阅读说明：本技术 散热控制与模型训练方法、设备、系统及存储介质 (Heat dissipation control and model training method, device, system and storage medium ) 是由 赵旭 李栈 卢毅军 宋军 奉有泉 陶原 陈钢 于 2019-09-05 设计创作，主要内容包括：一种散热控制与模型训练方法、设备、系统及存储介质。该方法包括：预先训练得到过热风险预测模型,通过该模型体现设备功率与制冷参数之间存在的过热风险关系,进而,在该过热风险预测模型的基础上,可根据指定空间内待散热设备的功率变化情况,动态调整制冷系统的制冷参数,达到动态散热控制的目的,有利于降低制冷系统的能耗,节约电能资源。(A method, device, system and storage medium for heat dissipation control and model training. The method comprises the following steps: the method comprises the steps of obtaining an overheating risk prediction model through pre-training, reflecting the overheating risk relation existing between equipment power and refrigeration parameters through the model, and further dynamically adjusting the refrigeration parameters of the refrigeration system according to the power change condition of equipment to be cooled in a specified space on the basis of the overheating risk prediction model to achieve the purpose of dynamic heat dissipation control, thereby being beneficial to reducing the energy consumption of the refrigeration system and saving electric energy resources.)

A method for controlling heat dissipation, comprising:

acquiring actual power information of at least one device to be radiated in a specified space every time a radiation control condition is triggered;

inputting actual power information of the at least one device to be cooled into an overheating risk prediction model to obtain the probability of overheating risk of the designated space under at least one candidate refrigeration parameter;

determining a target refrigeration parameter according to the probability of overheating risk of the designated space under at least one candidate refrigeration parameter;

and controlling a refrigerating system to dissipate heat of at least one device to be dissipated in the designated space according to the target refrigerating parameter.

The method of claim 1, wherein obtaining actual power information of at least one device to be cooled within a specified space whenever a cooling control condition is triggered comprises at least one of:

acquiring actual power information of the at least one device to be radiated when a radiation control cycle is reached;

when the total power change amplitude of the at least one device to be cooled is monitored to be larger than a first amplitude threshold value, acquiring actual power information of the at least one device to be cooled;

and acquiring the actual power information of the at least one device to be cooled when the situation that the power variation amplitude of the device to be cooled is larger than the second amplitude threshold value is monitored.

The method according to claim 1 or 2, wherein acquiring actual power information of at least one device to be cooled in the designated space whenever the cooling control condition is triggered comprises:

when the heat dissipation control condition is triggered, respectively collecting the power value of the at least one device to be dissipated at the moment when the heat dissipation control condition is triggered, and using the power value as actual power information of the at least one device to be dissipated; or

And when the heat dissipation control condition is triggered, respectively obtaining the power average value of the at least one device to be dissipated during the current heat dissipation control and the last heat dissipation control as the actual power information of the at least one device to be dissipated.

The method of claim 1, wherein inputting actual power information of the at least one device to be cooled into an overheating risk prediction model to obtain a probability that the designated space is at an overheating risk under at least one candidate refrigeration parameter comprises:

determining the at least one candidate refrigeration parameter according to the range of the refrigeration parameters used by the overheating risk prediction model in the training phase;

and for each candidate refrigeration parameter, inputting the actual power information of the at least one device to be cooled and the candidate refrigeration parameter into the overheating risk prediction model to obtain the probability of the overheating risk of the designated space under the candidate refrigeration parameter.

The method of claim 1, wherein determining a target refrigeration parameter based on a probability of a risk of overheating the designated space at the at least one candidate refrigeration parameter comprises:

selecting a target probability that is less than a corresponding probability threshold for the designated space for the probability of overheating risk from the probabilities of the designated space experiencing overheating risk under the at least one candidate refrigeration parameter;

and taking the refrigeration parameter corresponding to the target probability in the at least one candidate refrigeration parameter as the target refrigeration parameter.

The method of claim 5, wherein selecting a target probability from the probabilities of the designated space being at risk of overheating at the at least one candidate refrigeration parameter that is less than the threshold probability of overheating for the designated space comprises:

selecting as the target probability a maximum probability that is less than the overheat risk probability threshold from the probabilities that the designated space is at risk of overheating at the at least one candidate refrigeration parameter.

