Data center partition air conditioning temperature control energy-saving system based on computing power load dynamic prediction

Abstract

The application discloses a data center partition air conditioning temperature control energy-saving system based on computing power load dynamic prediction, belongs to the technical field of data center heat dissipation management and air conditioning system control, and is characterized in that a computing power load dynamic prediction mechanism is introduced to fuse and predict current computing power load, environment temperature and air conditioning heat source temperature of each target partition, output predicted computing power load sequence and predicted air conditioning heat source temperature sequence, convert the predicted computing power load into an IT equipment heat generation power prediction value through a thermodynamic equivalent model, convert the predicted air conditioning heat source temperature combined with a future environment temperature prediction value into an air conditioning cooling power prediction value, form a predicted heat load map containing a supply-demand relationship, and generate a corresponding feedforward control strategy based on map deviation analysis, so that the refrigeration output of the air conditioning system is dynamically matched with actual heat generation demand, the inherent thermal hysteresis of traditional feedback control is eliminated, the energy consumption of the data center air conditioning system is significantly reduced, and the power utilization efficiency is optimized.

Description

Technical Field

[0001] This invention relates to the field of data center heat dissipation management and air conditioning system control technology, and more specifically, to a data center zone air conditioning temperature control and energy-saving system based on dynamic prediction of computing load. Background Technology

[0002] Air conditioning temperature control systems are widely used in buildings and data centers. Conventional data center air conditioning systems usually adopt single-point temperature control or simple room-level group control. For example, the intelligent control system of air conditioning temperature controller disclosed in Chinese invention patent CN115654681A collects the current indoor temperature obtained by the indoor temperature detection unit through the wireless communication module and compares it with the preset temperature. When the temperature difference exceeds the set range, the microcontroller links to switch the electric valve, start the fan coil thermostat or the floor heating thermostat to adjust the heating or cooling, thereby realizing the automatic control of the indoor temperature.

[0003] The system also integrates functions such as remote control via APP, metering and billing, and cloud data recording to solve the problems of messy wiring and decentralized control in traditional wired control methods. However, the above-mentioned features and most similar air conditioning temperature control systems all adopt feedback control logic based on the current temperature deviation. This control method has the following inherent defects: 1. Control lag: The system only responds after the temperature has risen or fallen. In scenarios with high dynamic computing load, such as data centers, the temperature of server chips can rise sharply within seconds. The thermal inertia of the air conditioning system causes the cooling response to lag by several minutes, which can easily lead to local overheating or even server crashes.

[0004] 2. Disconnect between load and cooling: Traditional control is based solely on ambient temperature, which is completely decoupled from the real-time computing load of IT equipment (such as CPU / GPU utilization). It is impossible to predict the heat generation trend caused by business fluctuations, resulting in a mismatch between cooling capacity allocation and actual demand.

[0005] 3. Low energy efficiency: Due to the lack of prediction of future heat load, the system often adopts an "overcooling" strategy to deal with sudden hot spots, resulting in a lot of energy waste and making it difficult to meet the requirements of green data centers for low PUE (Power Usage Effectiveness).

[0006] Therefore, there is an urgent need to provide an air conditioning temperature control system that can overcome the above-mentioned defects in order to achieve precise coordination between cooling and computing load in data center scenarios. Summary of the Invention

[0007] The purpose of this invention is to address practical technical deficiencies, and it provides a data center zone air conditioning temperature control and energy-saving system based on dynamic prediction of computing load.

[0008] The objective of this invention can be achieved through the following technical solution: a data center zone air conditioning temperature control and energy-saving system based on dynamic prediction of computing load, comprising: a thermal management zone data acquisition module, deployed in each target zone of the data center, used to collect real-time computing load data, real-time ambient temperature data and air conditioning heat source temperature data of the IT equipment associated with each target zone at the same timestamp; The load and environment coupling prediction module has a built-in multivariate time series prediction model, which is used to output the predicted computing load sequence and the predicted air conditioning heat source temperature sequence of the target partition in the future preset time window based on the computing load data, real-time ambient temperature data and air conditioning heat source temperature data of the current operating status of each target partition. The heat load mapping module is used to receive the predicted computing power load sequence and the predicted air conditioning heat source temperature sequence, and simultaneously obtain the predicted value of future ambient temperature. Based on the thermodynamic equivalent model, it performs heat load mapping and outputs the predicted heat load map of the target partition within a preset time window in the future. The zone temperature control module is used to analyze the temperature compensation requirements based on the predicted heat load map and the current real-time ambient temperature of the target zone, generate temperature compensation requirement signals for each zone's air conditioner, and generate feedforward control commands for the air conditioner terminal based on each signal and send them to the corresponding air conditioner terminal actuator.

