CN-122018573-A - Liquid cooling system self-adaptive flow regulation and control method and device based on AI prediction

CN122018573ACN 122018573 ACN122018573 ACN 122018573ACN-122018573-A

Abstract

The disclosure belongs to the technical field of adaptive control and artificial intelligence, and provides a liquid cooling system adaptive flow regulation and control method and device based on AI prediction, which specifically comprises the following steps: firstly, identifying each flow control site from a liquid cooling system, respectively acquiring record state data, establishing a flow cooling benefit model according to historical state data of each flow control site, then inputting current state data, obtaining a flow cooling benefit array through the flow cooling benefit model, finally calculating cooling gain flow cost of the flow control site through the flow cooling benefit array, and carrying out self-adaptive flow regulation by combining each cooling gain flow cost. The realization degree of effective cooling conversion of each cooling branch in the multi-cooling node liquid cooling system is effectively quantized under the condition that the same unit flow is increased currently, so that the flow utilization effect of each flow control site is evaluated in a forward view angle, and the flow set value of each cooling branch is subjected to discrete step type self-adaptive adjustment.

Inventors

WANG XIAOMING
LI KONGZHENG
CHEN XIAOFENG
HUANG JIARONG

Assignees

广东百德朗科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260403

Claims (9)

1. The liquid cooling system self-adaptive flow regulation and control method based on AI prediction is characterized by comprising the following steps: S100, identifying each flow control site from the liquid cooling system, and respectively collecting and recording state data; S200, establishing a flow cooling benefit model according to historical state data of each flow control site; s300, inputting current state data and obtaining a flow cooling benefit array through a flow cooling benefit model; s400, calculating the cooling gain flow cost of the flow control site through the flow cooling benefit array; s500, carrying out self-adaptive flow regulation and control by combining the flow cost of each cooling gain; The S400 method is that for any flow control site, the maximum value of the opposite number and zero of the temperature amplitude is taken as effective cooling response quantity at each prediction point, the maximum value of all flow amplitude is taken as maximum flow amplitude, the ratio of the flow amplitude to the maximum flow amplitude is normalized flow excitation degree, the product of the flow amplitude and the effective cooling response quantity is taken as coupling response intensity, the ratio of the serial number of the prediction point to the total number of the prediction point is taken as attenuation weight kernel, the coupling response intensity is weighted and averaged to obtain weighted coupling response total quantity, the square of the normalized flow excitation degree is weighted and averaged to obtain nonlinear flow consumption and divergence, and cooling gain flow cost is calculated according to the weighted coupling response total quantity and the nonlinear flow consumption and divergence.
2. The method for adaptively controlling flow of the liquid cooling system based on AI prediction as set forth in claim 1, wherein in step S100, each flow control site is identified from the liquid cooling system, and the recorded status data is collected and recorded by the method comprising the steps of, the liquid cooling system comprising a plurality of flow control sites, each flow control site corresponding to an adjustable cooling branch, the status data being collected and recorded in real time by the flow control sites, wherein the status data comprises at least a flow value, a valve opening, a pump rotation speed, a water inlet temperature and a water outlet temperature of the cooling branch, a heat load power, a pressure or a pressure difference of a service object of the cooling branch, a total water supply temperature and a total water return temperature of the liquid cooling system, a total flow of the liquid cooling system and an environmental temperature of a machine room; The control period TR is preset, and status data is collected from all flow control sites once every interval TR.
3. The method for establishing the flow cooling benefit model according to the historical state data of each flow control site in the step S200 is characterized in that the historical state data sequences are obtained by arranging the state data according to time sequence, the historical state data sequences of all the flow control sites are used as monitoring characteristics; And taking the pair of the monitoring characteristic and the predicted target characteristic as a training sample set, and marking an AI prediction model obtained by training by adopting a machine learning regression algorithm as a flow cooling benefit model.
4. The method for obtaining a flow cooling benefit array by inputting current state data through a flow cooling benefit model in step S300 is to obtain the current flow cooling benefit array by taking the current state data as input of the flow cooling benefit model, wherein the current state data is a set of monitoring characteristics corresponding to all flow control sites in a control period.
5. The method for calculating the cooling gain flow cost of the flow control site through the flow cooling benefit array in the step S400 is characterized in that for any flow control site, the maximum value of the opposite number and zero of the temperature amplitude of each predicted point is recorded as effective cooling response, the maximum value of all the flow amplitudes of the flow control site is recorded as maximum flow amplitude, the ratio of the flow amplitude of the flow control site to the maximum flow amplitude is recorded as normalized flow excitation, the product of the effective cooling response and the normalized flow excitation is recorded as coupling response strength, and the ratio of the serial number value of the predicted point in the monitoring period to the total number of the predicted points in the monitoring period is recorded as attenuation weight core; The method comprises the steps of calculating a weighted average value of all coupling response intensities by taking an attenuation weight core as a weight for any flow control site, marking the weighted average value as a weighted coupling response total amount Qorsd, calculating the weighted average value by taking the attenuation weight core as the weight after squaring all normalized flow excitation degrees, marking the weighted average value as nonlinear flow dissipation Phedr, and calculating the cooling gain flow cost according to the nonlinear flow dissipation and the weighted coupling response total amount for each flow control site of the current prediction point.
