Heat flow dynamic tracking AI prediction type server temperature control system and device

Abstract

The invention discloses a heat flow dynamic tracking AI prediction type server temperature control system and device. The system comprises an AI prediction model module, a dynamic temperature control execution module and a data acquisition and feedback module. The AI prediction model module is connected with the output end of the data acquisition and feedback module through a data bus, and the input end of the dynamic temperature control execution module is connected with the output end of the AI prediction model module; the input end of the data acquisition and feedback module is connected with a temperature sensor group deployed in a server, the input data of the AI prediction model module comprises a server real-time load, an environment temperature and a historical temperature time sequence, and the output data of the AI prediction model module is a temperature prediction value in the future 5 minutes. According to the method, the time advantage is established through AI prediction, space buffering is established through PCM cold storage, a three-dimensional regulation and control mechanism which is precisely executed is achieved through pump valve cooperation, optimization is achieved in the heat dissipation stability, the energy efficiency, the transient response, the self-adaptive capacity and the operation and maintenance mode, and intelligent temperature control is provided for the high-density computing power center.

Description

Technical Field

[0001] The present invention relates to the technical field of liquid-cooled server heat dissipation, and particularly to an AI prediction-based server temperature control system and device for dynamic heat flow tracking. Background Art

[0002] The technical background of the server temperature control system and device stems from the sharp contradiction between the explosive growth of computing power in data centers and the bottleneck of thermal management. With the popularization of high-density computing scenarios such as cloud computing and artificial intelligence, the power density of server chips has been continuously climbing. The traditional extensive heat dissipation mode based on air conditioning refrigeration has exposed three fundamental defects. The air-cooling heat dissipation efficiency has approached the physical limit and cannot effectively conduct the kilowatt-level heat flow instantaneously generated by core components such as GPUs / CPUs, resulting in local high temperatures triggering hardware protective frequency reduction. The lagging response mechanism makes the cooling system passively chase the change of heat load, and the temperature oscillation exacerbates the equipment aging and downtime risks. The proportion of heat dissipation energy consumption has climbed to more than 40% of the total power consumption of the data center. The industry needs to break through the triangular dilemma of "late warning of thermal runaway", "mismatch of cooling resources", and "low energy efficiency conversion". By integrating dynamic thermal perception, intelligent prediction algorithms, and precise execution control, a new generation of temperature control paradigm deeply adapted to high-computing power scenarios is constructed. In this context, an intelligent temperature control system integrating efficient heat transfer of liquid cooling media, transient heat storage and release of phase change materials, and dynamic decision-making of artificial intelligence has become an inevitable technical path to solve the collaborative optimization of heat dissipation efficiency and energy efficiency.

[0003] Currently, when cooling the heat-generating components of a server, such as the cold plates of CPUs or GPUs, the traditional temperature control system cannot accurately predict the server load and heat generation trend, resulting in low cooling efficiency. Therefore, we propose an AI prediction-based server temperature control system and device for dynamic heat flow tracking to solve this problem. Summary of the Invention

[0004] The purpose of the present invention is to provide an AI prediction-based server temperature control system and device for dynamic heat flow tracking to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions: An AI prediction-based server temperature control system for dynamic heat flow tracking, comprising: an AI prediction model module, a dynamic temperature control execution module, and a data acquisition and feedback module; The output end of the data acquisition and feedback module is connected to the AI prediction model module through a data bus, and the input end of the dynamic temperature control execution module is connected to the output end of the AI prediction model module; The input end of the data acquisition and feedback module is connected to a temperature sensor group deployed in the server.

[0006] Preferably, the input data of the AI prediction model module includes the server's real-time load, environmental temperature, and historical temperature time series. The output data is the temperature prediction value within the next 5 minutes, with a prediction accuracy of ±1°C. The AI prediction model uses an LSTM neural network architecture, and the training dataset contains 10^6 groups of server operating condition samples.

[0007] Preferably, the dynamic temperature control execution module includes a flow rate adjustment unit, a pump speed control unit, and an execution response unit. The flow rate adjustment unit adjusts the coolant flow rate in the range of 1 L / min to 5 L / min. The pump speed control unit adjusts the rotational speed of the circulation pump in the range of 1000 rpm to 5000 rpm. The execution response unit has a response time ≤ 10 seconds and generates a PID control signal based on the temperature prediction value output by the AI prediction model.

[0008] Preferably, the flow rate adjustment unit and the pump speed control unit implement a collaborative control strategy.