[0009] Furthermore, the computing load data includes at least one of GPU utilization, memory usage, and disk I / O throughput; the air conditioning heat source temperature data includes at least one of air conditioning supply air temperature, return air temperature, chilled water supply temperature, and chilled water return temperature; and the real-time ambient temperature data includes at least one of rack air inlet temperature, cold aisle temperature, and hot aisle temperature.

[0010] Furthermore, the multivariate time series prediction model is a long short-term memory network (LSTM) or a gated recurrent unit (GRU) with an encoder-decoder architecture. Its input data includes: computing load sequence, ambient temperature sequence, and air conditioning heat source temperature sequence within the historical time window of each target partition. Its output data includes: predicted computing load sequence and predicted air conditioning heat source temperature sequence within the future preset time window of each target partition.

[0011] Furthermore, the process of obtaining the predicted heat load map includes: Establish a thermodynamic equivalent model for each target zone. The thermodynamic equivalent model is used to describe the dynamic balance mapping relationship between the heat generation power of IT equipment and the cooling power of air conditioning terminals in the zone, as well as the temperature change law of the zone over time. The mapping relationship based on the thermodynamic equivalent model converts the received predicted computing power load sequence into a predicted value sequence of IT equipment heat generation power in the target partition within a future preset time window. Obtain the predicted future ambient temperature of the zone, and convert the predicted air conditioning heat source temperature sequence and the predicted future ambient temperature into a predicted air conditioning cooling power sequence for the target zone within a preset time window based on the mapping relationship of the thermodynamic equivalent model. The predicted heat load map for this region is generated by associating the predicted heat output sequence of IT equipment with the predicted cooling output sequence of air conditioning.

[0012] Furthermore, the thermodynamic equivalent model is either a heat balance model based on physical mechanisms or a heat response model based on data-driven principles. The dynamic equilibrium mapping relationship includes power mapping relationship and cooling power mapping relationship.

[0013] Furthermore, the specific process of analyzing temperature control compensation requirements includes: Extract the predicted power generation sequence of IT equipment and the predicted power generation sequence of air conditioning from the predicted heat load map of the partition, and calculate the power supply and demand deviation sequence at each time step within the future preset time window; Obtain the real-time ambient temperature of the target zone and the preset target temperature, and calculate the current temperature deviation; Based on the trend characteristics of the power supply and demand deviation sequence and the magnitude of the current temperature deviation, the temperature control compensation demand is analyzed according to the preset threshold comparison rule, and a temperature control compensation demand signal for the target partition within a preset time window is generated.

[0014] Furthermore, the preset threshold comparison rules include: The power supply and demand deviation is compared with a preset first threshold and a second threshold, and the current temperature deviation is compared with a preset first threshold and a second threshold. Based on the comparison results, a temperature control compensation demand signal for the partition is generated. When the power supply and demand deviation is greater than the first threshold and the current temperature deviation is greater than the first threshold, it is determined as an "emergency compensation demand" and an emergency strong cooling demand signal is generated. When the power supply and demand deviation is greater than the first threshold and the current temperature deviation is less than the second threshold, it is determined to be a "preventive compensation demand" and a forward-looking pre-cooling demand signal is generated. When the power deviation between supply and demand is less than the second threshold and the current temperature deviation is greater than the first threshold, it is determined to be "reduced supply demand" and a transient demand maintenance signal is generated. When the power supply and demand deviation is less than the second threshold and the current temperature deviation is less than the second threshold, it is determined to be "maintain or fine-tune demand", and an energy-saving and supply reduction demand signal is generated. When the power supply and demand deviation is between the first threshold and the second threshold, and the current temperature deviation is between the first threshold and the second threshold, it is determined that there is "no adjustment demand" and a steady-state maintenance demand signal is generated.

