A method and system for collaborative control of immersion liquid cooling and waste heat recovery

Abstract

The application discloses a kind of immersion liquid cooling and waste heat recovery collaborative control method and system, method includes with fixed sampling period obtains the real-time power consumption sequence of chip side and the real-time outlet temperature sequence measured by liquid cooling cabinet;Based on real-time power consumption sequence and real-time outlet temperature sequence, extract heat transfer characteristic;Based on heat transfer characteristic and real-time power consumption sequence, obtain the outlet temperature component only by chip instant heat contribution, recorded as instant temperature component;Based on instant temperature component and real-time outlet temperature sequence, determine the temperature component of cooling liquid heat storage release contribution, recorded as heat storage temperature component;According to instant temperature component and heat storage temperature component, calculate heat flow distribution ratio by preset distribution parameter;According to heat flow distribution ratio, generate heat dissipation and waste heat recovery collaborative control strategy.The application can realize the collaborative optimization control of heat dissipation and waste heat recovery by effectively identifying heat component.

Description

Technical Field

[0001] This application relates to the field of data center heat dissipation technology, and in particular to a method and system for the coordinated control of immersion liquid cooling and waste heat recovery. Background Technology

[0002] Immersion liquid cooling technology achieves efficient heat dissipation by directly immersing servers in coolant. To further improve energy efficiency, more and more immersion liquid cooling systems are beginning to integrate waste heat recovery units to use the heat generated by servers for heating or industrial applications.

[0003] Existing immersion liquid cooling systems typically employ a temperature-based multi-point monitoring and control strategy. Among these strategies, the outlet temperature (the temperature at which the coolant flows out of the cabinet) is one of the important indicators for judging the overall thermal state of the system. The traditional control method is to set an outlet temperature threshold. When the outlet temperature exceeds the upper threshold, the system increases its heat dissipation capacity and limits waste heat recovery. When the outlet temperature is below the lower threshold, the system strengthens waste heat recovery and reduces active heat dissipation.

[0004] However, due to the heat capacity buffering effect of the coolant, the instantaneous heat generation of the chip and the heat release stored in the liquid are mixed in the temperature signal. For example, the reason for the increase in outlet temperature may be that the chip is currently under increased load and is generating a lot of instantaneous heat, or it may be that the chip load has decreased, but the heat previously stored in the coolant is being released. Traditional control methods cannot effectively distinguish between the two, which can easily lead to a contradiction between low recovery efficiency and sluggish heat dissipation response.

[0005] For example, when the coolant releases accumulated heat, causing the outlet temperature to rise, it is mistakenly judged as an increase in chip load, and instead, the heat recovery is reduced and the heat dissipation is strengthened, wasting the high-quality heat that should have been recovered. Conversely, when the chip load suddenly increases but the heat has not yet been transferred to the outlet, the heat dissipation action lags behind the actual heat demand of the chip because there is a significant time delay between the chip heating up and the outlet temperature response. Summary of the Invention

[0006] The purpose of this application is to provide a method and system for the coordinated control of immersion liquid cooling and waste heat recovery, so as to achieve synergistic optimization of heat dissipation and waste heat recovery.

[0007] In a first aspect, this application provides a method for the coordinated control of immersion liquid cooling and waste heat recovery, comprising: The real-time power consumption sequence of the chip side and the real-time outlet temperature sequence of the liquid cooling cabinet are obtained with a fixed sampling period. Based on real-time power consumption sequence and real-time outlet temperature sequence, heat transfer features are extracted. These heat transfer features are used to characterize the outlet temperature response pattern caused by changes in unit chip power consumption. The heat transfer features include response amplitude, response delay, and response duration. Based on heat transfer characteristics and real-time power consumption sequence, the outlet temperature component contributed solely by the chip's instantaneous heat generation is obtained and denoted as the instantaneous temperature component. Based on the instantaneous temperature component and the real-time outlet temperature sequence, the temperature component contributing to the heat storage release of the coolant is determined and denoted as the heat storage temperature component. The heat flow distribution ratio is calculated based on the instantaneous temperature component and the heat storage temperature component using preset distribution parameters. Based on the heat flow distribution ratio, a coordinated control strategy for heat dissipation and waste heat recovery is generated.

[0008] The above technical solution extracts the heat transfer characteristics that characterize the dynamic relationship between chip power consumption and outlet temperature. The real-time outlet temperature is decomposed into an instantaneous temperature component contributed by the chip's instantaneous heat generation and a heat storage temperature component contributed by the coolant's heat storage and release. This accurately identifies the heat source and calculates the heat flow distribution ratio based on the dynamic relationship between the two temperature components. This enables coordinated control of heat dissipation and waste heat recovery, solving the problem of control conflicts caused by the inability of traditional methods to distinguish heat components. It provides a precise decision-making basis for the coordinated control of heat dissipation and waste heat recovery.

[0009] Optionally, heat transfer features are extracted based on real-time power consumption sequences and real-time outlet temperature sequences, including: Based on the real-time power consumption sequence and the real-time outlet data sequence, the real-time power consumption and real-time outlet temperature within the current sliding window are selected according to the preset sliding window. The selected real-time power consumption and real-time outlet temperature are cross-correlated to generate a temperature response sequence, which contains the temperature response values ​​at each time point. Heat transfer features are extracted based on temperature response sequences.

[0010] Optionally, obtaining the outlet temperature component contributed solely by the chip's instantaneous heat generation based on heat transfer characteristics and real-time power consumption sequence includes: Using the current moment as the endpoint, and combining the response duration in the heat transfer characteristics, a target time period is generated; Select power consumption data for each moment within the target time period from the real-time power consumption sequence; The time difference between each moment within the target time period and the current moment is taken as the lag time. The temperature response value of the lag time is selected from the temperature response sequence and recorded as the target response value. The power consumption data at each moment within the target time period is weighted and summed with the corresponding target response value to generate an outlet temperature component contributed solely by the chip's instantaneous heat generation.

