Data center cooling control method and system based on IT load dynamic change

Abstract

The invention discloses a data center cooling control method and system based on IT load dynamic change. The method comprises the following steps: acquiring IT load data, environment temperature data and equipment state data to obtain initial data; performing preprocessing, data management and feature engineering on the initial data to obtain load gradient features; determining a load mode according to the load gradient characteristics and a preset working mode; a feedforward compensation or feedback control strategy is selected according to the load mode, processing is carried out in combination with the space thermal mapping matrix, output of the two strategies is dynamically fused to obtain a suggested control quantity, and the thermal mapping matrix and algorithm parameters are updated through initial calibration and online learning; and generating a driving instruction according to the suggested control quantity, performing safety verification through dew point protection and mechanical characteristic protection, and issuing the driving instruction. By implementing the method, the thermal hysteresis effect can be effectively counteracted, the overall energy consumption of the data center is remarkably reduced, and the energy utilization efficiency is improved.

Description

Technical Field

[0001] This invention relates to data centers, and more specifically to a data center cooling control method and system based on dynamic changes in IT load. Background Technology

[0002] Currently, data center cooling systems face several key challenges. First, traditional precision air conditioning control systems rely on return air or cold aisle temperature sensors to regulate temperature. Because heat transfer from server chips to these sensors involves multiple physical processes—including heatsinks, chassis, and air convection—this results in significant thermal hysteresis, a delay of several minutes between load increase and temperature detection. Second, to address the risk of sudden, localized overheating caused by this hysteresis, existing technologies typically set a lower overall temperature. While this effectively prevents overheating, it also leads to energy inefficiency, increases power efficiency ratings, and causes unnecessary energy waste.

[0003] To address these challenges, the widely adopted solution is a temperature-feedback-based PID (Proportional-Integral-Derivative) control strategy. This method involves installing temperature sensors at the air conditioning return vents or in the cold aisles of server racks to monitor the temperature in real time and adjust the air conditioning fan speed or chilled water valve opening based on the deviation from the set target temperature. However, this method has significant limitations: its response speed is slow; by the time the sensor detects a temperature rise, the server chip may have already been operating at high temperatures for several minutes. Furthermore, the lack of predictive ability for future temperature trends forces maintenance personnel to maintain a lower overall ambient temperature to ensure safety, making it difficult to raise the temperature setpoint even during low-load periods, further exacerbating energy waste.

[0004] Therefore, it is necessary to design a new method that can not only effectively offset the thermal hysteresis effect, achieve precise on-demand cooling, and eliminate local hot spots, but also significantly reduce the overall energy consumption of data centers and improve energy efficiency. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a data center cooling control method and system based on dynamic changes in IT load.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a data center cooling control method based on dynamic changes in IT load, comprising:

[0007] Acquire IT load data, ambient temperature data, and equipment status data to obtain initial data; The initial data is preprocessed, data-governed, and feature-engineered to obtain load gradient features; The load mode is determined based on the load gradient characteristics and the preset operating mode; Based on the load mode, a feedforward compensation or feedback control strategy is selected and processed in conjunction with the spatial heat mapping matrix. The outputs of the two are dynamically fused to obtain the suggested control quantity. The heat mapping matrix and algorithm parameters are updated through initial calibration and online learning. The drive command is generated based on the suggested control quantity, and safety verification is performed through dew point protection and mechanical characteristic protection before the drive command is issued.

[0008] Its further technical solutions include: In the event of sensor failure or network outage, virtual interpolation, conservative mode, or local takeover measures will be adopted.

[0009] The further technical solution is as follows: the preprocessing, data governance, and feature engineering of the initial data to obtain load gradient features includes: By using a circular buffer and linear interpolation, the IT load and ambient temperature data at different sampling frequencies of the initial data are synchronized to the same point in time to obtain the preprocessing results. The preprocessed results are then subjected to Kalman filtering for noise reduction, and the load power data is smoothed to obtain the data processing results. The rate of change of load is calculated based on the data governance results to obtain the load gradient characteristics.

[0010] The further technical solution is as follows: The preprocessing result is subjected to Kalman filtering for noise reduction, and the load power data is smoothed to obtain the data governance result, including: The preprocessed results are smoothed using a one-dimensional discrete Kalman filter to obtain the data governance results.

[0011] The further technical solution is as follows: The calculation of the load change rate based on the data governance results to obtain load gradient characteristics includes: use Calculate the rate of change of load to obtain the load gradient characteristics, where, The rate of change of load, This refers to the data governance result at time k. This refers to the data governance result at time k-1; It refers to the time difference between time k and time k-1.

[0012] The further technical solution is as follows: the preset working modes include steady-state working mode, slow-change working mode and thermal shock working mode, wherein the working mode of the load with a load change rate not greater than the minimum threshold is the steady-state working mode; the working mode of the load with a load change rate between the minimum threshold and the maximum threshold is the slow-change working mode; and the working mode of the load with a load change rate greater than the maximum threshold is the thermal shock working mode.

[0013] The further technical solution is as follows: The method of selecting a feedforward compensation or feedback control strategy based on the load mode, and processing it in conjunction with a spatial heat mapping matrix, dynamically fusing the outputs of both to obtain a suggested control quantity, includes: For loads operating in thermal shock mode, the feedforward control increment is calculated by combining the spatial thermal mapping matrix with the thermal hysteresis model and the lead-lag compensator. For loads operating in steady-state and slowly varying modes, an incremental PID feedback control algorithm is applied to eliminate feedforward control errors and adjust to the set temperature. At the same time, an anti-integral saturation strategy is adopted to obtain the feedback quantity. The feedforward control increment and the feedback quantity are dynamically weighted according to the load mode to obtain the suggested control quantity.