[0015] Furthermore, the feedforward control commands specifically include: When an emergency strong cooling demand signal is generated, feedforward control commands are generated to significantly reduce the supply air temperature setpoint, significantly increase the fan speed, and significantly increase the electronic expansion valve opening. When a forward-looking pre-cooling demand signal is generated, feedforward control commands are generated to slightly reduce the supply air temperature setpoint, slightly increase the fan speed, and slightly increase the opening of the electronic expansion valve. When a transient maintenance demand signal is generated, feedforward control commands are generated to maintain the current supply air temperature setpoint, maintain the current fan speed, and maintain the current electronic expansion valve opening. When an energy-saving and supply-reduction demand signal is generated, feedforward control commands are generated to increase the supply air temperature setpoint, reduce the fan speed, and decrease the opening of the electronic expansion valve. When a steady-state hold-up demand signal is generated, a feedforward control command is generated to keep the current control parameters unchanged.

[0016] Compared with the prior art, the advantages of this invention are: This invention introduces a dynamic prediction mechanism for computing load and constructs a multivariate time-series prediction model. It integrates and predicts the current computing load, ambient temperature, and air conditioning heat source temperature, outputting a predicted computing load sequence and a predicted air conditioning heat source temperature sequence within a preset future time window. The predicted computing load is converted into a predicted value of the heat generation power of IT equipment through a thermodynamic equivalent model, and the predicted air conditioning heat source temperature is combined with the future temperature prediction value to convert into a predicted value of the air conditioning cooling power, forming a predicted heat load map that includes a dynamic balance mapping relationship between supply and demand. Based on the map, the power supply and demand deviation is calculated, and a comprehensive deviation analysis is performed by combining the deviation between the current actual temperature and the target temperature. Feedforward control commands are generated, which fundamentally eliminates the inherent thermal lag of traditional feedback control, realizes precise synchronization between cooling supply and heating demand, avoids over-cooling and ineffective cooling caused by response lag, and significantly reduces the operating energy consumption of the air conditioning system.

[0017] This invention also employs a partitioned independent processing architecture, configuring independent prediction models and control logic for each thermal management target partition. This enables precise identification of load characteristic differences between partitions. In the predicted thermal load map, a corresponding feedforward control strategy is matched based on the supply-demand deviation and the current temperature deviation, allowing the cooling output of the air conditioning system to dynamically match the actual heat demand. This avoids fixed adjustments and redundant cooling in traditional control. At the same time, because the temperature change trend is effectively suppressed, the server operating environment becomes more stable, reducing thermal shock caused by temperature fluctuations and extending the service life of IT equipment and air conditioning equipment. Attached Figure Description

[0018] Figure 1 This is a system module block diagram of the present invention; Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] 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 embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] Example 1: This invention discloses a data center zone air conditioning temperature control and energy-saving system based on dynamic prediction of computing load. Please refer to [link to relevant documentation]. Figures 1-2 It includes a thermal management zone data acquisition module, a load and environment coupling prediction module, a thermal load mapping module, and a zone temperature control module.

[0021] The thermal management partition data acquisition module is deployed in each thermal management partition of the data center. It marks the thermal management partition as the target partition and is used to collect real-time computing load data, real-time ambient temperature data and air conditioning heat source temperature data of the IT equipment associated with each target partition at the same timestamp. The computing load data includes at least one of GPU utilization, memory usage, and disk I / O throughput; the air conditioning heat source temperature data includes at least one of air conditioning supply air temperature, return air temperature, chilled water supply temperature, and chilled water return temperature; and the real-time ambient temperature data includes at least one of rack air inlet temperature, cold aisle temperature, and hot aisle temperature. An auxiliary ambient temperature sensor is deployed in or near each thermal management zone. The auxiliary ambient temperature sensor is deployed upstream of the air conditioning supply air path of the zone or at the boundary between the zone and the non-air-conditioned area to collect real-time ambient temperature data. The data collection cycle for each target partition is the same, and the collection time of various types of data within the same partition is kept synchronized to ensure that the time series data has inherent physical consistency. Through multi-source data collected independently and synchronously by partition, high-quality training samples are provided for subsequent prediction models, ensuring the integrity and time series alignment of the data in each partition, and laying the foundation for accurate prediction.

