysis model is a demand model used to describe the temperature-flow rate. This embodiment of the invention does not impose any limitations.

[0060] 104. Based on the determined adaptive optimization algorithm, determine the optimal flow instruction according to the predicted flow data.

[0061] In this embodiment of the invention, the adaptive optimization algorithm is constructed based on the MPC (Model Predictive Control) algorithm, and the adaptive optimization algorithm is a multi-objective optimization algorithm. The objectives to be optimized by the adaptive optimization algorithm include at least the temperature control objective and the energy efficiency optimization objective.

[0062] In this embodiment of the invention, optionally, the optimal flow rate command is used to instruct the liquid cooling system to reach the optimal flow rate in the current cooling control cycle.

[0063] 105. Control the liquid cooling system to perform cooling operations for the intelligent computing data center according to the optimal flow command.

[0064] In this embodiment of the invention, step 105 may optionally include the following operations:

[0065] The intelligent computing cluster in the intelligent computing data center is divided into multiple cooling zones;

[0066] Based on a pre-defined command issuance strategy, the optimal flow command is issued to each cooling zone according to the optimal flow command; wherein, the command issuance strategy can be a zone control strategy.

[0067] For each cooling zone, cooling control parameters corresponding to the zone flow command are generated based on the zone flow command. Based on the cooling control parameters, the liquid cooling system is controlled to perform cooling operations for that cooling zone.

[0068] In this embodiment of the invention, for example, a partitioned control strategy is adopted for traffic allocation of a multi-GPU cluster: the intelligent computing cluster is divided into multiple cooling zones by rack, each zone is configured with an independent flow controller, and a dedicated zone flow command is output for each zone through the MPC algorithm to avoid cross-zone flow interference (e.g., high load in one zone leads to insufficient flow in other zones); the zone flow command is transmitted to the distributed flow controller of the liquid cooling system (e.g., one per server rack) through industrial Ethernet (Profinet protocol). The flow controller converts the flow command into a pump speed control signal to adjust the coolant flow of the corresponding pipeline to achieve cooling of the cooling zone.

[0069] As can be seen, the method described in the embodiments of the present invention can obtain the predicted flow data required by the intelligent computing data center by collecting multi-dimensional real-time data, analyzing the preprocessed real-time data based on the operation and maintenance coupling analysis model, and then determining the optimal flow command based on the adaptive optimization algorithm and the predicted flow data. The liquid cooling system is then controlled to cool the intelligent computing data center based on the optimal flow command. This enables intelligent adaptive adjustment of the liquid cooling system. Through multi-objective optimization and optimal flow calculation, the operation and maintenance adjustment scheme of the liquid cooling system can be made more adaptable to the high volatility of the intelligent computing load while reducing the energy consumption of the liquid cooling system. This improves the control accuracy of the liquid cooling system, thereby improving the cooling precision and efficiency of the liquid cooling system for the intelligent computing data center, and thus improving the heat dissipation efficiency of the intelligent computing data center. This effectively avoids GPU frequency reduction in the intelligent computing data center due to excessive temperature, which in turn helps the intelligent computing data center to execute intelligent computing tasks more efficiently and improve the operational stability of the intelligent computing data center.

[0070] In an optional embodiment, the method may further include the following operations:

[0071] During the process of controlling the liquid cooling system to perform cooling operations on the intelligent computing data center, the operating status data is monitored; the operating status data includes the first operating status data corresponding to the liquid cooling system and the second operating status data corresponding to the intelligent computing data center.

[0072] Based on the operating status data, determine whether the liquid cooling system meets the abnormal operating conditions corresponding to at least one preset abnormality type; all preset abnormality types include one or more of the following: flow regulation abnormality type, core temperature abnormality type, and pump group abnormality type;

[0073] When it is determined that the liquid cooling system meets the abnormal operating conditions corresponding to at least one preset abnormality type, each preset abnormality type met by the liquid cooling system is identified as a target abnormality type, and according to each target abnormality type, the liquid cooling system is controlled to perform the emergency control operation corresponding to that target abnormality type.

[0074] Optionally, the first operating status data may include the actual flow rate of the liquid cooling system and the actual equipment operating information of the liquid cooling system; the second operating status data may include the actual GPU core temperature of the intelligent computing data center, which is not limited in this embodiment of the invention.

[0075] As can be seen, this optional embodiment can monitor the operating status during the cooling process of the intelligent computing data center by the liquid cooling system, and determine whether the abnormal operating conditions corresponding to at least one preset abnormality type are met. If it is determined that at least one abnormal operating condition is met, the liquid cooling system is controlled to perform the emergency control operation corresponding to the abnormality type. This can realize the full automation of data acquisition, model calculation, instruction execution, and abnormality handling, and improve the timeliness and comprehensiveness of abnormality detection in the cooling process of the liquid cooling system. This is conducive to taking more timely and targeted effective abnormality handling measures, thereby improving the control accuracy of the liquid cooling system and the operational reliability of the liquid cooling system and the intelligent computing data center.

[0076] In this optional embodiment, determining whether the liquid cooling system meets the abnormal operating conditions corresponding to at least one preset abnormality type based on the operating status data may include the following operations:

[0077] Calculate the flow difference between the actual flow and the command flow corresponding to the optimal flow command;

[0078] Determine whether the traffic difference is greater than a preset traffic difference threshold, and whether the duration for which the traffic difference is greater than the traffic difference threshold is greater than or equal to a preset abnormal duration threshold.

[0079] When it is determined that the flow difference is greater than the flow difference threshold, and the duration of the flow difference being greater than the flow difference threshold is greater than or equal to the abnormal duration threshold, the liquid cooling system is determined to meet the abnormal operating conditions corresponding to the abnormal flow regulation type.

[0080] And / or,

[0081] Determine whether the actual GPU core temperature is greater than a preset core temperature threshold.

[0082] When it is determined that the actual GPU core temperature is greater than the core temperature threshold, the liquid cooling system is determined to meet the abnormal operating conditions corresponding to the core temperature abnormality type.

[0083] And / or,

[0084] Based on the actual equipment operating information, determine whether there is a fault in the currently operating pump set of the liquid cooling system;

[0085] When it is determined that the currently operating pump set of the liquid cooling system is faulty, the liquid cooling system is confirmed to meet the abnormal operating conditions corresponding to the pump set abnormality type.

[0086] For example, assuming the flow difference threshold is 0.2 L / min and the abnormal duration threshold is 3 seconds, when the flow difference is > 0.2 L / min and lasts for 3 seconds, it can be determined that the liquid cooling system meets the abnormal operating conditions corresponding to the abnormal flow regulation type. This embodiment of the invention does not impose any limitations.

[0087] For example, assuming the core temperature threshold is 85°C, when the actual GPU core temperature is greater than 85°C, it can be determined that the liquid cooling system meets the abnormal operating conditions corresponding to the core temperature abnormality type. This embodiment of the invention does not impose any limitations.