6. The adaptive flow regulation method of the liquid cooling system based on AI prediction as set forth in claim 1, wherein in step S400, the method for calculating the cooling gain flow cost of the flow control site by the flow cooling benefit array is to record the number of all the predicted points in the monitoring period as NK, arrange all the flow cooling benefit arrays of any one of the flow control sites in time sequence, fit the curve of the flow amplitude and the temperature amplitude changing with time by using a cubic spline difference, record as the flow response track and the temperature response track respectively; Calculating a value interval corresponding to the temperature amplitude of any flow control site, dividing the value interval into NK/2 uniform intervals, calculating the ratio of the number of the temperature amplitudes in each interval to NK, marking the ratio as temperature response probability, calculating entropy values of probability distribution corresponding to all the temperature response probabilities by using a Shannon entropy formula, and marking the entropy values as response stability entropy values Hents; for any flow control locus, calculating a corresponding delay time distance when a discrete cross correlation function of the flow response locus and the temperature response locus reaches the maximum value, dividing the obtained delay time distance by TR, and rounding down to obtain a lag interval; The method comprises the steps of updating the corresponding flow amplitude of any predicted point into the flow amplitude of the predicted point corresponding to the first lag interval in the reverse time direction, recording the flow amplitude of the predicted point as the reconstructed flow amplitude, converting the temperature amplitude by using a linear rectification function, recording the temperature amplitude as effective cooling efficiency, calculating the absolute value of the ratio of the effective cooling efficiency to the reconstructed flow amplitude as effective marginal efficiency Mefft, recording the average value and the range of all the reconstructed flow amplitudes as the reconstructed flow average value and the reconstructed flow range respectively, recording the ratio of the reconstructed flow amplitude to the reconstructed flow range after the difference is made between the reconstructed flow amplitude and the reconstructed flow average value as the return flow amplitude, recording the cube value of the return flow amplitude as nonlinear flow resistance dissipation potential Ediss, and calculating the cooling gain flow cost according to the nonlinear flow resistance dissipation potential and the effective marginal efficiency at each flow control site of the current predicted point.
7. The method for adaptively controlling flow of a liquid cooling system based on AI prediction as set forth in claim 1, wherein in step S500, the method for adaptively controlling flow in combination with each cooling gain flow cost is to read a current total flow Q_sys of the liquid cooling system and compare the current total flow Q_sys with a preset total flow upper limit Q_max, when Q_sys < Q_max, increase a flow set value for a flow control site with a smaller cooling gain flow cost, and the sum of the flow increases of each flow control site does not exceed the difference between Q_max and Q_sys, when Q_sys is not less than Q_max, decrease the flow set value for the flow control site with a larger cooling gain flow cost, increase the flow set value for the flow control site with a smaller cooling gain flow cost, and the flow decrease and the increase of each flow control site are equal in value.
8. The method for adaptively controlling flow of a liquid cooling system based on AI prediction according to claim 1, wherein in step S500, the method for adaptively controlling flow in combination with the temperature-decreasing gain flow cost is characterized in that the difference between the actual temperature of the service object of each cooling branch and the corresponding target temperature is obtained, the temperature deviation of each cooling branch is obtained, the temperature deviation change rate of each cooling branch in several adjacent control periods is calculated, the maximum value of the temperature deviation change rate is recorded as the cooling demand indication amount, if the cooling demand indication amount is smaller than the first preset threshold value and the temperature deviation change rate is smaller than the second preset threshold value, the flow-decreasing or flow-limiting is performed on the flow control sites with larger cooling gain flow cost and smaller temperature deviation, the slow flow-increasing is performed on the rest flow control sites, and otherwise, the flow-decreasing or current flow-maintaining is performed on the flow control sites with larger temperature deviation and smaller cooling gain flow cost.
9. The liquid cooling system self-adaptive flow regulating device based on the AI prediction is characterized by comprising a processor, a memory and a computer program which is stored in the memory and can run on the processor, wherein the step in the liquid cooling system self-adaptive flow regulating method based on the AI prediction is realized when the processor executes the computer program, and the liquid cooling system self-adaptive flow regulating device based on the AI prediction runs in computing equipment of a desktop computer, a notebook computer, a palm computer and a cloud data center.