[0022] The load and environment coupling prediction module has a built-in multivariate time series prediction model, which is used to output the predicted computing load sequence and the predicted air conditioning heat source temperature sequence of the target partition in the future preset time window based on the computing load data, real-time ambient temperature data and air conditioning heat source temperature data of the current operating status of each target partition. In the partitioned load and environment coupling prediction module, each thermal management partition is independently configured with a multivariate time series prediction model instance, or configured with a shared model but with the partition identifier as the input feature. The multivariate time series prediction model is a long short-term memory network (LSTM) or gated recurrent unit (GRU) with an encoder-decoder architecture to handle the temporal dependency between multivariate inputs and multi-step outputs. The encoder is used to extract feature representations of historical time series data, and the decoder is used to generate future multi-step predictions. Its input data includes computing load sequences, ambient temperature sequences, and air conditioning heat source temperature sequences within the historical time window of each target partition. Its output data includes predicted computing load sequences and predicted air conditioning heat source temperature sequences within a future preset time window for each target partition. The load and environment coupling prediction module is also equipped with an online model update unit, which is used to calculate the load prediction error based on the predicted computing power load sequence and the historical computing power load sequence, and to calculate the heat source temperature prediction error based on the predicted air conditioning heat source temperature sequence and the historical air conditioning heat source temperature sequence. When the prediction error exceeds a preset threshold within a preset number of consecutive time steps, the model is triggered to update. An incremental learning method is adopted to fine-tune the model using new data within the most recent preset time window. This module uses a multivariate time-series forecasting model to fuse and predict computing load, ambient temperature, and air conditioning heat source temperature, outputting fundamental data for both future supply and demand sides, providing a time margin for feedforward control. Independent model instances and update strategies ensure that each partition can adapt to changes in load patterns, avoiding mutual interference.

[0023] The heat load mapping module is used to receive the predicted computing power load sequence and the predicted air conditioning heat source temperature sequence, and simultaneously obtain the predicted value of future ambient temperature. Based on the thermodynamic equivalent model, it performs heat load mapping and outputs the predicted heat load map of the target partition within a preset time window in the future. The specific process of obtaining the predicted heat load map includes: Establish thermodynamic equivalent models for each target zone. The thermodynamic equivalent models are used to describe the dynamic balance mapping relationship between the heat generation power of IT equipment and the cooling power of air conditioning terminals in the zone, as well as the temperature change law of the zone over time. Based on the thermodynamic equivalent model, the received predicted computing load sequence is converted into a predicted value sequence of IT equipment heating power in the target partition within a future preset time window through the pre-stored heating power mapping relationship. The heating power mapping relationship is pre-calibrated according to the model, operating frequency and voltage parameters of the IT equipment, and characterizes the correspondence between computing load and heating power. The predicted future ambient temperature of the zone is obtained. The predicted future ambient temperature is used to characterize the impact of outdoor meteorological conditions and heat diffusion from adjacent zones on the thermal environment of the zone. The predicted air conditioning heat source temperature sequence and the predicted future ambient temperature are converted into the predicted air conditioning cooling power sequence of the target zone within a preset time window through a pre-stored cooling power mapping relationship. The predicted future ambient temperature is obtained as follows: the predicted outdoor ambient temperature is obtained, and combined with the preset heat transfer coefficient of the building envelope corresponding to the target zone, the outdoor influence component is obtained; based on the predicted air conditioning heat source temperature of adjacent zones and the pre-stored airflow organization model, the influence component of heat diffusion of adjacent zones on this zone is calculated. The outdoor influence component is superimposed with the thermal diffusion influence component of the adjacent zone, and combined with the historical ambient temperature trend of the target zone, the predicted future ambient temperature value of the zone is obtained. The predicted heat load map for this region is generated by associating the predicted heat power sequence of IT equipment with the predicted cooling power sequence of air conditioning. The thermodynamic equivalent model is either a heat balance model based on physical mechanisms or a heat response model based on data-driven principles. The dynamic equilibrium mapping relationship includes power mapping relationship and cooling power mapping relationship. Thermodynamic equivalent model update mechanism: Utilizing the error between the actual ambient temperature collected by the temperature sensor in the zone and the theoretical temperature calculated by the thermodynamic equivalent model, the thermal resistance parameter, heat capacity parameter and heat transfer efficiency parameter in the thermodynamic equivalent model are periodically or triggered online to compensate for the impact of equipment aging and environmental changes on the model accuracy. This module converts predicted data into physical power using a thermodynamic equivalent model, forming a predicted heat load map that includes the dynamic balance between supply and demand, providing a basis for feedforward control decisions. An online correction mechanism ensures the model continuously adapts to changes in the physical environment.