[0088] For example, determining whether the currently operating pump group of the liquid cooling system is faulty based on the actual equipment operation information can be specifically as follows: if the actual equipment operation information detects that the speed of the currently operating pump group of the liquid cooling system is unresponsive and / or the current is abnormal, then it can be determined that the currently operating pump group of the liquid cooling system is faulty. This embodiment of the invention does not limit this.

[0089] As can be seen, this optional embodiment can also determine that the liquid cooling system has an abnormal flow regulation when the flow difference is greater than the flow difference threshold for an extended period of time; it can also determine that the core temperature is abnormal when the GPU core temperature is too high; and it can also determine that the liquid cooling system has a pump malfunction when the currently running pump group is faulty. This improves the flexibility and accuracy of determining whether the liquid cooling system has an operational abnormality, thereby improving the accuracy of determining the necessary emergency control operations. This, in turn, facilitates more accurate and timely handling of abnormalities that occur during the cooling process, thereby improving the operational reliability of the liquid cooling system and the intelligent computing data center.

[0090] In this optional embodiment, optionally, according to each target anomaly type, controlling the liquid cooling system to perform the emergency control operation corresponding to that target anomaly type may include the following operations:

[0091] When the target anomaly type is the flow regulation anomaly type, the cooling mode of the liquid cooling system is switched to the standby regulation mode, and the liquid cooling system is controlled to perform the standby cooling operation corresponding to the standby regulation mode, and an alarm signal is output.

[0092] When the target anomaly type is a core temperature anomaly, the corresponding emergency cooling mechanism of the liquid cooling system is triggered to control the liquid cooling system to increase the flow rate corresponding to the cooling operation to the maximum flow rate of the pump group, and to control the intelligent computing data center to suspend the GPU from executing the target load task until the actual GPU core temperature drops to less than or equal to the target core temperature; the target load task is an intelligent computing task with a load higher than the preset load threshold.

[0093] When the target anomaly type is a pump set anomaly, the control liquid cooling system will switch the currently running pump set to the standby pump set so that the corresponding standby pump set of the liquid cooling system can continue to perform cooling operations.

[0094] The backup regulation mode is a temporary regulation mode based on the PID algorithm. Furthermore, the PID algorithm is a proportional, integral, and derivative control algorithm. Specifically, the proportional stage directly outputs the corresponding control quantity based on the deviation between the current actual flow rate and the commanded flow rate, reacting quickly to the deviation; the integral stage is used to eliminate long-term deviations, gradually adjusting the control quantity through integral calculations of historical deviations, so that the system can eventually stabilize near the commanded flow rate; the derivative stage can predict the trend of deviation changes and output the control quantity in advance based on the rate of change of the deviation, thereby reducing overshoot and oscillations during the regulation process. Through the synergistic effect of these three stages, the controller continuously calculates and outputs appropriate control signals to adjust the flow regulation components of the liquid cooling system (such as the pump speed), making the actual flow rate as close as possible to the commanded flow rate to maintain the normal cooling function of the liquid cooling system.

[0095] For example, assuming the core temperature threshold is 85°C, the target core temperature is usually set to 80°C to reserve a 5°C buffer space. This embodiment of the invention does not limit this.

[0096] Further optional, when the backup pump group is unavailable, a computing power limiting emergency strategy can be adopted: by limiting the computing power utilization of high-load tasks (e.g., from 90% to 60%), the heat generation of the GPU is reduced, thereby maintaining temperature stability when the flow is insufficient, sacrificing some computing power but ensuring that the system is not interrupted.

[0097] As can be seen, this optional embodiment can also switch the cooling mode of the liquid cooling system to the standby regulation mode and control the liquid cooling system to perform standby cooling operations when the target anomaly type is a flow regulation anomaly type, and output an alarm signal; when the target anomaly type is a core temperature anomaly type, it triggers the corresponding emergency cooling mechanism of the liquid cooling system to control the liquid cooling system to increase the flow rate corresponding to the cooling operation to the maximum flow rate of the pump group, and controls the intelligent computing data center to suspend the GPU from executing the target load task until the actual GPU core temperature drops to less than or equal to the target core temperature; when the target anomaly type is a pump group anomaly type, it controls the liquid cooling system to switch to the standby pump group to continue performing the cooling operation. It can efficiently take targeted emergency control operations for different operational anomalies, thereby further improving the timeliness and accuracy of anomaly handling, and improving the operational reliability of the liquid cooling system and the intelligent computing data center.

[0098] Example 2

[0099] Please see Figure 2 , Figure 2 This is a flowchart illustrating an adaptive operation and maintenance control method for a liquid cooling system in an intelligent computing data center, as disclosed in an embodiment of the present invention. Figure 2 The described adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers can be applied to adaptive operation and maintenance control devices for liquid cooling systems in intelligent computing data centers. These devices may include one of the following: intelligent devices, intelligent terminals, intelligent systems, and servers. The servers may include local servers or cloud servers; this embodiment of the invention does not limit the specific type of server. Figure 2 As shown, the adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers can include the following operations:

[0100] 201. Collect multi-dimensional real-time data related to the liquid cooling system.

[0101] In this embodiment of the invention, the multi-dimensional real-time data includes computing load data, real-time temperature data, and real-time flow data corresponding to the intelligent computing data center for which the liquid cooling system needs heat dissipation; the real-time temperature data includes first real-time temperature data corresponding to the intelligent computing data center and second real-time temperature data corresponding to the liquid cooling system.

[0102] 202. Perform preprocessing operations on multi-dimensional real-time data to obtain preprocessed real-time data.

[0103] In this embodiment of the invention, the preprocessing operations include outlier removal, data standardization, and data fusion.

[0104] 203. Based on the heat generation analysis model, determine the heat generation data corresponding to the intelligent computing data center according to the computing load data.

[0105] In this embodiment of the invention, optionally, the heat generation data corresponding to the intelligent computing data center may include the heat generation data corresponding to at least one GPU, and this embodiment of the invention does not limit it.

[0106] In this embodiment of the invention, optionally, future changes in computing load can be predicted using LSTM (Long Short-Term Memory); alternatively, exponential smoothing or ARIMA (AutoRegressive Integrated Moving Average) models can be used instead, and this embodiment of the invention does not impose any limitations.

[0107] 204. Based on the temperature distribution analysis model, determine the predicted temperature data according to the calorific value data and real-time temperature data.

[0108] In this embodiment of the invention, the predicted temperature data may include the predicted core temperature of the GPU in the intelligent computing data center and the predicted outlet temperature of the coolant flowing through the intelligent computing data center in the liquid cooling system.

[0109] 205. Based on the flow analysis model, determine the predicted flow data required for the intelligent computing data center according to the predicted temperature data.

[0110] In this embodiment of the invention, the predicted flow data required by the intelligent computing data center may include the minimum flow rate of coolant and / or the optimal flow rate of coolant.

[0111] 206. Based on the determined adaptive optimization algorithm, determine the optimal flow instruction according to the predicted flow data.

[0112] In this embodiment of the invention, the adaptive optimization algorithm is constructed based on the MPC algorithm, and the adaptive optimization algorithm is a multi-objective optimization algorithm. The objectives to be optimized by the adaptive optimization algorithm include at least the temperature control objective and the energy efficiency optimization objective.

