WO2022257267A1 - 环境参数控制设备集群控制方法、装置、设备及存储介质 - Google Patents

环境参数控制设备集群控制方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2022257267A1
WO2022257267A1 PCT/CN2021/112714 CN2021112714W WO2022257267A1 WO 2022257267 A1 WO2022257267 A1 WO 2022257267A1 CN 2021112714 W CN2021112714 W CN 2021112714W WO 2022257267 A1 WO2022257267 A1 WO 2022257267A1
Authority
WO
WIPO (PCT)
Prior art keywords
environmental parameter
control cycle
energy efficiency
model
sample
Prior art date
Application number
PCT/CN2021/112714
Other languages
English (en)
French (fr)
Inventor
刘敬民
冯晓波
李星
颜泽波
周薛继
王静
Original Assignee
维谛技术(西安)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 维谛技术(西安)有限公司 filed Critical 维谛技术(西安)有限公司
Publication of WO2022257267A1 publication Critical patent/WO2022257267A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H05—ELECTRIC TECHNIQUES NOT OTHERWISE PROVIDED FOR
    • H05K—PRINTED CIRCUITS; CASINGS OR CONSTRUCTIONAL DETAILS OF ELECTRIC APPARATUS; MANUFACTURE OF ASSEMBLAGES OF ELECTRICAL COMPONENTS
    • H05K7/00—Constructional details common to different types of electric apparatus
    • H05K7/20—Modifications to facilitate cooling, ventilating, or heating
    • H05K7/20709—Modifications to facilitate cooling, ventilating, or heating for server racks or cabinets; for data centers, e.g. 19-inch computer racks
    • H05K7/20718—Forced ventilation of a gaseous coolant
    • H05K7/20745—Forced ventilation of a gaseous coolant within rooms for removing heat from cabinets, e.g. by air conditioning device
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • H—ELECTRICITY
    • H05—ELECTRIC TECHNIQUES NOT OTHERWISE PROVIDED FOR
    • H05K—PRINTED CIRCUITS; CASINGS OR CONSTRUCTIONAL DETAILS OF ELECTRIC APPARATUS; MANUFACTURE OF ASSEMBLAGES OF ELECTRICAL COMPONENTS
    • H05K7/00—Constructional details common to different types of electric apparatus
    • H05K7/20—Modifications to facilitate cooling, ventilating, or heating

Definitions

  • the invention relates to the technical field of automation, in particular to a cluster control method, device, device and storage medium of environmental parameter control equipment.
  • a data center usually consists of multiple cabinets, multiple servers, uninterruptible power supply (Uninterruptible Power Supply, UPS), multiple air conditioners, temperature and humidity sensors, and so on.
  • UPS Uninterruptible Power Supply
  • the environment of the data center is always affected by various factors such as the heat generated by the server and environmental changes, and the air conditioner is used to adjust the temperature of the data center to ensure a constant temperature of the data center.
  • the air conditioner consumes a lot of electric energy during use, and the efficiency of converting electric energy into cooling capacity is different under different working powers. Therefore, how to control multiple air conditioners to reduce the power usage efficiency of the data center (Power Usage Effectiveness) , PUE) has become the general trend.
  • Power Usage Effectiveness Power Usage Effectiveness
  • the group control method of air conditioners in traditional data centers is relatively simple. Most of the group control of air conditioners is based on manual experience design rules, and the effect is not ideal.
  • Embodiments of the present invention provide a cluster control method, device, device, and storage medium for environmental parameter control equipment, to solve the problem in the prior art that using the existing model to control the air conditioner group alone, the effect of equipment security and energy consumption control is not good. good question.
  • An embodiment of the present invention provides a cluster control method for environmental parameter control equipment, which is applied to energy-saving adjustment and environmental parameter adjustment of a data center, including:
  • each time the control cycle is reached multiple candidate environmental parameter set points and multiple candidate start-up numbers are determined, and based on the state parameters and candidate start-up numbers collected when the current control cycle arrives, use the UCB model to predict the corresponding Based on the state parameters and candidate environmental parameter set points, use the regression model to predict the corresponding environmental parameter predicted values, based on the energy efficiency score predicted value and environmental parameter predicted values to decide the number of startups and environmental parameter set points and run;
  • the sample characteristics in the energy efficiency sample include the state parameters and the number of startups of the data center obtained when the previous control period arrives
  • the sample label is the energy efficiency score calculated when the current control period arrives
  • the sample characteristics in the environmental parameter sample include the above The status parameters and environmental parameter set points of the data center obtained when the control cycle arrives
  • the sample tag includes the measured values of the environmental parameters collected when the current control cycle arrives.
  • the energy efficiency score is a score determined according to the energy efficiency index of the data center
  • the energy efficiency index includes the total power of the data center or the total power of the environmental parameter control equipment cluster or the power usage efficiency PUE of the data center; wherein:
  • the energy efficiency score is calculated according to the energy efficiency index of the data center using the first formula
  • the energy efficiency score is calculated according to the energy efficiency index of the data center using the second formula
  • the energy efficiency score calculated by using the first formula is greater than the energy efficiency score calculated by the second formula.
  • the sample features in the environmental parameter samples also include: the number of startups in the last control cycle;
  • UCB model Based on the state parameters collected when the current control cycle arrives and the number of candidate start-ups, use the UCB model to predict the corresponding energy efficiency score prediction value, and use the regression model to predict the corresponding environmental parameter prediction value based on the state parameters and candidate environmental parameter set points , based on the predicted value of the energy efficiency score and the predicted value of the environmental parameter to decide the number of startups and the set point of the environmental parameter, including:
  • the environmental parameter set point is determined according to the predicted value of the environmental parameter.
  • sample characteristics in the energy efficiency sample further include: the environmental parameter set point of the last control cycle;
  • UCB model Based on the state parameters collected when the current control cycle arrives and the number of candidate start-ups, use the UCB model to predict the corresponding energy efficiency score prediction value, and use the regression model to predict the corresponding environmental parameter prediction value based on the state parameters and candidate environmental parameter set points , based on the predicted value of the energy efficiency score and the predicted value of the environmental parameter to decide the number of startups and the set point of the environmental parameter, including:
  • the power-on number is determined according to the predicted value of the energy efficiency score.
  • the UCB model is used to predict the corresponding energy efficiency score prediction value, and based on the state parameters and candidate environmental parameter set points, the regression model is used to predict the corresponding Predicted values of environmental parameters, based on the predicted value of the energy efficiency score and the predicted value of the environmental parameters, the number of startups and the set point of the environmental parameters are determined, including:
  • the environmental parameter set point is determined according to the predicted value of the environmental parameter.
  • the decision on the number of startups is made according to the predicted value of the energy efficiency score, including:
  • Decision-making of the environmental parameter set point according to the predicted value of the environmental parameter includes:
  • the candidate environmental parameter set point corresponding to the predicted value of the environmental parameter closest to the target environmental parameter is determined as the environmental parameter set point.
  • trigger model training when a control period is reached
  • the sample features in the energy efficiency index sample set are input into the upper confidence UCB model with context, and the model training is carried out with the goal of outputting the sample labels in the energy efficiency sample set, including:
  • the sample features in the environmental parameter sample set are input into the regression model, and the model training is carried out with the goal of outputting the label in the environmental parameter sample set, including:
  • the sample features of all environmental parameter samples in the environmental parameter sample set are sequentially input into the regression model, and the model training is performed with the goal of outputting corresponding sample labels.
  • the method also includes:
  • the number of startups and the set point of the environmental parameter are adjusted and run;
  • every time a control cycle is reached the measured value of the environmental parameter of the data center environment is obtained, and the number of startups and the set point of the environmental parameter are adjusted according to the difference between the measured value of the environmental parameter and the target environmental parameter, include:
  • the difference ⁇ E is less than the first threshold, it is determined that the environmental parameter set point of the current control cycle is the environmental parameter set point of the previous control cycle and the preset environmental parameter adjustment value is increased, and the number of startups in the current control cycle is the previous one.
  • the number of start-ups in the control cycle reduces the adjustment value of the number of start-ups;
  • the difference ⁇ E is greater than the second threshold, it is determined that the environmental parameter set point of the current control cycle is the environmental parameter set point of the previous control cycle and the preset environmental parameter adjustment value is reduced, and the number of startups in the current control cycle is the previous one.
  • the number of start-ups in the control cycle increases the preset start-up number adjustment value
  • the environmental parameter set point of the current control cycle is the environmental parameter set point of the previous control cycle, randomly increasing or decreasing the preset environmental parameter adjustment value, the number of start-ups in the current control cycle is the number of start-ups in the previous control cycle, which randomly increases or decreases the preset number of start-up adjustments;
  • EnvirMeasure is the environmental parameter measurement value collected when the current control cycle arrives
  • EnvirTarget is the target environmental parameter
  • determining a plurality of candidate environmental parameter set points and a plurality of candidate start-up numbers includes:
  • Determining an alternative environmental parameter set point that meets the range of the environmental parameter set point among the plurality of alternative environmental parameter set points, and the environmental parameter set point of the previous control cycle is the candidate environmental parameter set point
  • the method also includes:
  • the state parameters of the data center are collected, and the environmental parameters of the data center are triggered to be detected.
  • the safe environment parameter range is exceeded, the number of startups and the environmental parameter set point are adjusted according to the set adjustment range;
  • the length of one control cycle is equal to an integer multiple of the interval between two adjacent state parameter sampling moments.
  • the state parameters of the data center are collected, and the detection of the environmental parameters of the data center is triggered, and when the safe environment parameter range is exceeded, the number of startups and the environmental parameters are adjusted according to the set adjustment range set points, including:
  • the environmental parameter set point of the current control period is increased by the preset environmental parameter adjustment value, and the number of startups in the current control period is reduced by the preset adjustment value of the startup number;
  • the environmental parameter set point of the current control cycle is reduced by the preset environmental parameter adjustment value, and the number of startups in the current control cycle is increased by the preset startup number adjustment value;
  • EnvirMeasure is the measured value of the environmental parameter collected when the current control period arrives
  • EnvirTarget is the target environmental parameter
  • DeadLine is the difference value of the preset environmental parameter and is a positive number.