Description

Liquid cooling system self-adaptive flow regulation and control method and device based on AI prediction Technical Field The disclosure belongs to the technical field of adaptive control and artificial intelligence, and particularly relates to a liquid cooling system adaptive flow regulation and control method and device based on AI prediction. Background In a data center liquid cooling system, a parallel system consisting of a plurality of cooling branches is generally adopted to meet the heat dissipation requirements of a plurality of cabinets and a plurality of servers, and each branch realizes independent adjustment of local flow through a valve or a branch pump and other adjusting and controlling units. In the operation process of the liquid cooling system, limited cooling capacity needs to be dynamically distributed according to the actual heat dissipation load and the temperature state of each cooling node, so that the temperature control of key components is ensured, and the total energy consumption is reduced as much as possible. However, in the prior engineering practice, the flow distribution of the parallel cooling nodes is often controlled according to temperature feedback, the system generally decides how to adjust the flow of the nodes according to the temperature deviation between the current temperature and the target temperature of each node, and in order to simplify the control logic, the flow adjustment between each node is mostly performed independently or according to preset weights, and the effective comparison of which node has a more cooling effect in the flow adjustment is often lacking. In summary, the existing control method generally only feeds back which nodes have higher temperatures, and cannot judge in advance which node the additionally increased unit flow can obtain higher temperature improvement benefits. Further, the flow temperature response characteristics of each parallel cooling node are extremely unstable due to the fact that the flow temperature response characteristics of each parallel cooling node are often influenced by various time-varying factors, including load distribution differences corresponding to different nodes, chip package and cold plate structure differences, pipeline length and local resistance differences and the like, and in addition, the local heat exchange coefficient and pressure drop characteristics can be changed again due to factors such as scaling or material aging and the like along with the running life of the system. Under the combined action of the factors, even under the same target temperature deviation, the cooling benefits brought by unit flow in the actual application of different nodes at a certain moment often have obvious dynamic differences, and even the cooling capability of the same node at different moments can be obviously changed. If the temperature feedback control of the traditional liquid cooling system cannot identify the inherent thermal inertia and measurement hysteresis of the temperature response at all, because after the flow of a certain node is adjusted, the corresponding changes of the temperature of the cooling medium, the surface temperature of the cold plate and the junction temperature of the chip can be fully displayed after a certain time, the prior art usually indirectly evaluates the effect of the adjustment by observing the temperature change in a period after the completion of the flow adjustment, and then initiates the next round of adjustment according to a new temperature state. The complementary adjusting mode of firstly distributing, observing and then correcting is deployed, and the real-time cooling efficiencies of a plurality of parallel nodes are difficult to transversely compare and optimally distribute in a decision period, so that the limited cooling capacity cannot be preferentially distributed to the nodes with efficiency priority for effective cooling under the current working condition, and the problems of long temperature recovery time, increased local overheat risk, higher energy consumption of the whole pump and the like occur. Therefore, there is a need for a method and a device for adaptive flow control of a liquid cooling system based on AI prediction. Disclosure of Invention The disclosure aims to provide a liquid cooling system self-adaptive flow regulation method and device based on AI prediction, so as to solve one or more technical problems in the prior art, and at least provide a beneficial selection or creation condition. In order to achieve the above object, according to an aspect of the present disclosure, there is provided a liquid cooling system adaptive flow regulation method based on AI prediction, the method including the steps of: S100, identifying each flow control site from the liquid cooling system, and respectively collecting and recording state data; S200, establishing a flow cooling benefit model according to historical state data of each flow control site;