[0113] 207. Control the liquid cooling system to perform cooling operations for the intelligent computing data center according to the optimal flow command.

[0114] In this embodiment of the invention, for other detailed descriptions of steps 201-202 and steps 206-207, please refer to the detailed descriptions of steps 101-102 and steps 104-105 in Embodiment 1. These will not be repeated in this embodiment of the invention.

[0115] As can be seen, the method described in the embodiments of the present invention can obtain the predicted flow data required by the intelligent computing data center by collecting multi-dimensional real-time data, analyzing the preprocessed real-time data based on the operation and maintenance coupling analysis model, and then determining the optimal flow command based on the adaptive optimization algorithm and the predicted flow data. The liquid cooling system is then controlled to cool the intelligent computing data center based on the optimal flow command. This enables intelligent adaptive adjustment of the liquid cooling system. Through multi-objective optimization and optimal flow calculation, the operation and maintenance adjustment scheme of the liquid cooling system can be made more adaptable to the high volatility of the intelligent computing load while reducing the energy consumption of the liquid cooling system. This improves the control accuracy of the liquid cooling system, thereby improving the cooling precision and efficiency of the liquid cooling system for the intelligent computing data center, and thus improving the heat dissipation efficiency of the intelligent computing data center. This effectively avoids GPU frequency reduction in the intelligent computing data center due to excessive temperature, which in turn helps the intelligent computing data center to execute intelligent computing tasks more efficiently and improve the operational stability of the intelligent computing data center. Furthermore, it can determine the heat generation data corresponding to the computing load data based on the heat generation analysis model, then determine the predicted temperature data corresponding to the heat generation data based on the temperature distribution analysis model, and finally determine the required predicted flow rate data based on the predicted temperature data based on the flow rate analysis model. This enables coupled analysis of computing load, temperature, and flow rate, thereby improving the accuracy of predicting the heat generation of intelligent computing data centers and the accuracy of predicting the flow rate required for the predicted heat generation. This, in turn, helps improve the accuracy of generating subsequent optimal flow rate commands and improves the control accuracy of the liquid cooling system, thus enhancing the cooling precision of the liquid cooling system for intelligent computing data centers.

[0116] In an optional embodiment, determining the heat generation data corresponding to the intelligent computing data center based on the heat generation analysis model and the computing load data may include the following operations:

[0117] Determine the idle power consumption and full load power consumption of the GPU;

[0118] Determine the computational efficiency corresponding to the task type based on the task type of the intelligent computing task corresponding to the intelligent computing data center;

[0119] The actual power consumption of the GPU is determined based on the idle power consumption, full load power consumption, and the determined comprehensive computing power load index.

[0120] The heat output of the GPU is determined based on the actual power consumption and the computing efficiency.

[0121] The computational efficiency η is related to the task type of the intelligent computing task. For example, the training task involves a large number of matrix operations, so η is relatively high (60-70%); while the inference task has a small amount of data processing, so η is relatively low (40-50%). Optionally, a table of η values ​​for different task types can be obtained through experimental statistics, and the model can automatically match the η value according to the real-time task type. This embodiment of the invention does not limit this.

[0122] Optionally, the actual power consumption of the GPU is linearly related to the overall computing load index; further optionally, the formula for calculating the actual power consumption of the GPU can be as follows:

[0123]

[0124] in, This represents the actual power consumption of the GPU (in watts). This refers to the idle power consumption of the GPU (in watts, typically 50-80W). This refers to the full-load power consumption of the GPU (in watts, typically 300-500W).

[0125] Furthermore, since the electrical energy consumed by the computing load of intelligent computing tasks is partly converted into effective computing power and partly into heat energy (i.e., heat generation), the formula for calculating the heat generation of a GPU can be:

[0126]

[0127] Where Q represents the heat output of the GPU (in W), and η represents the computational efficiency.

[0128] As can be seen, this optional embodiment can utilize a heat analysis model to calculate the actual power consumption of the GPU based on the GPU idle power consumption, GPU full-load power consumption, and the determined comprehensive computing load index. Then, by combining the computing efficiency corresponding to the intelligent computing task type, the heat generation of the GPU can be calculated. This can improve the accuracy of the assessment of the actual power consumption of the GPU and the accuracy of the determination of the GPU heat generation, thereby improving the accuracy of the determination of heat generation data in intelligent computing data centers, and further improving the accuracy of subsequent prediction of GPU core temperature.

[0129] In an optional embodiment, the second real-time temperature data corresponding to the liquid cooling system includes the real-time inlet temperature of the coolant;

[0130] The process of determining predicted temperature data based on a temperature distribution analysis model, using calorific value data and real-time temperature data, may include the following operations:

[0131] The convective heat transfer coefficient between the coolant and the GPU surface, the contact area between the coolant and the GPU, and the conduction temperature difference from the GPU core to the surface are obtained.

[0132] The predicted core temperature of the GPU is calculated based on the real-time inlet temperature of the coolant, heat generation data, convective heat transfer coefficient, contact area and conduction temperature difference.

[0133] Obtain the material parameters corresponding to the coolant; the material parameters include material density and specific heat capacity.

[0134] The predicted outlet temperature of the coolant is calculated based on the real-time inlet temperature, calorific value, material parameters, and real-time flow rate data.

[0135] Optionally, the predicted core temperature of the GPU can be determined using the following formula:

[0136]

[0137] in, This refers to the predicted core temperature of the GPU. This refers to the real-time inlet temperature of the coolant. h Let be the convective heat transfer coefficient between the coolant and the GPU surface, A be the contact area between the GPU and the coolant, and Q be the heat generated by the GPU. This refers to the conduction temperature difference from the GPU core to the surface (in °C, typically 2-3 °C).

[0138] Optionally, the predicted outlet temperature of the coolant can be determined by the following formula:

[0139]

[0140] in, For the predicted outlet temperature of the coolant, This refers to the real-time inlet temperature of the coolant. The specific heat capacity of the coolant (J / (kg)) ρ is the material density of the coolant, F is the real-time flow rate of the coolant, and Q is the heat generated by the GPU.

[0141] As can be seen, this optional embodiment can calculate the predicted core temperature of the GPU based on the real-time inlet temperature of the coolant, heat generation data, convective heat transfer coefficient between the coolant and the GPU surface, contact area between the coolant and the GPU, and conduction temperature difference from the GPU core to the surface. Then, by combining the material density and specific heat capacity of the coolant, the real-time inlet temperature of the coolant, heat generation data, and real-time flow rate data, the predicted outlet temperature of the coolant can be calculated. This can improve the accuracy of predicting the GPU core temperature and the coolant outlet temperature, thereby improving the accuracy of subsequent determination of the required predicted flow rate data.

[0142] In an optional embodiment, determining the predicted traffic data required by the intelligent computing data center based on the traffic analysis model and the predicted temperature data may include the following operations:

[0143] When the predicted core temperature of the GPU is greater than the preset core temperature threshold, the minimum flow rate of coolant required for the intelligent computing data center is calculated based on the real-time inlet temperature of the coolant, the material parameters of the coolant, the core temperature threshold, and the heat generation data.

