Data center building heat dissipation optimization method, system and equipment based on AI intelligent control and computer readable storage medium

Abstract

The invention discloses a data center building heat dissipation optimization method, system and device based on AI intelligent control and a computer readable storage medium, and belongs to the technical field of data center heat management and intelligent control, and the method comprises the following steps: S1, data acquisition and thermal characteristic modeling; s2, dynamic thermal resistance regulation and control; s3, intelligent adjustment of the three-dimensional air duct; s4, cooperative control of the mixed cooling system; s5, thermal load prediction and adaptive adjustment; and S6, waste heat recovery intelligent management and control. According to the data center building heat dissipation optimization method, system and equipment based on AI intelligent control and the computer readable storage medium, dynamic self-adaptive regulation and control of the whole heat dissipation process are achieved, linkage control of an enclosure structure, airflow organization and a cooling system is optimized, heat load changes are accurately pre-judged and adjusted in advance, and the heat dissipation efficiency is improved. And waste heat is intelligently recycled and utilized in a gradient mode.

Description

Technical Field

[0001] This invention belongs to the field of data center thermal management and intelligent control technology, specifically relating to a data center building heat dissipation optimization method, system, equipment, and computer-readable storage medium based on AI intelligent control. Background Technology

[0002] With the rapid development of the digital economy, the computing power density of data centers continues to increase, and the heat flux density of server racks has reached 200~300W / ㎡. The energy consumption and heat dissipation of data centers have become the core factors restricting their green development. As a key indicator for measuring the green and low-carbon level of data centers, the need to reduce the PUE (Power Usage Effectiveness) value is becoming increasingly urgent.

[0003] Current data center building cooling methods generally adopt a combination of "fixed building insulation and mechanical cooling as the main methods". This method has the following technical problems in practical engineering applications and cannot adapt to the dynamic cooling requirements of high-density data centers: The heat dissipation control adopts a responsive adjustment method, which lacks the ability to predict changes in heat load. The control response is lagging, which can easily cause excessive output of cooling capacity or overheating of local hot spots, resulting in low heat dissipation efficiency and a high proportion of ineffective energy consumption. The building envelope, air duct system, and cooling system operate independently, with no data exchange or linkage control between the systems. This results in poor coordination, making it impossible to achieve adaptive control of the entire heat dissipation process and difficult to match the dynamic changes in heat load. The switching of cooling modes relies on a fixed threshold set manually, which lacks flexibility and cannot dynamically adjust the cooling capacity distribution ratio according to real-time changes in outdoor ambient temperature and indoor heat load, further increasing the energy consumption of mechanical refrigeration. The waste heat recovery unit is disconnected from the heat dissipation system, there is no targeted waste heat cascade utilization strategy, the waste heat recovery utilization rate is low, and it cannot meet the development needs of comprehensive energy utilization in data centers. The accuracy of basic data acquisition for heat dissipation control is insufficient, and there is a lack of accurate three-dimensional thermal field model support, making it impossible to accurately locate hot spots and weak points in heat dissipation. This results in control commands lacking specificity and poor local heat dissipation effect.

[0004] Therefore, a new method is urgently needed. Summary of the Invention

[0005] The purpose of this invention is to provide a data center building heat dissipation optimization method, system, equipment, and computer-readable storage medium based on AI intelligent control. This method achieves dynamic adaptive regulation of the entire heat dissipation process, optimizes the linkage control of the building envelope, airflow organization, and cooling system, accurately predicts and adjusts heat load changes in advance, and intelligently recovers and utilizes waste heat in stages.

[0006] To achieve the above objectives, the present invention provides a method, system, device, and computer-readable storage medium for optimizing heat dissipation in data center buildings based on AI intelligent control, comprising the following steps: S1. Real-time data collection of structural temperature, indoor and outdoor temperature and humidity, cabinet intake and return air temperature, air duct pressure, cooling system energy consumption, PUE real-time value, GPU core temperature, server power consumption, and computing load rate are obtained through a distributed sensor array. Combined with CFD fluid simulation and infrared thermal imaging detection, a three-level three-dimensional thermal field model of building-computing zone-cabinet is constructed to locate hot spots and weak heat dissipation nodes. S2. Based on the real-time data and thermal field model collected by S1, the AI ​​control module outputs control commands to control the thermal resistance switching of the building's composite envelope structure and the graphene thermal resistance control plate of the computing power partition wall; the thermal conductivity is adaptively adjusted according to the changes in outdoor ambient temperature and indoor heat load to achieve bidirectional thermal resistance control of the building and computing power partition. S3. By analyzing airflow distribution data in real time through the AI ​​control module, the opening degree of the damper in the horizontal closed cold / hot channel and the speed of the negative pressure exhaust device in the vertical through-type exhaust shaft are adjusted to construct a vertical-horizontal integrated three-dimensional cold and hot airflow separation organization, and to achieve coordinated control of air-cooled and liquid-cooled airflow in combination with liquid cooling heat dissipation requirements. Based on the heat load prediction results, the S4 AI control module dynamically switches the operation mode of the cooling system, which consists of a natural cooling module, a mechanical refrigeration module, and a dual liquid cooling module consisting of a cold plate and a spray system. It sets the mode switching threshold and the cooling capacity allocation ratio according to the heat load characteristics of the computing center, so as to realize the on-demand allocation of cooling capacity, mainly liquid cooling. The S5 and AI control modules adopt the LSTM time-series prediction algorithm. Based on historical operating data, real-time acquisition parameters of S1 and computing power task scheduling plan, GPU cluster load rate, and server start-stop records, they predict the trend of heat load change, start the cooling system pre-adjustment in advance, and adapt to adjust liquid cooling flow, air duct air volume and thermal resistance of the building envelope. The S6 and AI control modules monitor waste heat emission parameters in real time. Through a waste heat recovery system consisting of a plate heat exchanger, a waste heat storage tank, a low-temperature waste heat generator, and a waste heat direct supply air conditioning module, the waste heat is utilized in stages. The waste heat recovery control parameters and utilization rate data are fed back to S1 and S4 to optimize the three-dimensional thermal field model and the cooling mode switching logic, thus completing the closed-loop control of the entire heat dissipation optimization process.

