Data center liquid cooling global optimization control method and device

Abstract

The invention relates to a data center liquid cooling global optimization control method and device. The method comprises the following steps: S1, acquiring operation data of a primary side and a secondary side of the data center liquid cooling system, and preprocessing the operation data; and S2, training a load prediction model of the liquid cooling system based on the preprocessed operation data. Moreover, the energy consumption models of the equipment at the primary side and the secondary side are correspondingly trained based on the preprocessed operation data, and a total energy consumption model of the system is obtained according to the trained energy consumption models of the equipment. And S3, based on the load prediction model and the system total energy consumption model, calculating total energy consumption values of different operation parameter combinations under different load conditions according to an optimization algorithm, and determining the operation parameter combination corresponding to the lowest total energy consumption value as an optimal operation parameter combination. According to the invention, the data of the primary side and the secondary side can be uniformly processed, and the energy-saving potential of the liquid cooling system can be maximized based on load prediction and total energy consumption calculation of the system.

Description

Technical Field

[0001] The present invention relates to the field of data center control technology, and in particular to a data center liquid cooling global optimization control method and device. Background Art

[0002] Liquid cooling technology uses liquid as a heat exchange medium to perform heat exchange and cooling near the heat source. It does not need indirect cooling through air like air cooling. Since liquid has a relatively high specific heat capacity, its cooling effect and energy efficiency are much higher than air cooling. It has obvious energy-saving advantages in high-density, large-scale and high-heat dissipation demand data centers. At present, with the continuous emergence of diversified high-computing power demand scenarios such as high-tech, digital transformation and terminal consumption, the number of data centers using liquid cooling technology has gradually increased and has achieved rapid development. However, the current control of the liquid cooling system cannot ensure the lowest operating energy consumption of the entire liquid cooling system due to the independent closed control of the primary and secondary sides. On the other hand, due to the large inertia and large hysteresis of the liquid cooling system, it cannot quickly respond to the rapid changes in the load of the data center, and the system is prone to delayed cooling and lagging cooling. This limits the performance of the liquid cooling system. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a data center liquid cooling global optimization control method and device.

[0004] The technical solution adopted by the present invention to solve the technical problem is: a global optimization control method for liquid cooling in a data center, the method comprising the following steps:

[0005] S1. Obtaining operation data of the primary side and the secondary side of the liquid cooling system of the data center, and preprocessing the operation data;

[0006] S2. Training a load prediction model of the liquid cooling system based on the preprocessed operation data; and, training energy consumption models of each device on the primary side and the secondary side based on the preprocessed operation data, and obtaining a total energy consumption model of the system based on the trained energy consumption models of each device;

[0007] S3. Based on the load prediction model and the system total energy consumption model, calculate the total energy consumption values ​​of different operating parameter combinations under different load conditions according to the optimization algorithm, and determine the corresponding operating parameter combination with the lowest total energy consumption value as the optimal operating parameter combination.

[0008] Further, in the data center liquid cooling global optimization control method described in the present invention, in step S1, the preprocessing includes at least one of filling in part of the missing data by interpolation, replacing data whose error exceeds a preset error range by approximate environmental data, eliminating data whose missing time period exceeds a preset allowable time period, and smoothing the data; or

[0009] The method further includes:

[0010] S4. Sending and setting the optimal parameter combination to the field controllers of the corresponding primary and secondary side devices.

[0011] Further, in the data center liquid cooling global optimization control method of the present invention, the load prediction model is:

[0012] Q predict_t+T ,Q predict_t+2T ,……,Q predict_t+mT =f(t otd ,RH otd ,T servers ,P servers ,Q t ,Q t-T ,Q t-2T ,Q t-3T ,Q t-4T ,……Q t-nT )

[0013] Among them, P servers =P servers_total ·R servers , P servers_total is the total power of the liquid cooling cabinet at full load, R servers is the planned shelf rate of liquid-cooled cabinet servers, Q t is the current load, Q t-nT is the load at n·T moments before the current moment, Q predict_t+mT is the predicted load at m·T moments in the future from the current moment, T is the load prediction control time period constant, t otd is the outdoor temperature, RH otd is the outdoor relative humidity, T servers Plan the shelf time for liquid-cooled servers, n = 1, 2, 3, ..., m = 1, 2, 3, ...

[0014] Further, in the data center liquid cooling global optimization control method of the present invention, in step S2, obtaining the system total energy consumption model according to the trained energy consumption models of each device includes:

[0015] The energy consumption models of the various devices are added together to obtain the total energy consumption model of the system.

[0016] Further, in the data center liquid cooling global optimization control method of the present invention, the optimization algorithm includes preset constraints, and step S3 includes:

[0017] S311, using the system total energy consumption model as the objective function of the optimization algorithm, and using the load value at a future moment predicted by the load prediction model as the load condition of the optimization algorithm;

[0018] S312, determining the parameter setting value of each device that meets the preset constraint condition, and forming a plurality of different operating parameter combinations; each operating parameter combination includes the parameter setting value of each device on the primary side and the secondary side;

[0019] S313. Calculate the total energy consumption values ​​of different operating parameter combinations under different load conditions, and determine the corresponding operating parameter combination with the lowest total energy consumption value as the optimal operating parameter combination.

[0020] Furthermore, in the data center liquid cooling global optimization control method of the present invention, the method further includes:

[0021] The load change rate is calculated based on the predicted load value at the future time obtained by the load prediction model;

[0022] It is determined whether the load change rate is within a preset range. If so, it is allowed to change the operating parameter setting value of the data center liquid cooling system.