[0144] When the predicted core temperature of the GPU is less than or equal to the core temperature threshold, the target core temperature of the GPU is obtained; where the target core temperature is less than the core temperature threshold.

[0145] Based on the real-time inlet temperature of the coolant, the corresponding material parameters of the coolant, the target core temperature, and the heat generation data, the optimal flow rate of the coolant required for the intelligent computing data center is calculated.

[0146] By determining the minimum coolant flow rate, the actual coolant flow rate can be ensured to be no less than this value, thus preventing undercooling in the intelligent computing data center. Optionally, the minimum coolant flow rate can be determined by the following formula:

[0147]

[0148] In the formula, 85 refers to a core temperature threshold of 85℃. Let ρ be the specific heat capacity of the coolant, ρ be the material density of the coolant, and Q be the heat generated by the GPU. This is the real-time inlet temperature of the coolant.

[0149] By determining the optimal coolant flow rate, overcooling in the intelligent computing data center can be avoided, and pump energy consumption can be reduced. Optionally, the optimal coolant flow rate can be determined by the following formula:

[0150]

[0151] in, For the optimal flow rate, ρ is the material density of the coolant. This refers to the specific heat capacity of the coolant. Q represents the target core temperature, and Q represents the heat output of the GPU.

[0152] As can be seen, this optional embodiment can calculate the minimum coolant flow rate required for the intelligent computing data center based on the real-time inlet temperature of the coolant, the material parameters of the coolant, the core temperature threshold, and the heat generation data if the GPU predicts a core temperature greater than the core temperature threshold. This allows the intelligent computing data center to avoid undercooling by setting the minimum coolant flow rate. Furthermore, if the GPU predicts a core temperature less than or equal to the core temperature threshold, the target core temperature of the GPU is obtained, and the optimal coolant flow rate required for the intelligent computing data center is calculated based on the real-time inlet temperature of the coolant, the material parameters of the coolant, the target core temperature, and the heat generation data. This allows the intelligent computing data center to avoid overcooling while reducing the energy consumption of the pump group in the liquid cooling system. This facilitates efficient cooling temperature control while optimizing cooling energy consumption, further improving the reliability of the liquid cooling system's regulation and its environmental friendliness.

[0153] In an optional embodiment, determining the optimal traffic command based on the determined adaptive optimization algorithm and the predicted traffic data may include the following operations:

[0154] Through the aforementioned steps, the predicted flow data, comprehensive computing load index, heat generation data, and predicted core temperature of the GPU are obtained for the control cycle corresponding to the future preset cycle number.

[0155] The Sequential Quadratic Programming (SQP) algorithm is used. Based on the predicted flow data, comprehensive computing load index, heat generation data and the predicted core temperature of the GPU, the pre-constructed optimization objective function is solved in each control cycle to obtain the optimal flow sequence for the control cycle corresponding to the preset number of future cycles and the optimal flow command for the current control cycle.

[0156] For example, the preset number of cycles can be 5, but this embodiment of the invention does not limit it.

[0157] Optionally, the objective function for optimization can be as follows:

[0158]

[0159] in, For the energy consumption of liquid cooling pump units, The optimal flow command for the current control cycle. This represents the actual flow rate in the previous control cycle.

[0160] Where ω1 is the weight of the temperature deviation term, used to minimize the difference between the predicted core temperature and the target core temperature. The deviation is used to ensure temperature stability; ω2 is the weight of the pump group energy consumption term, used to minimize the energy consumption of the liquid cooling system and achieve energy efficiency optimization; ω3 is the weight of the flow rate change term, used to suppress frequent flow fluctuations, avoid system oscillations, and improve stability.

[0161] For example, the weight of ω1 can be 0.6, the weight of ω2 can be 0.3, and the weight of ω3 can be 0.1.

[0162] The energy consumption of the liquid cooling pump unit can be determined by the following formula:

[0163]

[0164] Where k is the pump set characteristic coefficient, which can be determined experimentally, and F can be the actual flow rate.

[0165] Alternatively, the constraints corresponding to the optimization objective function can be as follows:

[0166] Set physical constraints for flow regulation, including ( (This refers to the maximum output flow rate of the pump unit, typically 10 L / min) ( The maximum flow rate adjustment for a single operation is set to 0.5 L / min, ensuring that the adjustment command is within the range that the hardware can implement.

[0167] As can be seen, this optional embodiment can use an adaptive optimization algorithm and a sequential quadratic programming algorithm to solve a pre-built optimization objective function in each control cycle based on the predicted flow data, comprehensive computing load index, heat generation data, and predicted core temperature of the GPU corresponding to the future preset number of control cycles. This yields the optimal flow sequence for the future preset number of control cycles and the optimal flow command for the current control cycle. This improves the accuracy of determining the optimal flow command for the current control cycle, thereby improving the control accuracy of the liquid cooling system and further enhancing the cooling precision and efficiency of the liquid cooling system for intelligent computing data centers.

[0168] In this optional embodiment, the method may further include:

[0169] The difference between the actual GPU core temperature and the predicted core temperature in the current control cycle is input into the operation and maintenance coupling analysis model to correct the prediction parameters for the next control cycle. This reduces prediction errors and achieves adaptive correction.

[0170] In this embodiment of the invention, for example, the adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers can be applied to an adaptive operation and maintenance control system for liquid cooling systems in intelligent computing data centers. The architecture of this system can be found in [reference needed]. Figure 3 , Figure 3 This is a schematic diagram of the architecture of an adaptive operation and maintenance control system for a liquid cooling system in an intelligent computing data center, as disclosed in an embodiment of the present invention. Wherein, as... Figure 3 As shown, the system may include a data acquisition layer, a model calculation layer, and a decision execution layer; further, the data acquisition layer may include a GPU computing power load acquisition module, a temperature acquisition module, and a flow and pump group parameter acquisition module; the model calculation layer may include a data preprocessing module coupled with a model module and an MPC optimization algorithm module; the decision processing layer may include a flow controller, a pump group actuator, and a status monitoring and anomaly handling module, which are not limited in this embodiment of the invention.

[0171] Furthermore, exemplarily, when this method is applied to the above system, the steps and flow of this method can be found in [reference needed]. Figure 4 , Figure 4 This is a flowchart illustrating another adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers, as disclosed in an embodiment of the present invention. Wherein, as... Figure 4 As shown, the method may include the following operations:

[0172] After initializing the parameters, real-time data from the liquid cooling system and intelligent computing data are collected, and then the real-time data is preprocessed. Then, load and temperature are predicted based on the preprocessed real-time data, and an optimization function is constructed and solved based on the predicted data. This controls the liquid cooling system to execute the flow regulation command obtained from the optimization function (i.e., the above-mentioned optimal flow command). During the flow regulation process, the data is analyzed to provide feedback and check for abnormalities. If there are abnormalities, abnormality handling is performed. If there are no abnormalities, the regulation process can be terminated normally. This embodiment of the invention does not limit the scope of the invention.