[0007] Preferably, in S1, the acquisition error of the distributed sensor array The formula for calculating the acquisition error is: ; In the formula, For real-time sensor data acquisition values, This is the standard calibration value; The collected data is synchronously transmitted to the AI ​​control module via industrial Ethernet, and simultaneously stored on a local server for historical data retrieval and 3D thermal field model optimization. CFD fluid simulation boundary conditions are set to outdoor ambient temperature -15℃ to 42℃ and indoor cabinet heat flux density. The infrared thermal imaging detection accuracy is ±2℃.

[0008] Preferably, in S2, the building's composite envelope structure consists of an outer layer of color steel plate, a middle graphene composite thermal resistance regulating layer, and an inner insulation board; the formula for calculating the real-time thermal conductivity of the envelope structure is: ; In the formula, For real-time thermal conductivity, The maximum thermal conductivity in the low-resistivity state is taken as 0.1 W / (m·K). The minimum thermal conductivity in the high-resistivity state is taken as 0.03 W / (m·K). , The temperature threshold is controlled by thermal resistance. , ; This refers to the real-time outdoor ambient temperature. The thermal resistance switching control rule is as follows: When the outdoor temperature is ≥28℃, the building envelope is switched to a high-resistance state, and the thermal conductivity is adjusted to 0.025~0.03W / (m・K); When the outdoor temperature is ≤10℃, the building envelope is switched to a low-resistance state, and the thermal conductivity is adjusted to 0.1~0.12W / (m・K); When the outdoor temperature is 10℃~28℃, the thermal conductivity can be continuously adjusted from 0.03~0.1W / (m・K) by adjusting the power supply voltage of the graphene composite thermal resistance control layer from 0~24V. When the heat load of any computing power partition increases suddenly, the thermal resistance of the graphene thermal resistance control plate of the partition wall of that partition is increased separately.

[0009] Preferably, in S3, the damper is an electric damper with an adjustment range of 0~90°; the negative pressure exhaust device is a centrifugal fan with a speed range of 800~1800 r / min; The intelligent air duct control rules are as follows: When the rack return air temperature is ≥35℃ or the GPU core temperature is ≥80℃, the opening of the cold aisle damper in the corresponding area is increased by 20%, and the centrifugal fan speed is increased from 1200r / min to 1600r / min, until the return air temperature drops below 30℃ and the GPU core temperature is ≤80℃. When the air pressure deviation in the air duct exceeds ±5%, the damper opening is automatically adjusted, and the control response time is ≤5s. When the liquid cooling auxiliary heat dissipation of a zone is activated, the damper opening of the cold aisle of that zone is increased by 10%, and the centrifugal fan speed is increased by 100r / min.

[0010] Preferably, in S4, the rated heat load of the artificial intelligence computing center is 960kW, and the operation mode switching and cooling capacity distribution rules of the hybrid cooling system are as follows: When the outdoor temperature is ≤12℃ and the indoor heat load is ≤70% of the rated load, the pure natural cooling and liquid cooling micro-operation mode will be activated, with natural cooling handling 80% of the cooling capacity and liquid cooling module handling 20% ​​of the cooling capacity. When the outdoor temperature is 12℃~25℃ or the indoor heat load is 70%~85% of the rated load, the natural cooling, mechanical refrigeration and liquid cooling combined mode will be activated. Natural cooling will handle 40% of the cooling capacity, mechanical refrigeration will handle 20% of the cooling capacity and liquid cooling module will handle 40% of the cooling capacity. When the outdoor temperature is >25℃ or the indoor heat load is >85% of the rated load, the liquid cooling as the main mode and the mechanical refrigeration as the auxiliary mode will be activated, with the liquid cooling module undertaking 70% of the cooling capacity and the mechanical refrigeration undertaking 30% of the cooling capacity. The hybrid cooling system with 6 computing power zones enables independent mode switching and independent allocation of cooling capacity, with the liquid cooling module prioritizing the supply of cooling capacity to the GPU cluster area.

[0011] Preferably, in S5, the LSTM time series prediction model is trained based on the TensorFlow framework, with 1000 training batches and a learning rate of 0.001. Heat load prediction error The formula for calculating the prediction error is: ; In the formula, The heat load prediction value output by the LSTM time series prediction model. The measured value of heat load is obtained from real-time monitoring by a distributed sensor array. The adaptive pre-adjustment rule is: When the predicted heat load increase is ≥10%, increase the liquid cooling module liquid supply flow rate by 20%, adjust the air duct air volume by 30%, and adapt and adjust the thermal resistance of the corresponding partition enclosure structure and partition wall to ensure that the cabinet air intake temperature is stable at 18~27℃ and the GPU core temperature is ≤85℃. When the predicted heat load decreases by ≥10%, the liquid cooling flow rate, airflow in the duct, and mechanical refrigeration load should be reduced simultaneously.