[0011] Optionally, the preset allocation parameters include a preset baseline allocation coefficient and a preset weighting coefficient. The step of calculating the heat flow distribution ratio based on the instantaneous temperature component and the heat storage temperature component using the preset allocation parameters includes: Based on the real-time temperature component, a real-time correction factor is calculated by presetting the upper limit of the real-time component. Based on the heat storage temperature component, the heat storage correction factor is calculated by setting the upper limit of the heat storage component. Based on the instantaneous correction factor and the heat storage correction factor, the heat flow distribution ratio is calculated through the preset baseline allocation coefficient and the preset weight coefficient.

[0012] Optionally, before calculating the heat flow distribution ratio based on the instantaneous correction factor and the heat storage correction factor, using a preset baseline allocation coefficient and a preset weighting coefficient, the following steps are included: Obtain the real-time temperature of the coolant and the current ambient temperature; Based on the real-time temperature of the coolant, the current ambient temperature, the heat storage temperature component, and the real-time power consumption sequence, real-time operating condition features are extracted. These real-time operating condition features include chip real-time load features, heat dissipation condition features, residual heat intensity features, and heat storage change trend features. Based on real-time operating conditions, the preset baseline allocation coefficients and preset weight coefficients are corrected to generate corrected baseline allocation coefficients and corrected weight coefficients.

[0013] Optionally, the step of correcting the preset baseline allocation coefficient and preset weight coefficient based on real-time operating condition characteristics to generate corrected baseline allocation coefficients and corrected weight coefficients includes: Based on the characteristics of heat dissipation conditions and residual heat intensity, a benchmark influence coefficient is generated; Based on the benchmark influence coefficient, the preset benchmark allocation coefficient is corrected to generate the corrected benchmark allocation coefficient; Based on the chip's real-time load characteristics and heat storage change trend characteristics, a weighted influence coefficient is generated; Based on the weight influence coefficient, the preset weight coefficient is corrected to generate the corrected weight coefficient.

[0014] Optionally, before calculating the heat flow distribution ratio based on the instantaneous temperature component and the heat storage temperature component using preset distribution parameters, the following steps are included: Based on the instantaneous temperature component and the heat storage temperature component, the current system operation mode is determined, including heat dissipation-dominated mode, waste heat-dominated mode and balance mode. If the operating mode is heat dissipation-driven, a control strategy prioritizing heat dissipation will be generated. If the operating mode is waste heat-dominated, a control strategy prioritizing recovery will be generated. If the operating mode is balanced, the heat flow distribution ratio is calculated based on the instantaneous temperature component and the heat storage temperature component using preset distribution parameters.

[0015] Optionally, determining the current system operating mode based on the instantaneous temperature component and the heat storage temperature component includes: Determine whether the instantaneous temperature component is greater than a preset first threshold; If the instantaneous temperature component is greater than the preset first threshold, the current system operation mode is determined to be the heat dissipation-dominated mode. If the instantaneous temperature component is not greater than the preset first threshold, then determine whether the heat storage temperature component is greater than the preset second threshold. If the heat storage temperature component is greater than the preset second threshold, the current system operation mode is determined to be waste heat-dominated mode. If the heat storage temperature component is not greater than the preset second threshold, then the current system operation mode is determined to be the balanced mode.

[0016] Secondly, this application provides a synergistic control system for immersion liquid cooling and waste heat recovery, comprising: Data acquisition module 101 is used to acquire the real-time power consumption sequence of the chip side and the real-time outlet temperature sequence of the liquid cooling cabinet at a fixed sampling period; The heat transfer feature extraction module 102 is used to extract heat transfer features based on the real-time power consumption sequence and the real-time outlet temperature sequence. The heat transfer features are used to characterize the outlet temperature response pattern caused by the change in power consumption per unit chip. The heat transfer features include response amplitude, response delay and response duration. The temperature component identification module 103 is used to obtain the outlet temperature component contributed only by the instantaneous heating of the chip based on the heat transfer characteristics and the real-time power consumption sequence, and to determine the temperature component contributed by the heat storage release of the coolant based on the instantaneous temperature component and the real-time outlet temperature sequence, and to record the heat storage temperature component. The heat flow distribution calculation module 104 is used to calculate the heat flow distribution ratio based on the instantaneous temperature component and the heat storage temperature component, using preset distribution parameters. The collaborative control strategy generation module 105 is used to generate a collaborative control strategy for heat dissipation and waste heat recovery based on the heat flow distribution ratio.

[0017] Thirdly, this application provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above for a synergistic control method of immersion liquid cooling and waste heat recovery.

[0018] In summary, this application first extracts heat transfer characteristics to decompose the outlet temperature into an instantaneous heat generation component and a heat storage and release component, providing a reliable basis for achieving precise coordinated control of heat dissipation and waste heat recovery. Furthermore, by uniformly characterizing the corresponding proportions of heat dissipation and waste heat recovery through the heat flow distribution ratio, a dynamic balance between heat dissipation and recovery is achieved, avoiding the conflict between the two in traditional methods. In addition, the allocation parameters are dynamically corrected based on real-time operating conditions, enabling the control strategy to adapt to changes in ambient temperature, chip load, and other operating conditions, thus improving the robustness of the entire coordinated control method. Attached Figure Description

[0019] Figure 1 This is a flowchart of a synergistic control method for immersion liquid cooling and waste heat recovery provided in an embodiment of this application; Figure 2 This is a flowchart illustrating the extraction of heat transfer characteristics based on real-time power consumption sequence and real-time outlet temperature sequence provided in this application embodiment; Figure 3 This is a flowchart provided in an embodiment of the present application for obtaining the outlet temperature component contributed solely by the instantaneous heating of the chip based on heat transfer characteristics and real-time power consumption sequence; Figure 4 This is a flowchart provided in an embodiment of the present application, which calculates the heat flow distribution ratio based on the instantaneous temperature component and the heat storage temperature component using preset distribution parameters; Figure 5 This is a schematic diagram of a synergistic control system for immersion liquid cooling and waste heat recovery provided in an embodiment of this application. Detailed Implementation

[0020] The following is in conjunction with the appendix Figure 1 -Appendix Figure 5 This application will be described in further detail below.