[0009] Preferably, the system is physically connected to the cold plate pipeline of the liquid-cooled server. The flow rate adjustment unit is integrated into the electric proportional valve at the cold plate inlet. The pump speed control unit drives an external magnetic levitation centrifugal pump. The temperature sensor group includes PT1000 platinum resistance probes and an infrared thermal imager.

[0010] Preferably, it further includes an embedded control board and a human-machine interaction terminal. The embedded control board is equipped to execute the AI prediction model. The human-machine interaction terminal is used to display the real-time heat flow distribution map and the predicted temperature rise curve.

[0011] An AI prediction-based server temperature control device for dynamic heat flow tracking, comprising: A cold plate, on one side output end of the cold plate, a first Hall flowmeter is installed, and on the upper end face of the first Hall flowmeter, a first temperature sensor is installed. On the other side input end of the cold plate, a second Hall flowmeter is installed, and on the upper end face of the second Hall flowmeter, a second temperature sensor is installed. The output end of the first Hall flowmeter is installed with an integrated pipe one through a connecting pipe. The input end of the second Hall flowmeter is installed with an electromagnetic control valve through a connecting pipe, and the other end of the electromagnetic control valve is installed with an integrated pipe two; A stepless variable frequency pump; A cold storage module, between the cold storage module and the stepless variable frequency pump, a second liquid guide pipe is installed, and between the cold storage module and the integrated pipe two, a third liquid guide pipe is installed.

[0012] Preferably, a first liquid guide pipe is installed between the integrated pipe one and the stepless variable frequency pump. On the upper end face of the cold storage module, heat dissipation fins are installed, and on the upper end face of the heat dissipation fins, a heat dissipation fan is installed.

[0013] Preferably, the electromagnetic control valve is an electric proportional valve, which implements a cooperative control strategy with a stepless variable frequency pump to dynamically match the flow regulation with the pump speed.

[0014] The beneficial effects of the present invention are as follows: 1. In the present invention, a three-dimensional control mechanism is established through AI prediction to build time advantages, PCM cold storage to construct spatial buffers, and pump-valve coordination to achieve precise execution. Optimization is achieved in terms of heat dissipation stability, energy efficiency, transient response, adaptability, and operation and maintenance mode, providing intelligent temperature control for high-density computing centers.

[0015] 2. In the present invention, through the millisecond-level analysis of multi-dimensional data such as real-time load, environmental temperature, and historical temperature change curves by the LSTM neural network, the system can accurately predict the temperature change trend 5 minutes in advance. Combining with infrared thermal imaging technology to construct a spatial heat flow map, when the predicted temperature approaches the threshold, the dynamic temperature control module synchronously adjusts the opening of the flow valve and the pump speed based on the cooperative algorithm to form a "prediction-execution" closed-loop control. Compared with the traditional lag response mechanism, the temperature fluctuation of this system is suppressed within a very narrow range of ±1.5°C, avoiding the risk of server frequency reduction or downtime caused by local overheating, and significantly improving the operation stability and lifespan of hardware in high-computing scenarios. 3. In the present invention, through the dual-level control architecture of AI prediction + PCM cold storage, on the one hand, the LSTM model prospectively plans the cooling resource requirements, enabling the pump and valve to output power only as needed, eliminating the energy redundancy caused by the constant large flow in the traditional solution. On the other hand, the cold storage module pre-cools the returned flow through phase change materials, significantly reducing the peak cooling demand under sudden loads, and solving the problem of excessive heat dissipation energy consumption ratio in data centers while ensuring the heat dissipation efficiency. 4. In the present invention, due to the continuous learning characteristics of the embedded AI model, data such as the server aging curve, environmental parameter drift, and new load patterns are collected in real time, and the system automatically optimizes the prediction weights and control parameters. This dynamic evolution ability enables the temperature control system to break through the limitations of static design, actively adapt to complex variables such as hardware iteration, computer room relocation, and seasonal climate change, and maintain a prediction accuracy of ±1°C and a temperature control stability of ±1.5°C in the long term. Compared with the traditional solution that requires manual parameter adjustment, the operation and maintenance complexity of this system is significantly reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Schematic diagram of the AI prediction-based server temperature control system for thermal flow dynamic tracking proposed by the present invention; Figure 2 Schematic diagram of the process of the AI prediction-based server temperature control system for thermal flow dynamic tracking proposed by the present invention; Figure 3 Schematic diagram of the three-dimensional front view structure of the AI prediction-based server temperature control device for thermal flow dynamic tracking in the present invention; Figure 4 This is a schematic diagram of the three-dimensional rear view structure of the AI prediction type server temperature control device for thermal flow dynamic tracking in this specification.