[0012] Preferably, in S6, the waste heat recovery system consists of a plate heat exchanger, a waste heat storage tank, and a low-temperature waste heat generator, achieving a waste heat recovery utilization rate of [missing information]. The formula for calculating the waste heat recovery rate is: ; In the formula, The actual total amount of waste heat recovered. This is the total amount of waste heat discharged by the cooling system; The waste heat cascade utilization control rules are as follows: When the outdoor temperature is ≤5℃, the waste heat will be used first for heating the computing center park to maintain the heating temperature at 20~22℃. The remaining waste heat will be used to heat domestic hot water to 40~50℃. When the outdoor temperature is between 5℃ and 25℃, the waste heat is used first to heat domestic hot water to 40℃ to 50℃. The remaining waste heat is used to drive a low-temperature waste heat generator to supply electricity and to cool the maintenance area through a waste heat direct supply air conditioner. When the outdoor temperature is >25℃, the waste heat should be prioritized for low-temperature waste heat power generation and domestic hot water preparation. The control parameters and utilization rate data of waste heat recovery are fed back to S1 and S4.

[0013] This invention also provides a system for optimizing heat dissipation in data center buildings based on AI intelligent control, comprising: The data acquisition and thermal characteristic modeling module is used to execute step S1, complete multi-dimensional data acquisition including GPU core temperature, server power consumption, and computing load rate, construct a three-level refined three-dimensional thermal field model of building-computing zone-rack and locate hot spots and weak heat dissipation nodes, and communicate with the AI ​​control module. The dynamic thermal resistance control module is communicatively connected to the data acquisition and thermal characteristic modeling module and the AI ​​control module. It is used to execute step S2, receive real-time data and three-dimensional thermal field model data, and realize adaptive control of thermal resistance of the building composite envelope structure and the graphene thermal resistance control plate of the computing power partition wall. The three-dimensional air duct intelligent adjustment module is communicatively connected to the data acquisition and thermal characteristic modeling module and the AI ​​control module. It is used to execute step S3, receive airflow distribution data, and adjust the damper opening and centrifugal fan speed to achieve coordinated control of air-cooled and liquid-cooled airflow. The heat load prediction and adaptive adjustment module is communicatively connected to the data acquisition and thermal characteristic modeling module and the AI ​​control module. It is used to execute step S5, integrate relevant data of computing power tasks to complete the prediction of heat load trends, and output the advance adjustment instructions of each control module. The hybrid cooling system collaborative control module is communicatively connected to the heat load prediction and adaptive adjustment module and the AI ​​control module. It is used to execute step S4, receive the heat load prediction results, and realize dynamic switching of the liquid cooling mode and on-demand allocation of cooling capacity. The waste heat recovery intelligent management and control module is communicatively connected to the hybrid cooling system collaborative control module and the AI ​​control module. It is used to execute step S6 to realize the closed loop of the entire process of waste heat cascade utilization and heat dissipation optimization, and to feed back the waste heat recovery data to the data acquisition and thermal characteristic modeling module and the hybrid cooling system collaborative control module.

[0014] Therefore, the present invention employs the above-mentioned AI-based intelligent control-based data center building heat dissipation optimization method, system, equipment, and computer-readable storage medium. Compared with the prior art, the technical solution of the present invention has the following beneficial effects: (1) This invention uses the LSTM time-series prediction algorithm to predict the heat load change trend in the next 1 to 2 hours. The heat load prediction error is ≤5%. Based on the prediction results, the cooling system is pre-adjusted 20 minutes in advance, which solves the lag problem of existing responsive heat dissipation control. It can ensure the stability of the cabinet air intake temperature, effectively avoid the waste of cooling capacity and local hot spots overheating, greatly reduce ineffective energy consumption, and improve heat dissipation efficiency. (2) The present invention takes the AI ​​control module as the core and constructs a full-process linkage control system of "data acquisition - thermal resistance regulation - air duct adjustment - load prediction - cooling control - waste heat recovery". The modules realize real-time data interaction through industrial Ethernet, which solves the problem of independent operation and poor coordination of existing heat dissipation systems, constructs a full-process data closed loop, and improves the accuracy and timeliness of heat dissipation control. (3) This invention achieves high-precision acquisition of multi-dimensional data through a distributed sensor array with an acquisition error of ≤2%. It combines CFD fluid simulation (boundary conditions -10℃~40℃) and infrared thermal imaging detection (accuracy ±2℃) to construct a three-dimensional thermal field model of the building, which can accurately locate hot spots and weak heat dissipation nodes, providing accurate data support for intelligent control. Furthermore, the model is continuously optimized through data feedback to further improve the control accuracy. (4) The present invention regulates the power supply voltage of the graphene composite thermal resistance regulating layer through the AI ​​control module, so as to achieve the continuous adjustment of the thermal conductivity of the composite envelope structure from 0.025 to 0.12 W / (m·K). Combined with the physical isolation of hot and cold airflow in the three-dimensional air duct, it solves the problem of low heat dissipation efficiency caused by the existing fixed heat insulation method and chaotic airflow mixing. It can reduce outdoor heat penetration and use natural low temperature to assist heat dissipation, thereby improving the basic heat dissipation capacity from the perspective of building structure and airflow organization.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an embodiment of the data center building heat dissipation optimization method, system, device, and computer-readable storage medium based on AI intelligent control according to the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used in the present invention should have the ordinary meaning understood by those skilled in the art.