  • the UCB model is a linearly confident LinUCB model or a Gaussian UCB model.
  • the regression model is any of the following:
  • xgboost model random forest RF model, support vector machine SVM model, neural network model.
  • the state parameters include at least one of the following: load power, average air supply environment parameters, average return air environment parameters, average environment parameters on the hot aisle side, and average environment parameters on the cold aisle side.
  • the environmental parameter is temperature or humidity.
  • an embodiment of the present invention also provides a cluster control device for environmental parameter control equipment, which is applied to energy-saving adjustment and environmental parameter adjustment of a data center, including:
  • the sample collection module is used to collect environmental parameter samples and energy efficiency samples every time the control cycle is reached, and correspondingly update the environmental parameter sample set and energy efficiency sample set;
  • the model training module is used to trigger model training, input the sample features in the energy efficiency sample set into the upper confidence UCB model with context, and perform model training with the target of outputting the sample labels in the energy efficiency sample set, and set the environmental parameters
  • the sample features in the sample set are input into the regression model, and the model training is performed with the goal of outputting the sample label in the environmental parameter sample set;
  • the recommendation module is used to determine that when the current control period is in the recommended stage, each time the control period is reached, multiple candidate environmental parameter set points and multiple candidate start-up numbers are determined, based on the state parameters and candidate start-up numbers collected when the current control period arrives, using all
  • the UCB model predicts the corresponding predicted value of the energy efficiency score, based on the state parameter and the set point of the candidate environmental parameter, uses the regression model to predict the predicted value of the corresponding environmental parameter, and based on the predicted value of the energy efficiency score and the predicted value of the environmental parameter, decides the number of boots and Environmental parameter set point and run;
  • the sample characteristics in the energy efficiency sample include the state parameters and the number of startups of the data center obtained when the previous control period arrives
  • the sample label is the energy efficiency score calculated when the current control period arrives
  • the sample characteristics in the environmental parameter sample include the above The status parameters and environmental parameter set points of the data center obtained when the control cycle arrives
  • the sample tag includes the measured values of the environmental parameters collected when the current control cycle arrives.
  • the sample features in the environmental parameter samples also include: the number of startups in the last control cycle;
  • UCB model Based on the state parameters collected when the current control cycle arrives and the number of candidate start-ups, use the UCB model to predict the corresponding energy efficiency score prediction value, and use the regression model to predict the corresponding environmental parameter prediction value based on the state parameters and candidate environmental parameter set points , based on the predicted value of the energy efficiency score and the predicted value of the environmental parameter to decide the number of startups and the set point of the environmental parameter, including:
  • the environmental parameter set point is determined according to the predicted value of the environmental parameter.
  • sample characteristics in the energy efficiency sample further include: the environmental parameter set point of the last control cycle;
  • UCB model Based on the state parameters collected when the current control cycle arrives and the number of candidate start-ups, use the UCB model to predict the corresponding energy efficiency score prediction value, and use the regression model to predict the corresponding environmental parameter prediction value based on the state parameters and candidate environmental parameter set points , based on the predicted value of the energy efficiency score and the predicted value of the environmental parameter to decide the number of startups and the set point of the environmental parameter, including:
  • the power-on number is determined according to the predicted value of the energy efficiency score.
  • the UCB model is used to predict the corresponding energy efficiency score prediction value, and based on the state parameters and candidate environmental parameter set points, the regression model is used to predict the corresponding Predicted values of environmental parameters, based on the predicted value of the energy efficiency score and the predicted value of the environmental parameters, the number of startups and the set point of the environmental parameters are determined, including:
  • the environmental parameter set point is determined according to the predicted value of the environmental parameter.
  • the device also includes:
  • the initial control module is used to determine that when the model decision-making conditions are not satisfied, each time a control cycle is reached, the measured value of the environmental parameters collected when the current control cycle of the data center environment arrives;
  • the number of startups and the set point of the environmental parameter are adjusted and run;
  • every time a control cycle is reached the measured value of the environmental parameter of the data center environment is obtained, and the number of startups and the set point of the environmental parameter are adjusted according to the difference between the measured value of the environmental parameter and the target environmental parameter, include:
  • the difference ⁇ E is less than the first threshold, it is determined that the environmental parameter set point of the current control cycle is the environmental parameter set point of the previous control cycle and the preset environmental parameter adjustment value is increased, and the number of startups in the current control cycle is the previous one.
  • the number of start-ups in the control cycle reduces the adjustment value of the number of start-ups;
  • the difference ⁇ E is greater than the second threshold, it is determined that the environmental parameter set point of the current control cycle is the environmental parameter set point of the previous control cycle and the preset environmental parameter adjustment value is reduced, and the number of startups in the current control cycle is the previous one.
  • the number of start-ups in the control cycle increases the preset start-up number adjustment value
  • the environmental parameter set point of the current control cycle is the environmental parameter set point of the previous control cycle, randomly increasing or decreasing the preset environmental parameter adjustment value, the number of start-ups in the current control cycle is the number of start-ups in the previous control cycle, which randomly increases or decreases the preset number of start-up adjustments;
  • EnvirMeasure is the environmental parameter measurement value collected when the current control cycle arrives
  • EnvirTarget is the target environmental parameter
  • the device also includes:
  • the safety maintenance module is used to collect the status parameters of the data center every time the data sampling time is reached, and trigger the detection of the environmental parameters of the data center, and adjust the number of boots and Environmental parameter set points;
  • the length of one control cycle is an integer multiple of adjacent data sampling time intervals.
  • the state parameters of the data center are collected, and the detection of the environmental parameters of the data center is triggered, and when the safe environment parameter range is exceeded, the number of startups and the environmental parameters are adjusted according to the set adjustment range set points, including:
  • the environmental parameter set point of the current control period is increased by the preset environmental parameter adjustment value, and the number of startups in the current control period is reduced by the preset adjustment value of the startup number;
  • the environmental parameter set point of the current control cycle is reduced by the preset environmental parameter adjustment value, and the number of startups in the current control cycle is increased by the preset startup number adjustment value;
  • EnvirMeasure is the measured value of the environmental parameter collected when the current control period arrives
  • EnvirTarget is the target environmental parameter
  • DeadLine is the difference value of the preset environmental parameter and is a positive number.
  • an embodiment of the present invention also provides an electronic device, including: a processor and a memory for storing instructions executable by the processor;
  • the processor is configured to execute the instructions, so as to realize the environment parameter control device cluster control method.
  • an embodiment of the present invention further provides a storage medium, the computer storage medium stores a computer program, and the computer program is used to implement the method for controlling an environment parameter control device cluster.
  • the environmental parameter control equipment cluster control method, device, equipment, and storage medium provided by the embodiments of the present invention combine the UCB model algorithm with context with the regression model algorithm to recommend data center configuration, and use the combination model to make the air conditioner configuration decision-making action Space decoupling reduces the action space by nearly ten times and accelerates the learning convergence speed.
  • the main model adopts UCB algorithm, which also has faster convergence speed and online learning ability, and has higher accuracy. Adjust the number of start-ups of the air conditioner and the set points of environmental parameters with a certain control cycle, thereby improving the environmental parameters of the data center and reducing the PUE, and it is safe and reliable.
  • FIG. 1 is one of the flow charts of the method for controlling an environmental parameter control device cluster provided by an embodiment of the present invention
  • Fig. 2 is the effect diagram of training UCB model and regression model in the embodiment of the present invention
  • Fig. 3 is the second flow chart of the method for controlling the cluster of environmental parameter control equipment provided by the embodiment of the present invention.
  • Figure 4-1 is one of the input and output diagrams for recommending using the UCB model and the regression model in the embodiment of the present invention
  • Figure 4-2 is the second schematic diagram of input and output using UCB model and regression model for recommendation in the embodiment of the present invention
  • Figure 4-3 is the third schematic diagram of the input and output of recommendations using the UCB model and the regression model in the embodiment of the present invention.
  • Fig. 5 is a schematic diagram of the effect of the cluster control method of the environmental parameter control equipment in the embodiment of the present invention.
  • Fig. 6 is the third flowchart of the environmental parameter control device cluster control method provided by the embodiment of the present invention.
  • FIG. 7 is the fourth flowchart of the environmental parameter control device cluster control method provided by the embodiment of the present invention.
  • Figure 8-1 is a schematic diagram of the specific input and output of the UCB model shown in Figure 4-1;
  • Figure 8-2 is a schematic diagram of the specific input and output of the regression model shown in Figure 4-1;
  • Fig. 9 is the fifth flowchart of the environmental parameter control device cluster control method provided by the embodiment of the present invention.
  • FIG. 10 is a schematic structural diagram of an environmental parameter control device cluster control device provided by an embodiment of the present invention.
  • Fig. 11 is a schematic structural diagram of an electron provided by an embodiment of the present invention.
  • FIG. 12 is a schematic structural diagram of a data center control system using the electronic device illustrated in FIG. 11 .
  • the environmental parameter is temperature or humidity.
  • the following will take the environmental parameter as temperature as an example for illustration.
  • the implementation of the environmental parameter as humidity is basically the same as the implementation of temperature, so reference can be made to the embodiment of temperature, which will not be repeated below.
  • Temperature set point The temperature set by the temperature control device cluster during operation. For example, if the temperature control device is an air conditioner, then the temperature set point of the air conditioner is the air outlet temperature of the air conditioner.
  • Temperature measurement value the ambient temperature of the data center to be controlled by the temperature control device cluster collected by the temperature sensor.
  • Target temperature the expected ambient temperature finally reached by the data center controlled by the temperature control device cluster.