[0017] In the figure: 1. Cold plate; 2. Stepless variable frequency pump; 3. Cold storage module; 4. Hall flowmeter I; 5. Temperature sensor I; 6. Hall flowmeter II; 7. Temperature sensor II; 8. Integrated pipe I; 9. Electromagnetic control valve; 10. Integrated pipe II; 11. Liquid guide pipe I; 12. Liquid guide pipe II; 13. Liquid guide pipe III; 14. Heat dissipation fins; 15. Heat dissipation fan. Specific implementation manners

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments.

[0019] Refer to Figure 1 - Figure 4 , the AI prediction type server temperature control system for thermal flow dynamic tracking, includes: AI prediction model module, dynamic temperature control execution module, and data acquisition and feedback module; The AI prediction model module is connected to the output end of the data acquisition and feedback module through a data bus, and the input end of the dynamic temperature control execution module is connected to the output end of the AI prediction model module; The input end of the data acquisition and feedback module is connected to the temperature sensor group deployed in the server.

[0020] In this embodiment, the input data of the AI prediction model module includes the real-time load of the server, the ambient temperature, and the historical temperature time series. The output data is the temperature prediction value within the next 5 minutes, and the prediction accuracy is ±1°C. The AI prediction model adopts the LSTM (Long Short-Term Memory) neural network architecture, and the training data set contains 10^6 groups of server operating condition samples.

[0021] In this embodiment, the dynamic temperature control execution module includes a flow rate adjustment unit, a pump speed control unit, and an execution response unit. The flow rate adjustment unit adjusts the coolant flow rate range from 1 L / min to 5 L / min, the pump speed control unit adjusts the circulating pump speed range from 1000 rpm to 5000 rpm, and the execution response unit response time ≤ 10 seconds, generating a PID control signal according to the temperature prediction value output by the AI prediction model.

[0022] In this embodiment, the flow rate adjustment unit and the pump speed control unit implement a cooperative control strategy; When the predicted temperature exceeds the threshold T_th, it satisfies the functional relationship: Q = k1·(T_pred - T_th) + Q_min N = k2·(T_pred - T_th) + N_min Wherein, Q is the coolant flow rate (L / min), N is the pump speed (rpm), k1 = 0.8 L / (min·°C), k2 = 800 rpm / °C, Q_min = 1 L / min, N_min = 1000 rpm.

[0023] In this embodiment, the system is physically connected to the cold plate pipeline of the liquid-cooled server. The flow rate adjustment unit is integrated into the electric proportional valve at the cold plate inlet. The pump speed control unit drives an external magnetic levitation centrifugal pump. The temperature sensor group includes PT1000 platinum resistance probes and an infrared thermal imager.

[0024] In this embodiment, it further includes an embedded control board and a human-computer interaction terminal. The embedded control board is equipped to execute the AI prediction model, and the human-computer interaction terminal is used to display the real-time heat flow distribution map and the predicted temperature rise curve. The temperature sensor group (PT1000 platinum resistance + infrared thermal imager) collects the temperature points and heat flow distribution of the key areas of the server at a frequency of ≥10 Hz, combines with a Hall flowmeter to monitor the coolant flow rate (1–5 L / min), and the ambient temperature probe captures external heat interference. The data acquisition module integrates the real-time load (CPU / GPU power consumption), ambient temperature, historical temperature time series (≥1 minute granularity), and the temperature difference between the inlet and outlet of the coolant, and transmits it to the embedded control board through the EtherCAT bus. The LSTM time series modeling is based on an LSTM network trained with 10^6 sets of working conditions, analyzes the correlation between the temperature change trend and the load, outputs the temperature prediction value for the next 5 minutes (accuracy ±1°C), the model updates the prediction every 10 seconds, and supports online learning to adapt to hardware aging. The infrared thermal imaging data assists in locating local hot spots, and combines with the prediction results to generate a temperature rise risk map (such as: "The GPU area will exceed the threshold in 2 minutes").