[0018] Example 1 like Figure 1 As shown, this embodiment provides a data center building heat dissipation optimization method, system, device, and computer-readable storage medium based on AI intelligent control. It should be understood that the specific parameters, models, and protocols mentioned in this embodiment are merely examples to help those skilled in the art understand the present invention, and are not intended to limit the present invention.

[0019] The data center used in this embodiment is a single-story, large-span steel structure building with a building area of ​​[missing information]. The facility is divided into six computing power zones (100 server racks per zone). The core computing power equipment consists of GPU cluster servers, with a single server having a thermal power consumption of 8kW. The building is designed as a sealed server room, with an applicable outdoor ambient temperature range of -15℃ to 42℃. The rated thermal load of the computing center is 960kW. The specific implementation steps are as follows: S1. Deploy a distributed sensor array within the data center building. The sensors are arranged according to the principle of "densification at core hotspots and standardization in ordinary areas," specifically: Each AI computing power rack is equipped with one DS18B20 temperature sensor at the air inlet and outlet, one surface-mount temperature sensor is embedded in the GPU core, and one power consumption sensor is installed in each rack, with the sampling frequency increased to 2 times / minute. One PT100 temperature sensor is installed for every 20㎡ of the building envelope, and two differential pressure / flow sensors are installed at the air duct inlet / outlet and the liquid cooling supply / return water inlet. Indoor and outdoor temperature and humidity, cooling system energy consumption, and real-time PUE data collection shall be performed in accordance with the original method and standard. Real-time data collection of structural temperature, indoor and outdoor temperature and humidity (accuracy ±0.5℃, ±2%RH), inlet and outlet air temperature, air duct pressure (range 0-1000Pa), cooling system energy consumption, and real-time PUE value; The formula for calculating data acquisition accuracy is as follows: ; In the formula, To account for data acquisition error, For real-time sensor data acquisition values, It is the standard calibration value and meets the requirements. To ensure the accuracy of the collected data; Using ANSYS Fluent CFD fluid simulation software, the building drawings of the data center and the initial data collected by the distributed sensor array were imported. The simulation boundary conditions were set as outdoor ambient temperature -15℃~42℃ and indoor cabinet heat flux density 300-400W / ㎡, and steady-state thermal simulation was carried out. By integrating real-time sensor data, CFD simulation results, and infrared thermal imaging detection data, a three-dimensional thermal field model of the computing center building, computing zones, and server racks is constructed. The model can display the temperature distribution of a single GPU, a single server rack, and the entire zone in real time, and accurately locate hot spots (GPU cluster area) and weak heat dissipation nodes (the end of the server rack row, the corner of the building air duct). The output real-time acquisition parameters, three-dimensional thermal field model, hot spot area and weak heat dissipation node data serve as the basic data support for intelligent control in each step, and are also stored on the local server for subsequent historical data retrieval and three-dimensional thermal field model optimization. S2, the AI ​​control module (using an STM32F407 main control chip with a computing frequency of 168MHz) receives data collected by sensors and thermal field model analysis results in real time. The building's composite envelope structure consists of an outer layer of color steel plate, a middle graphene composite thermal resistance control layer, and an inner insulation board. The AI ​​control module adjusts the thermal resistance of the building's composite envelope structure according to preset control logic. At the same time, graphene composite thermal resistance control plates are added to the partition walls of the six computing power zones to achieve physical thermal isolation between the zones and avoid sudden changes in the heat load of a single zone from affecting other zones. The material and structure of the partition wall thermal resistance control plates are consistent with the building envelope structure. The formula for dynamically adjusting the thermal conductivity is as follows: ; In the formula, This refers to the real-time thermal conductivity. The maximum thermal conductivity in the low-resistivity state is 0.1 W / (m·K); The minimum thermal conductivity in the high-resistivity state is taken as 0.03 W / (m·K). , These are the temperature thresholds for thermal resistance control. , ; When the outdoor temperature is ≥28℃ (during the high-temperature period in summer), the graphene composite thermal resistance control layer is powered on, so that the building's composite envelope structure switches to a high resistance state, and the thermal conductivity is adjusted to 0.025-0.03W / (m·K), reducing outdoor heat penetration and enhancing summer heat insulation. When the outdoor temperature is ≤10℃ (during the low temperature period in winter), the power is cut off to control the graphene composite thermal resistance control layer, so that the building composite envelope structure switches to a low resistance state and the thermal conductivity is adjusted to 0.1-0.12W / (m·K), and the outdoor low temperature is used to assist in heat