[0021] This application provides a method for the coordinated control of immersion liquid cooling and waste heat recovery, see [link to relevant documentation]. Figure 1 This includes the following steps: S100: With a fixed sampling period, acquire the real-time power consumption sequence on the chip side and the real-time outlet temperature sequence measured by the liquid cooling cabinet.

[0022] S200 extracts heat transfer characteristics based on real-time power consumption sequence and real-time outlet temperature sequence.

[0023] S300: Based on heat transfer characteristics and real-time power consumption sequence, obtain the outlet temperature component contributed only by the chip's instantaneous heat generation, denoted as the instantaneous temperature component.

[0024] S400. Based on the instantaneous temperature component and the real-time outlet temperature sequence, determine the temperature component contributing to the heat storage and release of the coolant, denoted as the heat storage temperature component.

[0025] S500: Calculates the heat flow distribution ratio based on the instantaneous temperature component and the heat storage temperature component using preset distribution parameters.

[0026] S600 generates a coordinated control strategy for heat dissipation and waste heat recovery based on the heat flow distribution ratio.

[0027] The basic workflow of an immersion liquid cooling system is as follows: chip heats up → coolant absorbs heat → circulating pump sends high-temperature coolant out of the rack outlet → waste heat is recovered or discharged through the heat exchange unit → coolant returns to the rack inlet after cooling down → re-enters the rack to absorb heat.

[0028] In this embodiment of the application, firstly, a fixed sampling period is used. Real-time power consumption measured by the acquisition chip The real-time outlet temperature measured by the liquid cooling cabinet , where k is the sampling time number.

[0029] Real-time power consumption measured by the chip This information can be obtained through the server's built-in power consumption monitoring interface, reflecting the heat generation intensity at the current moment; the real-time outlet temperature measured by the liquid cooling cabinet. The temperature is obtained by a temperature sensor installed on the outlet pipe of the liquid-cooled cabinet, reflecting the thermal state of the coolant when it leaves the cabinet at the current moment.

[0030] According to the sampling period, a real-time power consumption sequence can be formed, for example... and real-time outlet temperature sequence, for example .

[0031] Since the chip's heat dissipation requirement is to remove heat as quickly as possible, while the waste heat recovery requirement is to release heat at the appropriate time and rate, they are physically coupled in the same coolant circuit. Therefore, in order to solve the optimal coordination problem under this coupling relationship, it is necessary to distinguish whether the change in coolant outlet temperature is caused by the chip generating heat or by the release of heat accumulated in the coolant.

[0032] Therefore, in this embodiment of the application, heat transfer features are extracted based on the real-time power consumption sequence and the real-time outlet temperature sequence. The heat transfer features are used to characterize the outlet temperature response mode caused by the change in power consumption per unit chip. The heat transfer features include response amplitude, response delay and response duration.

[0033] Specifically, see Figure 2 Based on real-time power consumption sequences and real-time outlet temperature sequences, heat transfer features are extracted, including the following steps: S210. Based on the real-time power consumption sequence and the real-time outlet data sequence, select the real-time power consumption and real-time outlet temperature within the current sliding window according to the preset sliding window.

[0034] S220: Perform cross-correlation processing on the selected real-time power consumption and real-time outlet temperature to generate a temperature response sequence.

[0035] S230: Extract heat transfer features based on temperature response sequences.

[0036] First, a preset sliding window is used to select the real-time power consumption and real-time outlet temperature within the current sliding window from the real-time power consumption sequence and the real-time outlet data sequence. For example, if the sliding window length is L, and L represents the number of sampling points, then the real-time power consumption within the current sliding window can be expressed as: The real-time outlet temperature within the current sliding window can be expressed as: .

[0037] Then, cross-correlation processing is performed on the selected real-time power consumption and real-time outlet temperature to generate a temperature response sequence. The temperature response sequence contains the temperature response value corresponding to each time point, which can also be understood as the temperature response value per unit power consumption. Of course, each time point here does not correspond to the time of real-time power consumption and real-time outlet temperature, but rather to the time lag.

[0038] Cross-correlation processing is mainly achieved through cross-correlation functions, which measure the similarity between two signal sequences under different relative shifts. These functions can be expressed as: in, The lag time, also known as the time offset, represents the delay time of the temperature response relative to the power input, which is the time required for heat to travel from the chip to the outlet. and These are the average values ​​of real-time power consumption and real-time outlet temperature within the sliding window, respectively.

[0039] Cross-correlation results This is equivalent to determining the correlation between real-time outlet temperature changes and real-time power consumption changes, and obtaining the cross-correlation results. Next, normalization processing is required, that is, using the cross-correlation results. Dividing the cross-correlation result by the variance of the real-time power consumption within the sliding window converts it into the temperature response caused by a change in unit power consumption, i.e., the temperature response value per unit power consumption. , The meaning can be understood as follows: if the chip increases its power consumption by one unit (1kW) at time 0, then at... At that moment, the outlet temperature will rise. ℃.