[0025] In this embodiment, the AI prediction type server temperature control device for thermal flow dynamic tracking includes: Cold plate 1, a Hall flowmeter 4 is installed at the output end on one side of the cold plate 1, and a temperature sensor 5 is installed on the upper end face of the Hall flowmeter 4. A Hall flowmeter 6 is installed at the input end on the other side of the cold plate 1, and a temperature sensor 7 is installed on the upper end face of the Hall flowmeter 6. The output end of the Hall flowmeter 4 is installed with an integrated pipe 8 through a connecting pipe. The input end of the Hall flowmeter 6 is installed with an electromagnetic control valve 9 through a connecting pipe, and the other end of the electromagnetic control valve 9 is installed with an integrated pipe 10. A stepless variable frequency pump 2, a cold storage module 3, a liquid guide pipe 12 is installed between the cold storage module 3 and the stepless variable frequency pump 2. A liquid guide pipe 13 is installed between the cold storage module 3 and the integrated pipe 10. A liquid guide pipe 11 is installed between the integrated pipe 8 and the stepless variable frequency pump 2. A heat dissipation fin 14 is installed on the upper end face of the cold storage module 3, and a heat dissipation fan 15 is installed on the upper end face of the heat dissipation fin 14. The electromagnetic control valve 9 is an electric proportional valve, and a collaborative control strategy is implemented with the stepless variable frequency pump 2 to make the flow regulation and the pump speed dynamically match; PT1000 probes and infrared thermal imagers are installed at the inlet and outlet of the cold plate 1; In the cold storage module 3, there is a PCM filling cavity. The material is expanded graphite-based composite paraffin (thermal conductivity ≥ 8W / (m·K)) or Ga-In-Sn alloy (melting point 13°C), the phase change latent heat is 180 - 220kJ / kg, and it can absorb 4.2kJ of heat (for every 100g of PCM to cool 5L of coolant by 2°C). The channel in the cold storage module 3 is a spiral coolant flow channel, a double-spiral titanium alloy flow channel (pitch 12mm / depth 4mm), which increases the contact area with the PCM by 300%, and the flow resistance ≤ 0.5kPa to avoid increasing the pump power burden. At the same time, a temperature sensor PT1000 platinum resistance (accuracy ±0.1°C) is installed in the cold storage module 3 to monitor the inlet liquid temperature in real time. The trigger threshold: 25–30°C is adjustable (matching the server load curve). The adiabatic outer shell of the cold storage module 3 is a polyimide foam layer (thermal conductivity 0.03W / (m·K)) to reduce environmental heat interference; Technical advantages of the cold storage module 3: Transient heat compensation Response time ≤ 10 seconds, suppressing the coolant temperature rise caused by sudden load within ±1.5°C (traditional solution > ±5°C); Pump power optimization Reduce the flow demand through precooling, and the pump power is reduced by 30% (example: under a 300W heat load, the flow rate changes from 5L / min → 3.5L / min); Before the coolant flows back to the cold plate, actively precool the coolant by the phase change material (PCM) melting and absorbing heat in 3 to suppress temperature fluctuations. Especially when the server has a sudden high load, the coolant temperature is monitored in real time through an embedded temperature sensor. When the detected inlet liquid temperature > the set threshold (such as 28°C), the PCM cold storage module is automatically activated to achieve dynamic temperature control.

[0026] In this embodiment, during use, the coolant is driven by the stepless variable frequency pump 2 and transported to the inlet of the cold plate 1 through the first liquid guide pipe 11. The inlet flow rate is monitored by the second Hall flowmeter 6, and at the same time, the second temperature sensor 7 collects the inlet liquid temperature in real time. After the coolant flows through the cold plate 1 and absorbs the heat of the server, the outflow rate is monitored by the first Hall flowmeter 4 at the outlet end, and the first temperature sensor 5 synchronously detects the outlet liquid temperature after the temperature rise. The heated coolant enters the third liquid guide pipe 13 of the cold storage module 3 through the first integrated pipe 8. When the embedded temperature sensor installed in the cold storage module 3 detects that the inlet liquid temperature exceeds the set threshold, the cold storage module 3 is automatically activated. The coolant fully contacts the phase change material (PCM) in the double helix titanium alloy flow channel of the cold storage module 3, and the PCM absorbs the excess heat in the coolant by melting and absorbing heat, realizing transient heat compensation. The pre-cooled coolant flows out of the cold storage module 3 and returns to the stepless variable frequency pump through the third liquid guide pipe 13 to form a closed-loop cycle; At the same time, the temperature sensor group (including the PT1000 probes at the inlet and outlet of the cold plate 1 and the infrared thermal imager) collects the server heat distribution data at a frequency of ≥10 Hz. Combining the ambient temperature and the real-time load information of the server, it is transmitted to the embedded control board through the EtherCAT bus. The control board runs the LSTM prediction model to analyze the temperature trend, generates a heat risk map and temperature prediction values for the next five minutes. If the predicted temperature exceeds the preset threshold, the response unit immediately generates a PID control signal, and the electro-hydraulic proportional valve dynamically adjusts the opening of the inlet of the cold plate 1 according to the signal to change the flow rate. At the same time, the stepless variable frequency pump 2 synchronously adjusts the speed to make the flow rate and the pump speed match according to the cooperation strategy. The cooling fan 15 automatically starts and stops based on the outlet temperature of the cold storage module 3, driving the air flow to discharge the heat stored in the PCM through the heat dissipation fins. The whole process dynamically displays the heat flow distribution and temperature rise curve through the human-machine interaction terminal, realizing a closed-loop temperature control from heat perception, dynamic prediction to cooperative execution.