[0023] Furthermore, in the data center liquid cooling global optimization control method of the present invention, the method further includes:

[0024] Preset multiple security activation events;

[0025] When a certain safety activation event occurs, the current optimization situation is determined to be abnormal. At this time, the optimal operating parameter combination corresponding to the most recent operation without abnormality is found from the historical data of the optimization calculation, and it is sent down for execution to the corresponding equipment of the liquid cooling system, or the preset backup operating parameter combination is sent down for execution to the corresponding equipment of the liquid cooling system.

[0026] Further, in the data center liquid cooling global optimization control method of the present invention, the safety assurance activation event includes safety assurance activation event 1:

[0027] Real-time monitoring of the deviation rate between the current total energy consumption value of the liquid cooling system and the total energy consumption value predicted at the last moment, wherein the duration for which the deviation rate is greater than the preset threshold value exceeds a first threshold value; and / or

[0028] The security activation event includes security activation event 2:

[0029] Monitoring the actual usage time of the optimization algorithm optimization calculation, where the actual usage time exceeds a second threshold; and / or

[0030] The safety assurance activation event includes safety assurance activation event 3:

[0031] A comprehensive failure occurs in the building automation control system used to monitor the primary side equipment of the liquid cooling system, or a comprehensive failure occurs in the dynamic environment system used to monitor the secondary side equipment of the liquid cooling system; and / or

[0032] The safety assurance activation event includes safety assurance activation event 4:

[0033] A preset emergency event occurred in the liquid cooling system.

[0034] Furthermore, in the data center liquid cooling global optimization control method of the present invention, the method further includes:

[0035] Monitor the accuracy of the load forecasting model in real time, and automatically retrain and re-deploy if the model accuracy does not meet the set requirements for a period of time exceeding a set period; and / or

[0036] The accuracy of the energy consumption model of each device is monitored in real time. If the model accuracy does not meet the set requirements for a period of time exceeding the set time, it will be automatically retrained and deployed.

[0037] In addition, the present invention also provides a data center liquid cooling global optimization control device, comprising:

[0038] The first communication module and the second communication module are used to receive the operation data transmitted from the primary side monitoring system and the secondary side monitoring system respectively;

[0039] Multiple data processing modules, used for preprocessing corresponding operation data;

[0040] A model training module is used to train a load prediction model of a liquid cooling system based on preprocessed operation data; and, based on the preprocessed operation data, to train energy consumption models of each device on the primary side and the secondary side, and to obtain a total energy consumption model of the system based on the trained energy consumption models of each device;

[0041] The optimization calculation module is used to calculate the total energy consumption values ​​of different operating parameter combinations under different load conditions based on the load prediction model and the system total energy consumption model according to the optimization algorithm, and determine the corresponding operating parameter combination with the lowest total energy consumption value as the optimal operating parameter combination.

[0042] The implementation of the data center liquid cooling global optimization control method and device thereof of the present invention has the following beneficial effects: the present invention can uniformly process the data on the primary and secondary sides, and based on load forecasting and system total energy consumption calculation, obtain the total energy consumption corresponding to different operating parameter combinations under different load conditions at future moments predicted in advance, thereby finding the operating parameter combination with the lowest total system energy consumption, which is conducive to solving the system's delayed cooling and lagging cooling defects, quickly responding to rapid changes in data center load, and maximizing the energy-saving potential of the liquid cooling system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0044] Figure 1 It is a flow chart of a global optimization control method for liquid cooling of a data center provided by an embodiment of the present invention;

[0045] Figure 2 is a schematic diagram of a data center liquid cooling system provided by an embodiment of the present invention;

[0046] Figure 3 It is a flow chart of a global optimization control method for liquid cooling of a data center provided by an embodiment of the present invention;

[0047] Figure 4 It is a flow chart of a global optimization control method for liquid cooling of a data center provided by an embodiment of the present invention;

[0048] Figure 5 It is a structural schematic diagram of a data center liquid cooling global optimization control device provided by an embodiment of the present invention;

[0049] Figure 6 It is a partial structural diagram of a data center liquid cooling global optimization control device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, the specific implementation methods of the present invention are now described in detail with reference to the accompanying drawings. In the following description, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed", "set" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral body; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. When an element is referred to as being "on" or "under" another element, the element can be "directly" or "indirectly" located on the other element, or there may be one or more intermediate elements. The terms "first", "second", "third", etc. are only for the convenience of describing the present technical solution, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first", "second", "third", etc. can explicitly or implicitly include one or more of the features. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0051] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0052] At present, with the continuous emergence of diversified high-computing demand scenarios such as high-tech, digital transformation and terminal consumption, the number of data centers using liquid cooling technology has gradually increased and has achieved rapid development. However, the control of the current liquid cooling system has the following two main problems that limit the performance of the liquid cooling system:

[0053] 1) The liquid cooling system consists of a primary side and a secondary side system, where the primary side is monitored by a building automation control system, also known as a BA system, and the secondary side is monitored by a power and environmental monitoring system, also known as a dynamic environment system. These two systems are usually developed and deployed by different manufacturers, and the cooling water temperature provided by the outdoor cooling equipment on the primary side to the secondary side will significantly affect the heat exchange efficiency of the liquid cooling terminal. The independent closed control of the primary and secondary sides cannot ensure the lowest energy consumption of the entire liquid cooling system.