dissipation. When the outdoor temperature is between 10-30℃, the power supply voltage of the graphene composite thermal resistance control layer is dynamically adjusted (adjustment range 0-24V) according to the indoor heat load data transmitted in step S1, so as to achieve a continuous adjustable thermal conductivity of 0.03-0.1W / (m·K) and achieve a dynamic balance between heat insulation and heat dissipation. When the heat load of any computing power zone increases suddenly, the thermal resistance of the graphene thermal resistance control plate of the partition wall of that zone is increased separately to achieve independent thermal resistance control of the zone. The control method is consistent with the building envelope structure. During the regulation process, the output includes real-time thermal conductivity regulation results and AI control command feedback data; at the same time, it assists in correcting the boundary conditions of the CFD three-dimensional thermal field model, improves the model accuracy, and feeds back the thermal conductivity regulation results to the AI ​​control module in real time for subsequent adaptive adjustment and optimization. S3. One set of exhaust shafts is configured for every two zones; the cold aisle is connected to the cooling air outlet of the liquid-cooled cabinet, and the hot aisle is directly connected to the return air end of the GPU cabinet. The adjustment range of the electric damper is 0-90°. The negative pressure exhaust device is a centrifugal fan with a speed range of 800-1800r / min; the horizontal cold aisle adopts a sealed design, with one set of electric dampers for each cold aisle. The hot aisle is directly connected to the vertical exhaust shaft, and one negative pressure exhaust device is installed every three floors in the exhaust shaft. The AI ​​control module analyzes airflow distribution data in real time. When it detects that the return air temperature of a server rack in a certain area of ​​the core computer room is ≥35℃ or the GPU core temperature is ≥80℃, the opening of the cold aisle damper in the corresponding area is increased by 20%, and the speed of the negative pressure exhaust device is increased from 1200r / min to 1600r / min (the speed increase ratio is about 25%, which is within the control range of 10%-20%) to accelerate the exhaust of hot air until the return air temperature drops below 30℃ and the GPU core temperature is ≤80℃. When the wind pressure deviation exceeds ±5%, the damper opening is automatically adjusted with a control response time of ≤5s to maintain the wind pressure in the duct at 200-300Pa, ensuring directional flow of hot and cold air and avoiding local hot spots from overheating. When any zone activates the liquid cooling auxiliary heat dissipation mode, the AI ​​control module automatically increases the opening of the cold aisle damper of that zone by 10% and increases the speed of the negative pressure exhaust device by 100 r / min to provide sufficient cooling air for the liquid cooling equipment and achieve airflow synergy between air cooling and liquid cooling. Output electric damper opening and negative pressure exhaust device speed control parameters, transmit them to AI control module, combine them with heat load prediction results to achieve coordinated adaptive adjustment, and feed back real-time airflow distribution data (including airflow velocity and wind pressure uniformity) to data acquisition and thermal characteristic modeling steps to optimize airflow field simulation parameters of CFD three-dimensional thermal field model. The S4 AI control module uses an LSTM time-series prediction algorithm to predict heat load changes and dynamically set the switching threshold for the hybrid cooling system. The hybrid cooling system consists of a natural cooling module, a mechanical refrigeration module, and a liquid cooling auxiliary module. The natural cooling module is an evaporative cooling unit (model ZSL-100) with a cooling capacity of 100kW. The mechanical refrigeration module is a screw chiller (model LSBLG130) with a cooling capacity of 130kW. The liquid cooling auxiliary module is a dual liquid cooling structure with cold plate liquid cooling as the primary method and spray liquid cooling as a secondary method, with a flow rate range of 5-10m³ / h. 3 / h; the liquid cooling medium is insulating heat-conducting oil; the natural cooling module is an air-water plate heat exchanger, and the mechanical refrigeration module is a screw chiller unit; Based on the rated heat load of 960kW for the computing center, the operating mode switching and cooling capacity distribution rules for the hybrid cooling system are as follows: When the outdoor temperature is ≤12℃ and the indoor heat load is ≤70% of the rated load (≤672kW), the pure natural cooling and liquid cooling micro-operation mode is activated. Natural cooling undertakes 80% of the cooling capacity, and the liquid cooling module only provides 20% of the cooling capacity for the GPU core. Mechanical cooling is turned off. When the outdoor temperature is 12-25℃ or the indoor heat load is 70%-85% of the rated load (672~816kW), the natural cooling, mechanical refrigeration and liquid cooling combined mode is activated, with natural cooling accounting for 40%, mechanical refrigeration accounting for 20% and liquid cooling module accounting for 40%. When the outdoor temperature is >25℃ or the indoor heat load is >85% of the rated load (>816kW), the liquid cooling as the main mode and mechanical refrigeration as the auxiliary mode is activated. The liquid cooling module (cold plate and spray) undertakes 70% of the cooling capacity and the mechanical refrigeration undertakes 30% of the cooling capacity, and natural cooling is turned off.