[0040] For each lag time , Calculate separately Furthermore, through normalization, the corresponding unit power consumption response value can be generated. Then all Arranged in order of lag time, a temperature response sequence can be generated. The temperature response sequence characterizes the response sequence of the liquid cooling cabinet outlet temperature as a function of unit power consumption excitation, which can be simply understood as how long after the chip generates heat will the outlet temperature respond and how large the response is.

[0041] After obtaining the temperature response sequence, heat transfer characteristics can be extracted from it. Specifically, the peak position of the temperature response sequence is used as the response delay. Response delay, characterizing how long it takes for the temperature to begin to change significantly after a change in power consumption, i.e., taking... Lag time corresponding to the maximum value , can be represented as: .

[0042] The peak amplitude of the temperature response sequence is defined as the response amplitude A. The response amplitude characterizes how much the temperature can rise under the same power consumption, i.e., the maximum value of the temperature response sequence, which can be expressed as: .

[0043] The duration of the temperature response sequence from excitation to decay to a preset ratio is defined as the response duration. The response duration characterizes how long a temperature change lasts before decaying. The preset percentage is a percentage threshold relative to the peak value, used to determine when the response essentially ends. It's typically set to 10%, meaning that starting from the peak, the hysteresis time corresponding to the first response value decaying to the preset percentage is found. Let the preset percentage be denoted as... The response duration is... It can be represented as: By using the temperature response sequence, a mapping relationship between chip heating and outlet temperature can be established. This allows us to deduce the temperature component contributed by chip heating at the current outlet temperature. In other words, based on heat transfer characteristics and real-time power consumption sequence, we can obtain the outlet temperature component contributed solely by the chip's instantaneous heating, which is denoted as the instantaneous temperature component.

[0044] Specifically, see Figure 3 Based on heat transfer characteristics and real-time power consumption sequences, the outlet temperature component contributed solely by the chip's instantaneous heat generation is obtained, including the following steps: S310. Using the current time as the endpoint, and combining the response duration in the heat transfer characteristics, generate the target time period.

[0045] S320. Select power consumption data for each moment within the target time period from the real-time power consumption sequence.

[0046] S330. Using the time difference between each moment in the target time period and the current moment as the lag time, select the temperature response value of the lag time from the temperature response sequence and record it as the target response value.

[0047] S340: The power consumption data at each moment within the target time period is weighted and summed with the corresponding target response value to generate an outlet temperature component contributed only by the chip's instantaneous heat generation.

[0048] The current outlet temperature is essentially the result of a weighted sum of all historical power consumption over different delay times; the weights are... ,express The extent to which the power consumption at the previous moment affects the current outlet temperature.

[0049] That is, the real-time outlet temperature at the current moment k. In fact, it can be expressed as: The outlet temperature at current time k It consists of two parts: the outlet temperature component contributed solely by the instantaneous heat generation of the chip, denoted as the instantaneous temperature component, and the temperature component contributed by the heat storage and release of the coolant, denoted as the heat storage temperature component.

[0050] The temperature response sequence shows that the impact of a single power consumption on the outlet temperature will last for a period of time, but it is not permanent. It can be understood that the impact of the recent power consumption (that is, within the response duration) is still ongoing, corresponding to the heat currently being generated by the chip; while the impact of the earlier power consumption has decayed to the point of being negligible, corresponding to the heat that has passed and is slowly released by the coolant.

[0051] Therefore, the contribution of recent power consumption is theoretically the instantaneous temperature component, which is the result of weighted summation of power consumption over different delay times during the response duration.

[0052] Therefore, firstly, taking the current time k as the endpoint, and combining it with the response duration in the heat transfer characteristics... Generate the target time period, that is, start from the current time k and work backwards. Each sampling period is used to select power consumption data for each moment within the target time period from the real-time power consumption sequence. .

[0053] Then, using the time difference between each moment within the target time period and the current moment as the lag time, the temperature response value at the lag time is selected from the temperature response sequence and recorded as the target response value. .

[0054] Finally, by weighting and summing the power consumption data at each moment within the target time period with the corresponding target response value, the instantaneous temperature component can be generated. This instantaneous heat component is denoted as... , can be represented as: Then, based on the instantaneous temperature component and the corresponding real-time outlet temperature, the heat storage temperature component can be determined, and denoted as the heat storage temperature component. ,but It can be represented as: Because the instantaneous temperature component reflects the chip's current real-time heat pressure, while the heat storage temperature component reflects the coolant's heat reserve status, the former focuses on heat dissipation needs, while the latter focuses on waste heat recovery. At the current moment, whether to prioritize heat dissipation or waste heat recovery, and how to set the intensity of heat dissipation and waste heat recovery, are issues that need to be addressed.

[0055] In this application embodiment, three operating modes are predefined: heat dissipation-dominated mode, waste heat-dominated mode, and balanced mode. Heat dissipation-dominated mode prioritizes chip safety; waste heat-dominated mode prioritizes waste heat recovery; and balanced mode is a collaborative optimization.

[0056] Then, based on the instantaneous temperature component and the heat storage temperature component, the current system operating mode is determined, and based on the operating mode, the corresponding control strategy is determined.

[0057] Specifically, after obtaining the instantaneous temperature component and the heat storage temperature component, the following steps are also included: S410. Determine the current system operating mode based on the instantaneous temperature component and the heat storage temperature component.

[0058] S420. If the operating mode is heat dissipation-driven, a control strategy prioritizing heat dissipation will be generated.

[0059] S430. If the operating mode is waste heat-dominated mode, a control strategy prioritizing recovery will be generated.