  • An embodiment of the present invention provides a cluster control method for environmental parameter control equipment, which is applied to energy-saving regulation and temperature regulation of a data center, as shown in FIG. 1 , including:
  • step S110 of the decision-making control part and step S210 of the model training part are executed;
  • step S110 If the result of step S110 is yes, execute step S120;
  • S120 collect temperature samples and energy efficiency samples every time the control cycle is reached, and update them to the temperature sample set and energy efficiency sample set correspondingly;
  • the sample characteristics in the energy efficiency sample include the state parameters and the number of startups of the data center obtained when the previous control cycle arrives, the sample label is the energy efficiency score calculated when the current control cycle is reached, and the sample characteristics in the temperature sample include the previous The state parameters and temperature set point of the data center obtained when the control period arrives, and the sample label includes the temperature measurement value collected when the current control period arrives;
  • step S130 If the result of the step S130 is yes, it is determined that it is currently in the recommendation stage, and the step S150 is executed;
  • each time a control cycle is reached determine a plurality of candidate temperature set points and a plurality of candidate start-up numbers, and use the UCB model to predict the corresponding energy efficiency score prediction value based on the state parameters and candidate start-up numbers collected when the current control cycle arrives, Based on the state parameters and candidate temperature set points, the regression model is used to predict the corresponding ambient temperature prediction value, and the number of power-on and temperature set points are determined and run based on the energy efficiency score prediction value and the ambient temperature prediction value.
  • step S210 If the result of step S210 is yes, execute step S220; if the result of step S210 is no, continue to wait until the result of step S210 is yes;
  • the UCB model is a UCB model with context.
  • the UCB model is a linear upper confidence (Linear Upper Confidence Bound, LinUCB) model or a Gaussian upper confidence (Gaussian Process Upper Confidence Bound, GPUCB) model.
  • the regression model is any of the following:
  • xgboost model Random Forest (Random Fores, RF) model, Support Vector Machine (Support Vector Machine, SVM) model, neural network model.
  • the regression model can also be other unmentioned models, which can be selected according to actual needs, and are not limited here.
  • the model training condition in the step S210 may be to trigger a model training in each control cycle, or to trigger a model training at intervals of multiple control cycles, or to trigger a model training when a certain condition is met ( For example, when multiple consecutive control cycles arrive, use the temperature set point and the number of startups determined in step S150 to control the environmental parameter control device cluster, and the data center data obtained when the next control cycle arrives The state parameter does not meet the preset state parameter range) triggers a model training, which is not limited here.
  • model training is triggered when a control period is reached. That is, model training is performed once in each control cycle.
  • the implementation of triggering a model training in each of the control cycles can refer to the flow chart shown in Figure 3, wherein the steps shown in Figure 3 are basically the same as those in Figure 1, and can be referred to the content described above, so it will not be repeated. .
  • the corresponding update to the temperature sample set and the energy efficiency sample set includes:
  • the number of samples in the temperature sample set and the energy efficiency sample set is equal to the capacity of the sample set, delete the temperature sample and energy efficiency sample corresponding to the earliest control cycle in the temperature sample set and the energy efficiency sample set, and replace the current control cycle corresponding to The temperature sample and the energy efficiency sample are correspondingly updated to the temperature sample set and the energy efficiency sample set.
  • the temperature sample and energy efficiency sample corresponding to the first control cycle in the two sample sets will be deleted, and the temperature sample corresponding to the 721st control cycle will be The samples and energy efficiency samples are correspondingly updated to the two sample sets. For subsequent control cycles, delete the temperature samples and energy efficiency samples corresponding to the second control cycle at the 722nd control cycle, update the temperature samples and energy efficiency samples corresponding to the 722nd control cycle to the two sample sets, and so on .
  • This application combines the UCB model algorithm with context and the regression model algorithm to recommend data center temperature and energy consumption configuration, which has high accuracy and convergence speed.
  • the UCB model and the regression model need to be correspondingly trained using the energy consumption sample set and the temperature sample set respectively.
  • the sample features in the temperature sample set are input into the regression model, and the model training is performed with the goal of outputting the labels in the temperature sample set, including:
  • the sample features of all temperature samples in the temperature sample set are sequentially input into the regression model, and the model training is performed with the goal of outputting corresponding sample labels.
  • the sample features in the energy efficiency index sample set are input into the upper confidence UCB model with context, and the model training is performed with the goal of outputting the sample labels in the energy efficiency sample set, and any of the following implementations can be adopted Way:
  • Mode A Input the sample features of all energy efficiency samples in the energy efficiency sample set into the UCB model in sequence, and perform model training with the goal of outputting corresponding sample labels.
  • Mode B When the control cycle is reached, model training is triggered.
  • the sample features in the energy efficiency samples of the current control period in the energy efficiency sample set are input into the UCB model, and the model training is performed with the goal of outputting the sample labels in the energy efficiency samples of the current control period.
  • the UCB model is trained using mode B, and only updated energy efficiency samples are used for training every time a control cycle is reached, which can reduce the amount of training data and speed up the training.
  • the method further includes:
  • step S130 If the result of the step S130 is no, it is determined that it is currently in the initialization control stage, and the step S141 is executed;
  • the location and environment of the data centers are different (for example, the temperature changes in different regions are different, so that the temperature and energy efficiency of the data centers also have corresponding changes. different) and other factors, so pre-set unified training samples to train the UCB model and the regression model, use the model trained in this way to carry out recommended control of environmental parameter control equipment clusters, its security and energy saving Effects can be problematic.
  • pre-set unified training samples to train the UCB model and the regression model use the model trained in this way to carry out recommended control of environmental parameter control equipment clusters, its security and energy saving Effects can be problematic.
  • the temperature samples and the energy efficiency samples are used to accumulate training samples for the UCB model and the regression model, so as to facilitate the subsequent use of the trained model for recommendation control.
  • adjusting the number of startups and the temperature set point according to the difference between the measured temperature value and the target temperature includes:
  • the difference ⁇ T is less than the first threshold, it is determined that the temperature set point of the current control cycle is the temperature set point of the previous control cycle and the preset temperature adjustment value is increased, and the number of startups in the current control cycle is that of the previous control cycle.
  • the number of boots reduces the adjustment value of the number of boots;
  • the difference ⁇ T is greater than the second threshold, it is determined that the temperature set point of the current control cycle is the temperature set point of the previous control cycle, and the preset temperature adjustment value is reduced, and the number of startups in the current control cycle is that of the previous control cycle.
  • the number of startups increases the preset adjustment value of the number of startups;
  • the difference ⁇ T is greater than or equal to the first threshold and less than or equal to the second threshold, it is determined that the temperature set point of the current control cycle is the temperature set point of the previous control cycle, randomly increasing or decreasing the preset temperature adjustment value, and the current The number of start-ups in the control cycle is the number of start-ups in the previous control cycle that randomly increases or decreases the preset number of start-ups;
  • TempMeasure is the temperature measurement value collected when the current control period arrives
  • TempTarget is the target temperature
  • the preset temperature adjustment value is 1°C
  • the preset power-on number adjustment value is 1
  • the first threshold value is -2°C
  • the second threshold value is 2°C.
  • the temperature set point is increased and the number of startups is reduced, so that the ambient temperature of the data center can be controlled; the temperature measurement value is significantly higher than the target temperature.
  • the target temperature is lowered and the number of startups is increased, the ambient temperature of the data center can be controlled to decrease, so that the ambient temperature of the data center can be adjusted to the target temperature as soon as possible.
  • randomly controlling the change of the temperature set point and the number of startups can enrich the training samples for the training of the UCB model and the regression model, and improve the efficiency after training. The reliability of the resulting model recommendations.
  • the energy efficiency score is a score determined according to the energy efficiency index of the data center
  • the energy efficiency index includes the total power of the data center or the total power of the environmental parameter control equipment cluster or the PUE of the data center; wherein:
  • the energy efficiency score is calculated according to the energy efficiency index of the data center using the first formula
  • the energy efficiency score is calculated according to the energy efficiency index of the data center using the second formula
  • the energy efficiency score calculated by using the first formula is greater than the energy efficiency score calculated by the second formula
  • TempMeasure is the temperature measurement value collected when the current control cycle arrives
  • Temp is the temperature set point of the previous control cycle
  • DeadLine is the preset temperature difference and is a positive number.
  • the energy efficiency index is taken as an example of PUE.
  • the first formula is:
  • the second formula is:
  • score is the energy efficiency score
  • the PUE in the first and second formulas above can also be replaced by the The total power of the data center and the environmental parameters control the total power of the equipment cluster, and the values of A and B are adjusted as required, which will not be repeated here.
  • first formula and the second formula are not limited to the above-mentioned inverse proportional relationship, and may also be other types of formulas, which are not limited here.
  • determining a plurality of candidate temperature set points and a plurality of candidate startup numbers includes:
  • Determining the candidate temperature set point that meets the temperature set point range among the plurality of candidate temperature set points, and the temperature set point of the previous control cycle is the candidate temperature set point
  • the temperature set point range is a preset range, for example, 10°C-30°C.
  • the alternative temperature set point will have values greater than 30°C and less than 30°C, but the alternative temperature set point greater than 30°C does not meet the stated
  • the temperature setpoint range that will be discarded.
  • the candidate temperature set points are finally determined to be the alternative temperature set points less than 30°C and 30°C.
  • the range of the number of boots it can be directly determined as 0 to the number of devices in the environmental parameter control device cluster, or it can be further set as a subset (such as half of the number of devices in the environmental parameter control device cluster to all).
  • the number n 1 of the candidate temperature set points, the numerical difference ⁇ 1 between the two candidate temperature set points adjacent to each other, and the alternative power-on The number n 2 of numbers, and the numerical difference ⁇ 2 between the two alternative starting numbers with adjacent values (n 1 and n 2 are both positive and even numbers).
  • the number n 1 of the candidate temperature set points is determined to be 4, and the numerical difference ⁇ 1 of the candidate temperature set points is 2°C.
  • the temperature set point of the last control cycle is 20°C
  • the The alternative temperature set points described are 16°C, 18°C, 22°C, 24°C.
  • the setting method of the number of alternate boots is the same, so it will not be repeated here.
  • determining multiple alternative temperature set points centered on the temperature set point of the previous control cycle includes:
  • the change value of the minimum temperature set point is 1°C
  • the alternative temperature set points are determined to be 19°C and 21°C.
  • the temperature change of the data center is more balanced by increasing the number of boots and the temperature set point at most one change unit each time.