[0020] The liquid cooling module prioritizes supplying cooling to the GPU cluster area, and the hybrid cooling system for the six computing power zones enables independent mode switching and independent allocation of cooling capacity. The output cooling mode switching command and cooling capacity allocation parameters are transmitted to the heat load prediction and adaptive adjustment and waste heat recovery intelligent management steps, providing cooling capacity operation data support for heat load adaptive adjustment and waste heat recovery management. At the same time, the cooling system operation status data (including unit energy consumption, cooling capacity, and operating pressure) are fed back to the data acquisition steps to assist in the calibration of data collected by the distributed sensor array, ensure data accuracy, realize on-demand cooling capacity allocation, and reduce energy consumption. The S5 and AI control modules employ the LSTM time-series prediction algorithm. The LSTM time-series prediction model is trained using the TensorFlow framework with 1000 training batches and a learning rate of 0.001. Based on historical operational data from the past three months, real-time data collected in step S1 (indoor and outdoor temperature and humidity, rack heat load, cooling system energy consumption), computing task scheduling plans, GPU cluster load rates, and server start / stop records, it predicts the heat load change trend in the next 15 minutes to 2 hours. The formula for calculating the heat load prediction error is: ; In the formula, For prediction error, The heat load prediction value (unit: kW) output by the LSTM time series prediction model. The measured value of heat load (unit: kW) is monitored in real time by the distributed sensor array in step S1, and the prediction error is controlled within 5%. When the predicted heat load increase is ≥10%, increase the liquid cooling module liquid supply flow rate by 20% and adjust the air duct air volume by 30% 20 minutes in advance, and simultaneously increase the thermal resistance of the corresponding zone enclosure structure and partition wall to ensure that the cabinet air intake temperature is stable at 18~27℃ and the GPU core temperature is ≤85℃. When the predicted heat load decreases by ≥10%, the liquid cooling flow rate, airflow in the duct, and mechanical refrigeration load should be reduced simultaneously to avoid ineffective energy consumption. The computing power scheduling system synchronizes computing power task data to the AI ​​control module in advance to achieve forward linkage between computing power load rate, thermal load and cooling system, and the linkage method is data interaction. The output heat load prediction results and adaptive adjustment commands are transmitted to the dynamic thermal resistance regulation control, three-dimensional air duct intelligent adjustment, and hybrid cooling system collaborative control steps through the bus communication of the AI ​​control module, guiding the dynamic regulation of each step. At the same time, the prediction error data is fed back to the data acquisition step to optimize the acquisition accuracy of the distributed sensor array and the CFD three-dimensional thermal field model. The S6 AI control module monitors waste heat emission parameters (waste heat temperature, emission amount) and the cooling system operation data (including cooling water temperature and flow rate) transmitted in real time during the hybrid cooling system collaborative control steps. The waste heat recovery system consists of a plate heat exchanger, a waste heat storage tank, and a low-temperature waste heat generator. The plate heat exchanger has a heat exchange area of ​​[area missing]. The volume of the waste heat storage tank is The power generation capacity of the low-temperature waste heat power generation is 50kW; All waste heat recovery equipment is arranged nearby along the liquid cooling return water pipeline and exhaust shaft of the computing center to achieve local waste heat recovery; The formula for calculating the waste heat recovery rate is: ; In the formula, To improve the waste heat recovery and utilization rate, The actual total amount of waste heat recovered (unit: kJ). The total waste heat discharged by the heat dissipation system (unit: kJ) should be measured to ensure a waste heat recovery and utilization rate of 90%. The waste heat cascade utilization control rules are as follows: When the outdoor temperature is ≤5℃, the waste heat will be used first for heating the office area and maintenance dormitory of the computing center park to maintain the heating temperature at 20~22℃. The remaining waste heat will be used to heat domestic hot water to 40~50℃. When the outdoor temperature is 5-20℃, the waste heat is used first to heat domestic hot water to 40-50℃. The remaining waste heat is divided into two paths: one path drives a low-temperature waste heat generator to power auxiliary equipment such as the environmental monitoring system, lighting, and fire inspection robot in the computing center; the other path cools the operation and maintenance area of ​​the computing center through a waste heat direct-supply air conditioner. When the outdoor temperature is >25℃, the waste heat is prioritized for low-temperature waste heat power generation and domestic hot water preparation to maximize the conversion of electrical energy. The control parameters of waste heat recovery (waste heat temperature, recovery amount, utilization rate) are fed back in real time to the S1 data acquisition and thermal characteristic modeling module (to optimize the boundary conditions of the three-dimensional thermal field model) and the S4 hybrid cooling system collaborative control module (to adjust the operating mode of the cooling system). The fault warning unit of the AI ​​control module monitors the operating status of each sensor, cooling equipment, air duct adjustment device and waste heat recovery device in real time. When the deviation of a certain sensor parameter exceeds the set threshold ±5%, it immediately issues an audible and visual warning signal and pushes the abnormal information to the operation and maintenance terminal. At the same time, the AI ​​control module has a preset backup control strategy. When the AI ​​control module itself fails, it automatically switches to the backup control strategy. The backup control strategy is as follows: operate according to the preset temperature threshold, airflow parameters and cooling mode switching rules. The preset temperature threshold, airflow parameters and cooling mode switching rules are consistent with the control thresholds of steps S1 to S6, until the AI ​​control module returns to normal, ensuring the continuous and stable operation of the heat dissipation system.