[0014] Its further technical solution is as follows: For the load in the thermal shock working mode, the feedforward control increment is calculated by combining the spatial thermal mapping matrix with the thermal hysteresis model and the lead-lag compensator, including: For loads operating in thermal shock mode, the corresponding responsible air conditioner is determined using a spatial thermal mapping matrix, and the feedforward control increment is calculated. In this process, a lead-lag compensator is used to reduce the thermal diffusion lag effect in the physical space, and programmable control is achieved through transfer functions and time-domain difference equations.

[0015] The further technical solution is as follows: The process of generating a drive command based on the suggested control quantity, performing safety verification through dew point protection and mechanical characteristic protection, and issuing the drive command includes: The drive command is generated based on the suggested control quantity, and the supply air temperature is prevented from falling below the dew point temperature by limiting the opening of the chilled water valve. The valve adjustment is reduced by setting a dead zone and the rate of change of the control fan speed is limited by using the slope. The drive command is then issued.

[0016] This invention also provides a data center cooling control system based on dynamic changes in IT load, comprising: The acquisition unit is used to acquire IT load data, ambient temperature data, and equipment status data to obtain initial data; The processing unit is used to preprocess, govern, and perform feature engineering on the initial data to obtain load gradient features; A mode determination unit is used to determine the load mode based on the load gradient characteristics and a preset working mode; The compensation and feedback fusion unit is used to select a feedforward compensation or feedback control strategy according to the load mode, and process it in combination with the spatial heat mapping matrix, dynamically fusing the outputs of the two to obtain the suggested control quantity. The heat mapping matrix and algorithm parameters are updated through initial calibration and online learning. The instruction generation, verification, and issuance unit is used to generate drive instructions based on the suggested control quantities, perform safety verification through dew point protection and mechanical characteristic protection, and issue the drive instructions.

[0017] The advantages of this invention compared to existing technologies are as follows: This invention acquires and processes initial data on IT load, ambient temperature, and equipment status; utilizes advanced preprocessing and feature engineering techniques to extract load gradient features; determines the operating mode accordingly; and employs feedforward compensation or feedback control strategies combined with a spatial heat mapping matrix to dynamically fuse and output suggested control quantities. Simultaneously, it continuously optimizes the heat mapping matrix and algorithm parameters through initial calibration and online learning. Based on this suggested control quantity, drive commands are generated and executed after safety checks for dew point protection and mechanical characteristic protection. This not only effectively offsets thermal hysteresis effects, achieving precise on-demand cooling and eliminating local hotspots, but also significantly reduces the overall energy consumption of the data center and greatly improves energy utilization efficiency. This systematic intelligent control method ensures that the data center maintains high-efficiency operation while minimizing energy consumption and improving overall operational efficiency.

[0018] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating a data center cooling control method based on dynamic changes in IT load, provided in an embodiment of the present invention. Figure 2 A schematic block diagram of a data center cooling control system based on dynamic changes in IT load, provided for an embodiment of the present invention; Figure 3 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0022] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0023] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0024] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0025] Please see Figure 1 , Figure 1 This is a flowchart illustrating a data center cooling control method based on dynamic IT load changes, provided in an embodiment of the present invention. This method is applied to servers. It calculates load gradient characteristics by acquiring and processing IT load, ambient temperature, and equipment status data. Based on these characteristics, it identifies the operating mode and employs feedforward compensation or feedback control strategies combined with a spatial heat mapping matrix for precise on-demand cooling. Specifically, for rapidly changing thermal shock states, a lead-lag compensator is used to reduce the thermal diffusion lag effect; while for stable or slowly changing operating modes, an incremental PID feedback control algorithm is applied to eliminate errors and adjust to the set temperature, while an anti-integral saturation strategy is employed. Furthermore, dew point protection prevents condensation short circuits, and mechanical characteristic protection reduces unnecessary valve adjustments and control fan speed changes, ensuring safe system operation while reducing energy consumption, improving energy efficiency, effectively offsetting thermal lag effects, and eliminating local hot spots, thereby achieving optimized control of the data center cooling system.

[0026] Figure 1 This is a flowchart illustrating a data center cooling control method based on dynamic changes in IT load, provided in an embodiment of the present invention. Figure 1As shown, the method includes the following steps S110 to S160.

[0027] S110: Obtain IT load data, ambient temperature data, and equipment status data to obtain initial data.

[0028] In this embodiment, initial data refers to a series of raw information collected from various components of the data center, which forms the basis for subsequent processing and decision-making. Specifically: IT load data: Through an independent out-of-band management network (OOB Network), using IPMI 2.0 or Redfish protocol, server telemetry metrics are acquired at a high sampling rate of 1Hz, including but not limited to CPU / GPU utilization, memory bandwidth, and instantaneous power of the entire machine. It can also be any operating parameter reflecting the heat generation trend of IT equipment (such as current, power, IO throughput, etc.). These data reflect the workload of the internal components of the server and are the core characterization for judging the intensity of heat sources.

[0029] Ambient temperature data: Using Modbus-TCP or BACnet / IP protocols, the operating status of the precision air conditioner (CRAC / CRAH), such as fan speed feedback, return air temperature, and supply air temperature, is read at a frequency of 0.2Hz. In addition, Zigbee / LoRa wireless sensor arrays are deployed at key locations (top, middle, and bottom) of the cabinet air inlet to monitor local microenvironmental temperature changes for more accurate temperature distribution.