[0173] For a more detailed explanation of the load and temperature prediction process, please refer to [link / reference needed]. Figure 5 , Figure 5 This is a flowchart illustrating another adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers, as disclosed in an embodiment of the present invention. Wherein, as... Figure 4 As shown, load and temperature prediction can include the following operations:

[0174] Based on the intelligent computing tasks that the intelligent computing data center needs to execute, the computing load of the GPU is calculated; then the computing load is input into the computing load-heat generation model to calculate the heat generation of the GPU; then the heat generation is input into the heat generation-temperature model to calculate the GPU core temperature and coolant outlet temperature; then the temperature data is input into the temperature-flow model to calculate the required optimal flow rate. This embodiment of the invention is not limited.

[0175] Furthermore, in such Figure 4 Based on the flowchart shown, the execution steps of the method are described in chronological order (control cycle 10 seconds), as an example:

[0176] (1) Step 1: Initialize parameters (executed once when the program starts): Set the MPC prediction period to 5 control cycles (50 seconds in total), optimize the objective function weights ω1=0.6, ω2=0.3, ω3=0.1; set flow constraints: =10L / min (maximum flow rate of pump unit) =0.5L / min (maximum adjustment in a single cycle); Initialize the LSTM prediction model (load the pre-trained weight file), connect to the InfluxDB database, and set up the Profinet communication interface;

[0177] (2) Step 2: Collect real-time data (executed at 0-1 second of each control cycle): Obtain the comprehensive computing load index, intelligent computing task type, and computing core frequency of each GPU through the GPU computing load acquisition module; obtain the GPU core temperature, coolant inlet temperature and / or coolant outlet temperature, and temperature field data of each GPU through the temperature acquisition module; obtain the real-time flow rate, pump speed, and power of each pipeline through the flow and pump group parameter acquisition module; write all data to InfluxDB through HTTPPOST request, and the data tag includes gpu_id and acquisition timestamp;

[0178] (3) Step 3: Data preprocessing (executed in the first 1-2 seconds of each control cycle): Outlier removal: Data with GPU core temperature >100℃ or <20℃ and real-time flow <0 or >10L / min are identified as outliers and removed using the 3σ criterion, and missing values ​​are filled in by linear interpolation; Data standardization: The computing load (0-100%), GPU core temperature (20-100℃), and real-time flow (0-10L / min) are standardized to the [0,1] range; Data fusion: The GPU core temperature (single point) and temperature field data (region) are fused by Kalman filtering, and the corrected single-point temperature data is output; Calculate the comprehensive computing load index CLI;

[0179] (4) Step 4: Load and temperature prediction (executed in the 2nd-4th second of each control cycle): Input the CLI sequence (30 data points) of the past 30 seconds into the LSTM model to predict the CLI prediction sequence for the next 50 seconds (5 control cycles); determine η based on the intelligent computing task type, and calculate the heat generation for the next 50 seconds in combination with the CLI prediction sequence; input the heat generation and inlet temperature (real-time acquisition value, assuming stability for the next 50 seconds) into the coupled model to predict the GPU core temperature prediction sequence for the next 50 seconds, and determine whether there is a case where the predicted core temperature is >85℃;

[0180] (5) Step 5: Construct and solve the optimization function (executed in the 4th-7th second of each control cycle): Construct a multi-objective optimization function J and set constraints, thereby calling the SQP algorithm to solve for the minimum value of J, and obtain the optimal flow sequence for the next 5 cycles. Only the optimal flow instruction of the current cycle is retained as the optimal flow instruction to be executed at the current time.

[0181] (6) Step 6: Execute the flow regulation command (executed at the 7th-8th second of each control cycle): Convert the optimal flow command into a 4-20mA current signal; send the current signal to the flow controller of the corresponding GPU rack through the Profinet protocol, and the controller drives the pump group to adjust the speed so that the actual flow approaches the command flow corresponding to the optimal flow command; record the optimal flow command and the command issuance timestamp, and write it to the "control_log" table of InfluxDB;

[0182] (7) Step 7: Feedback correction and anomaly handling (executed at the 8th-10th second of each control cycle):

[0183] Collect the actual flow rate for the current cycle and calculate the flow rate difference between the actual flow rate and the command flow rate corresponding to the optimal flow rate command. If the flow rate difference is >0.2℃, correct the output weights of the LSTM model to reduce the prediction error for the next cycle and switch to PID standby adjustment. If the actual GPU core temperature is >85℃, start emergency cooling and pause the GPU task. If the pump group current is abnormal, switch to the standby pump group. Determine whether the program needs to be terminated (e.g., if a stop command is received from the maintenance personnel). If not, return to step 2 and enter the next control cycle.

[0184] Example 3

[0185] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of an adaptive operation and maintenance control device for a liquid cooling system in an intelligent computing data center, as disclosed in an embodiment of the present invention. Figure 6The described adaptive operation and maintenance control device for liquid cooling systems in intelligent computing data centers may include one of the following: intelligent devices, intelligent terminals, intelligent systems, and servers. The servers may include local servers or cloud servers; this embodiment of the invention does not limit the scope. Figure 6 As shown, the adaptive operation and maintenance control device for the liquid cooling system of the intelligent computing data center may include:

[0186] The acquisition module 301 is used to acquire multi-dimensional real-time data associated with the liquid cooling system. The multi-dimensional real-time data includes computing load data, real-time temperature data, and real-time flow data of the liquid cooling system corresponding to the intelligent computing data center that the liquid cooling system needs to dissipate heat. The real-time temperature data includes the first real-time temperature data of the intelligent computing data center and the second real-time temperature data of the liquid cooling system.

[0187] Preprocessing module 302 is used to perform preprocessing operations on multi-dimensional real-time data to obtain preprocessed real-time data; the preprocessing operations include outlier removal, data standardization, and data fusion.

[0188] Analysis module 303 is used to analyze and preprocess real-time data based on the determined operation and maintenance coupling analysis model to obtain the predicted traffic data required by the intelligent computing data center;

[0189] The determination module 304 is used to determine the optimal flow command based on the determined adaptive optimization algorithm and the predicted flow data; the adaptive optimization algorithm is built based on the MPC algorithm and is a multi-objective optimization algorithm, and the objectives to be optimized by the adaptive optimization algorithm include at least the temperature control objective and the energy efficiency optimization objective;

[0190] The cooling control module 305 is used to control the liquid cooling system to perform cooling operations for the intelligent computing data center according to the optimal flow command.