The method of claim 5, further comprising, prior to selecting a target probability that is less than an overheating risk probability threshold corresponding to the designated space:

converting the thermal failure rate allowed by the application or service carried by the specified space into an overheating risk probability threshold corresponding to the specified space; the thermal fault rate indicates a maximum number of times that the designated space can be at risk of overheating within a certain time.

The method according to any one of claims 1, 2 and 4-7, further comprising, before inputting actual power information of the at least one device to be cooled into an overheating risk prediction model:

generating a plurality of groups of marked sample data by combining a sample generation mode based on real data and a sample generation mode based on CFD simulation;

carrying out deep neural network model training by using the multiple groups of marked sample data to obtain the overheating risk prediction model;

each group of marked sample data comprises at least one sample power information corresponding to the at least one device to be radiated, a sample refrigeration parameter corresponding to the refrigeration system and a marking result of whether the designated space has the overheating risk under the group of marked sample data.

The method of claim 8, wherein generating a plurality of sets of labeled sample data by combining the sample generation method based on the real data with the sample generation method based on the CFD simulation comprises:

generating at least one group of marked historical sample data according to the historical power information of the at least one device to be radiated and the historical refrigeration parameters of the refrigeration system; and

and performing simulation calculation between the power information and the refrigeration parameters by using the CFD model to generate at least one group of labeled simulation sample data.

The method of claim 9, wherein generating at least one set of labeled historical sample data based on historical power information of the at least one device to be cooled and historical refrigeration parameters of the refrigeration system comprises:

obtaining at least one group of unmarked historical sample data, wherein each group of unmarked historical sample data comprises historical power information of the at least one device to be radiated and historical refrigeration parameters of the refrigeration system at the same historical moment or within a historical period;

and marking whether the designated space has the overheating risk or not according to the temperature of the internal component of the at least one device to be radiated at the corresponding historical moment or the historical time period and the overheating temperature threshold corresponding to the internal component aiming at each group of unmarked historical sample data to obtain at least one group of marked historical sample data.

The method of claim 9, wherein performing a simulation calculation between the power information and the refrigeration parameters using the CFD model to generate at least one set of labeled simulation sample data comprises:

designing at least one group of unmarked simulation sample data, wherein each group of unmarked simulation sample data comprises at least one piece of simulation power information corresponding to the at least one piece of equipment to be radiated and simulation refrigeration parameters corresponding to the refrigeration system;

and aiming at each group of unmarked simulation sample data, simulating the unmarked simulation sample data by utilizing a CFD (computational fluid dynamics) model to obtain the temperature of the internal device of the at least one equipment to be radiated, and marking whether the designated space has the overheating risk or not by utilizing the temperature of the internal device of the at least one equipment to be radiated and an overheating temperature threshold value corresponding to an internal component to obtain at least one group of marked simulation sample data.

The method of claim 9, wherein before performing the simulation calculations between the power information and the refrigeration parameters using the CFD model to generate at least one set of labeled simulation sample data, further comprising:

performing parameter correction on the CFD model by using the at least one set of marked historical sample data.

The method according to claim 8, wherein in generating a plurality of sets of label sample data, further comprising:

aiming at a first group of marking sample data, acquiring the external atmospheric temperature of the first group of marking sample data at the corresponding moment as the sample atmospheric temperature, and adding the sample atmospheric temperature into the first group of marking sample data; wherein the first set of marker sample data is any one of the plurality of sets of marker sample data.

The method of claim 13, further comprising, each time a heat dissipation control condition is triggered: acquiring the corresponding external atmospheric temperature when the heat dissipation control condition is triggered;

inputting actual power information of the at least one device to be cooled into an overheating risk prediction model to obtain the probability of overheating risk of the designated space under at least one candidate refrigeration parameter, wherein the method comprises the following steps:

and inputting the actual power information of the at least one device to be radiated and the corresponding external atmospheric temperature when the radiation control condition is triggered into the overheating risk prediction model as input parameters to obtain the probability of the overheating risk of the designated space under at least one candidate refrigeration parameter.