  • the UCB model and the regression model can be used respectively
  • the model performs energy efficiency score prediction and ambient temperature prediction, and the two can also be used in combination, and the parameters obtained by using the prediction results of one model to recommend decisions are input into the other model to affect the prediction results and recommendation decisions of the other parameters.
  • UCB model based on the state parameters collected when the current control cycle arrives and the number of candidate start-ups, use the UCB model to predict the corresponding energy efficiency score, and use the regression model to predict the corresponding ambient temperature prediction based on the state parameters and candidate temperature set points value, based on the energy efficiency score and the predicted value of the ambient temperature to decide the number of boots and the temperature set point, including any of the following implementation methods:
  • the sample characteristics in the temperature samples further include: the number of startups in the last control period.
  • the temperature set point is determined based on the predicted value of the ambient temperature.
  • the sample characteristics in the energy efficiency sample further include: a temperature set point of a previous control cycle.
  • the candidate temperature set point and the state parameters collected when the current control cycle arrives are input into the regression model to predict the corresponding ambient temperature prediction value;
  • the power-on number is determined according to the predicted value of the energy efficiency score.
  • the number of candidate start-ups and the state parameters collected when the current control cycle arrives are input into the UCB model, and the corresponding energy efficiency score prediction values are respectively predicted;
  • the temperature set point is determined based on the predicted value of the ambient temperature.
  • the temperature set point corresponding to the target temperature closest to the target temperature in the predicted value of the ambient temperature is the temperature set point of the current control cycle. If according to the energy efficiency score calculated by the energy efficiency index, the higher the energy efficiency index is, the higher the energy efficiency score is, and the candidate starting number corresponding to the maximum value of the predicted value of the energy efficiency score is determined to be the starting number of the current control period.
  • mode 1 and mode 2 use two models to couple the decision-making of the power-on number and temperature set point, which reduces the recommendation space and greatly improves the convergence speed of the model.
  • Method 3 uses two models to determine the number of starts and temperature set points respectively, and the solution is relatively simple.
  • the above implementation method is used to adjust the number of startups and temperature set points of the environmental parameter control equipment cluster.
  • the cycle can also be set It detects and adjusts the permanent state parameter sampling time.
  • the method also includes:
  • step S160 If the result of the step S160 is yes, execute the step S170; if the result of the step S160 is no, return to the step S110;
  • step S170 If the result of step S170 is yes, execute step S180; if the result of step S170 is no, return to step S110;
  • the length of one control cycle is an integer multiple of adjacent data sampling time intervals.
  • the duration of the data sampling time may be set according to actual needs.
  • the control period is 1 hour
  • the data sampling time is 5 minutes.
  • the step S170 detecting the ambient temperature of the data center to determine whether it exceeds a safe temperature range, includes:
  • the step S180 adjusting the power-on number and temperature set point according to the set adjustment range, includes:
  • TempMeasure is the temperature measurement value collected when the current control cycle arrives
  • TempTarget is the target temperature
  • DeadLine is a preset temperature difference and is a positive number.
  • the DeadLine is that the preset temperature difference is 3°C, the preset temperature adjustment value is 1°C, and the preset power-on number adjustment value is 1.
  • the state parameters include at least one of the following: load power, average supply air temperature, average return air temperature, average temperature on the hot aisle side, and average temperature on the cold aisle side.
  • the temperature control device is an air conditioner
  • the control period is 1 hour
  • the time interval between adjacent data sampling is 5 minutes.
  • the UCB model is a LinUCB model or a GPUCB model
  • the regression model is an xgboost model.
  • the target temperature TempTarget 22°C
  • the preset temperature difference DeadLine 3°C
  • the number of devices in the air-conditioning cluster ranges from half to all of the number of devices in the air-conditioning cluster
  • the range of the temperature set point is [15°C , 20°C]
  • the sample set capacity of the temperature sample set and the energy efficiency sample set is 720.
  • the recommendation phase starts from the 241st control cycle, the first threshold is -2°C, and the second threshold is 2°C.
  • the method for controlling the control device cluster includes:
  • step S310 If the result of the step S310 is yes, execute the step S320; if the result of the step S310 is no, execute the step S380.
  • the sample characteristics in the energy efficiency sample include the state parameters of the data center obtained when the previous control period arrives, and the number of startups n
  • the sample label is the energy efficiency score calculated when the current control period arrives
  • the sample characteristics in the temperature sample include The state parameters of the data center, the temperature set point AcTempSet, and the number of startups n obtained when the previous control cycle arrives
  • the sample label includes the temperature measurement value TempMeasure collected when the current control cycle arrives.
  • the calculation method of the energy efficiency score score is:
  • step S321 If the result of the step S321 is yes, execute the step S322; if the result of the step S320 is no, execute the step S323.
  • step S340 If the result of the step S340 is yes, execute the step S360; if the result of the step S340 is no, execute the step S350.
  • step S351 If ⁇ T ⁇ -2°C, execute the step S351; if ⁇ T>2°C, execute the step S352; if -2°C ⁇ T ⁇ -2°C, execute the step S353.
  • n is randomly increased or decreased by 1
  • AcTempSet is randomly increased or decreased by 1°C. Execute the step S370.
  • step S380 If the result of the step S380 is yes, execute the step S381; if the result of the step S380 is no, execute the step S310.
  • step S382 If ⁇ T ⁇ -DeadLine, perform step S382; if ⁇ T>DeadLine, perform step S383; if -DeadLine ⁇ T ⁇ DeadLine, return to step S310.
  • Alpha/Delta is a parameter to adjust the prediction strategy in the LinUCB model/GPUCB model, and is used to adjust whether the model is inclined to exploit or explore.
  • the UCB model learns potential regularities from historical samples, and it can predict the expected value for the working conditions to be recommended.
  • the model learns potential regularities from historical samples, and it can predict the expected value for the working conditions to be recommended.
  • it is most appropriate for the model to choose the one with the largest expected value each time, and the benefit is the largest. But when the environment changes, the new environment state may not be included in the historical samples, so the model does not learn it, and the prediction made by the model is likely to be wrong.
  • the tendency to exploit means that the model tends to choose the one with the largest expected value
  • the tendency to explore means that the model tends to choose the one with the non-maximized expected value.
  • max_depth represents the maximum depth of the tree in the xgboost model. The larger the value is, the stronger the fitting ability of the sample is, but if it is too large, it is easy to fit to the noise and cause overfitting. So the value can neither be too large nor too small.
  • max_depth 5.
  • learning_rate represents the learning rate in the xgboost model, also known as the learning step size. The smaller the value, the more iterations of the weak learner are needed, and the better the generalization. However, if the value is too small, the fitting effect may be reduced.
  • an embodiment of the present invention also provides a cluster control device for environmental parameter control equipment, which is applied to energy-saving adjustment and environmental parameter adjustment of a data center, as shown in FIG. 10 , including:
  • the sample collection module M101 is used to collect environmental parameter samples and energy efficiency samples at each arrival control period, and update them to the environmental parameter sample set and energy efficiency sample set accordingly;
  • the model training module M102 is configured to input the sample features in the energy efficiency sample set into the upper-confidence UCB model with context when triggering model training, and perform model training with the goal of outputting the sample labels in the energy efficiency sample set.
  • the sample features in the parameter sample set are input into the regression model, and the model training is performed with the goal of outputting the sample label in the environmental parameter sample set;
  • the recommendation module M104 is used to determine the set points of multiple candidate environmental parameters and the number of candidate start-ups every time the control cycle is reached when it is currently in the recommendation stage. Based on the state parameters and the number of candidate start-ups collected when the current control cycle arrives, use The UCB model predicts the predicted value of the corresponding energy efficiency score, based on the state parameter and the set point of the candidate environmental parameter, uses the regression model to predict the predicted value of the corresponding environmental parameter, and decides the number of boots based on the predicted value of the energy efficiency score and the predicted value of the environmental parameter and environmental parameter set points and run;
  • the sample characteristics in the energy efficiency sample include the state parameters and the number of startups of the data center obtained when the previous control period arrives
  • the sample label is the energy efficiency score calculated when the current control period arrives
  • the sample characteristics in the environmental parameter sample include the above The status parameters and environmental parameter set points of the data center obtained when the control cycle arrives
  • the sample tag includes the measured values of the environmental parameters collected when the current control cycle arrives.
  • the energy efficiency score is a score determined according to the energy efficiency index of the data center
  • the energy efficiency index includes the total power of the data center or the total power of the environmental parameter control equipment cluster or the power usage efficiency PUE of the data center; wherein:
  • the energy efficiency score is calculated according to the energy efficiency index of the data center using the first formula
  • the energy efficiency score is calculated according to the energy efficiency index of the data center using the second formula
  • the energy efficiency score calculated by using the first formula is greater than the energy efficiency score calculated by the second formula.
  • the sample characteristics in the environmental parameter samples further include: the number of startups in the last control cycle; based on the state parameters and candidate startup numbers collected when the current control cycle arrives, use the UCB model to predict the corresponding energy efficiency score, based on The state parameter and the candidate environmental parameter set point use a regression model to predict the predicted value of the corresponding environmental parameter, and decide the number of startups and the environmental parameter set point based on the energy efficiency score and the predicted value of the environmental parameter, including:
  • the environmental parameter set point is determined according to the predicted value of the environmental parameter.
  • sample characteristics in the energy efficiency sample further include: the environmental parameter set point of the last control cycle;
  • the UCB model Based on the state parameters collected when the current control cycle arrives and the number of candidate start-ups, use the UCB model to predict the corresponding energy efficiency score, and based on the state parameters and candidate environmental parameter set points, use the regression model to predict the corresponding environmental parameter prediction value, based on The energy efficiency score and the predicted value of the environmental parameters determine the number of start-ups and the set points of the environmental parameters, including:
  • the power-on number is determined according to the predicted value of the energy efficiency score.
  • the UCB model is used to predict the corresponding energy efficiency score, and based on the state parameters and candidate environmental parameter set points, the regression model is used to predict the corresponding environmental parameters.