[0191] As can be seen, the apparatus described in the embodiments of the present invention can obtain the predicted flow data required by the intelligent computing data center by collecting multi-dimensional real-time data, analyzing the pre-processed real-time data based on the operation and maintenance coupling analysis model, and then determining the optimal flow command based on the adaptive optimization algorithm and the predicted flow data. The liquid cooling system is then controlled to cool the intelligent computing data center based on the optimal flow command. This enables intelligent adaptive adjustment of the liquid cooling system. Through multi-objective optimization and optimal flow calculation, the operation and maintenance adjustment scheme of the liquid cooling system can be made more adaptable to the high volatility of the intelligent computing load while reducing the energy consumption of the liquid cooling system. This improves the control accuracy of the liquid cooling system, thereby improving the cooling precision and efficiency of the liquid cooling system for the intelligent computing data center, and thus improving the heat dissipation efficiency of the intelligent computing data center. This effectively avoids GPU frequency reduction in the intelligent computing data center due to excessive temperature, which in turn helps the intelligent computing data center to execute intelligent computing tasks more efficiently and improve the operational stability of the intelligent computing data center.

[0192] In an optional embodiment, the operation and maintenance coupling analysis model includes a heat generation analysis model, a temperature distribution analysis model, and a flow analysis model;

[0193] The analysis module 303, based on the determined operation and maintenance coupling analysis model, analyzes and preprocesses real-time data to obtain the specific method for obtaining the predicted traffic data required by the intelligent computing data center. This may include:

[0194] Based on the heat generation analysis model, the heat generation data corresponding to the intelligent computing data center is determined according to the computing load data.

[0195] Based on the temperature distribution analysis model, the predicted temperature data is determined according to the heat generation data and real-time temperature data. The predicted temperature data includes the predicted core temperature of the GPU in the intelligent computing data center and the predicted outlet temperature of the coolant flowing through the intelligent computing data center in the liquid cooling system.

[0196] Based on the flow analysis model, the predicted flow data required for the intelligent computing data center is determined according to the predicted temperature data; the predicted flow data includes the minimum flow rate of coolant and / or the optimal flow rate of coolant.

[0197] As can be seen, the apparatus described in this optional embodiment can determine the heat generation data corresponding to the computing load data based on the heat generation analysis model, then determine the predicted temperature data corresponding to the heat generation data based on the temperature distribution analysis model, and then determine the corresponding predicted flow rate data based on the predicted temperature data according to the flow rate analysis model. This enables coupled analysis of computing load, temperature, and flow rate, thereby improving the accuracy of predicting the heat generation of the intelligent computing data center and the accuracy of predicting the flow rate required for the predicted heat generation. This, in turn, helps to improve the accuracy of generating subsequent optimal flow rate commands, and further helps to improve the control accuracy of the liquid cooling system, thereby improving the cooling precision of the liquid cooling system for the intelligent computing data center.

[0198] In this optional embodiment, optionally, the analysis module 303 determines the specific method of determining the heat generation data corresponding to the intelligent computing data center based on the heat generation analysis model and the computing load data, which may include:

[0199] Determine the idle power consumption and full load power consumption of the GPU;

[0200] Determine the computational efficiency corresponding to the task type based on the task type of the intelligent computing task corresponding to the intelligent computing data center;

[0201] The actual power consumption of the GPU is determined based on the idle power consumption, full load power consumption, and the determined comprehensive computing power load index.

[0202] The heat output of the GPU is determined based on the actual power consumption and the computing efficiency.

[0203] Among them, the heat generation data corresponding to the intelligent computing data center includes the heat generation of at least one GPU.

[0204] As can be seen, the apparatus described in this optional embodiment can also calculate the actual power consumption of the GPU by utilizing a heat analysis model, based on the GPU idle power consumption, GPU full-load power consumption, and the determined comprehensive computing load index. Combined with the computing efficiency corresponding to the intelligent computing task type, the heat generation of the GPU can be calculated. This can improve the accuracy of the assessment of the actual power consumption of the GPU and the accuracy of the determination of the GPU heat generation, thereby improving the accuracy of the determination of heat generation data in intelligent computing data centers, and further improving the accuracy of subsequent prediction of GPU core temperature.

[0205] In this optional embodiment, the second real-time temperature data corresponding to the liquid cooling system may include the real-time inlet temperature of the coolant.

[0206] The analysis module 303, based on a temperature distribution analysis model, determines the specific method for predicting temperature data according to calorific value data and real-time temperature data, which may include:

[0207] The convective heat transfer coefficient between the coolant and the GPU surface, the contact area between the coolant and the GPU, and the conduction temperature difference from the GPU core to the surface are obtained.

[0208] The predicted core temperature of the GPU is calculated based on the real-time inlet temperature of the coolant, heat generation data, convective heat transfer coefficient, contact area and conduction temperature difference.

[0209] Obtain the material parameters corresponding to the coolant; the material parameters include material density and specific heat capacity.

[0210] The predicted outlet temperature of the coolant is calculated based on the real-time inlet temperature, calorific value, material parameters, and real-time flow rate data.

[0211] As can be seen, the apparatus described in this optional embodiment can also calculate the predicted core temperature of the GPU based on the real-time inlet temperature of the coolant, heat generation data, convective heat transfer coefficient between the coolant and the GPU surface, contact area between the coolant and the GPU, and conduction temperature difference from the GPU core to the surface. Then, by combining the material density and specific heat capacity of the coolant, the real-time inlet temperature of the coolant, heat generation data, and real-time flow rate data, the predicted outlet temperature of the coolant can be calculated. This can improve the accuracy of predicting the GPU core temperature and the coolant outlet temperature, thereby improving the accuracy of subsequent determination of the required predicted flow rate data.

[0212] In this optional embodiment, optionally, the analysis module 303, based on the traffic analysis model and the predicted temperature data, determines the specific method by which it determines the predicted traffic data required by the intelligent computing data center, which may include:

[0213] When the predicted core temperature of the GPU is greater than the preset core temperature threshold, the minimum flow rate of coolant required for the intelligent computing data center is calculated based on the real-time inlet temperature of the coolant, the material parameters of the coolant, the core temperature threshold, and the heat generation data.

[0214] When the predicted core temperature of the GPU is less than or equal to the core temperature threshold, the target core temperature of the GPU is obtained; where the target core temperature is less than the core temperature threshold.

[0215] Based on the real-time inlet temperature of the coolant, the corresponding material parameters of the coolant, the target core temperature, and the heat generation data, the optimal flow rate of the coolant required for the intelligent computing data center is calculated.

[0216] As can be seen, the apparatus described in this optional embodiment can also calculate the minimum coolant flow rate required for the intelligent computing data center based on the real-time inlet temperature of the coolant, the material parameters of the coolant, the core temperature threshold, and the heat generation data if the GPU predicts a core temperature greater than the core temperature threshold. This allows the intelligent computing data center to avoid undercooling by setting the minimum coolant flow rate. Furthermore, if the GPU predicts a core temperature less than or equal to the core temperature threshold, the target core temperature of the GPU is obtained, and the optimal coolant flow rate required for the intelligent computing data center is calculated based on the real-time inlet temperature of the coolant, the material parameters of the coolant, the target core temperature, and the heat generation data. This allows the intelligent computing data center to avoid overcooling by setting the optimal coolant flow rate, while reducing the energy consumption of the pump group corresponding to the liquid cooling system. This is beneficial for achieving efficient cooling temperature control while optimizing cooling energy consumption, and for further improving the control reliability of the liquid cooling system while also improving the environmental friendliness of the liquid cooling system.