[0021] Therefore, the present invention adopts the above-mentioned AI-based intelligent control-based data center building heat dissipation optimization method, system, equipment and computer-readable storage medium. The method realizes dynamic adaptive control of the entire heat dissipation process, optimizes the linkage control of the building envelope, airflow organization and cooling system, accurately predicts heat load changes and adjusts them in advance, and intelligently recovers and utilizes waste heat in stages.

[0022] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0023] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A data center building heat dissipation optimization method based on AI intelligent control, characterized in that, Includes the following steps: S1. Real-time data collection of structural temperature, indoor and outdoor temperature and humidity, cabinet intake and return air temperature, air duct pressure, cooling system energy consumption, PUE real-time value, GPU core temperature, server power consumption, and computing load rate are obtained through a distributed sensor array. Combined with CFD fluid simulation and infrared thermal imaging detection, a three-level three-dimensional thermal field model of building-computing zone-cabinet is constructed to locate hot spots and weak heat dissipation nodes. S2. Based on the real-time data and thermal field model collected by S1, the AI ​​control module outputs control commands to control the thermal resistance switching of the building's composite envelope structure and the graphene thermal resistance control plate of the computing power partition wall; the thermal conductivity is adaptively adjusted according to the changes in outdoor ambient temperature and indoor heat load to achieve bidirectional thermal resistance control of the building and computing power partition. S3. By analyzing airflow distribution data in real time through the AI ​​control module, the opening degree of the damper in the horizontal closed cold / hot channel and the speed of the negative pressure exhaust device in the vertical through-type exhaust shaft are adjusted to construct a vertical-horizontal integrated three-dimensional cold and hot airflow organization, and to achieve coordinated control of air-cooled and liquid-cooled airflow in combination with liquid cooling heat dissipation requirements. Based on the heat load prediction results, the S4 AI control module dynamically switches the operation mode of the cooling system, which consists of a natural cooling module, a mechanical refrigeration module, and a dual liquid cooling module consisting of a cold plate and a spray system. It sets the mode switching threshold and the cooling capacity allocation ratio according to the heat load characteristics of the computing center, so as to realize the on-demand allocation of cooling capacity, mainly liquid cooling. The S5 and AI control modules adopt the LSTM time-series prediction algorithm. Based on historical operating data, real-time acquisition parameters of S1 and computing power task scheduling plan, GPU cluster load rate, and server start-stop records, they predict the trend of heat load change, start the cooling system pre-adjustment in advance, and adapt to adjust liquid cooling flow, air duct air volume and thermal resistance of the building envelope. The S6 and AI control modules monitor waste heat emission parameters in real time. Through a waste heat recovery system consisting of a plate heat exchanger, a waste heat storage tank, a low-temperature waste heat generator, and a waste heat direct supply air conditioning module, the waste heat is utilized in stages. The waste heat recovery control parameters and utilization rate data are fed back to S1 and S4 to optimize the three-dimensional thermal field model and the cooling mode switching logic, thus completing the closed-loop control of the entire heat dissipation optimization process.

2. The data center building heat dissipation optimization method based on AI intelligent control according to claim 1, characterized in that, In S1, the acquisition error of the distributed sensor array The formula for calculating the acquisition error is: ； In the formula, For real-time sensor data acquisition values, This is the standard calibration value; The collected data is synchronously transmitted to the AI ​​control module via industrial Ethernet, and simultaneously stored on a local server for historical data retrieval and 3D thermal field model optimization. CFD fluid simulation boundary conditions are set to outdoor ambient temperature -15℃ to 42℃ and indoor cabinet heat flux density. The infrared thermal imaging detection accuracy is ±2℃.

3. The data center building heat dissipation optimization method based on AI intelligent control according to claim 2, characterized in that, In S2, the building's composite envelope structure consists of an outer layer of color steel plate, a middle graphene composite thermal resistance regulating layer, and an inner insulation board; the formula for calculating the real-time thermal conductivity of the envelope structure is: ； In the formula, For real-time thermal conductivity, The maximum thermal conductivity in the low-resistivity state is taken as 0.1 W / (m·K). The minimum thermal conductivity in the high-resistivity state is taken as 0.03 W / (m·K). , The temperature threshold is controlled by thermal resistance. , ; This refers to the real-time outdoor ambient temperature. The thermal resistance switching control rule is as follows: When the outdoor temperature is ≥28℃, the building envelope is switched to a high-resistance state, and the thermal conductivity is adjusted to 0.025~0.03W / (m・K); When the outdoor temperature is ≤10℃, the building envelope is switched to a low-resistance state, and the thermal conductivity is adjusted to 0.1~0.12W / (m・K); When the outdoor temperature is 10℃~28℃, the thermal conductivity can be continuously adjusted from 0.03~0.1W / (m・K) by adjusting the power supply voltage of the graphene composite thermal resistance control layer from 0~24V. When the heat load of any computing power partition increases suddenly, the thermal resistance of the graphene thermal resistance control plate of the partition wall of that partition is increased separately.

4. The data center building heat dissipation optimization method based on AI intelligent control according to claim 3, characterized in that, In S3, the damper is an electric damper with an adjustment range of 0~90°; the negative pressure exhaust device is a centrifugal fan with a speed range of 800~1800r / min. The intelligent air duct control rules are as follows: When the rack return air temperature is ≥35℃ or the GPU core temperature is ≥80℃, the opening of the cold aisle damper in the corresponding area is increased by 20%, and the centrifugal fan speed is increased from 1200r / min to 1600r / min, until the return air temperature drops below 30℃ and the GPU core temperature is ≤80℃. When the air pressure deviation in the air duct exceeds ±5%, the damper opening is automatically adjusted, and the control response time is ≤5s. When the liquid cooling auxiliary heat dissipation of a zone is activated, the damper opening of the cold aisle of that zone is increased by 10%, and the centrifugal fan speed is increased by 100r / min.

5. The data center building heat dissipation optimization method based on AI intelligent control according to claim 4, characterized in that, In S4, the rated heat load of the artificial intelligence computing center is 960kW. The operating mode switching and cooling capacity distribution rules of the hybrid cooling system are as follows: When the outdoor temperature is ≤12℃ and the indoor heat load is ≤70% of the rated load, the pure natural cooling and liquid cooling micro-operation mode will be activated, with natural cooling handling 80% of the cooling capacity and liquid cooling module handling 20% ​​of the cooling capacity. When the outdoor temperature is 12℃~25℃ or the indoor heat load is 70%~85% of the rated load, the natural cooling, mechanical refrigeration and liquid cooling combined mode will be activated. Natural cooling will handle 40% of the cooling capacity, mechanical refrigeration will handle 20% of the cooling capacity and liquid cooling module will handle 40% of the cooling capacity. When the outdoor temperature is >25℃ or the indoor heat load is >85% of the rated load, the liquid cooling as the main mode and the mechanical refrigeration as the auxiliary mode will be activated, with the liquid cooling module undertaking 70% of the cooling capacity and the mechanical refrigeration undertaking 30% of the cooling capacity. The hybrid cooling system with 6 computing power zones enables independent mode switching and independent allocation of cooling capacity, with the liquid cooling module prioritizing the supply of cooling capacity to the GPU cluster area.