[0054] 2) Due to the large inertia and large hysteresis of the liquid cooling system, it is unable to respond quickly to rapid changes in the data center load, and the system is prone to delayed cooling and lagging cooling.

[0055] In a preferred embodiment, reference Figure 1 The data center liquid cooling global optimization control method of this embodiment includes the following steps:

[0056] S1. Obtain the operating data of the primary and secondary sides of the data center liquid cooling system and pre-process the operating data. Figure 2As shown, it is a schematic diagram of a typical data center liquid cooling system. The primary side is composed of outdoor cooling equipment, a cooling ring network and a cooling water pump. The secondary side is composed of a CDU, a refrigeration ring network and a liquid cooling cabinet. Among them, the Cooling Distribution Units (CDU) is a thermal management device based on liquid cooling technology. Its core task is to transfer the heat generated in the data center to the cooling medium, and then dissipate the heat of the cooling medium through the cooling system. The CDU consists of a plate changer and a chilled water pump (some CDUs also add a backup water pump as a redundant configuration). It can be understood that the acquired operating data can include at least two parts. The first part of the data is the BA system data that can be obtained through the communication module 1, that is, the operating data of the primary side of the liquid cooling system. The second part of the data is the dynamic ring system data that can be obtained through the communication module 2, that is, the operating data of the secondary side of the liquid cooling system. Exemplarily, the operating data on the primary side of the liquid cooling system may include:

[0057] 1) Outdoor temperature t otd , Outdoor relative humidity RH otd .

[0058] 2) Inlet water temperature of each cooling water pump t cw_pump_in , Cooling water pump outlet temperature t cw_pump_out , cooling water pump frequency f cw_pump , Cooling water pump operating power P cw_pump .

[0059] 3) Water inlet temperature of each outdoor cooling equipment t w_odu_in , outdoor cooling equipment outlet water temperature t w_odu_out , outdoor cooling equipment fan frequency f odu_fan , Outdoor cooling equipment inlet wind speed v a_odu_fan , Outdoor cooling equipment fan operating power P odu_fan .

[0060] 4) BA system comprehensive fault E1 (E1=0 represents the normal operation of the BA system, and E1=1 represents the existence of a comprehensive fault in the BA system).

[0061] The operating data of the secondary side of the liquid cooling system can include:

[0062] 1) The inlet water temperature of the plate exchanger on the chilled side of each CDU is t chw_exch_in , Plate exchanger refrigeration side water outlet temperature t chw_exch_out , Plate exchange cooling side water inlet temperature t cw_exch_in , Plate exchange cooling side water outlet temperature t cw_exch_out , Plate exchanger cooling side water flow m chw_exch , Plate exchange cooling side water flow m cw_exch .

[0063] 2) The chilled water pump inlet temperature t in each CDUchw_pump_in , Chilled water pump outlet temperature t chw_pump_out , chilled water pump frequency f chw_pump , Chilled water pump operating power P chw_pump .

[0064] 3) Water inlet temperature of each liquid cooling cabinet t w_lqc_in , Liquid cooling cabinet water outlet temperature t w_lqc_out , Liquid cooling cabinet water flow m w_lqc、 Liquid cooling cabinet server surface temperature t surface .

[0065] 4) Liquid cooling cabinet full load total power P servers_total ,Liquid Cooling Cabinet Server Shelf Plan Tab servers , which consists of two parameters (planned shelf rate R servers , Planned listing time T servers ).

[0066] 5) Comprehensive fault E2 of the dynamic environment system (E2=0 represents the normal operation of the dynamic environment system, and E2=1 represents the existence of a comprehensive fault in the dynamic environment system).

[0067] In some embodiments, if the corresponding algorithm model still lacks the necessary operating data, the operating data also includes a third part of data, and other digital or analog inputs (third part of data) can be obtained through the acquisition module 3 of the device of the present invention. Exemplarily, if the sensor data necessary for the corresponding algorithm model is lacking, the corresponding sensor needs to be added on site. For example, if there is a lack of outdoor humidity data, a new humidity sensor can be added, etc., especially if some sensors in the partially renovated data center are missing or damaged, the corresponding sensors can be re-installed on site and connected to the acquisition module 3 of the device of the present invention. That is, the data detected by the newly added sensor is obtained through the acquisition module 3. If the shelf plan Tab of the liquid-cooled cabinet server cannot be obtained from the dynamic environment system servers Parameters, which exist in some data center systems and can only be obtained from the data center IT monitoring system. The IT monitoring system needs to calculate the planned shelf rate R servers and planned launch time T servers Converted into digital or analog output, and connected to the acquisition module 3 by hard wiring. These two parameters are of great significance to the improvement of the accuracy and response speed of the load forecasting model, which will be described in detail below.

[0068] It can be understood that preprocessing includes using interpolation to fill in some missing data, using approximate environmental data to replace data with errors exceeding a preset error range, eliminating data with missing time periods exceeding a preset allowable time period, smoothing data, etc. In other words, the preprocessing method can include using interpolation to fill in some missing data, using approximate environmental data to replace environmental data with obvious errors, eliminating data with too many missing time periods for measuring point data, and appropriate smoothing, etc.

[0069] S2. Training a load prediction model of the liquid cooling system based on the preprocessed operation data. Moreover, training energy consumption models of each device on the primary side and the secondary side based on the preprocessed operation data, and obtaining a total energy consumption model of the system based on the trained energy consumption models of each device.