[0217] In an optional embodiment, such as Figure 7 As shown, the device may further include:

[0218] Monitoring module 306 is used to monitor the operating status data during the process of controlling the liquid cooling system to perform cooling operations on the intelligent computing data center; the operating status data includes the first operating status data corresponding to the liquid cooling system and the second operating status data corresponding to the intelligent computing data center;

[0219] The judgment module 307 is used to determine whether the liquid cooling system meets the abnormal operating conditions corresponding to at least one preset abnormality type based on the operating status data; all preset abnormality types include one or more of the following: flow regulation abnormality type, core temperature abnormality type, and pump group abnormality type.

[0220] The anomaly handling module 308 is used to determine each preset anomaly type met by the liquid cooling system as a target anomaly type when the judgment module 307 determines that the liquid cooling system meets the abnormal operating conditions corresponding to at least one preset anomaly type, and to control the liquid cooling system to perform the emergency control operation corresponding to the target anomaly type according to each target anomaly type.

[0221] As can be seen, the device described in this optional embodiment can monitor the operating status during the cooling process of the liquid cooling system for the intelligent computing data center, and determine whether the abnormal operating conditions corresponding to at least one preset abnormality type are met. If it is determined that at least one abnormal operating condition is met, the device controls the liquid cooling system to perform the emergency control operation corresponding to the abnormality type. This enables full automation of the data acquisition, model calculation, instruction execution, and abnormality handling process, and improves the timeliness and comprehensiveness of abnormality detection in the cooling process of the liquid cooling system. This facilitates more timely and targeted implementation of effective abnormality handling measures, thereby improving the control accuracy of the liquid cooling system and enhancing the operational reliability of the liquid cooling system and the intelligent computing data center.

[0222] In this optional embodiment, the first operating status data may include the actual flow rate of the liquid cooling system and the actual equipment operating information of the liquid cooling system; the second operating status data may include the actual GPU core temperature of the intelligent computing data center.

[0223] The specific method by which the judgment module 307 determines whether the liquid cooling system meets the abnormal operating conditions corresponding to at least one preset abnormality type based on the operating status data may include:

[0224] Calculate the flow difference between the actual flow and the command flow corresponding to the optimal flow command;

[0225] Determine whether the traffic difference is greater than a preset traffic difference threshold, and whether the duration for which the traffic difference is greater than the traffic difference threshold is greater than or equal to a preset abnormal duration threshold.

[0226] When it is determined that the flow difference is greater than the flow difference threshold, and the duration of the flow difference being greater than the flow difference threshold is greater than or equal to the abnormal duration threshold, the liquid cooling system is determined to meet the abnormal operating conditions corresponding to the abnormal flow regulation type.

[0227] And / or,

[0228] Determine whether the actual GPU core temperature is greater than a preset core temperature threshold.

[0229] When it is determined that the actual GPU core temperature is greater than the core temperature threshold, the liquid cooling system is determined to meet the abnormal operating conditions corresponding to the core temperature abnormality type.

[0230] And / or,

[0231] Based on the actual equipment operating information, determine whether there is a fault in the currently operating pump set of the liquid cooling system;

[0232] When it is determined that the currently operating pump set of the liquid cooling system is faulty, the liquid cooling system is confirmed to meet the abnormal operating conditions corresponding to the pump set abnormality type.

[0233] As can be seen, the apparatus described in this optional embodiment can also determine that the liquid cooling system has an abnormal flow regulation when the flow difference is greater than the flow difference threshold for an extended period of time; it can also determine that the core temperature is abnormal when the GPU core temperature is too high; and it can also determine that the liquid cooling system has a pump group abnormality when the currently operating pump group is faulty. This improves the flexibility and accuracy of determining whether the liquid cooling system has an operational abnormality, thereby improving the accuracy of determining the necessary emergency control operations. This, in turn, facilitates more accurate and timely handling of abnormalities that occur during the cooling process, thereby improving the operational reliability of the liquid cooling system and the intelligent computing data center.

[0234] Example 4

[0235] Please see Figure 8 , Figure 8 This is a schematic diagram of another adaptive operation and maintenance control device for a liquid cooling system in an intelligent computing data center, as disclosed in an embodiment of the present invention. Figure 8 As shown, the adaptive operation and maintenance control device for the liquid cooling system of the intelligent computing data center may include:

[0236] Memory 401 storing executable program code;

[0237] Processor 402 coupled to memory 401;

[0238] The processor 402 calls the executable program code stored in the memory 401 to execute some or all of the steps in the adaptive operation and maintenance control method for the liquid cooling system of the intelligent computing data center described in Embodiment 1 or Embodiment 2 of the present invention.

[0239] Example 5

[0240] This invention discloses a computer storage medium storing computer instructions. When these computer instructions are invoked, they are used to execute some or all of the steps in the adaptive operation and maintenance control method for a liquid cooling system in an intelligent computing data center described in Embodiment 1 or Embodiment 2 of this invention.

[0241] Example 6

[0242] This invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform some or all of the steps in the adaptive operation and maintenance control method for a liquid cooling system of an intelligent computing data center described in Embodiment 1 or Embodiment 2.

[0243] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0244] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0245] Finally, it should be noted that the adaptive operation and maintenance control method and apparatus for liquid cooling systems in intelligent computing data centers disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers, characterized in that, The method includes: Collect multi-dimensional real-time data associated with the liquid cooling system; the multi-dimensional real-time data includes computing load data, real-time temperature data, and real-time flow data of the liquid cooling system corresponding to the intelligent computing data center that the liquid cooling system needs to dissipate heat; the real-time temperature data includes first real-time temperature data corresponding to the intelligent computing data center and second real-time temperature data corresponding to the liquid cooling system; Preprocessing operations are performed on the multi-dimensional real-time data to obtain preprocessed real-time data; the preprocessing operations include outlier removal, data standardization, and data fusion. Based on the determined operation and maintenance coupling analysis model, the preprocessed real-time data is analyzed to obtain the predicted traffic data required by the intelligent computing data center; wherein, the operation and maintenance coupling analysis model includes a heat generation analysis model, a temperature distribution analysis model, and a traffic analysis model; Based on the determined adaptive optimization algorithm, the optimal flow command is determined according to the predicted flow data; the adaptive optimization algorithm is constructed based on the MPC algorithm, and the adaptive optimization algorithm is a multi-objective optimization algorithm, and the objectives to be optimized by the adaptive optimization algorithm include at least temperature control objectives and energy efficiency optimization objectives; According to the optimal flow command, the liquid cooling system is controlled to perform cooling operations on the intelligent computing data center; The step of analyzing the preprocessed real-time data based on the determined operation and maintenance coupling analysis model to obtain the predicted traffic data required by the intelligent computing data center includes: Based on the heat generation analysis model, the heat generation data corresponding to the intelligent computing data center is determined according to the computing load data. Based on the temperature distribution analysis model, the predicted temperature data is determined according to the heat generation data and the real-time temperature data; the predicted temperature data includes the predicted core temperature of the GPU in the intelligent computing data center and the predicted outlet temperature of the coolant flowing through the intelligent computing data center in the liquid cooling system. Based on the flow analysis model, and according to the predicted temperature data, the predicted flow data required by the intelligent computing data center is determined; the predicted flow data includes the minimum flow rate of coolant and / or the optimal flow rate of coolant. The step of determining the heat generation data corresponding to the intelligent computing data center based on the heat generation analysis model and the computing load data includes: Determine the idle power consumption and the full load power consumption of the GPU; The computational efficiency corresponding to the task type is determined based on the task type of the intelligent computing task corresponding to the intelligent computing data center. The actual power consumption of the GPU is determined based on the idle power consumption, the full-load power consumption, and the determined comprehensive computing power load index. The comprehensive computing power load index is obtained by weighted summation of the GPU's computing power utilization and core frequency, and is used to quantify the actual load level of the GPU. The core frequency includes the actual core frequency and the GPU's maximum core frequency. The actual power consumption of the GPU is linearly related to the comprehensive computing power load index. The heat generation of the GPU is determined based on the actual power consumption and the computing efficiency. The heat generation data corresponding to the intelligent computing data center includes the heat generation data corresponding to at least one of the GPUs.