[0060] S440. If the operating mode is balanced mode, the heat flow distribution ratio is calculated based on the instantaneous temperature component and the heat storage temperature component using preset distribution parameters.

[0061] First, the operating mode of the current system will be determined based on the instantaneous temperature component and the heat storage temperature component.

[0062] Specifically, the current system operating mode is determined based on the instantaneous temperature component and the heat storage temperature component, including the following steps: S411. Determine whether the instantaneous temperature component is greater than the preset first threshold.

[0063] S412. If the instantaneous temperature component is greater than the preset first threshold, then the current system operation mode is determined to be the heat dissipation-dominated mode.

[0064] S413. If the instantaneous temperature component is not greater than the preset first threshold, then determine whether the heat storage temperature component is greater than the preset second threshold.

[0065] S414. If the heat storage temperature component is greater than the preset second threshold, then the current system operation mode is determined to be the heat dissipation-dominated mode.

[0066] S415. If the heat storage temperature component is not greater than the preset second threshold, then the current system operation mode is determined to be the balanced mode.

[0067] The preset first threshold is the upper limit warning line for the safe operation of the chip. It can be determined based on the chip's highest operating temperature and with a safety margin. For example, the instantaneous temperature component value corresponding to 80% of the chip's junction temperature limit can be taken. If the instantaneous temperature component exceeds this value, it means that the chip's heat generation is approaching its temperature limit, and heat dissipation must be prioritized.

[0068] The preset second threshold is the upper limit warning line of the coolant's heat capacity. It can be determined based on the highest allowable temperature of the coolant and with a safety margin. For example, the heat storage temperature component value corresponding to 5°C below the boiling point of the coolant can be taken. If this value is exceeded, it means that the heat accumulated in the coolant is close to its safe storage limit, and the recovery must be increased to prevent the liquid from overheating or boiling.

[0069] Therefore, it is first determined whether the instantaneous temperature component is greater than the preset first threshold. If the instantaneous temperature component is greater than the preset first threshold, the current system operation mode is determined to be the heat dissipation-dominated mode.

[0070] If the instantaneous temperature component is not greater than the preset first threshold and the heat storage temperature component is greater than the preset second threshold, then the current system operation mode is determined to be the waste heat-dominated mode; otherwise, the current system operation mode is determined to be the balanced mode.

[0071] If the operating mode is heat dissipation-driven, a control strategy prioritizing heat dissipation will be generated. For example, the recovery intensity of the waste heat recovery unit will be set to the minimum value (e.g., valve opening 0%), the heat dissipation intensity of the heat dissipation unit will be set to the maximum value (e.g., cooling tower fully open), and the pumping intensity of the circulating pump will be set to the maximum value.

[0072] If the operating mode is waste heat-dominated, a control strategy prioritizing recovery is generated. For example, the recovery intensity of the waste heat recovery unit is set to the maximum value (e.g., valve opening 100%), and the heat dissipation intensity of the heat dissipation unit is set to the minimum value (e.g., cooling tower closed). The pumping intensity can be adjusted according to the difference between the instantaneous temperature component and the heat storage temperature component, because the difference reflects to some extent the urgency of the heat needing to be delivered, which can be matched with the intensity.

[0073] If the operating mode is balanced, the heat flow distribution ratio is calculated based on the instantaneous temperature component and the heat storage temperature component through preset distribution parameters. The heat flow distribution ratio represents the proportion of the total heat generated by the chip at the current moment that should be used for waste heat recovery. In other words, the heat flow distribution ratio is used to balance the needs of heat dissipation and waste heat recovery, so as to maximize waste heat recovery while meeting the heat dissipation needs.

[0074] Specifically, see Figure 4 Based on the instantaneous temperature component and the heat storage temperature component, the heat flow distribution ratio is calculated using preset distribution parameters, including the following steps: S510: Based on the real-time temperature component, calculate the real-time correction factor by presetting the upper limit of the real-time component.

[0075] S520. Based on the heat storage temperature component, calculate the heat storage correction factor by setting the upper limit of the heat storage component.

[0076] S530: Based on the instantaneous correction factor and the heat storage correction factor, the heat flow distribution ratio is calculated through the preset baseline allocation coefficient and the preset weight coefficient.

[0077] The preset allocation parameters include a preset baseline allocation coefficient and a preset weighting coefficient. The preset baseline allocation coefficient represents the default heat flow distribution ratio of the system in steady state. The preset weighting coefficient represents the importance of the instantaneous temperature component relative to the heat storage temperature component in steady state. Here, steady state represents an ideal state without emergencies or dynamic disturbances.

[0078] First, based on the instantaneous temperature component, the instantaneous correction factor is calculated by setting an upper limit for the instantaneous component. The upper limit for the instantaneous temperature component is the maximum possible value of the instantaneous temperature component. Usually, the instantaneous temperature component corresponding to the highest operating temperature of the chip is directly taken, mainly for normalization.

[0079] The preset upper limit of the instantaneous temperature component is: Then the immediate correction factor It can be represented as: The larger the real-time correction factor, the closer the chip is to the risk of overheating, and the more priority is needed to ensure heat dissipation.

[0080] Similarly, based on the heat storage temperature component, a heat storage correction factor is calculated by setting a preset upper limit for the heat storage component. This preset upper limit for the heat storage component is the maximum possible value of the heat storage temperature component. It is mainly used for normalization by directly taking the heat storage temperature component corresponding to the highest allowable temperature of the coolant.

[0081] The upper limit of the preset heat storage temperature component is recorded as follows: Then the heat storage correction factor It can be represented as: The larger the heat storage correction factor, the more heat is stored in the coolant, and the more valuable it is to recover.

