US-12621957-B2 - Fan control method and fan control device for controlling fans using a neural network to process characteristic variables

US12621957B2US 12621957 B2US12621957 B2US 12621957B2US-12621957-B2

Abstract

A fan control method for controlling a set of fans of a system includes collecting M first sets of characteristic variables of a first period; inputting the M first sets of characteristic variables to a neural network to generate N third sets of characteristic variables of a second period corresponding to a second set of characteristic variables; adjusting the second set of characteristic variables to generate P adjusted second sets of characteristic variables to accordingly generate Q adjusted third sets of characteristic variables; generating an optimized second set of characteristic variables according to the N third sets of characteristic variables and the Q adjusted third sets of characteristic variables; generating a set of weights according to the optimized second set of characteristic variables; and controlling the set of fans according to the set of weights. The first period precedes the second period. M, N, P, Q are positive integers.

Inventors

Chien-Ming Lee
Kai-Yang Tung
Hsin-Cheng Chu

Assignees

Inventec (Pudong) Technology Corp.
INVENTEC CORPORATION

Dates

Publication Date: 20260505
Application Date: 20221209
Priority Date: 20220610

Claims (10)

1 . A fan control method for controlling a set of fans of a system, comprising: collecting M first sets of characteristic variables of a first period; inputting the M first sets of characteristic variables to a neural network to generate N third sets of characteristic variables of a second period corresponding to a second set of characteristic variables; adjusting the second set of characteristic variables to generate P adjusted second sets of characteristic variables according to a non-minimum adjustment value of the second set of characteristic variables, so as to accordingly generate Q adjusted third sets of characteristic variables; generating an optimized second set of characteristic variables according to the N third sets of characteristic variables and the Q adjusted third sets of characteristic variables, wherein generating the optimized second set of characteristic variables comprises: generating a set of sums of absolute values according to differences of a predetermined value and each third set of characteristic variables of the N third sets of characteristic variables and the Q adjusted third sets of characteristic variables; selecting a plurality of sets of third set of characteristic variables from the N third sets of characteristic variables and the Q adjusted third sets of characteristic variables; and selecting a plurality of second sets of characteristic variables corresponding to the plurality of sets of third set of characteristic variables, wherein the plurality of second sets of characteristic variables are of a subset of the second set of characteristic variables and the P adjusted second sets of characteristic variables; and generating the optimized second set of characteristic variables according to at least the plurality of second sets of characteristic variables; generating a set of weights according to the optimized second set of characteristic variables; and controlling the set of fans according to the set of weights; wherein the first period precedes the second period, each first set of characteristic variables comprises a second set of characteristic variables and a third set of characteristic variables, M, N, P, Q are positive integers.
2 . The fan control method of claim 1 , wherein: each first set of characteristic variables comprises a power load of the system, fan speeds of the set of fans, and a first set of temperatures generated by measuring the system; each second set of characteristic variables comprises the fan speeds of the set of fans; each third set of characteristic variables comprises a second set of temperatures generated by measuring the system; and the first set of temperatures comprises the second set of temperatures.
3 . The fan control method of claim 1 , wherein the first period and the second period are equal in length.
4 . The fan control method of claim 1 , wherein: the plurality of sets of third set of characteristic variables are corresponding to lowest x % of the set of sums of absolute values, and 0<x<100; and generating the optimized second set of characteristic variables according to at least the plurality of second sets of characteristic variables comprises: generating the optimized second set of characteristic variables according to the plurality of second sets of characteristic variables and a minimum adjustment value of the second set of characteristic variables.
5 . The fan control method of claim 1 , wherein the neural network comprises 512 encoding neurons and 256 decoding neurons.
6 . The fan control method of claim 1 , wherein each first set of characteristic variables comprises a set of temperatures generated by measuring the system, and the set of temperatures comprises a first processor temperature, a second processor temperature, a first bus card temperature, a second bus card temperature, a third bus card temperature and an inlet temperature.
7 . The fan control method of claim 1 , wherein the M first sets of characteristic variables are inputted into the neural network to perform a multistage attention (MSA) algorithm.
8 . The fan control method of claim 1 , wherein each first set of characteristic variables comprises a power load of the system, fan speeds of the set of fans, and a first set of temperatures generated by measuring the system, and the fan control method further comprises: generating a plurality of system states according to the power load and the fan speeds of the set of fans; generating a plurality of pieces of training data according to the plurality of system states; and training the neural network according to the plurality of pieces of training data.
9 . A fan control method for controlling a set of fans of a system, comprising: collecting M first sets of characteristic variables of a first period; inputting the M first sets of characteristic variables to a neural network to generate N third sets of characteristic variables of a second period corresponding to a second set of characteristic variables; adjusting the second set of characteristic variables to generate P adjusted second sets of characteristic variables to accordingly generate Q adjusted third sets of characteristic variables; generating an optimized second set of characteristic variables according to the N third sets of characteristic variables and the Q adjusted third sets of characteristic variables; generating a set of weights according to the optimized second set of characteristic variables; and controlling the set of fans according to the set of weights; wherein the first period precedes the second period, each first set of characteristic variables comprises a second set of characteristic variables and a third set of characteristic variables, M, N, P, Q are positive integers; wherein: adjusting the second set of characteristic variables to generate the P adjusted second sets of characteristic variables to accordingly generate the Q adjusted third sets of characteristic variables comprises: adjusting the second set of characteristic variables to generate the P adjusted second sets of characteristic variables according to a minimum adjustment value of the second set of characteristic variables, so as to accordingly generate the Q adjusted third set of characteristic variables; and generating the optimized second set of characteristic variables according to the N third sets of characteristic variables and the Q adjusted third sets of characteristic variables comprises: generating a set of sums of absolute values according to differences of a predetermined value and each third set of characteristic variables of the N third sets of characteristic variables and the Q adjusted third set of characteristic variables; and generating the optimized second set of characteristic variables according to R third sets of characteristic variables which are selected from the N third sets of characteristic variables and the Q adjusted third sets of characteristic variables and are corresponding to a smallest sum of absolute values; wherein R is a positive integer.