  • the predicted value based on the energy efficiency score and the predicted value of the environmental parameter, determines the number of startups and the set point of the environmental parameter, including:
  • the environmental parameter set point is determined according to the predicted value of the environmental parameter.
  • the decision on the number of startups is made according to the predicted value of the energy efficiency score, including:
  • Decision-making of the environmental parameter set point according to the predicted value of the environmental parameter includes:
  • the candidate environmental parameter set point corresponding to the predicted value of the environmental parameter closest to the target environmental parameter is determined as the environmental parameter set point.
  • trigger model training when a control period is reached
  • the sample features in the energy efficiency index sample set are input into the upper confidence UCB model with context, and the model training is carried out with the goal of outputting the sample labels in the energy efficiency sample set, including:
  • the sample features in the environmental parameter sample set are input into the regression model, and the model training is carried out with the goal of outputting the label in the environmental parameter sample set, including:
  • the sample features of all environmental parameter samples in the environmental parameter sample set are sequentially input into the regression model, and the model training is performed with the goal of outputting corresponding sample labels.
  • the device also includes:
  • the initial control module M103 is used to determine that when the model decision-making conditions are not satisfied, each time a control cycle is reached, the measured value of the environmental parameters collected when the current control cycle of the data center environment arrives;
  • the number of startups and the set point of the environmental parameter are adjusted and run;
  • every time a control cycle is reached the measured value of the environmental parameter of the data center environment is obtained, and the number of startups and the set point of the environmental parameter are adjusted according to the difference between the measured value of the environmental parameter and the target environmental parameter, include:
  • the difference ⁇ E is less than the first threshold, it is determined that the environmental parameter set point of the current control cycle is the environmental parameter set point of the previous control cycle and the preset environmental parameter adjustment value is increased, and the number of startups in the current control cycle is the previous one.
  • the number of start-ups in the control cycle reduces the adjustment value of the number of start-ups;
  • the difference ⁇ E is greater than the second threshold, it is determined that the environmental parameter set point of the current control cycle is the environmental parameter set point of the previous control cycle and the preset environmental parameter adjustment value is reduced, and the number of startups in the current control cycle is the previous one.
  • the number of start-ups in the control cycle increases the preset start-up number adjustment value
  • the environmental parameter set point of the current control cycle is the environmental parameter set point of the previous control cycle, randomly increasing or decreasing the preset environmental parameter adjustment value, the number of start-ups in the current control cycle is the number of start-ups in the previous control cycle, which randomly increases or decreases the preset number of start-up adjustments;
  • EnvirMeasure is the environmental parameter measurement value collected when the current control cycle arrives
  • EnvirTarget is the target environmental parameter
  • determining a plurality of candidate environmental parameter set points and a plurality of candidate start-up numbers includes:
  • Determining an alternative environmental parameter set point that meets the range of the environmental parameter set point among the plurality of alternative environmental parameter set points, and the environmental parameter set point of the previous control cycle is the candidate environmental parameter set point
  • the device also includes:
  • the safety maintenance module M105 is used to collect the state parameters of the data center every time the data sampling time is reached, and trigger the detection of the environmental parameters of the data center. and environmental parameter set points;
  • the length of one control cycle is an integer multiple of adjacent data sampling time intervals.
  • the state parameters of the data center are collected, and the detection of the environmental parameters of the data center is triggered, and when the safe environment parameter range is exceeded, the number of startups and the environmental parameters are adjusted according to the set adjustment range set points, including:
  • the environmental parameter set point of the current control period is increased by the preset environmental parameter adjustment value, and the number of startups in the current control period is reduced by the preset adjustment value of the startup number;
  • the environmental parameter set point of the current control cycle is reduced by the preset environmental parameter adjustment value, and the number of startups in the current control cycle is increased by the preset startup number adjustment value;
  • EnvirMeasure is the measured value of the environmental parameter collected when the current control period arrives
  • EnvirTarget is the target environmental parameter
  • DeadLine is the difference value of the preset environmental parameter and is a positive number.
  • the UCB model is a linearly confident LinUCB model or a Gaussian UCB model.
  • the regression model is any of the following:
  • xgboost model random forest RF model, support vector machine SVM model, neural network model.
  • the state parameters include at least one of the following: load power, average air supply environment parameters, average return air environment parameters, average environment parameters on the hot aisle side, and average environment parameters on the cold aisle side.
  • the environmental parameter is temperature or humidity.