[0082] Finally, the heat flow distribution ratio is calculated based on the real-time correction factor and the heat storage correction factor, using preset baseline allocation coefficients and preset weighting coefficients.

[0083] Let the heat flow distribution ratio at the current moment be . ,but It can be represented as: in, As a preset baseline allocation coefficient, For preset weighting coefficients, As a penalty, if fever increases immediately, recycling should be reduced. The larger the value, the more attention is paid to heat dissipation requirements, and the greater the penalty. As a reward, when heat storage increases, recovery should be increased. The larger the number, the more concerned the person is about coolant overheating, and the greater the reward.

[0084] Because of the consideration of the preset baseline allocation coefficient and preset weighting coefficients These are all ideal values ​​set under steady-state conditions. However, in actual operation, the system conditions are dynamic. Fixed parameters can easily lead to deviations in the control strategy. Therefore, the preset baseline allocation coefficients and preset weight coefficients need to be corrected based on the current real-time operating conditions.

[0085] Therefore, in this embodiment of the application, before calculating the heat flow distribution ratio based on the instantaneous correction factor and the heat storage correction factor, using the preset baseline allocation coefficient and the preset weighting coefficient, the following steps are also included: S531. Obtain the real-time temperature of the coolant and the current ambient temperature.

[0086] S532. Extract real-time operating characteristics based on the real-time temperature of the coolant, the current ambient temperature, the heat storage temperature component, and the real-time power consumption sequence.

[0087] S533. Based on the real-time operating conditions, the preset baseline allocation coefficient and preset weight coefficient are corrected to generate the corrected baseline allocation coefficient and corrected weight coefficient.

[0088] Among them, the real-time operating condition characteristics include the chip's real-time load characteristics, heat dissipation condition characteristics, residual heat intensity characteristics, and heat storage change trend characteristics.

[0089] Chip Real-Time Load Characteristics This is an indicator reflecting the current heat intensity of the chip, i.e., the current power consumption. With the chip's rated power consumption The ratio of can be expressed as: .

[0090] Heat dissipation characteristics This is an indicator reflecting the ease with which the current heat dissipation system can expel heat through the coolant, and can be expressed as: in, The current real-time temperature of the coolant can be directly taken as the current real-time outlet temperature. , The current ambient temperature. The maximum allowable temperature of the coolant. The lowest possible ambient temperature (e.g., the lowest ambient temperature that occurred in the past year, depending on the environment in which the system is located).

[0091] Residual heat intensity characteristics This is a quality indicator reflecting the recoverable heat in the current coolant, specifically the ratio of the current heat storage temperature component to the preset upper limit of the heat storage component, which can be expressed as: It has the same definition as the heat storage correction factor, but the two act on different aspects.

[0092] Characteristics of heat storage change trend , is an indicator reflecting whether the heat storage temperature component is rising or falling, that is, the rate of change of the heat storage temperature component, which can be expressed as: in, This represents the heat storage temperature component from the previous moment. The sampling period.

[0093] After obtaining the real-time operating condition characteristics, the preset baseline allocation coefficient and preset weight coefficient can be corrected based on the real-time operating condition characteristics to generate the corrected baseline allocation coefficient and corrected weight coefficient.

[0094] Specifically, based on real-time operating conditions, the preset baseline allocation coefficients and preset weight coefficients are corrected to generate corrected baseline allocation coefficients and corrected weight coefficients, including the following steps: S5331. Based on the characteristics of heat dissipation conditions and residual heat intensity, a benchmark influence coefficient is generated.

[0095] S5332. Based on the benchmark influence coefficient, the preset benchmark allocation coefficient is corrected to generate the corrected benchmark allocation coefficient.

[0096] S5333 generates weighted influence coefficients based on the chip's real-time load characteristics and heat storage change trend characteristics.

[0097] S5334. Based on the weight influence coefficient, the preset weight coefficient is corrected to generate the corrected weight coefficient.

[0098] Since the preset baseline allocation coefficient is essentially used to determine the control tone, that is, the starting point, that is, whether the current operating condition is biased towards heat dissipation or waste heat recovery, the heat dissipation condition characteristics and waste heat intensity characteristics represent the background conditions and recovery value, respectively, and the changes are relatively slow, so they are suitable for determining the control tone.

[0099] The preset weighting coefficients are essentially used to determine the dynamic bias of the control, that is, the bias when encountering abnormal situations during the process. The real-time load characteristics and heat storage change trend characteristics of the chip represent instantaneous risks and dynamic early warnings, respectively. They change rapidly, so they are suitable for determining the dynamic bias of the control.

[0100] Therefore, in this embodiment of the application, firstly, based on the characteristics of heat dissipation conditions... and residual heat intensity characteristics A benchmark influence coefficient can be generated, denoted as . ,but It can be represented as: Where α and β are the influence weights, and their values ​​range from (0,1). This is a penalty item for heat dissipation conditions. The larger the value, the better the heat dissipation conditions, which means that heat dissipation needs to be strengthened and recycling reduced. Equivalent to a reward item, The larger the value, the higher the waste heat intensity, and the more waste heat should be recovered.

[0101] Then, based on the baseline influence coefficient Assigning coefficients to the preset benchmark Make corrections to generate the corrected baseline allocation coefficients. , can be represented as: .

[0102] Similarly, based on the chip's real-time load characteristics and characteristics of heat storage change trends This can generate a weighted influence coefficient, denoted as [the weighted influence coefficient is missing from the original text]. ,but It can be represented as: in, and To influence the weights, the value range is (0,1). As a reward item for chip load, The larger the value, the greater the need for heat dissipation of the chip; therefore, the weighting influence coefficient needs to be increased. ; As a penalty for the heat storage trend, A larger value indicates a faster increase in heat storage, and a greater concern about coolant overheating; therefore, the weighting influence coefficient needs to be reduced. .