[0070] Regarding the training of the load forecasting model, the preprocessed operating data can be randomly divided into a training data set and a validation data set, wherein the default ratio of the training data set to the validation data set can be 8:2, which can be adjusted through the data set segmentation ratio parameter. The training data set is used to train the equipment model, and the algorithm used for training is not specifically limited, and reference can be made to relevant technologies.

[0071] It should be noted that, as the data center changes in business demand during the entire cycle, its server shelf rate may change significantly. The commonly used load algorithm model does not introduce relevant parameters, so it cannot respond to this change quickly, resulting in a large deviation in the prediction result, and affecting the entire control method, causing system oscillation. In the load prediction model of the present invention, two key parameters are introduced for the first time: the planned shelf rate R of the liquid cooling cabinet server and the planned shelf rate R of the liquid cooling cabinet server. servers The planned launch time of the liquid cooling cabinet server is T servers , the specific load forecasting model is as follows:

[0072] Q predict_t+T ,Q predict_t+2T ,……,Q predict_t+mT =f(t otd ,RH otd ,T servers ,P servers ,Q t ,Q t-T ,Q t-2T ,Q t-3T ,Q t-4T ,……Q t-nT )

[0073] Among them, P servers =P servers_total ·R servers , P servers_total is the total power of the liquid cooling cabinet at full load, Rservers is the planned shelf rate of liquid-cooled cabinet servers, Q t is the current load, Q t-nT is the load at n·T moments before the current moment, n=1,2,3,..., the recommended value of n is 24. Q predict_t+mT is the predicted load at m·T moments in the future from the current moment, m=1,2,3,…, the recommended value of m is 3. T is the load prediction control time cycle constant, which is usually determined by the scale of the data center liquid cooling system. otd is the outdoor temperature, RH otd is the outdoor relative humidity, T servers Plan shelf time for liquid-cooled servers.

[0074] The accuracy of the model is verified using the validation data set, and the trained model is deployed after it meets the accuracy requirements. The accuracy of the load forecasting model module is monitored in real time, and when the model accuracy continues to fail to meet the set requirements, it is automatically retrained and deployed.

[0075] In this embodiment, the load forecasting model introduces the planned shelf rate R of the liquid cooling cabinet server servers The planned launch time of the liquid cooling cabinet server is T servers Parameters enable the algorithm to quickly respond to significant changes in server shelf rates, greatly improving the accuracy of load forecasting.

[0076] Regarding the training of each device energy consumption model, the data can be randomly divided into a training data set and a validation data set, wherein the default ratio of the training data set to the validation data set is 8:2, which can be adjusted by the data set segmentation ratio parameter. The training data set is used to train the device model, and the algorithm used for training is not specifically limited, and reference can be made to the relevant technology. Exemplarily, the device energy consumption model of the present invention includes:

[0077] Energy consumption model and heat exchange model of outdoor cooling unit:

[0078] P odu_fan_predict =f(t w_odu_in ,t w_odu_out ,t otd ,RH otd ,v a_odu_fan ,f odu_fan )

[0079] Q odu_fan_predict =f(t w_odu_in ,t w_odu_out ,t otd ,RH otd ,v a_odu_fan ,f odu_fan ,)

[0080] If there is no missing data or no corresponding sensor, v a_odu_fan No need,t w_odu_in Can be obtained by cw_exch_out Instead, w_odu_out Can be obtained by cw_exch_in replace.

[0081] Energy consumption model of cooling water pump:

[0082] P cw_pump_predict =f(t cw_pump_in ,t cw_pump_out ,m cw_exch ,f cw_pump )

[0083] If there is no missing data or no corresponding sensor, t cw_pump_in Can be obtained by cw_exch_out Instead of; cw_pump_out Can be obtained by cw_exch_in replace.

[0084] Heat transfer model of plate exchanger:

[0085] ε exch_predict =f(t chw_exch_in ,t chw_exch_out ,t cw_exch_in ,t cw_exch_out ,m chw_exch ,m cw_exch )

[0086] Among them, ε exch_predict is the heat transfer efficiency of the plate exchanger.

[0087] Energy consumption model of chilled water pump:

[0088] P chw_pump_predict =f(t chw_pump_in ,t chw_pump_out ,m chw_exch ,f chw_pump )

[0089] If there is no missing data or no corresponding sensor, t chw_pump_in Can be obtained by chw_exch_out Instead of; chw_pump_out Can be obtained by chw_exch_in replace.

[0090] Liquid cooling heat exchange model:

[0091] Q lqc_predict =f(t chw_pac_in ,t chw_pac_out ,m w_lqc ,t surface )

[0092] If there is no missing data or no corresponding sensor, mw_lqc It can be approximately calculated by the following formula:

[0093] m w_lqc =Σ i NCDU ·m chw_exch_i / N lqc_run

[0094] Where i is the number of the CDU, N lqc_run is the number of liquid cooling counters in operation, N CDU is the number of CDUs.

[0095] Use the validation data set to verify the accuracy of the model, and deploy the trained model after it meets the accuracy requirements. Monitor the accuracy of the device model in real time, and automatically retrain and deploy when the model accuracy continues to fail to meet the set requirements.

[0096] In step S2, obtaining the total energy consumption model of the system according to the trained energy consumption models of each device includes: adding the energy consumption models of each device to obtain the total energy consumption model of the system.