[0024] The zone temperature control module is used to analyze the temperature compensation requirements based on the predicted heat load map and the current real-time ambient temperature of the target zone, generate temperature compensation requirement signals for each zone's air conditioner, generate feedforward control commands for the air conditioner terminal of the zone based on each signal, and send the feedforward control commands to the corresponding zone's air conditioner terminal actuator before the zone temperature actually changes. The specific process of analyzing temperature control compensation requirements based on the predicted heat load map and the current real-time ambient temperature of the target zone includes: Extract the predicted power generation sequence of IT equipment and the predicted power generation sequence of air conditioning from the predicted heat load map of the region, and calculate the power supply and demand deviation sequence of each time step within the preset time window. This deviation sequence represents the cooling gap at each time step in the future (positive value indicates insufficient cooling, negative value indicates excessive cooling). Obtain the current real-time ambient temperature of the target partition and the preset target temperature, and calculate the current temperature deviation. This deviation represents the degree of actual temperature deviation at the current moment (positive value indicates current overheating, negative value indicates current undercooling). Based on the trend characteristics of the power supply and demand deviation sequence and the magnitude of the current temperature deviation, the temperature control compensation demand is analyzed according to the preset threshold comparison rules, and the temperature control compensation demand signal of the target zone in the future preset time window is generated. The preset threshold comparison rules include: The power supply and demand deviation is compared with the preset first threshold and second threshold. The current temperature deviation is compared with the preset first threshold and second threshold. If the first threshold is greater than the second threshold, the temperature control compensation demand signal for this zone is generated based on the comparison result. When the power supply and demand deviation exceeds the first threshold and the current temperature deviation exceeds the first threshold, it is determined as an "emergency compensation demand" and an emergency strong cooling demand signal is generated. This indicates that the power supply and demand deviation sequence shows a continuous positive trend, indicating insufficient cooling in the future, while the current ambient temperature is too hot. The generated emergency strong cooling demand signal includes: compensation start time (immediate), compensation intensity (high intensity, calculated by weighting the peak value of the power supply and demand deviation and the magnitude of the current temperature deviation), and compensation duration (covering the time window when the power supply and demand deviation is continuously positive). When the power supply and demand deviation is greater than the first threshold and the current temperature deviation is less than the second threshold, it is determined as "preventive compensation demand" and a forward-looking pre-cooling demand signal is generated, indicating that the cooling capacity needs to be increased in advance to prevent future overheating. The generated forward-looking pre-cooling demand signal includes: compensation start time (the advance amount before the power supply and demand deviation is expected to turn positive), compensation intensity (medium to low intensity, mainly calculated based on the predicted value of the power supply and demand deviation), and compensation duration (covering the time window when the power supply and demand deviation is positive). When the power supply and demand deviation is less than the second threshold and the current temperature deviation is greater than the first threshold, it is determined as "reduced supply demand" and a transient maintenance demand signal is generated, indicating that the cooling capacity needs to be reduced to avoid overcooling. The generated transient maintenance demand signal includes: the start time of reduced supply (immediately or earlier according to the deviation trend), the intensity of reduced supply (calculated by weighting the negative value of the power supply and demand deviation and the magnitude of the current temperature deviation), and the duration of reduced supply. When the power supply and demand deviation is less than the second threshold and the current temperature deviation is less than the second threshold, an energy-saving and supply reduction demand signal is generated, indicating that it is determined to be "maintain or fine-tune demand". If there is a cooling surplus but it is not cold at present, no adjustment can be made or only a small adjustment can be made. The generated energy-saving and supply reduction demand signal includes: maintaining the current cooling state, or only making a fine adjustment with an amplitude not exceeding the preset dead zone. When the power supply and demand deviation is between the first threshold and the second threshold and the current temperature deviation is between the first threshold and the second threshold, it is determined that there is "no adjustment requirement". A steady-state maintenance requirement signal is generated, indicating that the current cooling state is maintained normally and no new feedforward control command is generated. Feedforward control commands specifically include: When an emergency strong cooling demand signal is generated, feedforward control commands are generated to significantly reduce the supply air temperature setpoint, significantly increase the fan speed, and significantly increase the electronic expansion valve opening. When a forward-looking pre-cooling demand signal is generated, feedforward control commands are generated to slightly reduce the supply air temperature setpoint, slightly increase the fan speed, and slightly increase the opening of the electronic expansion valve. When a transient maintenance demand signal is generated, feedforward control commands are generated to maintain the current supply air temperature setpoint, maintain the current fan speed, and maintain the current electronic expansion valve opening. When an energy-saving and supply-reduction demand signal is generated, feedforward control commands are generated to increase the supply air temperature setpoint, reduce the fan speed, and decrease the opening of the electronic expansion valve. When a steady-state hold-up demand signal is generated, a feedforward control command is generated to keep the current control parameters unchanged.