10 . A fan control device for controlling a set of fans of a system, comprising: a set of sensors configured to measure a set of temperatures of the system; a fan speed control unit configured to control fan speeds of the set of fans using pulse width modulation; and a controller coupled to the set of sensors and the fan speed control unit, and configured to: collect M first sets of characteristic variables of a first period; input the M first sets of characteristic variables to a neural network to generate N third sets of characteristic variables of a second period corresponding to a second set of characteristic variables; adjust the second set of characteristic variables to generate P adjusted second sets of characteristic variables according to a non-minimum adjustment value of the second set of characteristic variables, so as to accordingly generate Q adjusted third sets of characteristic variables; generate an optimized second set of characteristic variables according to the N third sets of characteristic variables and the Q adjusted third sets of characteristic variables; generate a set of weights according to the optimized second set of characteristic variables; and control the set of fans according to the set of weights; wherein the first period precedes the second period, each first set of characteristic variables comprises a second set of characteristic variables and a third set of characteristic variables, each first set of characteristic variables comprises a power load of the system, the set of temperatures and the fan speeds of the set of fans, M, N, P, Q are positive integers; generating the optimized second set of characteristic variables comprises: generating a set of sums of absolute values according to differences of a predetermined value and each third set of characteristic variables of the N third sets of characteristic variables and the Q adjusted third sets of characteristic variables; selecting a plurality of sets of third set of characteristic variables from the N third sets of characteristic variables and the Q adjusted third sets of characteristic variables; and selecting a plurality of second sets of characteristic variables corresponding to the plurality of sets of third set of characteristic variables, wherein the plurality of second sets of characteristic variables are of a subset of the second set of characteristic variables and the P adjusted second sets of characteristic variables; and generating the optimized second set of characteristic variables according to at least the plurality of second sets of characteristic variables.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The disclosure is related to a fan control method and a fan control device, and more particularly, a fan control method and a fan control device for controlling fans using a neural network to process characteristic variables. 2. Description of the Prior Art Advanced technologies such as artificial Intelligence (AI), cloud service, 5G and 6G communications, and the internet of things (IoT) are driven by many data centers, which includes a large number of servers used for performing related operations, storage and communications. A large number of servers consume a lot of power, so power efficiency is a key factor in designing a server. For example, each server is usually equipped with 4 to 10 cooling fans, and the fan speed of each fan can be controlled separately. In order to achieve the best cooling effect, the speed of all fans can be turned to the maximum at present. However, this will cause the fan to consume too much energy, resulting in lower power efficiency. In addition, according to actual observations, it may not obtain the best heat dissipation effect by turning the fan speed to the maximum. The reason is, an excessive air volume may blow the heat energy from one element to another element, which is not conducive to the cooling effect. Therefore, a better solution for improving the performance of controlling fans is still in need in the field. SUMMARY OF THE INVENTION An embodiment provides a fan control method for controlling a set of fans of a system. The fan control method includes collecting M first sets of characteristic variables of a first period; inputting the M first sets of characteristic variables to a neural network to generate N third sets of characteristic variables of a second period corresponding to a second set of characteristic variables; adjusting the second set of characteristic variables to generate P adjusted second sets of characteristic variables to accordingly generate Q adjusted third sets of characteristic variables; generating an optimized second set of characteristic variables according to the N third sets of characteristic variables and the Q adjusted third sets of characteristic variables; generating a set of weights according to the optimized second set of characteristic variables; and controlling the set of fans according to the set of weights. The first period precedes the second period, each first set of characteristic variables comprises a second set of characteristic variables and a third set of characteristic variables, M, N, P, Q are positive integers. Another embodiment provides a fan control device for controlling a set of fans of a system. The fan control device includes a system power load unit configured to control a power load; a set of sensors configured to measure a set of temperatures of the system; a fan speed control unit configured to control fan speeds of the set of fans; and a controller coupled to the system power load unit, the set of sensors and the fan speed control unit. The controller is configured to collect M first sets of characteristic variables of a first period; input the M first sets of characteristic variables to a neural network to generate N third sets of characteristic variables of a second period corresponding to a second set of characteristic variables; adjust the second set of characteristic variables to generate P adjusted second sets of characteristic variables to accordingly generate Q adjusted third sets of characteristic variables; generate an optimized second set of characteristic variables according to the N third sets of characteristic variables and the Q adjusted third sets of characteristic variables; generate a set of weights according to the optimized second set of characteristic variables; and control the set of fans according to the set of weights. The first period precedes the second period, each first set of characteristic variables comprises a second set of characteristic variables and a third set of characteristic variables, each first set of characteristic variables comprises the power load, the set of temperatures and the fan speeds of the set of fans, M, N, P, Q are positive integers. These and other objectives of the present invention will no doubt become obvious to those of ordinary skill in the art after reading the following detailed description of the preferred embodiment that is illustrated in the various figures and drawings. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 illustrates a system according to an embodiment. FIG. 2 illustrates a fan control device according to an embodiment. FIG. 3 illustrates a flowchart of the fan control method according to an embodiment. FIG. 4 illustrates the characteristic variables mentioned in FIG. 3 corresponding to different times according to an embodiment. FIG. 5 illustrates a flowchart for generating the optimized second set of characteristic variables in FIG. 3. FIG. 6 illustrates an example of temperatures