[0097] S3. Based on the load prediction model and the system total energy consumption model, the total energy consumption values ​​of different operating parameter combinations under different load conditions are calculated according to the optimization algorithm, and the corresponding operating parameter combination with the lowest total energy consumption value is determined as the optimal operating parameter combination. This embodiment uniformly processes the data on the primary and secondary sides, and based on the load prediction model and the system total energy consumption model, calculates the total energy consumption corresponding to different operating parameter combinations under different load conditions at the predicted future moment, thereby finding the operating parameter combination with the lowest total energy consumption of the system, which is conducive to solving the system's delayed cooling and lagging cooling defects, quickly responding to the rapid changes in the data center load, and maximizing the energy-saving potential of the liquid cooling system.

[0098] Based on the above embodiments, reference Figure 3 , the optimization algorithm includes preset constraints, and step S3 includes: S311, taking the system total energy consumption model as the objective function of the optimization algorithm, and taking the load value at the future moment predicted by the load prediction model as the load condition of the optimization algorithm. S312, determining the parameter setting values ​​of each device that meets the preset constraints, and forming multiple groups of different operating parameter combinations. Each group of operating parameter combinations contains parameter setting values ​​for each device on the primary side and the secondary side. S313, calculating the total energy consumption values ​​of different operating parameter combinations under different load conditions, and determining the corresponding operating parameter combination with the lowest total energy consumption value as the optimal operating parameter combination.

[0099] Exemplarily, the total energy consumption model of the liquid cooling system is:

[0100]

[0101] Among them, the specific constraints of the optimization algorithm are as follows:

[0102] f chw_pump_min ≤f chw_pump ≤f chw_pump_max

[0103] f cw_pump_min ≤f cw_pump ≤f cw_pump_max

[0104] f odu_fan_min ≤f odu_fan ≤f odu_fan_max

[0105] t surface_min ≤t surface ≤t surface_max

[0106] N CDU_min ≤N CDU ≤N CDU_max

[0107] N odu_min ≤N odu ≤N odu_max

[0108] Q CDU_chw =m chw_exch ·(t chw_exch_in -t chw_exch_out )

[0109]

[0110] The optimization algorithm is used to calculate the operating parameter combination setting value at the lowest total energy consumption point under different load conditions:

[0111] 1) Number of outdoor cooling equipment opened N odu .

[0112] 2) Number of CDUs turned on N CDU .

[0113] 3) Chilled water pump frequency f chw_pump The setting value, unit is HZ.

[0114] 4) Cooling water pump frequency f cw_pump The setting value, unit is HZ.

[0115] 5) Fan frequency f of outdoor cooling equipment odu_fan , unit is HZ.

[0116] The upper and lower limits of the specific constraints of the optimization algorithm (ie, the optimization algorithm) can be set by the operation and maintenance personnel and synchronized to the data integration module 14. The specific optimization algorithm is not limited, and reference can be made to related technologies.

[0117] In this step, a specific cooling load is input to obtain the combination of operating parameter settings that minimize the energy consumption of the liquid cooling system under this cooling load condition [N odu ,N ct_fan ,f chw_pump ,f cw_pump ,f odu_fan 】.

[0118] In some embodiments, reference Figure 4 , the method further includes: S4, issuing and setting the optimal parameter combination to the field controllers of the corresponding primary and secondary side devices. That is to say, this embodiment trains and deploys the data center cooling load prediction model, and trains and deploys the energy consumption and heat exchange models of each primary and secondary side device through the operation data of the primary and secondary side devices of the liquid cooling system obtained. The optimal operating parameter combination that can minimize the overall energy consumption of the liquid cooling system under different load conditions is determined by the optimization algorithm, and is set to the field controllers of each primary and secondary side device.

[0119] In this embodiment, by uniformly processing the data on the primary and secondary sides, and based on load forecasting and system total energy consumption calculation, the total energy consumption corresponding to different operating parameter combinations under different load conditions in the future can be predicted in advance, thereby finding the operating parameter combination with the lowest total energy consumption of the system, that is, the optimal operating parameter combination. The optimal operating parameter combination at the future moment or the next moment is set to the corresponding field controller of each device, thereby effectively solving the system delayed cooling and lagging cooling defects caused by the large inertia and large lag characteristics of the liquid cooling system, quickly responding to the rapid changes in the data center load, and thus maximizing the energy-saving potential of the liquid cooling system.

[0120] In some embodiments, the method further includes: calculating a load change rate based on a predicted load value at a future time obtained by a load prediction model, and determining whether the load change rate is within a preset range, and if so, allowing the operating parameter setting value of the data center liquid cooling system to be changed.

[0121] It can be understood that in order to avoid system instability caused by frequent changes in the combination of operating parameters, that is, when the load changes are too small or too large, the risk of system oscillation caused by changing the combination of operating parameters is greater than the energy saving benefit it brings. predict_t+T Calculate the load change rate R at the future time T Q , the calculation formula is as follows:

[0122] RQ =(Q predict_t+T -Q t ) / Q t

[0123] If R Q_min ≤R Q ≤R Q_max , then the operating parameter setting value can be changed, that is, the trigger signal Chg sets =1, if not, then Chg sets =0.

[0124] For example, in multiple project cases that are actually running, R Q_min and R Q_max Can be set to R Q 5% and 40% can achieve better results.