[0030] Equipment status data: This includes status information for all cooling-related hardware, such as the valve opening and operating mode of the air conditioning unit. This type of data is crucial for understanding the actual operation of the current cooling system, helping to assess whether the existing configuration meets the requirements, and providing a basis for subsequent adjustments.

[0031] By acquiring the aforementioned multi-source heterogeneous data, the system can obtain a comprehensive data view at the beginning of each control cycle. However, due to the inconsistent sampling frequencies of data from different sources, direct calculation may lead to phase errors. Therefore, time alignment processing is required, for example, by using a 30-second circular buffer and applying linear interpolation resampling technology to ensure that all data can be mapped to the same moment, thereby providing accurate input for subsequent feature engineering and control algorithms. This process is the specific implementation of "acquiring IT load data, ambient temperature data, and equipment status data to obtain initial data" as described in step S110.

[0032] S120. The initial data is preprocessed, data-governed, and feature-engineered to obtain load gradient features.

[0033] In this embodiment, the load gradient characteristic refers to an index that measures the rate of load change by calculating the difference in load power after smoothing between adjacent time points.

[0034] In one embodiment, step S120 described above may include steps S121 to S123.

[0035] S121. Using a circular buffer and linear interpolation method, the IT load and ambient temperature data of the initial data at different sampling frequencies are synchronized to the same time point to obtain the preprocessing result.

[0036] In this embodiment, the preprocessing result refers to the time-aligned dataset obtained by synchronizing and linearly interpolating data at different sampling frequencies.

[0037] To ensure that data from different sources (such as IT load and ambient temperature) can be compared and analyzed at the same point in time, this embodiment employs a method based on a circular buffer and linear interpolation to achieve data synchronization. Specifically: Circular Buffer: The system maintains a circular buffer with a length of 30 seconds to store data at different sampling frequencies. For example, IT load data is collected once per second (1Hz), while ambient temperature data is collected once every 5 seconds (0.2Hz). This data is temporarily stored in the circular buffer.

[0038] Linear interpolation: At the beginning of each control cycle, all heterogeneous data will be mapped to the same time point using a linear interpolation method. .if If no measured value is available at a given moment, then the two sampling points before and after the current moment are used. and The estimated value at that moment is calculated by linear interpolation. In this way, even if the original data have different sampling rates, it is possible to obtain synchronized time series data, the so-called "preprocessed results".

[0039] S122. Perform Kalman filtering on the preprocessed results to remove noise and smooth the load power data to obtain the data processing results.

[0040] In this embodiment, the data governance result refers to the load power data after denoising and smoothing by a one-dimensional discrete Kalman filter, which is used to more accurately reflect the true "effective heat load".

[0041] Specifically, the preprocessing results are smoothed using a one-dimensional discrete Kalman filter to obtain data governance results.

[0042] Next, the preprocessing results need to be further processed to eliminate sensor electromagnetic interference and non-thermal load transient spikes (such as a CPU utilization of 100% lasting only one second). This step is accomplished by applying a one-dimensional discrete Kalman filter.

[0043] Prediction step: First, predict the value of the next state, i.e. ; Update step: Then, based on the actual measured values... The difference between the estimated value and the predicted value is adjusted using the following formula: ;in For measured values, Kalman gain. After filtering. It can more accurately reflect the "effective heat load".

[0044] S123. Calculate the rate of change of load based on the data governance results to obtain the load gradient characteristics.

[0045] The final step involves calculating the rate of change of the load, i.e., the load gradient characteristic, which is done by performing a differential operation on the data governance results.

[0046] In this embodiment, the following is adopted: Calculate the rate of change of load to obtain the load gradient characteristics, where, The rate of change of load, This refers to the data governance result at time k. This refers to the data governance result at time k-1; It refers to the time difference between time k and time k-1.

[0047] Through the above three steps—data synchronization and timing alignment, Kalman filtering for noise reduction, and load gradient calculation—key load gradient features can be effectively extracted from the original input data, thereby providing accurate basic information for the subsequent control system to make more precise cooling adjustment decisions.

[0048] S130. Determine the load mode based on the load gradient characteristics and the preset working mode.

[0049] In this embodiment, load mode refers to different operating states defined based on the rate of change of server load (i.e., load gradient). This process mainly involves calculating the first-order difference (rate of change) of load power and comparing it with a preset threshold to determine which operating mode the data center IT equipment is currently in.

[0050] The preset operating modes include steady-state operating mode, slow-change operating mode, and thermal shock operating mode. Among them, the operating mode of the load with a load change rate not greater than the minimum threshold is the steady-state operating mode; the operating mode of the load with a load change rate between the minimum threshold and the maximum threshold is the slow-change operating mode; and the operating mode of the load with a load change rate greater than the maximum threshold is the thermal shock operating mode.

[0051] Specifically, the system defines three operating modes based on Gradient: Steady Mode: This means that the heat output of IT equipment is relatively stable, without a significant increasing or decreasing trend. In this case, the system is more inclined to use feedback control to maintain temperature stability.

[0052] Ramp Mode: In this state, the load is gradually changing, but the rate of change is insufficient to trigger the feedforward control mechanism. Therefore, the system will combine feedforward and feedback control to cope with potential temperature fluctuations.

[0053] Thermal shock mode: (For example, >50W / s). At this point, the system determines that local overheating is about to occur. For example, if there is an instantaneous power increase exceeding 50W / s, the system will identify it as a thermal shock operating mode. In this case, since a rapid temperature rise is expected, the system will immediately activate the feedforward control mechanism to increase the cooling capacity of the corresponding area in advance to offset the impending thermal shock and prevent the formation of hot spots.