6. The data center building heat dissipation optimization method based on AI intelligent control according to claim 5, characterized in that, In S5, the LSTM time series prediction model is trained based on the TensorFlow framework with 1000 training batches and a learning rate of 0.

001. Heat load prediction error The formula for calculating the prediction error is: ； In the formula, The heat load prediction value output by the LSTM time series prediction model. The measured value of heat load is obtained from real-time monitoring by a distributed sensor array. The adaptive pre-adjustment rule is: When the predicted heat load increase is ≥10%, increase the liquid cooling module liquid supply flow rate by 20%, adjust the air duct air volume by 30%, and adapt and adjust the thermal resistance of the corresponding partition enclosure structure and partition wall to ensure that the cabinet air intake temperature is stable at 18~27℃ and the GPU core temperature is ≤85℃. When the predicted heat load decreases by ≥10%, the liquid cooling flow rate, airflow in the duct, and mechanical refrigeration load should be reduced simultaneously.

7. The data center building heat dissipation optimization method based on AI intelligent control according to claim 6, characterized in that, In S6, the waste heat recovery system consists of a plate heat exchanger, a waste heat storage tank, and a low-temperature waste heat generator, achieving a waste heat recovery utilization rate of [missing information]. The formula for calculating the waste heat recovery rate is: ； In the formula, This represents the actual total amount of waste heat recovered. This is the total amount of waste heat discharged by the cooling system; The waste heat cascade utilization control rules are as follows: When the outdoor temperature is ≤5℃, the waste heat will be used first for heating the computing center park to maintain the heating temperature at 20~22℃. The remaining waste heat will be used to heat domestic hot water to 40~50℃. When the outdoor temperature is between 5℃ and 25℃, the waste heat is used first to heat domestic hot water to 40℃ to 50℃. The remaining waste heat is used to drive a low-temperature waste heat generator to supply electricity and to cool the maintenance area through a waste heat direct supply air conditioner. When the outdoor temperature is >25℃, the waste heat should be prioritized for low-temperature waste heat power generation and domestic hot water preparation. The control parameters and utilization rate data of waste heat recovery are fed back to S1 and S4.

8. A system for implementing the data center building heat dissipation optimization method based on AI intelligent control as described in any one of claims 1-7, characterized in that, include: The data acquisition and thermal characteristic modeling module is used to execute step S1, complete multi-dimensional data acquisition including GPU core temperature, server power consumption, and computing load rate, construct a three-level refined three-dimensional thermal field model of building-computing zone-rack and locate hot spots and weak heat dissipation nodes, and communicate with the AI ​​control module. The dynamic thermal resistance control module is communicatively connected to the data acquisition and thermal characteristic modeling module and the AI ​​control module. It is used to execute step S2, receive real-time data and three-dimensional thermal field model data, and realize adaptive control of thermal resistance of the building composite envelope structure and the graphene thermal resistance control plate of the computing power partition wall. The three-dimensional air duct intelligent adjustment module is communicatively connected to the data acquisition and thermal characteristic modeling module and the AI ​​control module. It is used to execute step S3, receive airflow distribution data, and adjust the damper opening and centrifugal fan speed to achieve coordinated control of air-cooled and liquid-cooled airflow. The heat load prediction and adaptive adjustment module is communicatively connected to the data acquisition and thermal characteristic modeling module and the AI ​​control module. It is used to execute step S5, integrate relevant data of computing power tasks to complete the prediction of heat load trends, and output the advance adjustment instructions of each control module. The hybrid cooling system collaborative control module is communicatively connected to the heat load prediction and adaptive adjustment module and the AI ​​control module. It is used to execute step S4, receive the heat load prediction results, and realize dynamic switching of the cooling mode with liquid cooling as the main mode and on-demand allocation of cooling capacity. The waste heat recovery intelligent management and control module is communicatively connected to the hybrid cooling system collaborative control module and the AI ​​control module. It is used to execute step S6 to realize the closed loop of the entire process of waste heat cascade utilization and heat dissipation optimization, and to feed back the waste heat recovery data to the data acquisition and thermal characteristic modeling module and the hybrid cooling system collaborative control module.

9. A computer device, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and executable on the processor. The processor is coupled to the memory. When the processor reads and executes the computer program, it implements the data center building heat dissipation optimization method based on AI intelligent control as described in any one of claims 1 to 7. The processor is an STM32F407 main control chip with an operating frequency of 168MHz; the memory is a local server storage module used to store real-time data, historical operating data, three-dimensional thermal field model data, and operating parameters of each control module collected by the distributed sensor array.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is a non-volatile storage medium, including at least one of disk storage, CD-ROM, and optical storage, and the computer-readable storage medium stores computer program code. When the computer program code is run on a computer, it enables the computer to implement the data center building heat dissipation optimization method based on AI intelligent control as described in any one of claims 1 to 7.

Images