[0025] One or more of the aforementioned feedforward control commands are sent to the corresponding air conditioning terminal actuators in the zone before the actual temperature change. Through comprehensive analysis of the power supply and demand deviation and the current temperature deviation, a scenario-based temperature control compensation demand signal is generated, and feedforward control commands are generated accordingly. The adjustment is completed before the actual temperature change, completely eliminating thermal hysteresis and achieving precise matching between cooling and load.

[0026] It should be added that the article involves comparisons of various thresholds. Thresholds, preset values, preset ranges, etc., are set for result comparison and analysis to determine good or bad. The magnitude of these thresholds is determined by a combination of large-scale model analysis of sample data and human experience. They can also be appropriately adjusted based on seasonal or common-sense influences.

[0027] In summary, this system includes: a partitioned data acquisition module, which collects real-time computing load data, ambient temperature and air conditioning heat source temperature of IT equipment by partition; and a partitioned load and environment coupling prediction module, which has a built-in multivariate time series prediction model and outputs the predicted computing load sequence and predicted air conditioning heat source temperature sequence within a future preset time window based on the current collected data of each target partition. The partitioned heat load mapping module converts the predicted computing power load sequence into a predicted value sequence of IT equipment heat generation power based on a thermodynamic equivalent model, and converts the predicted air conditioning heat source temperature sequence into a predicted value sequence of air conditioning cooling power. The associated output forms a predicted heat load map containing both supply and demand sides. The zone air conditioning control module calculates the power supply and demand deviation based on the predicted heating power and cooling power in the predicted heat load map, and generates different temperature control compensation demand signals based on the current temperature deviation. Based on the converted feedforward control command, it sends the signal to the corresponding air conditioning terminal actuator before the actual temperature change in the zone. This invention constructs a thermal load map on both the supply and demand sides by dynamically predicting computing load and fusing multi-source data. It analyzes the temperature control compensation requirements based on the future supply and demand deviation and the current temperature deviation, and provides various feedforward control strategies based on the analysis results. This eliminates the thermal lag of traditional feedback control, avoids overcooling and ineffective cooling caused by response lag, achieves precise matching between cooling and load, significantly reduces the operating energy consumption of the air conditioning system, and improves the operational stability of the data center.

[0028] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto; any equivalent substitutions or modifications made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and its improved concept, should be covered within the scope of protection of the present invention.

Claims

1. A data center zoned air conditioning temperature control and energy-saving system based on dynamic prediction of computing load, characterized in that: include: The thermal management partition data acquisition module is deployed in each target partition of the data center to collect real-time computing load data, real-time ambient temperature data and real-time air conditioning heat source temperature data of the IT equipment in each target partition at the same timestamp. The load and environment coupling prediction module has a built-in multivariate time series prediction model, which is used to output the predicted computing load sequence and the predicted air conditioning heat source temperature sequence of the target partition in the future preset time window based on the computing load data, real-time ambient temperature data and air conditioning heat source temperature data of the current operating status of each target partition. The heat load mapping module is used to receive the predicted computing power load sequence and the predicted air conditioning heat source temperature sequence, and simultaneously obtain the predicted value of future ambient temperature. Based on the thermodynamic equivalent model, it performs heat load mapping and outputs the predicted heat load map of the target partition within a preset time window in the future. The process of obtaining the predicted heat load map includes: Establish a thermodynamic equivalent model for each target zone. The thermodynamic equivalent model is used to describe the dynamic balance mapping relationship between the heat generation power of IT equipment and the cooling power of air conditioning terminals in the zone, as well as the temperature change law of the zone over time. The mapping relationship based on the thermodynamic equivalent model is used to convert the predicted computing power load sequence into a sequence of predicted IT equipment heat generation power values ​​for the target partition within a future preset time window. Obtain the predicted future ambient temperature of the zone, and convert the predicted air conditioning heat source temperature sequence and the predicted future ambient temperature into a predicted air conditioning cooling power sequence for the target zone within a preset time window based on the mapping relationship of the thermodynamic equivalent model. The predicted heat load map for this region is generated by associating the predicted heat power sequence of IT equipment with the predicted cooling power sequence of air conditioning. The thermodynamic equivalent model is either a heat balance model based on physical mechanisms or a heat response model based on data-driven principles. The dynamic equilibrium mapping relationship includes power mapping relationship and cooling power mapping relationship. The zone temperature control module is used to analyze the temperature compensation requirements based on the predicted heat load map and the current real-time ambient temperature of the target zone, generate temperature compensation requirement signals for each zone's air conditioner, and generate feedforward control commands for the air conditioner terminal based on each signal and send them to the air conditioner terminal actuator. The specific process of analyzing temperature control compensation requirements includes: Extract the predicted power generation sequence of IT equipment and the predicted power generation sequence of air conditioning from the predicted heat load map of the partition, and calculate the power supply and demand deviation sequence at each time step within the future preset time window; Obtain the real-time ambient temperature of the target zone and the preset target temperature, and calculate the current temperature deviation; Based on the trend characteristics of the power supply and demand deviation sequence and the magnitude of the current temperature deviation, the temperature control compensation demand is analyzed according to the preset threshold comparison rule, and a temperature control compensation demand signal for the target partition within a preset time window is generated.