[0054] To accurately determine the aforementioned operating mode, the system first calculates the load gradient for each control cycle, typically by comparing the smoothed load power differences between adjacent time points. Next, the calculated load gradient value is compared to a set threshold to determine the current specific operating mode. This mechanism ensures that the cooling system responds promptly and appropriately, effectively managing the data center's temperature environment without sacrificing energy efficiency.

[0055] S140. Select a feedforward compensation or feedback control strategy based on the load mode, and process it in conjunction with the spatial heat mapping matrix. Dynamically fuse the outputs of the two to obtain the suggested control quantity. The heat mapping matrix and algorithm parameters are updated through initial calibration and online learning.

[0056] In this embodiment, the recommended control quantity refers to the final control signal used to regulate the air conditioning equipment, calculated by combining the dynamic weighted feedforward control increment and feedback quantity with the current load mode.

[0057] In one embodiment, step S140 described above may include steps S141 to S143.

[0058] S141. For loads operating in thermal shock mode, the feedforward control increment is calculated by combining the spatial thermal mapping matrix with the thermal hysteresis model and the lead-lag compensator.

[0059] In this embodiment, the feedforward control increment refers to the control adjustment amount calculated based on the thermal shock load condition, using the spatial thermal mapping matrix and the lead-lag compensator, to adjust the air conditioner in advance to cope with future temperature increases.

[0060] For loads operating in thermal shock mode, the corresponding responsible air conditioner is determined using a spatial thermal mapping matrix, and the feedforward control increment is calculated. In this process, a lead-lag compensator is used to reduce the thermal diffusion lag effect in the physical space, and programmable control is achieved through transfer functions and time-domain difference equations.

[0061] For loads operating in a thermal shock mode (i.e., when the load change rate exceeds the maximum threshold), the system first utilizes the spatial thermal mapping matrix. Identify the air conditioner responsible for that load. For example, if the server... If the system is under high load, then locate all air conditioners associated with it. And according to weight Calculate the corresponding responsible air conditioner. Introduce a Z-domain lead-lag compensator. This is to offset the lag in thermal diffusion in physical space.

[0062] Next, a thermal hysteresis model is used to predict the time constant of heat transfer from the server to the air conditioner, and a lead-lag compensator is applied to reduce the hysteresis effect of heat diffusion in the physical space. Specifically, this is implemented using a transfer function: ; and time-domain difference equations: ;in This causes the output signal to generate a brief "kick-start" at the moment of sudden load increase, driving the fan to quickly overcome rotational inertia and start operating at high air volume ahead of schedule.

[0063] S142. For loads in steady-state and slowly varying operating modes, an incremental PID feedback control algorithm is applied to eliminate feedforward control errors and adjust to the set temperature. At the same time, an anti-integral saturation strategy is adopted to obtain the feedback quantity.

[0064] In this embodiment, the feedback quantity refers to the control adjustment quantity calculated by the incremental PID feedback control algorithm for steady-state or slowly varying loads, used to eliminate the current temperature error and adjust to the set temperature.

[0065] For loads operating in steady-state or slowly varying modes, the system will employ an incremental PID feedback control algorithm to correct the error in the feedforward control and adjust it to the set temperature.

[0066] Used to eliminate feedforward model bias and steady-state error: ,in It is the actual temperature of the current air conditioner j, and It is the set target temperature.

[0067] Output increment of PID controller The calculation is as follows: ; Furthermore, to prevent sluggish response caused by the integral term reaching the physical limit (e.g., 100% speed), an anti-integral saturation strategy is adopted: when the actuator reaches its physical limit (e.g., 100% speed), it is forced to... This prevents the system from becoming sluggish during desaturation.

[0068] S143. The feedforward control increment and the feedback quantity are dynamically weighted according to the load mode to obtain the suggested control quantity.

[0069] In this embodiment, a control quantity is recommended. Not a simple feedforward control increment and feedback volume It is not a superposition, but a dynamic weighting based on load modes: ; The weight function Using the Sigmoid function: ; This represents the rate of change of the total load in the current region. This design ensures that the system tends to rely more on feedforward control when the load fluctuates drastically (at this time...). When the load is relatively stable, it relies more on feedback control (at which point...). ).

[0070] Furthermore, throughout the process, the spatial heat mapping matrix The system's parameters are continuously updated through initial calibration and online learning, thereby improving its adaptability and accuracy. This not only helps optimize cooling efficiency but also effectively reduces energy consumption and extends equipment lifespan.

[0071] In this embodiment, the spatial heat mapping matrix (The Thermal Coupling Matrix): This is a... sparse matrix ( Total number of servers (Total number of air conditioners). Physical meaning: Indicates server The generated heat load includes The proportion of air conditioning Air is drawn in through the return air vent. Constraints: (Energy conservation constraint). The initial value of this matrix is ​​generated by CFD simulation and corrected online during operation using LMS (Least Mean Square algorithm).

[0072] S150. Generate a drive command based on the suggested control quantity, perform safety verification through dew point protection and mechanical characteristic protection, and issue the drive command.

[0073] In this embodiment, a drive command is generated based on the suggested control quantity, and the supply air temperature is prevented from falling below the dew point temperature by limiting the opening of the chilled water valve. The valve adjustment is reduced by setting a dead zone, and the rate of change of the control fan speed is limited by using a slope. The drive command is then issued.