  • the specific working principle of the environmental parameter control equipment cluster control device is similar to the environmental parameter control equipment cluster control method, so it can be implemented correspondingly with reference to the specific implementation manner of the environmental parameter control equipment cluster control method, I won't repeat them here.
  • an embodiment of the present invention also provides an electronic device 100, as shown in FIG. 11 , including: a processor 110 and a memory 120 for storing instructions executable by the processor 110;
  • the controller 110 is configured to execute the instructions, so as to implement the environment parameter control device cluster control method.
  • the device 100 may have relatively large differences due to different configurations or performances, and may include one or more processors 110 and memory 120, and one or more storage media for storing application programs 131 or data 132 130.
  • the memory 120 and the storage medium 130 may be temporary storage or persistent storage.
  • the application program 131 stored in the storage medium 130 may include one or more than one units (not shown in FIG. 11 ), and each module may include a series of instruction operations for the environment parameter control device cluster control device.
  • the processor 110 may be configured to communicate with the storage medium 130 , and execute a series of instruction operations in the storage medium 130 on the device 100 .
  • the device 100 may also include one or more power supplies (not shown in FIG.
  • one or more network interfaces 140 the network interface 140 including a wired network interface 141 or a wireless network interface 142; one or more input Output interface 143; And/or, one or more operating systems 133, such as Windows, Mac OS, Linux, IOS, Android, Unix, FreeBSD etc.
  • FIG. 12 schematically illustrates a data center control system composed of electronic devices 100 provided by an embodiment of the present invention.
  • the data center control system includes the electronic equipment 100 , data center equipment 200 and monitoring system equipment 300 .
  • the data center equipment 200 includes a temperature control equipment cluster and/or a humidity control equipment cluster, a cabinet installed with a data center server equipment, a temperature sensor and/or a humidity sensor, and the like.
  • the monitoring system equipment 300 is at least one, and is used to control the operation status of the data center equipment.
  • the electronic device 100 receives the status parameters collected by the data center processed and forwarded by the monitoring system device 300, and decides the setting point of the environment parameters and the number of boots according to the status parameters, and the monitoring system device 300 according to the electronic The decision result of the device 100 controls the corresponding data center device 200 .
  • an embodiment of the present invention further provides a computer storage medium, where the computer storage medium stores a computer program, and the computer program is used to implement the method for controlling an environment parameter control device cluster.
  • the temperature and humidity control equipment cluster control method, device, equipment, and storage medium provided by the embodiments of the present invention combine the UCB model algorithm with context with the regression model algorithm to recommend data center configuration, and have high accuracy and convergence speed. Adjust the number of start-ups and temperature set points of the air conditioner with a certain control cycle, so as to improve the temperature of the data center and reduce the PUE, and it is safe and reliable.
  • the embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.
  • These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions
  • the device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Software Systems (AREA)
  • Pure & Applied Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Operations Research (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Thermal Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Hardware Design (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Medical Informatics (AREA)
  • Strategic Management (AREA)
  • Microelectronics & Electronic Packaging (AREA)
  • Algebra (AREA)
  • Development Economics (AREA)
  • Educational Administration (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Automation & Control Theory (AREA)
  • Marketing (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)

Abstract

本发明实施例提供了一种环境参数控制设备集群控制方法、装置、设备及存储介质，所述方法包括：每到达控制周期，采集环境参数样本和能效样本，并对应更新至环境参数样本集和能效样本集中；利用能效样本集训练带上下文的UCB模型进行模型训练，利用环境参数样本集训练回归模型；在推荐阶段时，确定多个候选环境参数设定点和候选开机数，基于当前控制周期到达时采集的状态参数和候选开机数，利用UCB模型预测对应的能效评分预测值，基于状态参数、候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于能效评分预测值和环境参数预测值决策开机数和环境参数设定点并运行。

Description

环境参数控制设备集群控制方法、装置、设备及存储介质
本申请要求于2021年6月7日提交中国专利局、申请号为202110632624.5、发明名称为“环境参数控制设备集群控制方法、装置、设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及自动化技术领域，尤指一种环境参数控制设备集群控制方法、装置、设备及存储介质。
背景技术
伴随着大数据的迅猛发展，数据处理量不断增长，数据中心也迅速增多。数据中心通常由多个机柜、多台服务器、不间断电源(Uninterruptible Power Supply，UPS)、多台空调、温湿度传感器等等组成。数据中心的环境时刻受到服务器产生的热量、环境变化等多种因素影响，而空调正是用来调节数据中心的温度等，从而保证数据中心的温度恒定。空调作为大功率电器在使用过程中消耗了大量电能，且在不同的工作功率下电能转化成冷量的效率不同，因此如何对多台空调进行控制，降低数据中心的电源使用效率(Power Usage Effectiveness，PUE)成为大势所趋。
传统数据中心空调的群控方式比较简单，大多都是根据人工经验设计规则对空调进行群控，效果不够理想。
近年来陆续有科研单位和科研人员研究机器学习的空调群控方法，研究方向基本围绕着强化学习模型、神经网络模型和回归模型。其中单一的回归模型准确率稍有不足，在线学习能力相对较弱。强化学习模型和神经网络模型的学习能力较为优秀，但调参困难且收敛速度较慢。因此，在实际使用过程中，使用这些模型进行空调群控，设备安全性和能耗控制的效果不好。
发明内容
本发明实施例提供一种环境参数控制设备集群控制方法、装置、设备及存储介质，用以解决现有技术中存在使用现有模型单独进行空调群控，设备安全性和能耗控制的效果不好问题。
本发明实施例提供了一种环境参数控制设备集群控制方法，应用于数据中心的节能调节和环境参数调节，包括：
每到达控制周期，采集环境参数样本和能效样本，并对应更新至环境参数样本集和能效样本集中；
触发模型训练时，将所述能效样本集中的样本特征输入带上下文的UCB模型，以输出所述能效样本集中的样本标签为目标进行模型训练，将所述环境参数样本集中样本特征输入回归模型，以输出所述环境参数样本集中样本标签为 目标进行模型训练；
确定当前处于推荐阶段时，每到达控制周期，确定多个候选环境参数设定点和多个候选开机数，基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点并运行；
其中，所述能效样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、开机数，样本标签为到达当前控制周期时计算的能效评分，所述环境参数样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、环境参数设定点，样本标签包括当前控制周期到达时采集的环境参数测量值。
可选地，所述能效评分为根据所述数据中心的能效指标确定的评分；
所述能效指标包括所述数据中心的总功率或所述环境参数控制设备集群的总功率或所述数据中心的电源使用效率PUE；其中：
若当前控制周期到达时采集的环境参数测量值与上一个控制周期的环境参数设定点小于等于预设环境参数偏差，所述能效评分根据所述数据中心的能效指标利用第一公式计算得到；