[0125] In some embodiments, the method further includes: presetting multiple safety assurance activation events. When a certain safety assurance activation event occurs, it is determined that the current optimization situation is abnormal. At this time, the optimal operating parameter combination corresponding to the most recent operation without abnormality is searched from the historical data of the optimization calculation, and it is sent down for execution to the corresponding device of the liquid cooling system, or the preset standby operating parameter combination is sent down for execution to the corresponding device of the liquid cooling system. In other words, when the following safety assurance activation events occur, it is determined that the current optimization situation is abnormal, the optimization calculation process in the optimization calculation module 8 is jumped out in an emergency, and it is rolled back to the correct parameter setting combination without abnormality in the last operation or a default expert preset parameter setting combination is executed (this combination can be set and modified by the operation and maintenance expert). This embodiment can effectively handle the optimization control situation with abnormalities and ensure the stability of the operation of the liquid cooling system.

[0126] Specifically, the security activation events may include:

[0127] Safety assurance activation event 1: real-time monitoring of the deviation rate between the current total energy consumption value of the liquid cooling system and the total energy consumption value predicted at the last moment, the deviation rate is greater than the preset threshold for a duration exceeding the first threshold. In other words, the real-time monitoring of the deviation rate between the current total energy consumption value of the liquid cooling system and the total energy consumption value predicted at the last moment, when the deviation rate is continuously greater than the set threshold (usually set to 4%) for more than the set time (usually set to 10 times the data collection cycle), the event is activated.

[0128] Safety assurance activation event 2: monitor the actual usage time of the optimization algorithm optimization calculation, and the actual usage time exceeds the second threshold. That is, when the optimization calculation module 8 fails, that is, when the time of the real-time optimization calculation exceeds the set threshold (usually set to 2 minutes), this event is activated.

[0129] Safety assurance activation event 3: A comprehensive fault occurs in the building automation control system used to monitor the primary side equipment of the liquid cooling system, or a comprehensive fault occurs in the dynamic environment system used to monitor the secondary side equipment of the liquid cooling system. In other words, if the BA system comprehensive fault E1 = 1 or the dynamic environment system comprehensive fault E2 = 1, this event is activated.

[0130] Safety assurance activation event 4: A preset emergency event occurs in the liquid cooling system. In other words, this event is activated when another preset emergency event occurs in the liquid cooling system.

[0131] This embodiment can maximize the energy-saving potential of the liquid cooling system by uniformly processing the data on the primary side and the secondary side based on load prediction and system total energy consumption calculation.

[0132] In another preferred embodiment, the data center liquid cooling global optimization control device of this embodiment includes: a first communication module and a second communication module, which are used to receive the operation data transmitted from the primary side monitoring system and the secondary side monitoring system respectively. A plurality of data processing modules are used to preprocess the corresponding operation data. A model training module is used to train the load prediction model of the liquid cooling system based on the preprocessed operation data. Moreover, the energy consumption models of each device on the primary side and the secondary side are correspondingly trained based on the preprocessed operation data, and the total energy consumption model of the system is obtained based on the trained energy consumption models of each device. An optimization calculation module is used to calculate the total energy consumption values ​​of different operating parameter combinations under different load conditions according to the optimization algorithm based on the load prediction model and the total energy consumption model of the system, and determine the corresponding operating parameter combination with the lowest total energy consumption value as the optimal operating parameter combination.

[0133] This embodiment uniformly processes the data on the primary and secondary sides, and based on load forecasting and calculation of the total system energy consumption, obtains the total energy consumption corresponding to different operating parameter combinations under different load conditions at future moments predicted in advance, thereby finding the operating parameter combination with the lowest total system energy consumption. This is beneficial to solving the defects of delayed cooling and lagging cooling in the system, and quickly responds to rapid changes in the data center load, thereby obtaining the optimal operating parameter combination, which can maximize the energy-saving potential of the liquid cooling system.

[0134] In a specific embodiment, the device can obtain the operating data of the primary and secondary side devices of the liquid cooling system through the built-in communication module. The built-in communication module can perform tasks such as data processing, model training and algorithm operation. Figure 5As shown, the device includes 15 modules, namely: communication module 1, communication module 2, acquisition module 3, primary side data processing module 4, secondary side data processing module 5, acquisition data processing module 6, safety assurance module 7, optimization calculation module 8, equipment model module 9, optimization judgment module 10, equipment model training module 11, load prediction model module 12, load prediction model training module 13, data integration module 14, and communication module 15.

[0135] The primary side data processing module 4 processes the primary side operation data of the liquid cooling system obtained by the communication module 1, and the processing methods include using interpolation to fill in part of the missing data, replacing environmental data with obvious errors with approximate environmental data, eliminating data in time periods where the measurement point data is missing too much, and appropriate smoothing, etc. The secondary side data processing module 5 processes the secondary side operation data of the liquid cooling system obtained by the communication module 2, and the processing methods include using interpolation to fill in part of the missing data, replacing environmental data with obvious errors with approximate environmental data, eliminating data in time periods where the measurement point data is missing too much, and appropriate smoothing, etc. The acquisition data processing module 6 processes the data obtained by the acquisition module 3, and the processing methods include using interpolation to fill in part of the missing data, replacing environmental data with obvious errors with approximate environmental data, eliminating data in time periods where the measurement point data is missing too much, and appropriate smoothing, etc.