[0074] In this embodiment, the system first generates specific air conditioning equipment drive commands based on the calculated suggested control quantities (i.e., the fusion result of feedforward control increments and feedback control quantities). These commands aim to adjust the operating state of the precision air conditioner (CRAC / CRAH), such as adjusting fan speed and valve opening, to achieve precise temperature control. However, before executing these commands, the system performs a series of safety checks to ensure the safety and effectiveness of the operation: Dew point protection: To prevent the risk of short circuits caused by condensation inside the cabinet, the system reads the current ambient humidity and calculates the corresponding dew point temperature. If the predicted supply air temperature is lower than the dew point temperature plus a safety margin (e.g., 1.5°C), the system will automatically limit the opening of the chilled water valve to avoid overcooling.

[0075] Mechanical protection: To avoid damage to valves and fans caused by frequent small adjustments, and to reduce the risk of current surges and belt slippage, the system is equipped with a deadband and a rate limiter. When the parameter to be adjusted changes by less than 1%, the system will not issue an adjustment command; and for changes in fan speed, the change will not exceed 5% per second, thus ensuring the smooth operation of the equipment.

[0076] After completing the above security checks, the system will finally send the processed drive command to the corresponding air conditioning equipment to achieve the purpose of providing cooling on demand.

[0077] S160. In the event of sensor failure or network interruption, take measures such as virtual interpolation, conservative mode or local takeover to handle the situation.

[0078] To improve the system's reliability and fault tolerance, this embodiment also includes a mechanism for handling abnormal situations: Sensor Failure Handling: If a thermal aisle sensor fails to provide valid data, the system will use data from adjacent sensors for "virtual interpolation" to estimate the temperature of the missing area. If the number of failed sensors exceeds 30%, the system will automatically switch to conservative mode, lowering the setpoints of all air conditioners by 2°C to ensure the safe operation of the data center.

[0079] Network Outage Handling (Split-Brain Protection): In the event of a network connection interruption lasting longer than 60 seconds, to prevent data center cooling from spiraling out of control, the local controller at the air conditioning unit will automatically take over control and revert to the factory-set simple return air temperature PID control mode. This strategy ensures that even without AI intelligent control, the data center can maintain basic cooling functions, preventing downtime.

[0080] Through the above mechanism, this embodiment not only improves the response speed and energy efficiency of the data center cooling system, but also enhances the stability and reliability of the system.

[0081] In this embodiment, by collecting real-time load data from IT equipment and utilizing a "spatial mapping relationship" and a "thermal hysteresis model," the future temperature rise trend is predicted. Before the actual temperature increase, the cooling capacity of the corresponding area is increased in advance to offset the thermal hysteresis effect, achieving "on-demand cooling" and significantly reducing energy consumption while eliminating hotspots. Specifically, a "time-series feedforward-feedback dual decoupled cooling control system based on heterogeneous telemetry data fusion" is constructed. This system runs on the edge computing node of Data Center Infrastructure Management (DCIM) or a standalone AI control server. By establishing a dynamic spatiotemporal mapping relationship between IT load and cooling equipment, and utilizing the physical characteristic that load changes precede temperature changes, feedforward control intervenes in advance before heat causes a significant temperature rise in the atmosphere, and feedback control corrects steady-state errors.

[0082] The system is logically structured into four layers: the physical sensing layer, the data governance layer, the decision computing layer, and the execution-driven layer. The physical sensing layer includes the southbound interface (IT side) which reads server telemetry metrics via an independent out-of-band management network; the southbound interface (facilities side) which reads the operating status of precision air conditioners via Modbus-TCP or BACnet / IP protocols; and the environmental sensing layer which deploys Zigbee / LoRa wireless sensor arrays to cover key locations at the cabinet air inlets. The system memory maintains enhanced server state vectors, air conditioning control vectors, and a spatial heat mapping matrix. and thermal hysteresis eigenvector Key data objects are specified to ensure programmability and precise control in the implementation. The system's main control cycle is set to ΔT = 10 seconds.

[0083] Specifically, to ensure the programmability and precise control of the embodiments, the following key data objects are maintained in the system memory: Enhanced Server State Vector It not only includes numerical values, but also introduces timestamps and confidence levels to handle network packet loss or latency. ;in, Instantaneous power of the whole machine (as the core characterization of heat source intensity). Data quality confidence level. When packet loss occurs for more than 3 consecutive periods, If the value decreases, the system will automatically reduce the weight of that data.

[0084] Air conditioning control vector : ; Spatial heat mapping matrix ; Thermal hysteresis eigenvector : ,in This represents heat transfer from the server to the air conditioner. The physical time constant (TimeConstant). This value changes dynamically with the current speed of the fan: .

[0085] Furthermore, before final issuance, the system undergoes "pruning" and "correction" through physical constraint limitations and dew point protection logic to ensure that the final generated ACT command remains within physical safety limits. To ensure safety, two safety checks are required: dew point protection and mechanical characteristic protection. The system initialization and online adaptive process is responsible for initializing and updating the spatial heat mapping matrix. The algorithm parameters, belonging to the "parameter plane" or "meta-learning" layer of the control loop, do not directly participate in the real-time 10-second control loop, but they determine the accuracy of the control algorithm. Anomaly handling and fault tolerance mechanisms cover various situations, from sensor failures to network interruptions, ensuring the stability and reliability of the system.

[0086] Specifically, during the initial system deployment, a "step response test" is performed. Each rack is run at full load for 10 minutes sequentially, and the return air temperature response curves of each air conditioner are recorded. The initial matrix is ​​then calculated through data fitting. and time constant .

[0087] Online adaptive correction (Online Learning): Taking into account changes in data center equipment (such as server removal or blind plate detachment), a built-in LMS (Least Mean Square) estimator is provided.