The method of claim 8, further comprising at least one of:

performing model training on the overheating risk prediction model again every time a model updating period is reached;

performing model training on the overheating risk prediction model again every time the number of the devices to be radiated in the designated space changes;

and performing model training on the overheating risk prediction model again every time the topological structure among the devices to be radiated in the specified space changes.

The method as claimed in any one of claims 1, 2 and 4-7, wherein controlling the refrigeration system to dissipate heat of at least one device to be dissipated in the designated space according to the target refrigeration parameter comprises:

sending the target refrigeration parameter to the refrigeration system so that the refrigeration system can carry out refrigeration work according to the target refrigeration parameter; or

And when the target refrigeration parameter is different from the currently used refrigeration parameter of the refrigeration system, sending the target refrigeration parameter to the refrigeration system so that the refrigeration system can carry out refrigeration work according to the target refrigeration parameter.

The method of any one of claims 1, 2, and 4-7, wherein the designated space is a machine room, a data center, an office building, an office area in a corporate environment, or a designated area in a home environment.

A method of model training, comprising:

generating a plurality of groups of marked sample data by combining a sample generation mode based on real data and a sample generation mode based on CFD simulation;

carrying out deep neural network model training by using the multiple groups of marked sample data to obtain an overheating risk prediction model;

each group of marked sample data comprises at least one sample power information corresponding to the at least one device to be radiated, a sample refrigeration parameter corresponding to the refrigeration system and a marking result of whether the designated space has the overheating risk under the group of marked sample data.

The method of claim 18, wherein generating a plurality of sets of labeled sample data by combining the sample generation method based on the real data with the sample generation method based on the CFD simulation comprises:

generating at least one group of marked historical sample data according to the historical power information of the at least one device to be radiated and the historical refrigeration parameters of the refrigeration system; and

and performing simulation calculation between the power information and the refrigeration parameters by using the CFD model to generate at least one group of labeled simulation sample data.

The method of claim 19, wherein generating at least one set of labeled historical sample data based on historical power information of the at least one device to be cooled and historical refrigeration parameters of the refrigeration system comprises:

obtaining at least one group of unmarked historical sample data, wherein each group of unmarked historical sample data comprises historical power information of the at least one device to be radiated and historical refrigeration parameters of the refrigeration system at the same historical moment or within a historical period;

and marking whether the designated space has the overheating risk or not according to the temperature of the internal component of the at least one device to be radiated at the corresponding historical moment or the historical time period and the overheating temperature threshold corresponding to the internal component aiming at each group of unmarked historical sample data to obtain at least one group of marked historical sample data.

The method of claim 19, wherein performing a simulation calculation between the power information and the refrigeration parameters using the CFD model to generate at least one set of labeled simulation sample data comprises:

designing at least one group of unmarked simulation sample data, wherein each group of unmarked simulation sample data comprises at least one piece of simulation power information corresponding to the at least one piece of equipment to be radiated and simulation refrigeration parameters corresponding to the refrigeration system;

and aiming at each group of unmarked simulation sample data, simulating the unmarked simulation sample data by utilizing a CFD (computational fluid dynamics) model to obtain the temperature of the internal device of the at least one equipment to be radiated, and marking whether the designated space has the overheating risk or not by utilizing the temperature of the internal device of the at least one equipment to be radiated and an overheating temperature threshold value corresponding to an internal component to obtain at least one group of marked simulation sample data.

The method of any of claims 19-21, further comprising, prior to performing a simulation calculation between the power information and the refrigeration parameters using the CFD model to generate at least one set of labeled simulation sample data:

performing parameter correction on the CFD model by using the at least one set of marked historical sample data.

A heat dissipation control apparatus, comprising: a memory and a processor;

the memory for storing a computer program;

the processor, coupled with the memory, to execute the computer program to:

acquiring actual power information of at least one device to be radiated in a specified space every time a radiation control condition is triggered;

inputting actual power information of the at least one device to be cooled into an overheating risk prediction model to obtain the probability of overheating risk of the designated space under at least one candidate refrigeration parameter;

determining a target refrigeration parameter according to the probability of overheating risk of the designated space under at least one candidate refrigeration parameter;

and controlling a refrigerating system to dissipate heat of at least one device to be dissipated in the designated space according to the target refrigerating parameter.