若当前控制周期到达时采集的环境参数测量值与上一个控制周期的环境参数设定点大于预设环境参数偏差，所述能效评分根据所述数据中心的能效指标利用第二公式计算得到；
其中，当所述能效指标相同时，利用所述第一公式计算得到的所述能效评分大于所述第二公式计算得到的所述能效评分。
可选地，所述环境参数样本中样本特征还包括：上一控制周期的开机数；
基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点，包括：
将所述候选开机数、所述当前控制周期到达时采集的状态参数输入所述UCB模型中，分别预测对应的能效评分预测值；
根据所述能效评分预测值决策所述开机数；
将所述开机数、所述当前控制周期到达时采集的状态参数和所述候选环境参数设定点输入所述回归模型中，预测对应的环境参数预测值；
根据所述环境参数预测值决策所述环境参数设定点。
可选地，所述能效样本中样本特征还包括：上一控制周期的环境参数设定点；
基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点，包括：
将当前控制周期到达时采集的状态参数和候选环境参数设定点输入所述回 归模型中，预测对应的环境参数预测值；
根据所述环境参数预测值决策所述环境参数设定点；
将所述候选开机数、所述环境参数设定点、当前控制周期到达时采集的状态参数输入所述UCB模型中，分别预测对应的能效评分预测值；
根据所述能效评分预测值决策所述开机数。
可选地，基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点，包括：
将所述候选开机数、所述当前控制周期到达时采集的状态参数输入所述UCB模型中，分别预测对应的能效评分预测值；
根据所述能效评分预测值决策所述开机数；
将所述当前控制周期到达时采集的状态参数和所述候选环境参数设定点输入所述回归模型中，分别预测对应的环境参数预测值；
根据所述环境参数预测值决策所述环境参数设定点。
可选地，若所述能效评分越大，所述环境参数控制设备集群的能耗效率越高，则根据所述能效评分预测值决策所述开机数，包括：
将最大的所述能效评分预测值对应的候选开机数决策为所述开机数；
根据所述环境参数预测值决策所述环境参数设定点，包括：
将最接近目标环境参数的所述环境参数预测值对应的候选环境参数设定点决策为所述环境参数设定点。
可选地，当到达控制周期时，触发模型训练；
将所述能效指标样本集中的样本特征输入带上下文的上置信UCB模型，以输出能效样本集中的样本标签为目标进行模型训练，包括：
将所述能效样本集中当前控制周期的能效样本中的样本特征输入UCB模型，以输出所述当前控制周期的能效样本中的样本标签为目标进行模型训练；或
将所述能效样本集中所有能效样本的样本特征依次输入上置信UCB模型，以输出对应的样本标签为目标进行模型训练；
将所述环境参数样本集中样本特征输入回归模型，以输出所述环境参数样本集中标签为目标进行模型训练，包括：
将所述环境参数样本集中所有环境参数样本的样本特征依次输入回归模型，以输出对应的样本标签为目标进行模型训练。
可选地，所述的方法还包括：
确定不满足模型决策条件时，每达到一个控制周期，获取数据中心环境的当前控制周期到达时采集的环境参数测量值；
根据环境参数测量值与目标环境参数的差值调整开机数和环境参数设定点并运行；
确定满足模型决策条件时，确定进入推荐阶段。
可选地，确定不满足模型决策条件时，每达到一个控制周期，获取数据中 心环境的环境参数测量值，根据环境参数测量值与目标环境参数的差值调整开机数和环境参数设定点，包括：
确定不满足模型决策条件时，每到达一个控制周期时，计算差值ΔE＝EnvirMeasure-EnvirTarget；
若所述差值ΔE小于第一阈值，确定当前控制周期的环境参数设定点为上一个控制周期的环境参数设定点增大预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数减少开机数调整值；
若所述差值ΔE大于第二阈值，确定当前控制周期的环境参数设定点为上一个控制周期的环境参数设定点减小预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数增大预设开机数调整值；
若所述差值ΔE大于等于第一阈值且小于等于第二阈值，确定当前控制周期的环境参数设定点为上一个控制周期的环境参数设定点随机增大或减小预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数随机增大或减小预设开机数调整值；
其中，EnvirMeasure为当前控制周期到达时采集的环境参数测量值，EnvirTarget为所述目标环境参数。
可选地，确定多个候选环境参数设定点和多个候选开机数，包括：
确定以上一个控制周期的环境参数设定点为中心的多个备选环境参数设定点；
确定所述多个备选环境参数设定点中符合环境参数设定点范围的备选环境参数设定点，及所述上一个控制周期的环境参数设定点为候选环境参数设定点；
确定以上一个控制周期的开机数为中心的多个备选开机数；
确定所述多个备选开机数中符合开机数范围的备选开机数，及所述上一个控制周期的开机数为候选开机数。
可选地，所述的方法还包括：
每到达数据采样时间，采集所述数据中心的状态参数，并触发对数据中心的环境参数进行检测，在超出安全环境参数范围时，按照设定的调整幅度调整开机数和环境参数设定点；
其中，一个所述控制周期的长度等于相邻两个所述状态参数采样时刻的间隔的整数倍。
可选地，每到达数据采样时间，采集所述数据中心的状态参数，并触发对数据中心的环境参数进行检测，在超出安全环境参数范围时，按照设定的调整幅度调整开机数和环境参数设定点，包括：
每到达状态参数采样时间，并触发检测数据中心的环境参数EnvirMeasure；
计算ΔE＝EnvirMeasure-EnvirTarget；
若ΔE＜-DeadLine，将当前控制周期的环境参数设定点增大预设环境参数调整值，当前控制周期的开机数减小预设开机数调整值；
若ΔE＞DeadLine，将当前控制周期的环境参数设定点减小预设环境参数调整值，当前控制周期的开机数增大预设开机数调整值；
其中，EnvirMeasure为当前控制周期到达时采集的环境参数测量值，EnvirTarget为所述目标环境参数，DeadLine为预设环境参数差值且为正数。
可选地，所述UCB模型为线性上置信LinUCB模型或高斯UCB模型。
可选地，所述回归模型为如下任一种：
xgboost模型、随机森林RF模型、支持向量机SVM模型、神经网络模型。
可选地，所述状态参数包括如下至少一种：负载功率、平均送风环境参数、平均回风环境参数、热通道侧平均环境参数、冷通道侧平均环境参数。
可选地，所述环境参数为温度或者湿度。
基于同一发明构思，本发明实施例还提供了一种环境参数控制设备集群控制装置，应用于数据中心的节能调节和环境参数调节，包括：
样本采集模块，用于每到达控制周期，采集环境参数样本和能效样本，并对应更新至环境参数样本集和能效样本集中；
模型训练模块，用于触发模型训练时，将所述能效样本集中的样本特征输入带上下文的上置信UCB模型，以输出所述能效样本集中的样本标签为目标进行模型训练，将所述环境参数样本集中样本特征输入回归模型，以输出所述环境参数样本集中样本标签为目标进行模型训练；
推荐模块，用于确定当前处于推荐阶段时，每到达控制周期，确定多个候选环境参数设定点和多个候选开机数，基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点并运行；
其中，所述能效样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、开机数，样本标签为到达当前控制周期时计算的能效评分，所述环境参数样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、环境参数设定点，样本标签包括当前控制周期到达时采集的环境参数测量值。
可选地，所述环境参数样本中样本特征还包括：上一控制周期的开机数；
基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点，包括：
将所述候选开机数、所述当前控制周期到达时采集的状态参数输入所述UCB模型中，分别预测对应的能效评分预测值；
根据所述能效评分预测值决策所述开机数；
将所述开机数、所述当前控制周期到达时采集的状态参数和所述候选环境参数设定点输入所述回归模型中，预测对应的环境参数预测值；
根据所述环境参数预测值决策所述环境参数设定点。
可选地，所述能效样本中样本特征还包括：上一控制周期的环境参数设定点；
基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点，包括：
将所述候选环境参数设定点和所述当前控制周期到达时采集的状态参数输入所述回归模型中，预测对应的环境参数预测值；
根据所述环境参数预测值决策所述环境参数设定点；
将所述候选开机数、所述当前控制周期到达时采集的状态参数和所述环境参数设定点输入所述UCB模型中，分别预测对应的能效评分预测值；
根据所述能效评分预测值决策所述开机数。
可选地，基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点，包括：
将所述候选开机数和所述当前控制周期到达时采集的状态参数输入所述UCB模型中，分别预测对应的能效评分预测值；
根据所述能效评分预测值决策所述开机数；
将所述候选环境参数设定点和所述当前控制周期到达时采集的状态参数输入所述回归模型中，分别预测对应的环境参数预测值；
根据所述环境参数预测值决策所述环境参数设定点。
可选地，所述的装置还包括：
初始控制模块，用于确定不满足模型决策条件时，每达到一个控制周期，获取数据中心环境的当前控制周期到达时采集的环境参数测量值；
根据环境参数测量值与目标环境参数的差值调整开机数和环境参数设定点并运行；
确定满足模型决策条件时，确定进入推荐阶段。
可选地，确定不满足模型决策条件时，每达到一个控制周期，获取数据中心环境的环境参数测量值，根据环境参数测量值与目标环境参数的差值调整开机数和环境参数设定点，包括：
确定不满足模型决策条件时，每到达一个控制周期时，计算差值ΔE＝EnvirMeasure-EnvirTarget；
若所述差值ΔE小于第一阈值，确定当前控制周期的环境参数设定点为上一个控制周期的环境参数设定点增大预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数减少开机数调整值；
若所述差值ΔE大于第二阈值，确定当前控制周期的环境参数设定点为上一个控制周期的环境参数设定点减小预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数增大预设开机数调整值；
若所述差值ΔE大于等于第一阈值且小于等于第二阈值，确定当前控制周期的环境参数设定点为上一个控制周期的环境参数设定点随机增大或减小预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数随机增大或减小预设开机数调整值；
其中，EnvirMeasure为当前控制周期到达时采集的环境参数测量值，EnvirTarget为所述目标环境参数。
可选地，所述的装置还包括：
安全维护模块，用于每到达数据采样时间，采集所述数据中心的状态参数，并触发对数据中心的环境参数进行检测，在超出安全环境参数范围时，按照设定的调整幅度调整开机数和环境参数设定点；
其中，一个所述控制周期的长度为相邻数据采样时间间隔的整数倍。
可选地，每到达数据采样时间，采集所述数据中心的状态参数，并触发对数据中心的环境参数进行检测，在超出安全环境参数范围时，按照设定的调整幅度调整开机数和环境参数设定点，包括：
每到达数据采样时间，采集所述数据中心的状态参数，并触发检测数据中心的环境参数EnvirMeasure；
计算ΔE＝EnvirMeasure-EnvirTarget；
若ΔE＜-DeadLine，将当前控制周期的环境参数设定点增大预设环境参数调整值，当前控制周期的开机数减小预设开机数调整值；
若ΔE＞DeadLine，将当前控制周期的环境参数设定点减小预设环境参数调整值，当前控制周期的开机数增大预设开机数调整值；
其中，EnvirMeasure为当前控制周期到达时采集的环境参数测量值，EnvirTarget为所述目标环境参数，DeadLine为预设环境参数差值且为正数。
基于同一发明构思，本发明实施例还提供了一种电子设备，包括：处理器和用于存储所述处理器可执行指令的存储器；
其中，所述处理器被配置为执行所述指令，以实现所述的环境参数控制设备集群控制方法。
基于同一发明构思，本发明实施例还提供了一种存储介质，所述计算机存储介质存储有计算机程序，所述计算机程序被用于实现所述的环境参数控制设备集群控制方法。
本发明有益效果如下：
本发明实施例提供的环境参数控制设备集群控制方法、装置、设备及存储介质，通过将带上下文的UCB模型算法与回归模型算法相结合进行数据中心配置推荐，采用组合模型将空调配置决策的动作空间解耦，使得动作空间缩小近十倍，加快了学习收敛速度，主模型采用UCB算法同样具有较快的收敛速度和在线学习能力，同时具有较高的准确性。以一定的控制周期对空调的开机数和环境参数设定点进行调节，从而改良数据中心的环境参数并降低PUE，且安全可靠。
附图说明
图1为本发明实施例提供的环境参数控制设备集群控制方法的流程图之一；
图2为本发明实施例中训练UCB模型和回归模型的效果示意图；
图3为本发明实施例提供的环境参数控制设备集群控制方法的流程图之二；
图4-1为本发明实施例中使用UCB模型和回归模型进行推荐的输入输出示意图之一；
图4-2为本发明实施例中使用UCB模型和回归模型进行推荐的输入输出示意图之二；
图4-3为本发明实施例中使用UCB模型和回归模型进行推荐的输入输出示意图之三；
图5为本发明实施例中环境参数控制设备集群控制方法的效果示意图；
图6为本发明实施例提供的环境参数控制设备集群控制方法的流程图之三；
图7为本发明实施例提供的环境参数控制设备集群控制方法的流程图之四；
图8-1为图4-1示意的UCB模型具体输入输出示意图；
图8-2为图4-1示意的回归模型具体输入输出示意图；
图9为本发明实施例提供的环境参数控制设备集群控制方法的流程图之五；
图10为本发明实施例提供的环境参数控制设备集群控制装置的结构示意图；
图11为本发明实施例提供的电子的结构示意图；
图12为应用了图11示意的电子设备的数据中心控制系统的结构示意图。
具体实施方式
为使本发明的上述目的、特征和优点能够更为明显易懂，下面将结合附图和实施例对本发明做进一步说明。然而，示例实施方式能够以多种形式实施，且不应被理解为限于在此阐述的实施方式；相反，提供这些实施方式使得本发明更全面和完整，并将示例实施方式的构思全面地传达给本领域的技术人员。在图中相同的附图标记表示相同或类似的结构，因而将省略对它们的重复描述。本发明中所描述的表达位置与方向的词，均是以附图为例进行的说明，但根据需要也可以做出改变，所做改变均包含在本发明保护范围内。本发明的附图仅用于示意相对位置关系不代表真实比例。