2. The adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers according to claim 1, characterized in that, The second real-time temperature data corresponding to the liquid cooling system includes the real-time inlet temperature of the coolant; The step of determining the predicted temperature data based on the temperature distribution analysis model, according to the calorific value data and the real-time temperature data, includes: The convective heat transfer coefficient between the coolant and the surface of the GPU, the contact area between the coolant and the GPU, and the conduction temperature difference from the core to the surface of the GPU are obtained. The predicted core temperature of the GPU is calculated based on the real-time inlet temperature of the coolant, the heat generation data, the convective heat transfer coefficient, the contact area, and the conduction temperature difference. Obtain the material parameters corresponding to the coolant; the material parameters include material density and specific heat capacity; The predicted outlet temperature of the coolant is calculated based on the real-time inlet temperature of the coolant, the calorific value data, the material parameters, and the real-time flow rate data.

3. The adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers according to claim 2, characterized in that, The step of determining the predicted traffic data required by the intelligent computing data center based on the traffic analysis model and the predicted temperature data includes: When the predicted core temperature of the GPU is greater than the preset core temperature threshold, the minimum flow rate of the coolant required by the intelligent computing data center is calculated based on the real-time inlet temperature of the coolant, the material parameters of the coolant, the core temperature threshold, and the heat generation data. When the predicted core temperature of the GPU is less than or equal to the core temperature threshold, the target core temperature of the GPU is obtained; wherein the target core temperature is less than the core temperature threshold. The optimal flow rate of the coolant required for the intelligent computing data center is calculated based on the real-time inlet temperature of the coolant, the material parameters corresponding to the coolant, the target core temperature, and the heat generation data.

4. The adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers according to any one of claims 1-3, characterized in that, The method further includes: During the process of controlling the liquid cooling system to perform the cooling operation on the intelligent computing data center, the operating status data is monitored; the operating status data includes first operating status data corresponding to the liquid cooling system and second operating status data corresponding to the intelligent computing data center; Based on the operating status data, it is determined whether the liquid cooling system meets the abnormal operating conditions corresponding to at least one preset abnormality type; all preset abnormality types include one or more of the following: flow regulation abnormality type, core temperature abnormality type, and pump group abnormality type. When it is determined that the liquid cooling system meets at least one of the preset abnormality types, each of the preset abnormality types met by the liquid cooling system is identified as a target abnormality type, and according to each target abnormality type, the liquid cooling system is controlled to perform the emergency control operation corresponding to that target abnormality type.

5. The adaptive operation and maintenance control method for liquid cooling systems in intelligent computing data centers according to claim 4, characterized in that, The first operating status data includes the actual flow rate of the liquid cooling system and the actual equipment operating information of the liquid cooling system; the second operating status data includes the actual GPU core temperature of the intelligent computing data center. The step of determining whether the liquid cooling system meets the abnormal operating conditions corresponding to at least one preset abnormality type based on the operating status data includes: Calculate the flow difference between the actual flow and the command flow corresponding to the optimal flow command; Determine whether the traffic difference is greater than a preset traffic difference threshold, and whether the duration for which the traffic difference is greater than the traffic difference threshold is greater than or equal to a preset abnormal duration threshold. When it is determined that the flow difference is greater than the flow difference threshold, and the duration for which the flow difference is greater than the flow difference threshold is greater than or equal to the abnormal duration threshold, it is determined that the liquid cooling system meets the abnormal operating conditions corresponding to the abnormal flow regulation type. And / or, Determine whether the actual GPU core temperature is greater than a preset core temperature threshold. When it is determined that the actual GPU core temperature is greater than the core temperature threshold, it is determined that the liquid cooling system meets the abnormal operating conditions corresponding to the core temperature anomaly type. And / or, Based on the actual equipment operating information, determine whether the currently operating pump set of the liquid cooling system is faulty; When it is determined that the fault exists in the currently operating pump set of the liquid cooling system, it is determined that the liquid cooling system meets the abnormal operating conditions corresponding to the abnormal type of the pump set.

6. An adaptive operation and maintenance control device for a liquid cooling system in an intelligent computing data center, characterized in that, The apparatus is used to execute the adaptive operation and maintenance control method for a liquid cooling system in an intelligent computing data center as described in any one of claims 1-5, and the apparatus comprises: The acquisition module is used to acquire multi-dimensional real-time data associated with the liquid cooling system; the multi-dimensional real-time data includes computing load data, real-time temperature data, and real-time flow data corresponding to the intelligent computing data center that the liquid cooling system needs to dissipate heat from; the real-time temperature data includes first real-time temperature data corresponding to the intelligent computing data center and second real-time temperature data corresponding to the liquid cooling system; The preprocessing module is used to perform preprocessing operations on the multi-dimensional real-time data to obtain preprocessed real-time data; the preprocessing operations include outlier removal, data standardization, and data fusion. The analysis module is used to analyze the preprocessed real-time data based on the determined operation and maintenance coupling analysis model to obtain the predicted traffic data required by the intelligent computing data center. The determination module is used to determine the optimal flow instruction based on the predicted flow data and the determined adaptive optimization algorithm; the adaptive optimization algorithm is constructed based on the MPC algorithm and is a multi-objective optimization algorithm, and the objectives to be optimized by the adaptive optimization algorithm include at least temperature control objectives and energy efficiency optimization objectives; The cooling control module is used to control the liquid cooling system to perform cooling operations on the intelligent computing data center according to the optimal flow command.

7. An adaptive operation and maintenance control device for a liquid cooling system in an intelligent computing data center, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the adaptive operation and maintenance control method for the liquid cooling system of the intelligent computing data center as described in any one of claims 1-5.

8. A computer storage medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked, are used to execute the adaptive operation and maintenance control method for the liquid cooling system of the intelligent computing data center as described in any one of claims 1-5.

Images