2. The data center zone air conditioning temperature control and energy-saving system based on dynamic prediction of computing load according to claim 1, characterized in that: The computing load data includes at least one of GPU utilization, memory usage, and disk I / O throughput; the air conditioning heat source temperature data includes at least one of air conditioning supply air temperature, return air temperature, chilled water supply temperature, and chilled water return temperature; the real-time ambient temperature data includes at least one of rack air inlet temperature, cold aisle temperature, and hot aisle temperature.

3. The data center zone air conditioning temperature control and energy-saving system based on dynamic prediction of computing load according to claim 2, characterized in that: The multivariate time series prediction model is a long short-term memory network (LSTM) or a gated recurrent unit (GRU) with an encoder-decoder architecture. Its input data includes: computing load sequence, ambient temperature sequence, and air conditioning heat source temperature sequence within the historical time window of each target partition. Its output data includes: predicted computing load sequence and predicted air conditioning heat source temperature sequence within the future preset time window of each target partition.

4. The data center zone air conditioning temperature control and energy-saving system based on dynamic prediction of computing load according to claim 1, characterized in that: The preset threshold comparison rules include: comparing the power supply and demand deviation with preset first threshold and second threshold, comparing the current temperature deviation with preset first threshold and second threshold, and generating a temperature control compensation demand signal for the partition based on the comparison results; When the power supply and demand deviation is greater than the first threshold and the current temperature deviation is greater than the first threshold, it is determined as an "emergency compensation demand" and an emergency strong cooling demand signal is generated. When the power supply and demand deviation is greater than the first threshold and the current temperature deviation is less than the second threshold, it is determined to be a "preventive compensation demand" and a forward-looking pre-cooling demand signal is generated. When the power deviation between supply and demand is less than the second threshold and the current temperature deviation is greater than the first threshold, it is determined as "reduced supply and demand", and a transient demand maintenance signal is generated. When the power supply and demand deviation is less than the second threshold and the current temperature deviation is less than the second threshold, it is determined to be "maintain or fine-tune demand", and an energy-saving and supply reduction demand signal is generated. When the power supply and demand deviation is between the first threshold and the second threshold, and the current temperature deviation is between the first threshold and the second threshold, it is determined that there is "no adjustment demand" and a steady-state maintenance demand signal is generated.

5. The data center zone air conditioning temperature control and energy-saving system based on dynamic prediction of computing load according to claim 4, characterized in that: Feedforward control commands specifically include: When an emergency strong cooling demand signal is generated, feedforward control commands are generated to significantly reduce the supply air temperature setpoint, significantly increase the fan speed, and significantly increase the electronic expansion valve opening. When a forward-looking pre-cooling demand signal is generated, feedforward control commands are generated to slightly reduce the supply air temperature setpoint, slightly increase the fan speed, and slightly increase the opening of the electronic expansion valve. When a transient maintenance demand signal is generated, feedforward control commands are generated to maintain the current supply air temperature setpoint, maintain the current fan speed, and maintain the current electronic expansion valve opening. When an energy-saving and supply-reduction demand signal is generated, feedforward control commands are generated to increase the supply air temperature setpoint, reduce the fan speed, and decrease the opening of the electronic expansion valve. When a steady-state hold-up demand signal is generated, a feedforward control command is generated to keep the current control parameters unchanged.

Images