[0088] Continuously monitor the residual between the "predicted temperature" and the "actual temperature": .

[0089] When the residual is consistently large, the matrix elements are automatically fine-tuned. : This gives the system the ability to "learn for life" without the need for manual recalibration.

[0090] The core protection point of this embodiment lies in using real-time load data from IT equipment as a feedforward signal to adjust the cooling equipment control method in advance before temperature sensor changes occur, thus solving the problem of response lag in traditional PID control. Through the above technical solution, this embodiment not only improves the response speed and energy efficiency of the data center cooling system but also reduces mechanical wear and electrical grid impact, possessing significant economic and technical value. Furthermore, compared to alternatives such as liquid cooling technology and infrared thermal imaging monitoring, this embodiment is lower in cost and does not require large-scale replacement of existing hardware, making it more suitable for the optimization and upgrade of air-cooled data centers.

[0091] This embodiment achieves significant optimization of the data center cooling system by employing a feedforward predictive control method. Firstly, in terms of safety, the system responds 1-5 minutes faster than traditional temperature control, effectively preventing instantaneous overheating and frequency throttling of servers due to sudden high loads, thus eliminating hotspot risks. Secondly, from an economic perspective, the more precise control eliminates the fear of "lag," allowing the base operating temperature of the data center to be set higher (e.g., from 22°C to 24°C). According to industry data, for every 1°C increase in return air temperature, cooling energy consumption can be reduced by approximately 3%-5%, resulting in significant energy savings. Furthermore, stability is also enhanced. By predicting future heat loads, the air conditioning system can accelerate linearly and slowly, rather than abruptly accelerating after the temperature exceeds limits, reducing mechanical wear on equipment and grid impact.

[0092] As an alternative to the technical solution, liquid cooling technology (including cold plate or immersion cooling) is a purely hardware solution. However, while liquid cooling can solve the problem of slow heat dissipation, it requires a complete replacement of server hardware and data center infrastructure, resulting in extremely high costs and making it unsuitable for retrofitting existing air-cooled data centers. In contrast, this embodiment proposes a purely software / control-level optimization that is extremely low-cost and easy to implement. Another alternative is to use infrared cameras to monitor rack temperature. Although this method is faster than air sensors and provides quicker feedback, it still lags behind changes in "load," and due to obstruction issues, it is difficult to accurately monitor the actual heat sources inside the rack, making it less effective than the method in this embodiment.

[0093] To better understand this technology, it's helpful to refer to some industry trends and practices. With the integration of the Internet of Things (IoT) and machine learning (ML) technologies, the industry is beginning to leverage real-time data and predictive models to optimize data center cooling systems, moving beyond traditional PID control. For example, AI-driven predictive HVAC control systems can learn the complex relationships between IT load, environmental changes, and cooling system response, enabling proactive adjustments to cooling response to improve energy efficiency. A case study of intelligent cooling control systems is Cooling Optimize, which uses a closed-loop AI engine to continuously analyze the air inlet temperatures of air conditioners and servers, automatically adjusting fans and cooling units to achieve on-demand cooling. Furthermore, numerous blog posts highlight that key optimization strategies for data center cooling lie in airflow management, load analysis, and the appropriate setting of temperature setpoints, emphasizing the importance of dynamic matching between control strategies and cooling demands.

[0094] In related research and technology comparisons, studies applying Model Predictive Control (MPC) to data center cooling have shown that predicting future temperature responses based on system dynamic models and adjusting HVAC settings in advance can reduce the hysteresis effects caused by thermal inertia, significantly outperforming traditional PID control. Other research indicates that using deep learning (such as 3D models and neural networks) to rapidly predict the thermal distribution field within data centers enables control systems to predict temperature distribution at the millisecond level, achieving more refined on-demand cooling strategies. Furthermore, offline reinforcement learning has been used to construct schedulable HVAC control strategies. These methods train strategies using extensive historical data, significantly reducing cooling energy consumption while ensuring equipment safety constraints.

[0095] Finally, a comparative analysis with existing technologies reveals that traditional PID control relies heavily on temperature feedback, resulting in issues such as response lag and overshoot risk, particularly pronounced in ventilation and cooling systems with significant thermal inertia. In contrast, predictive control or AI control overcomes thermal lag by anticipating load thermal effects and adjusting cooling output accordingly. This embodiment, through the fusion of feedforward and feedback loops, combined with load derivative judgment and dynamic weight adjustment, represents a fundamental difference in control strategy compared to traditional static PID or single predictive methods. Furthermore, while some systems on the market employ "influence models" or "empirical thermal coupling relationships," these are typically only used to estimate cooling effects. This embodiment, however, specifically describes the spatial thermal contribution through a weight matrix of server load and air conditioning response, providing a more formal mathematical modeling method.

[0096] This embodiment of the method, by collecting real-time load data from IT equipment and utilizing "spatial mapping relationships" and "thermal hysteresis models," can accurately predict future temperature rise trends. Before the actual temperature increase, it proactively increases cooling capacity in relevant areas to offset the hysteresis effect caused by thermal inertia. This method not only achieves on-demand cooling, effectively preventing the formation of hotspots, but also significantly reduces energy consumption and improves the efficiency and stability of the data center cooling system. This forward-looking control strategy ensures that heat dissipation requirements are met while minimizing unnecessary energy loss, demonstrating a perfect combination of intelligent optimization and energy conservation.