[0136] The data integration module 14 is used to store and integrate pre-processed data, data set segmentation ratio parameters related to model training, and hyperparameters involved in the algorithm used for training. That is to say, the optimized data are transmitted and synchronized to the data integration module 14 respectively, and the data set segmentation ratio parameters related to model training and the hyperparameters involved in the algorithm used for training are also synchronized to the data integration module 14. It can be understood that the data integration module 14 uploads the relevant operating data obtained by the primary side data processing module 4, the secondary side data processing module 5 and the acquisition data processing module 6 to the load forecasting model training module 13 for model training, and uploads it to the equipment model training module 11 for model training. The load forecasting model module 12 obtains the trained load forecasting model that meets the accuracy requirements from the load forecasting model training module 13 for deployment. Similarly, the equipment model module 9 obtains the trained energy consumption models of each device that meet the accuracy requirements from the equipment model training module 11 for deployment. Figure 6 As shown, it is the specific equipment energy consumption model deployed by the equipment model module 9 and the equipment model specifically trained by the equipment model training module 11.

[0137] The optimization calculation module 8 obtains the energy consumption model of each device on the primary side and the secondary side from the equipment model module 9, and adds them together to obtain the total energy consumption model of the liquid cooling system. The optimization algorithm is used to calculate the operating parameter combination setting value (that is, the optimal operating parameter combination) when the total energy consumption is the lowest under different load conditions. The upper and lower limits of the specific constraints of the optimization algorithm can be set by the operation and maintenance personnel and synchronized to the data integration module 14. The specific optimization algorithm of the optimization calculation module is not limited, and the hyperparameters involved in the algorithm are synchronized to the data integration module 14.

[0138] In this embodiment, unlike the previous control method of separately controlling the primary and secondary side devices of the liquid cooling system, the global optimization control device of the present invention can uniformly process and control the data of the primary and secondary sides to maximize the energy saving potential of the liquid cooling system. In addition, the load forecasting model introduces the planned shelf rate R of the liquid cooling cabinet server servers The planned launch time of the liquid cooling cabinet server is T servers Parameters enable the algorithm to quickly respond to significant changes in server shelf rates, greatly improving the accuracy of load forecasting.

[0139] In order to avoid system instability and oscillation caused by frequent changes in the combination of operating parameters. That is, when the load changes are too small or too large, in order to avoid the risk of system oscillation caused by changing the combination of operating parameters being greater than the energy saving benefits brought by it, this embodiment sets an optimization judgment module 10. The optimization judgment module 10 is based on the data Q output by the load prediction model module 12. predict_t+T Calculate the load change rate R at the future time T Q , if R Q_min ≤R Q ≤R Q_max , then the operating parameter setting value can be changed, that is, the trigger signal Chg sets =1, if not, then Chg sets = 0. In this embodiment, the optimization judgment module 10 effectively avoids frequent changes in the combination settings of the operating parameters, and further improves the safety and stability of the global optimization control method during the operation process.

[0140] When one of several preset safety activation events occurs, the safety module 7 can determine that the current optimization situation is abnormal, and can jump out of the optimization calculation process in the optimization calculation module 8 in an emergency, and roll back to the correct parameter setting combination that ran without abnormality last time, or execute a default expert preset parameter setting combination (this combination can be set and modified by the operation and maintenance expert). In this embodiment, the safety module 7 can effectively handle the optimization control situation with abnormalities, and ensure the stability of the liquid cooling system operation.

[0141] It can be understood that the load value predicted by the load forecasting model module 12 from the future time T to m·T provides forward-looking suggestions to the operation and maintenance personnel and the optimization system. The optimal operating parameter combination at the future time can be sent to the data integration module 14 in advance for standard processing by manual operation or feedforward control or model-based MPC control, and then sent to the BA system and the dynamic environment system through communication module 1 and communication module 2 respectively, and then set to the field controllers of the corresponding primary and secondary side devices respectively. The relevant setting parameters in all modules will be synchronized to the data integration module 14. The data in the data integration module 14 is provided to the northbound device or system for setting according to the privacy authority and actual project requirements.

[0142] It should be noted that in the technical solution of this embodiment: 1) The data preprocessing algorithm in the primary data processing module 4, the secondary data processing module 5 and the acquisition data processing module 6 is not limited to a specific algorithm, and is not specifically limited here. 2) The training algorithm adopted by the equipment model training module 11 and the load forecasting model training module 13 is not limited to a specific algorithm, and is not specifically limited here. 3) The optimization algorithm used in the optimization calculation module 8 is not limited to a specific algorithm, and is not specifically limited. 4) The sensors that need to be supplemented by the acquisition module 3 depend on the specific project, and are not specifically limited here. 5) The relevant setting parameter constants depend on the specific project, and are not specifically limited here. All constraints and configurations of the present invention can be opened to on-site operation and maintenance experts, realizing the controlled freedom and configuration freedom of the control method.

[0143] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0144] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0145] It can be understood that the above embodiments only express the preferred implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the patent scope of the present invention. It should be pointed out that, for ordinary technicians in this field, the above technical features can be freely combined without departing from the concept of the present invention, and several deformations and improvements can be made, which all belong to the protection scope of the present invention. Therefore, all equivalent changes and modifications made to the scope of the claims of the present invention should belong to the coverage of the claims of the present invention.

Claims

1. A global optimization control method for liquid cooling in a data center, characterized in that: The method comprises the following steps: S1. Obtaining operation data of the primary side and the secondary side of the liquid cooling system of the data center, and preprocessing the operation data; S2. Training a load prediction model of the liquid cooling system based on the preprocessed operation data; and, training energy consumption models of each device on the primary side and the secondary side based on the preprocessed operation data, and obtaining a total energy consumption model of the system based on the trained energy consumption models of each device; S3. Based on the load prediction model and the system total energy consumption model, calculate the total energy consumption values ​​of different operating parameter combinations under different load conditions according to the optimization algorithm, and determine the corresponding operating parameter combination with the lowest total energy consumption value as the optimal operating parameter combination.