[0027] The above has introduced in detail the AI prediction type server temperature control system and device for thermal flow dynamic tracking provided by the present invention. Specific embodiments are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of the present invention, several improvements and modifications can still be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. An AI prediction-based server temperature control system for thermal flow dynamic tracking, characterized in that, Including: AI prediction model module, dynamic temperature control execution module, and data acquisition and feedback module; The AI prediction model module is connected to the output end of the data acquisition and feedback module through a data bus, and the input end of the dynamic temperature control execution module is connected to the output end of the AI prediction model module; The input end of the data acquisition and feedback module is connected to a temperature sensor group deployed in the server.

2. The AI prediction-based server temperature control system for thermal flow dynamic tracking according to claim 1, characterized in that, The input data of the AI prediction model module includes the server's real-time load, ambient temperature, and historical temperature time series. The output data is the temperature prediction value within the next 5 minutes, with a prediction accuracy of ±1°C. The AI prediction model adopts an LSTM neural network architecture, and the training dataset contains 10^6 sets of server operating condition samples.

3. The AI prediction-based server temperature control system for thermal flow dynamic tracking according to claim 1, characterized in that, The dynamic temperature control execution module includes a flow rate adjustment unit, a pump speed control unit, and an execution response unit. The flow rate adjustment unit adjusts the coolant flow rate in the range of 1 L / min to 5 L / min. The pump speed control unit adjusts the rotational speed of the circulation pump in the range of 1000 rpm to 5000 rpm. The execution response unit has a response time ≤ 10 seconds and generates a PID control signal according to the temperature prediction value output by the AI prediction model.

4. The AI prediction-based server temperature control system for thermal flow dynamic tracking according to claim 3, wherein The flow rate adjustment unit and the pump speed control unit implement a coordinated control strategy.

5. The AI prediction-based server temperature control system for thermal flow dynamic tracking according to claim 3, wherein, The system is physically connected to the cold plate pipeline of the liquid-cooled server. The flow rate adjustment unit is integrated into the electric proportional valve at the cold plate inlet. The pump speed control unit drives an external magnetic levitation centrifugal pump. The temperature sensor group includes PT1000 platinum resistance probes and an infrared thermal imager.

6. The AI prediction-based server temperature control system for thermal flow dynamic tracking according to claim 1, characterized in that, It also includes an embedded control board and a human-computer interaction terminal. The embedded control board is equipped to execute the AI prediction model. The human-computer interaction terminal is used to display the real-time heat flow distribution map and the predicted temperature rise curve.

7. The AI prediction-based server temperature control device for thermal flow dynamic tracking is applicable to the AI prediction-based server temperature control system for thermal flow dynamic tracking described in any one of claims 1-6, and is characterized in that Including: Cold plate (1), on one side output end of the cold plate (1), a first Hall flowmeter (4) is installed, and on the upper end face of the first Hall flowmeter (4), a first temperature sensor (5) is installed. On the other side input end of the cold plate (1), a second Hall flowmeter (6) is installed, and on the upper end face of the second Hall flowmeter (6), a second temperature sensor (7) is installed. The output end of the first Hall flowmeter (4) is installed with an integrated pipe one (8) through a connecting pipe. The input end of the second Hall flowmeter (6) is installed with an electromagnetic control valve (9) through a connecting pipe, and the other end of the electromagnetic control valve (9) is installed with an integrated pipe two (10); Stepless variable frequency pump (2); Cool storage module (3), between the cool storage module (3) and the stepless variable frequency pump (2), a second liquid guide pipe (12) is installed, and between the cool storage module (3) and the integrated pipe two (10), a third liquid guide pipe (13) is installed.

8. The AI prediction-based server temperature control device for thermal flow dynamic tracking according to claim 7, characterized in that, Between the integrated pipe one (8) and the stepless variable frequency pump (2), a first liquid guide pipe (11) is installed. On the upper end face of the cool storage module (3), heat dissipation fins (14) are installed, and on the upper end face of the heat dissipation fins (14), a heat dissipation fan (15) is installed.

9. The AI prediction-based server temperature control device for thermal flow dynamic tracking according to claim 7, characterized in that, The electromagnetic control valve (9) is an electric proportional valve and implements a coordinated control strategy with the stepless variable frequency pump (2) to dynamically match the flow rate adjustment and the pump speed.

Images