[0097] The aforementioned data center cooling control method based on dynamic IT load changes acquires and processes initial data on IT load, ambient temperature, and equipment status. It then utilizes advanced preprocessing and feature engineering techniques to extract load gradient characteristics and determine the operating mode. A feedforward compensation or feedback control strategy combined with a spatial heat mapping matrix is ​​used to dynamically fuse and output suggested control quantities. Simultaneously, the heat mapping matrix and algorithm parameters are continuously optimized through initial calibration and online learning. Based on this suggested control quantity, drive commands are generated and executed after safety checks for dew point protection and mechanical characteristic protection. This not only effectively offsets thermal hysteresis effects, achieving precise on-demand cooling and eliminating localized hotspots, but also significantly reduces the overall energy consumption of the data center and greatly improves energy efficiency. This systematic intelligent control method ensures that the data center maintains high-efficiency operation while minimizing energy consumption and improving overall operational efficiency.

[0098] Figure 2 This is a schematic block diagram of a data center cooling control system 300 based on dynamic changes in IT load, provided in an embodiment of the present invention. Figure 2 As shown, corresponding to the above-described data center cooling control method based on dynamic IT load changes, the present invention also provides a data center cooling control system 300 based on dynamic IT load changes. This data center cooling control system 300 includes a unit for executing the above-described data center cooling control method based on dynamic IT load changes, and the system can be configured in a server. Specifically, please refer to... Figure 2 The data center cooling control system 300 based on dynamic changes in IT load includes an acquisition unit 301, a processing unit 302, a modality determination unit 303, a compensation feedback fusion unit 304, and an instruction generation, verification, and issuance unit 305.

[0099] The acquisition unit 301 is used to acquire IT load data, ambient temperature data, and equipment status data to obtain initial data; the processing unit 302 is used to preprocess, govern, and perform feature engineering on the initial data to obtain load gradient characteristics; the mode determination unit 303 is used to determine the load mode based on the load gradient characteristics and preset working modes; the compensation feedback fusion unit 304 is used to select a feedforward compensation or feedback control strategy based on the load mode, and process it in conjunction with the spatial heat mapping matrix, dynamically fusing the two outputs to obtain a suggested control quantity, wherein the heat mapping matrix and algorithm parameters are updated through initial calibration and online learning; the instruction generation, verification, and issuance unit 305 is used to generate drive instructions based on the suggested control quantity, perform safety verification through dew point protection and mechanical characteristic protection, and issue the drive instructions.

[0100] It also includes a fault handling unit 302, which is used to handle sensor failures or network interruptions by taking virtual interpolation, conservative mode or local takeover measures.

[0101] In one embodiment, the processing unit 302 includes: The preprocessing subunit is used to synchronize IT load and ambient temperature data at different sampling frequencies of the initial data to the same time point through a circular buffer and linear interpolation method to obtain preprocessing results; the data governance subunit is used to perform Kalman filtering to denoise the preprocessing results and smooth the load power data to obtain data governance results; the calculation subunit is used to calculate the load change rate based on the data governance results to obtain load gradient characteristics.

[0102] In one embodiment, a data governance subunit is used to smooth the load power data using a one-dimensional discrete Kalman filter on the preprocessing result to obtain a data governance result.

[0103] In one embodiment, the computing subunit is used to employ Calculate the rate of change of load to obtain the load gradient characteristics, where, The rate of change of load, This refers to the data governance result at time k. This refers to the data governance result at time k-1; It refers to the time difference between time k and time k-1.

[0104] In one embodiment, the compensation feedback fusion unit 304 includes: The feedforward subunit is used to calculate the feedforward control increment for loads operating in thermal shock mode by combining the spatial thermal mapping matrix with the thermal hysteresis model and the lead-lag compensator. The feedback subunit is used to eliminate the feedforward control error and adjust to the set temperature by applying an incremental PID feedback control algorithm for loads operating in steady-state and slowly varying modes, while adopting an anti-integral saturation strategy to obtain the feedback quantity. The fusion subunit is used to dynamically weight the feedforward control increment and the feedback quantity according to the load mode to obtain the suggested control quantity.

[0105] In one embodiment, the feedforward subunit is used to determine the corresponding responsible air conditioner using a spatial thermal mapping matrix for a load in a thermal shock operating mode, and to calculate the feedforward control increment. The lead-lag compensator is used to reduce the thermal diffusion lag effect in the physical space, and programmable control is achieved through transfer functions and time-domain difference equations.

[0106] In one embodiment, the instruction generation, verification, and issuance unit 305 is used to generate drive instructions based on the suggested control quantity, prevent the supply air temperature from falling below the dew point temperature by limiting the opening of the chilled water valve, reduce valve adjustment by setting a dead zone and limit the rate of change of the control fan speed by using a slope, and issue the drive instructions.

[0107] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned data center cooling control system 300 based on dynamic changes in IT load and each unit can be referred to the corresponding description in the foregoing method embodiments. For the sake of convenience and brevity, it will not be repeated here.

[0108] The aforementioned data center cooling control system 300 based on dynamic changes in IT load can be implemented as a computer program, which can, for example... Figure 3 It runs on the computer device shown.

[0109] Please see Figure 3 , Figure 3 This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device 500 can be a server, wherein the server can be a standalone server or a server cluster composed of multiple servers.

[0110] See Figure 3 The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a system bus 501. The memory may include a non-volatile storage medium 503 and internal memory 504.

[0111] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform a data center cooling control method based on dynamic changes in IT load.

[0112] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0113] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a data center cooling control method based on dynamic changes in IT load.

[0114] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 500 to which the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0115] The processor 502 is used to run a computer program 5032 stored in the memory to implement all the steps of the data center cooling control method based on dynamic changes in IT load.