2. The data center liquid cooling global optimization control method according to claim 1 is characterized in that: In step S1, the preprocessing includes at least one of filling in part of the missing data by interpolation, replacing data with an error exceeding a preset error range by using approximate environmental data, eliminating data with a missing time period exceeding a preset allowable time period, and smoothing the data; or The method further includes: S4. Sending and setting the optimal parameter combination to the field controllers of the corresponding primary and secondary side devices.

3. The data center liquid cooling global optimization control method according to claim 1, characterized in that: The load forecasting model is: Q predict_t+T ,Q predict_t+2T ,……,Q predict_t+mT =f(t otd ,RH otd ,T servers ,P servers ,Q t ,Q t-T ,Q t-2T , Q t-3T ,Q t-4T ,……Q t-nT ) Among them, P servers =P servers_total ·R servers , P servers_total is the total power of the liquid cooling cabinet at full load, R servers is the planned shelf rate of liquid-cooled cabinet servers, Q t is the current load, Q t-nT is the load at n·T moments before the current moment, Q predict_t+mT is the predicted load at m·T moments in the future from the current moment, T is the load prediction control time period constant, t otd is the outdoor temperature, RH otd is the outdoor relative humidity, T servers Plan the shelf time for liquid-cooled servers, n = 1, 2, 3, ..., m = 1, 2, 3, ...

4. The data center liquid cooling global optimization control method according to claim 1, characterized in that: In step S2, obtaining the system total energy consumption model according to the trained energy consumption models of each device includes: The energy consumption models of the various devices are added together to obtain the total energy consumption model of the system.

5. The data center liquid cooling global optimization control method according to claim 4 is characterized in that: The optimization algorithm includes preset constraints, and step S3 includes: S311, using the system total energy consumption model as the objective function of the optimization algorithm, and using the load value at a future moment predicted by the load prediction model as the load condition of the optimization algorithm; S312, determining the parameter setting value of each device that meets the preset constraint condition, and forming a plurality of different operating parameter combinations; each operating parameter combination includes the parameter setting value of each device on the primary side and the secondary side; S313. Calculate the total energy consumption values ​​of different operating parameter combinations under different load conditions, and determine the corresponding operating parameter combination with the lowest total energy consumption value as the optimal operating parameter combination.

6. The data center liquid cooling global optimization control method according to claim 1, characterized in that: The method further includes: The load change rate is calculated based on the predicted load value at the future time obtained by the load prediction model; It is determined whether the load change rate is within a preset range. If so, it is allowed to change the operating parameter setting value of the data center liquid cooling system.

7. The data center liquid cooling global optimization control method according to claim 1, characterized in that: The method further includes: Preset multiple security activation events; When a certain safety activation event occurs, the current optimization situation is determined to be abnormal. At this time, the optimal operating parameter combination corresponding to the most recent operation without abnormality is found from the historical data of the optimization calculation, and it is sent down for execution to the corresponding equipment of the liquid cooling system, or the preset backup operating parameter combination is sent down for execution to the corresponding equipment of the liquid cooling system.

8. The data center liquid cooling global optimization control method according to claim 7, characterized in that: The security activation event includes security activation event 1: Real-time monitoring of the deviation rate between the current total energy consumption value of the liquid cooling system and the total energy consumption value predicted at the last moment, wherein the duration for which the deviation rate is greater than the preset threshold value exceeds a first threshold value; and / or The security activation event includes security activation event 2: Monitoring the actual usage time of the optimization algorithm optimization calculation, where the actual usage time exceeds a second threshold; and / or The safety assurance activation event includes safety assurance activation event 3: A comprehensive failure occurs in the building automation control system used to monitor the primary side equipment of the liquid cooling system, or a comprehensive failure occurs in the dynamic environment system used to monitor the secondary side equipment of the liquid cooling system; and / or The safety assurance activation event includes safety assurance activation event 4: A preset emergency event occurred in the liquid cooling system.

9. The data center liquid cooling global optimization control method according to claim 1, characterized in that: The method further includes: Monitor the accuracy of the load forecasting model in real time, and automatically retrain and re-deploy if the model accuracy does not meet the set requirements for a period of time exceeding a set period; and / or The accuracy of the energy consumption model of each device is monitored in real time. If the model accuracy does not meet the set requirements for a period of time exceeding the set time, it will be automatically retrained and deployed.

10. A data center liquid cooling global optimization control device, characterized in that: include: The first communication module and the second communication module are used to receive the operation data transmitted from the primary side monitoring system and the secondary side monitoring system respectively; Multiple data processing modules, used for preprocessing corresponding operation data; A model training module is used to train a load prediction model of a liquid cooling system based on preprocessed operation data; and, based on the preprocessed operation data, to train energy consumption models of each device on the primary side and the secondary side, and to obtain a total energy consumption model of the system based on the trained energy consumption models of each device; The optimization calculation module is used to calculate the total energy consumption values ​​of different operating parameter combinations under different load conditions based on the load prediction model and the system total energy consumption model according to the optimization algorithm, and determine the corresponding operating parameter combination with the lowest total energy consumption value as the optimal operating parameter combination.

Images