The device of claim 23, wherein the processor is further configured to:

generating a plurality of groups of marked sample data by combining a sample generation mode based on real data and a sample generation mode based on CFD simulation;

carrying out deep neural network model training by using the multiple groups of marked sample data to obtain the overheating risk prediction model;

each group of marked sample data comprises at least one sample power information corresponding to the at least one device to be radiated, a sample refrigeration parameter corresponding to the refrigeration system and a marking result of whether the designated space has the overheating risk under the group of marked sample data.

A model training apparatus, comprising: a memory and a processor;

the memory for storing a computer program;

the processor, coupled with the memory, to execute the computer program to:

generating a plurality of groups of marked sample data by combining a sample generation mode based on real data and a sample generation mode based on CFD simulation;

carrying out deep neural network model training by using the multiple groups of marked sample data to obtain an overheating risk prediction model;

each group of marked sample data comprises at least one sample power information corresponding to the at least one device to be radiated, a sample refrigeration parameter corresponding to the refrigeration system and a marking result of whether the designated space has the overheating risk under the group of marked sample data.

A machine room system, comprising: the system comprises a machine room, and at least one device to be cooled, a refrigeration system and cooling control equipment which are positioned in the machine room;

the heat dissipation control equipment is used for acquiring the actual power information of the at least one equipment to be dissipated every time when a heat dissipation control condition is triggered, and inputting the actual power information of the at least one equipment to be dissipated into an overheating risk prediction model to obtain the probability of overheating risk of the machine room system under at least one candidate refrigeration parameter; determining a target refrigeration parameter according to the probability of overheating risk of the machine room system under at least one candidate refrigeration parameter; controlling the refrigeration system to dissipate heat of the at least one device to be dissipated according to the target refrigeration parameter;

and the refrigerating system is used for dissipating heat of at least one device to be dissipated in the machine room under the control of the heat dissipation control device.

The machine room system according to claim 26, wherein the refrigeration system is an air conditioning system, and the target refrigeration parameter is at least one of an operating temperature, an operating wind speed, and an operating mode of the air conditioning system; or

The refrigerating system is a water-cooling system, and the target refrigerating parameter is at least one of the water outlet temperature, the return water temperature, the water flow speed and the water flow of the water-cooling system.

A data center system, comprising: at least one machine room; each computer room comprises: the system comprises at least one device to be cooled, a refrigeration system and cooling control equipment;

the heat dissipation control equipment is used for acquiring actual power information of at least one equipment to be dissipated in the machine room to which the heat dissipation control equipment belongs every time when a heat dissipation control condition is triggered, and inputting the actual power information of the at least one equipment to be dissipated into an overheating risk prediction model to obtain the probability of overheating risk of the machine room under at least one candidate refrigeration parameter; determining target refrigeration parameters according to the probability of overheating risks of the machine room under at least one candidate refrigeration parameter; controlling a refrigerating system in the machine room to dissipate heat of the at least one device to be dissipated according to the target refrigerating parameter;

and the refrigerating system is used for dissipating heat of at least one device to be dissipated in the machine room to which the refrigerating system belongs under the control of the heat dissipation control device.

A data center system, comprising: at least one machine room, a refrigeration system and a heat dissipation control device; wherein each machine room comprises at least one device to be cooled, and the refrigeration system comprises refrigeration devices deployed in each machine room;

the heat dissipation control equipment is used for acquiring actual power information of at least one equipment to be dissipated in each machine room when a heat dissipation control condition is triggered, and inputting the actual power information of the at least one equipment to be dissipated in the machine room into an overheating risk prediction model to obtain the probability of overheating risk of the machine room under at least one candidate refrigeration parameter; determining target refrigeration parameters according to the probability of overheating risks of the machine room under at least one candidate refrigeration parameter; and controlling the refrigeration equipment in the machine room to dissipate heat of at least one device to be cooled in the machine room according to the target refrigeration parameters.

A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the processor to carry out the steps of the method according to any one of claims 1-17.

A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the processor to carry out the steps of the method of any one of claims 18-22.