[0116] It should be understood that in the embodiments of this application, the processor 502 may be a central processing unit (CPU) 302, or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0117] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0118] Therefore, the present invention also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein when executed by a processor, the computer program causes the processor to perform all steps of the data center cooling control method based on dynamic changes in IT load.

[0119] The storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0120] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0121] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of each unit is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0122] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the system of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit 302, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0123] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0124] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A data center cooling control method based on dynamic changes in IT load, characterized in that, include: Acquire IT load data, ambient temperature data, and equipment status data to obtain initial data; The initial data is preprocessed, data-governed, and feature-engineered to obtain load gradient features; The load mode is determined based on the load gradient characteristics and the preset operating mode; Based on the load mode, a feedforward compensation or feedback control strategy is selected and processed in conjunction with the spatial heat mapping matrix. The outputs of the two are dynamically fused to obtain the suggested control quantity. The heat mapping matrix and algorithm parameters are updated through initial calibration and online learning. The drive command is generated based on the suggested control quantity, and safety verification is performed through dew point protection and mechanical characteristic protection before the drive command is issued.

2. The data center cooling control method based on dynamic changes in IT load according to claim 1, characterized in that, Also includes: In the event of sensor failure or network outage, virtual interpolation, conservative mode, or local takeover measures will be adopted.

3. The data center cooling control method based on dynamic changes in IT load according to claim 1, characterized in that, The process of preprocessing, data governance, and feature engineering the initial data to obtain load gradient features includes: By using a circular buffer and linear interpolation, the IT load and ambient temperature data at different sampling frequencies of the initial data are synchronized to the same point in time to obtain the preprocessing results. The preprocessed results are then subjected to Kalman filtering for noise reduction, and the load power data is smoothed to obtain the data processing results. The rate of change of load is calculated based on the data governance results to obtain the load gradient characteristics.

4. The data center cooling control method based on dynamic changes in IT load according to claim 3, characterized in that, The process of performing Kalman filtering to denoise the preprocessed results and smoothing the load power data to obtain the data processing results includes: The preprocessed results are smoothed using a one-dimensional discrete Kalman filter to obtain the data governance results.

5. The data center cooling control method based on dynamic changes in IT load according to claim 1, characterized in that, The step of calculating the rate of change of load based on the data governance results to obtain load gradient characteristics includes: use Calculate the rate of change of load to obtain the load gradient characteristics, where, The rate of change of load, This refers to the data governance result at time k. This refers to the data governance result at time k-1; It refers to the time difference between time k and time k-1.

6. The data center cooling control method based on dynamic changes in IT load according to claim 5, characterized in that, The preset operating modes include steady-state operating mode, slow-change operating mode, and thermal shock operating mode. Among them, the operating mode of the load with a load change rate not greater than the minimum threshold is the steady-state operating mode; the operating mode of the load with a load change rate between the minimum threshold and the maximum threshold is the slow-change operating mode; and the operating mode of the load with a load change rate greater than the maximum threshold is the thermal shock operating mode.

7. The data center cooling control method based on dynamic changes in IT load according to claim 6, characterized in that, The step of selecting either feedforward compensation or feedback control strategy based on the load mode, processing it in conjunction with the spatial heat mapping matrix, and dynamically fusing the outputs of both to obtain the suggested control quantity includes: For loads operating in thermal shock mode, the feedforward control increment is calculated by combining the spatial thermal mapping matrix with the thermal hysteresis model and the lead-lag compensator. For loads operating in steady-state and slowly varying modes, an incremental PID feedback control algorithm is applied to eliminate feedforward control errors and adjust to the set temperature. At the same time, an anti-integral saturation strategy is adopted to obtain the feedback quantity. The feedforward control increment and the feedback quantity are dynamically weighted according to the load mode to obtain the suggested control quantity.

8. The data center cooling control method based on dynamic changes in IT load according to claim 7, characterized in that, For the load operating in the thermal shock mode, the feedforward control increment is calculated using a thermal hysteresis model and a lead-lag compensator, based on the spatial thermal mapping matrix, including: For loads operating in thermal shock mode, the corresponding responsible air conditioner is determined using a spatial thermal mapping matrix, and the feedforward control increment is calculated. In this process, a lead-lag compensator is used to reduce the thermal diffusion lag effect in the physical space, and programmable control is achieved through transfer functions and time-domain difference equations.

9. The data center cooling control method based on dynamic changes in IT load according to claim 1, characterized in that, The process of generating drive commands based on the suggested control values, performing safety checks through dew point protection and mechanical characteristic protection, and issuing the drive commands includes: The drive command is generated based on the suggested control quantity, and the supply air temperature is prevented from falling below the dew point temperature by limiting the opening of the chilled water valve. The valve adjustment is reduced by setting a dead zone and the rate of change of the control fan speed is limited by using the slope. The drive command is then issued.

10. A data center cooling control system based on dynamic changes in IT load, characterized in that, include: The acquisition unit is used to acquire IT load data, ambient temperature data, and equipment status data to obtain initial data; The processing unit is used to preprocess, govern, and perform feature engineering on the initial data to obtain load gradient features; A mode determination unit is used to determine the load mode based on the load gradient characteristics and a preset working mode; The compensation and feedback fusion unit is used to select a feedforward compensation or feedback control strategy according to the load mode, and process it in combination with the spatial heat mapping matrix, dynamically fusing the outputs of the two to obtain the suggested control quantity. The heat mapping matrix and algorithm parameters are updated through initial calibration and online learning. The instruction generation, verification, and issuance unit is used to generate drive instructions based on the suggested control quantities, perform safety verification through dew point protection and mechanical characteristic protection, and issue the drive instructions.

Images