[0103] Then, based on the weighted influence coefficient For the preset weight coefficients Make corrections to generate revised weighting coefficients. , can be represented as: .

[0104] Based on the current real-time operating conditions, after generating the corrected baseline allocation coefficient and weighting coefficient, the heat flow distribution ratio can be generated based on the corrected baseline allocation coefficient and weighting coefficient.

[0105] Finally, based on the heat flow distribution ratio, a coordinated control strategy for heat dissipation and waste heat recovery is generated. That is, the intensity of waste heat recovery and heat dissipation are determined, and then the intensity is converted into control execution parameters of the corresponding actuators, such as valve opening and closing size.

[0106] This application also provides a synergistic control system for immersion liquid cooling and waste heat recovery, see [link to relevant documentation]. Figure 5 The system includes: a data acquisition module 101, a heat transfer feature extraction module 102, a temperature component identification module 103, a heat flow distribution calculation module 104, and a collaborative control strategy generation module 105.

[0107] The data acquisition module 101 is used to acquire the real-time power consumption sequence of the chip side and the real-time outlet temperature sequence of the liquid cooling cabinet at a fixed sampling period.

[0108] The heat transfer feature extraction module 102 is used to extract heat transfer features based on the real-time power consumption sequence and the real-time outlet temperature sequence. The heat transfer features are used to characterize the outlet temperature response mode caused by the change in power consumption per unit chip. The heat transfer features include response amplitude, response delay and response duration.

[0109] The temperature component identification module 103 is used to obtain the outlet temperature component contributed only by the instantaneous heating of the chip based on the heat transfer characteristics and the real-time power consumption sequence, which is denoted as the instantaneous temperature component. Based on the instantaneous temperature component and the real-time outlet temperature sequence, the temperature component contributed by the heat storage and release of the coolant is determined and denoted as the heat storage temperature component.

[0110] The heat flow distribution calculation module 104 is used to calculate the heat flow distribution ratio based on the instantaneous temperature component and the heat storage temperature component, using preset distribution parameters.

[0111] The collaborative control strategy generation module 105 is used to generate a collaborative control strategy for heat dissipation and waste heat recovery based on the heat flow distribution ratio.

[0112] In this embodiment, the data acquisition module 101 is specifically used to acquire the real-time power consumption sequence of the chip side and the real-time outlet temperature sequence of the liquid cooling cabinet at a fixed sampling period.

[0113] The heat transfer feature extraction module 102 is specifically used to extract heat transfer features based on the real-time power consumption sequence and real-time outlet temperature sequence obtained by the data acquisition module 101. The heat transfer features are used to characterize the outlet temperature response mode caused by the change in power consumption per unit chip. The heat transfer features include response amplitude, response delay and response duration.

[0114] The temperature component identification module 103 is specifically used to obtain the outlet temperature component contributed only by the instantaneous heating of the chip based on the heat transfer characteristics and real-time power consumption sequence obtained by the heat transfer feature extraction module 102, which is denoted as the instantaneous temperature component. Based on the instantaneous temperature component and the real-time outlet temperature sequence, the temperature component contributed by the heat storage and release of the coolant is determined and denoted as the heat storage temperature component.

[0115] The heat flow distribution calculation module 104 is specifically used to calculate the heat flow distribution ratio based on the instantaneous temperature component and heat storage temperature component generated by the temperature component identification module 103 and through preset distribution parameters.

[0116] The collaborative control strategy generation module 105 is specifically used to generate a collaborative control strategy for heat dissipation and waste heat recovery based on the heat flow distribution ratio calculated by the heat flow distribution calculation module 104.

[0117] This application also provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed by any of the above-described immersion liquid cooling and waste heat recovery coordinated control methods.

[0118] The embodiments described in this application are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the principles of this application should be included within the scope of protection of this application.

Claims

1. A method for coordinated control of immersion liquid cooling and waste heat recovery, characterized in that, include: The real-time power consumption sequence of the chip side and the real-time outlet temperature sequence of the liquid cooling cabinet are obtained with a fixed sampling period. Based on real-time power consumption sequence and real-time outlet temperature sequence, heat transfer features are extracted. These heat transfer features are used to characterize the outlet temperature response pattern caused by changes in unit chip power consumption. The heat transfer features include response amplitude, response delay, and response duration. Based on heat transfer characteristics and real-time power consumption sequence, the outlet temperature component contributed solely by the chip's instantaneous heat generation is obtained and denoted as the instantaneous temperature component. Based on the instantaneous temperature component and the real-time outlet temperature sequence, the temperature component contributing to the heat storage release of the coolant is determined and denoted as the heat storage temperature component. The heat flow distribution ratio is calculated based on the instantaneous temperature component and the heat storage temperature component using preset distribution parameters. Based on the heat flow distribution ratio, a coordinated control strategy for heat dissipation and waste heat recovery is generated.

2. The method for coordinated control of immersion liquid cooling and waste heat recovery according to claim 1, characterized in that, Based on real-time power consumption sequences and real-time outlet temperature sequences, heat transfer features are extracted, including: Based on the real-time power consumption sequence and the real-time outlet data sequence, the real-time power consumption and real-time outlet temperature within the current sliding window are selected according to the preset sliding window. The selected real-time power consumption and real-time outlet temperature are cross-correlated to generate a temperature response sequence, which contains the temperature response values ​​at each time point. Heat transfer features are extracted based on temperature response sequences.

3. The method for coordinated control of immersion liquid cooling and waste heat recovery according to claim 2, characterized in that, The process of obtaining the outlet temperature component contributed solely by the chip's instantaneous heat generation, based on heat transfer characteristics and real-time power consumption sequences, includes: Using the current moment as the endpoint, and combining the response duration in the heat transfer characteristics, a target time period is generated; Select power consumption data for each moment within the target time period from the real-time power consumption sequence; The time difference between each moment within the target time period and the current moment is taken as the lag time. The temperature response value of the lag time is selected from the temperature response sequence and recorded as the target response value. The power consumption data at each moment within the target time period is weighted and summed with the corresponding target response value to generate an outlet temperature component contributed solely by the chip's instantaneous heat generation.

4. The method for coordinated control of immersion liquid cooling and waste heat recovery according to claim 1, characterized in that, The preset allocation parameters include a preset baseline allocation coefficient and a preset weighting coefficient. The calculation of the heat flow distribution ratio based on the instantaneous temperature component and the heat storage temperature component using the preset allocation parameters includes: Based on the real-time temperature component, a real-time correction factor is calculated by presetting the upper limit of the real-time component. Based on the heat storage temperature component, the heat storage correction factor is calculated by setting the upper limit of the heat storage component. Based on the instantaneous correction factor and the heat storage correction factor, the heat flow distribution ratio is calculated through the preset baseline allocation coefficient and the preset weight coefficient.

5. The method for coordinated control of immersion liquid cooling and waste heat recovery according to claim 4, characterized in that, Before calculating the heat flow distribution ratio based on the instantaneous correction factor and the heat storage correction factor, using a preset baseline allocation coefficient and a preset weighting coefficient, the following steps are included: Obtain the real-time temperature of the coolant and the current ambient temperature; Based on the real-time temperature of the coolant, the current ambient temperature, the heat storage temperature component, and the real-time power consumption sequence, real-time operating condition features are extracted. These real-time operating condition features include chip real-time load features, heat dissipation condition features, residual heat intensity features, and heat storage change trend features. Based on real-time operating conditions, the preset baseline allocation coefficients and preset weight coefficients are corrected to generate corrected baseline allocation coefficients and corrected weight coefficients.

6. The method for coordinated control of immersion liquid cooling and waste heat recovery according to claim 5, characterized in that, The step of correcting the preset baseline allocation coefficient and preset weight coefficient based on real-time operating condition characteristics to generate corrected baseline allocation coefficients and corrected weight coefficients includes: Based on the characteristics of heat dissipation conditions and residual heat intensity, a benchmark influence coefficient is generated; Based on the benchmark influence coefficient, the preset benchmark allocation coefficient is corrected to generate the corrected benchmark allocation coefficient; Based on the chip's real-time load characteristics and heat storage change trend characteristics, a weighted influence coefficient is generated; Based on the weight influence coefficient, the preset weight coefficient is corrected to generate the corrected weight coefficient.

7. The method for coordinated control of immersion liquid cooling and waste heat recovery according to claim 1, characterized in that, Before calculating the heat flow distribution ratio based on the instantaneous temperature component and the heat storage temperature component using preset distribution parameters, the following steps are included: Based on the instantaneous temperature component and the heat storage temperature component, the current system operation mode is determined, including heat dissipation-dominated mode, waste heat-dominated mode and balance mode. If the operating mode is heat dissipation-driven, a control strategy prioritizing heat dissipation will be generated. If the operating mode is waste heat-dominated, a control strategy prioritizing recovery will be generated. If the operating mode is balanced, the heat flow distribution ratio is calculated based on the instantaneous temperature component and the heat storage temperature component using preset distribution parameters.

8. The method for coordinated control of immersion liquid cooling and waste heat recovery according to claim 7, characterized in that, The step of determining the current system operating mode based on the instantaneous temperature component and the heat storage temperature component includes: Determine whether the instantaneous temperature component is greater than a preset first threshold; If the instantaneous temperature component is greater than the preset first threshold, the current system operation mode is determined to be the heat dissipation-dominated mode. If the instantaneous temperature component is not greater than the preset first threshold, then determine whether the heat storage temperature component is greater than the preset second threshold. If the heat storage temperature component is greater than the preset second threshold, the current system operation mode is determined to be waste heat-dominated mode. If the heat storage temperature component is not greater than the preset second threshold, then the current system operation mode is determined to be the balanced mode.

9. A synergistic control system for immersion liquid cooling and waste heat recovery, characterized in that, include: The data acquisition module (101) is used to acquire the real-time power consumption sequence of the chip side and the real-time outlet temperature sequence of the liquid cooling cabinet at a fixed sampling period. The heat transfer feature extraction module (102) is used to extract heat transfer features based on the real-time power consumption sequence and the real-time outlet temperature sequence. The heat transfer features are used to characterize the outlet temperature response mode caused by the change in power consumption per unit chip. The heat transfer features include response amplitude, response delay and response duration. The temperature component identification module (103) is used to obtain the outlet temperature component contributed only by the instantaneous heating of the chip based on the heat transfer characteristics and the real-time power consumption sequence, which is denoted as the instantaneous temperature component. Based on the instantaneous temperature component and the real-time outlet temperature sequence, the temperature component contributed by the heat storage release of the coolant is determined and denoted as the heat storage temperature component. The heat flow distribution calculation module (104) is used to calculate the heat flow distribution ratio based on the instantaneous temperature component and the heat storage temperature component, using preset distribution parameters. The collaborative control strategy generation module (105) is used to generate a collaborative control strategy for heat dissipation and waste heat recovery based on the heat flow distribution ratio.

10. A computer-readable storage medium storing a computer program capable of being loaded by a processor and executing a synergistic control method for immersion liquid cooling and waste heat recovery as described in any one of claims 1 to 8.

Images