需要说明的是，在以下描述中阐述了具体细节以便于充分理解本发明。但是本发明能够以多种不同于在此描述的其它方式来实施，本领域技术人员可以在不违背本发明内涵的情况下做类似推广。因此本发明不受下面公开的具体实施方式的限制。说明书后续描述为实施本申请的较佳实施方式，然所述描述乃以说明本申请的一般原则为目的，并非用以限定本申请的范围。本申请的保护范围当视所附权利要求所界定者为准。
在本发明实施例中，所述环境参数为温度或者湿度。下文将以所述环境参数为温度为例进行说明，所述环境参数为湿度的实施方式与温度的实施方式基 本相同，故可以参考温度的实施例，下文不再赘述。
在介绍本发明实施例之前，首先对下文将要出现的名词进行解释。
温度设定点：温度控制设备集群在运行过程中所设置的温度。例如温度控制设备为空调，那么空调的温度设定点即为空调的出风温度。
温度测量值：通过温度传感器采集的温度控制设备集群所要控制的数据中心的环境温度。
目标温度：期望的通过所述温度控制设备集群控制所述数据中心所最终达到的环境温度。
下面结合附图，对本发明实施例提供的环境参数控制设备集群控制方法、装置、设备及存储介质进行具体说明。
第一方面：
本发明实施例提供了一种环境参数控制设备集群控制方法，应用于数据中心的节能调节和温度调节，如图1所示，包括：
开始环境参数控制设备集群控制时，各自进行决策控制与模型训练两个部分的步骤；首先执行决策控制部分的步骤S110与模型训练部分的步骤S210；
S110、判断是否到达控制周期；
若所述步骤S110的结果为是，执行步骤S120；
如图2所示，S120、每到达控制周期，采集温度样本和能效样本，并对应更新至温度样本集和能效样本集中；
其中，所述能效样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、开机数，样本标签为到达当前控制周期时计算的能效评分，所述温度样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、温度设定点，样本标签包括当前控制周期到达时采集的温度测量值；
S130、判断是否满足模型决策条件；
若所述步骤S130的结果为是，确定当前处于推荐阶段，执行所述步骤S150；
S150、每到达控制周期，确定多个候选温度设定点和多个候选开机数，基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数、候选温度设定点，利用回归模型预测对应环境温度预测值，基于所述能效评分预测值和环境温度预测值决策开机数和温度设定点并运行。
S210、判断是否满足模型训练条件；
若所述步骤S210的结果为是，执行步骤S220；若所述步骤S210的结果为否，继续等待直至所述步骤S210的结果为是；
S220、将所述能效样本集中的样本特征输入带上下文的上置信(Upper Confidence Bound，UCB)模型，以输出所述能效样本集中的样本标签为目标进行模型训练，将所述温度样本集中样本特征输入回归模型，以输出所述温度样本集中样本标签为目标进行模型训练。
在具体实施过程中，所述UCB模型是带上下文的UCB模型。可选地，所述UCB模型为线性上置信(Linear Upper Confidence Bound，LinUCB)模型或高斯上置信(Gaussian Process Upper Confidence Bound，GPUCB)模型。
在具体实施过程中，可选地，所述回归模型为如下任一种：
xgboost模型、随机森林(Random Fores，RF)模型、支持向量机(Support Vector Machine，SVM)模型、神经网络模型。
所述回归模型也可以为其它未提及的模型，可以根据实际需要选择，在此不作限定。
在具体实施过程中，所述步骤S210中的模型训练条件可以为每个所述控制周期中触发一次模型训练，也可以为间隔多个控制周期触发一次模型训练，还可以为满足一定条件时(例如，连续多个所述控制周期到达时，使用所述步骤S150决策得到的温度设定点与开机数控制所述环境参数控制设备集群，在下一个所述控制周期到达时所获取的数据中心的状态参数不符合预设状态参数范围)触发一次模型训练，在此不作限定。
作为一种可选的实施方式，当到达控制周期时，触发模型训练。即每个所述控制周期都进行一次模型训练。其中每个所述控制周期中触发一次模型训练的实施方式可以参见图3示意的流程图，其中图3中示意的步骤与图1基本一致，可以参见上文所述的内容，故不再赘述。
在具体实施过程中，可选地，所述步骤S120中，对应更新至温度样本集和能效样本集中，包括：
若所述温度样本集和所述能效样本集的样本数量等于样本集容量时，删除所述温度样本集和所述能效样本集中最早的控制周期对应的温度样本和能效样本，将当前控制周期对应的所述温度样本和所述能效样本对应更新至所述温度样本集和所述能效样本集中。
例如，所述控制周期的时间为1小时，所述温度样本集和所述能效样本集的样本集容量设置为30天的样本数量，那么所述样本集容量为24×30＝720个样本。当所述环境参数控制设备集群运行至第721个小时的时候，将删除所述两个样本集中第1个控制周期对应的温度样本和能效样本，并将第721个控制周期对应的所述温度样本和能效样本对应更新至所述两个样本集中。之后的控制周期，在第722个控制周期时删除第2个控制周期对应的温度样本和能效样本，将第722个控制周期对应的温度样本和能效样本对应更新至所述两样本集，依次类推。
本申请通过带上下文的UCB模型算法与回归模型算法相结合进行数据中心温度和能耗配置推荐，具有较高的准确性和收敛速度。
在实施过程中，需要分别使用所述能耗样本集和所述温度样本集对应训练所述UCB模型和所述回归模型。
对于所述回归模型的训练过程，可选地，将所述温度样本集中样本特征输入回归模型，以输出所述温度样本集中标签为目标进行模型训练，包括：
将所述温度样本集中所有温度样本的样本特征依次输入回归模型，以输出对应的样本标签为目标进行模型训练。
而对于所述UCB模型的训练过程，将所述能效指标样本集中的样本特征输入带上下文的上置信UCB模型，以输出能效样本集中的样本标签为目标进行模型训练，可以采用如下任一种实施方式：
方式A：将所述能效样本集中所有能效样本的样本特征依次输入所述UCB模型，以输出对应的样本标签为目标进行模型训练。
方式B：当到达控制周期时，触发模型训练。将所述能效样本集中当前控制周期的能效样本中的样本特征输入所述UCB模型，以输出所述当前控制周期的能效样本中的样本标签为目标进行模型训练。
这样，使用方式B进行对所述UCB模型进行训练，每到达控制周期时仅使用更新的能效样本进行训练，可以减少训练数据的数量，加快训练速度。
可选地，如图1和图3所示，所述的方法还包括：
若所述步骤S130的结果为否，确定当前处于初始化控制阶段，执行所述步骤S141；
S141、每达到一个控制周期，获取数据中心环境的当前控制周期到达时采集的温度测量值；
S142、根据温度测量值与目标温度的差值调整开机数和温度设定点并运行。
在具体实施过程中，由于受到部分数据中心不具备定制化数据中心的特点，数据中心所在位置环境不一(例如不同地区的气温变化规律不同，使得数据中心的温度与能效的变化规律也相应有所不同)等因素影响，因此预先设置统一的训练样本对所述UCB模型和所述回归模型进行训练，使用这种方式训练得到的模型进行环境参数控制设备集群的推荐控制，其安全性和节能效果可能都存在问题。那么，在事先没有数据供模型训练的情况下，通过上述实施方式中首先按照温度测量值与目标温度的差值调整开机数和温度设定点，并在每个所述控制周期到达时采集所述温度样本和所述能效样本，从而为所述UCB模型和所述回归模型积累训练样本，便于后续使用训练后的模型进行推荐控制。
可选地，所述步骤S142中，根据温度测量值与目标温度的差值调整开机数和温度设定点，包括：
确定不满足模型决策条件时，每到达一个控制周期时，计算差值ΔT＝TempMeasure-TempTarget；
若所述差值ΔT小于第一阈值，确定当前控制周期的温度设定点为上一个控制周期的温度设定点增大预设温度调整值，当前控制周期的开机数为上一个控制周期的开机数减少开机数调整值；
若所述差值ΔT大于第二阈值，确定当前控制周期的温度设定点为上一个控制周期的温度设定点减小预设温度调整值，当前控制周期的开机数为上一个控制周期的开机数增大预设开机数调整值；
若所述差值ΔT大于等于第一阈值且小于等于第二阈值，确定当前控制周期的温度设定点为上一个控制周期的温度设定点随机增大或减小预设温度调整值，当前控制周期的开机数为上一个控制周期的开机数随机增大或减小预设开机数调整值；
其中，TempMeasure为当前控制周期到达时采集的温度测量值，TempTarget为所述目标温度。
例如，所述预设温度调整值为1℃，所述预设开机数调整值为1，所述第一阈值为-2℃，所述第二阈值为2℃。那么，当所述差值ΔT＜-2℃时，所述环境参数控制设备集群将所述温度设定点增大1℃，将所述开机数减少1台；当所述差值ΔT＞2℃时，所述环境参数控制设备集群将所述温度设定点减小1℃，将所述开机数增加1台；当所述差值-2℃≤ΔT≤2℃时，所述环境参数控制设备集群将所述温度设定点随机增大或减小1℃，将所述开机数随机增加或减少1台。
这样，将所述温度测量值大幅低于所述目标温度时提高所述温度设定点并减少开机数，能够控制所述数据中心的环境温度升高；将所述温度测量值大幅高于所述目标温度时降低所述温度设定点并增加开机数，能够控制所述数据中心的环境温度降低，从而使数据中心的环境温度尽快调整至所述目标温度。而在所述温度测量值与所述目标温度相近时随机控制所述温度设定点和所述开机数的变化，能够丰富对所述UCB模型和所述回归模型训练的训练样本，提高训练后得到的模型推荐的可靠性。
可选地，所述能效评分为根据所述数据中心的能效指标确定的评分；
所述能效指标包括所述数据中心的总功率或所述环境参数控制设备集群的总功率或所述数据中心的PUE；其中：
若|TempMeasure-Temp|≤DeadLine，所述能效评分根据所述数据中心的能效指标利用第一公式计算得到；
若|TempMeasure-Temp|＞DeadLine，所述能效评分根据所述数据中心的能效指标利用第二公式计算得到；
其中，当所述能效指标相同时，利用所述第一公式计算得到的所述能效评分大于所述第二公式计算得到的所述能效评分；
TempMeasure为当前控制周期到达时采集的温度测量值，Temp为上一个控制周期的温度设定点，DeadLine为预设温度差值且为正数。
在具体实施过程中，以所述能效指标为PUE为例。可选地，所述第一公式为：
Figure PCTCN2021112714-appb-000001
所述第二公式为：
Figure PCTCN2021112714-appb-000002
其中score为所述能效评分，A＞B。例如，A＝1，B＝0.8。
由于所述数据中心的总功率、所述环境参数控制设备集群的总功率与所述数据中心的PUE具有相似的变化规律，也可以将上述的第一和第二公式中的 PUE替换为所述数据中心的总功率和所述环境参数控制设备集群的总功率，并根据需要调整A和B的取值，此处不再赘述。
当然，所述第一公式与所述第二公式不局限于上述的反比例关系，也可以为其它类型的公式，此处不作限定。
这样，通过不同的公式计算温度测量值与上一个控制周期的温度设定点的能效评分，能够对两者数值相差较大的情况给予更低的能效评分，从而能够使所述UCB模型在进行推荐决策时考虑到温度的影响。
可选地，在所述步骤S150中，确定多个候选温度设定点和多个候选开机数，包括：
确定以上一个控制周期的温度设定点为中心的多个备选温度设定点；
确定所述多个备选温度设定点中符合温度设定点范围的备选温度设定点，及所述上一个控制周期的温度设定点为候选温度设定点；
确定以上一个控制周期的开机数为中心的多个备选开机数；
确定所述多个备选开机数中符合开机数范围的备选开机数，及所述上一个控制周期的开机数为候选开机数。
在具体实施过程中，所述温度设定点范围为预先设置的一个范围，例如10℃-30℃。当上一个控制周期的温度设定点为30℃时，所述备选温度设定点将存在大于30℃和小于30℃的数值，但大于30℃的备选温度设定点不符合所述温度设定点范围，将被舍弃。最终确定所述候选温度设定点为小于30℃的备选温度设定点和30℃。对于所述开机数范围而言，可以直接确定为0至所述环境参数控制设备集群的设备数，也可以进一步地设置为其子集合(例如所述环境参数控制设备集群的设备数的一半至全部)。若直接使用0至所述环境参数控制设备集群的设备数的范围，当所述备选开机数小于0或大于所述环境参数控制设备集群的设备数时，此备选开机数将被舍弃。
在具体实施过程中，可以根据需要确定所述备选温度设定点的数量n 1，及数值相邻的两个所述备选温度设定点的数值差Δ 1，及所述备选开机数的数量n 2，及数值相邻的两个所述备选开机数的数值差Δ 2(n 1、n 2均为正偶数)。例如，确定所述备选温度设定点的数量n 1＝4，备选温度设定点的数值差Δ 1＝2℃，当上一个控制周期的温度设定点为20℃时，确定所述备选温度设定点为16℃、18℃、22℃、24℃。备选开机数的设置方式同理，故不再赘述。
这样，能够在每个控制周期进行推荐决策调整时，从所述候选温度设定点与所述候选开机数中最终确定的温度设定点与开机数在上一个控制周期的温度设定点与开机数附近，避免这两个设置参数剧烈变化导致温度控制设备运行异常。
作为一种优选的实施方式，确定以上一个控制周期的温度设定点为中心的多个备选温度设定点，包括：
确定以上一个控制周期的温度设定点为中心的2个备选温度设定点，并且所述备选温度设定点与上一个控制周期的温度设定点相差最小温度设定点变化 值。
例如，所述最小温度设定点变化值为1℃，当上一个控制周期的温度设定点为20℃时，确定所述备选温度设定点为19℃、21℃。
确定以上一个控制周期的开机数为中心的多个备选开机数，包括：
确定以上一个控制周期的开机数为中心的2个备选开机数，并且所述备选开机数与上一个控制周期的开机数相差1。
这样，通过每次开机数和温度设定点最多变多一个变化单位，使得数据中心温度变化更加平衡。
在推荐阶段，在完成对所述UCB模型和所述回归模型的训练后，分别使用所述UCB模型和所述回归模型预测能效评分和环境温度时，可以分别使用所述UCB模型和所述回归模型进行能效评分预测和环境温度预测，也可以将两者结合使用，利用其中一个模型的预测结果推荐决策得到的参数输入至另一个模型中以影响另一个参数的预测结果和推荐决策。
可选地，基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分，基于所述状态参数和候选温度设定点，利用回归模型预测对应环境温度预测值，基于所述能效评分和环境温度预测值决策开机数和温度设定点，包括如下任一种实施方式：
方式1：
在方式1中，所述温度样本中样本特征还包括：上一控制周期的开机数。
如图4-1所示，将所述候选开机数和所述当前控制周期到达时采集的状态参数输入所述UCB模型中，分别预测对应的能效评分预测值；
根据所述能效评分预测值决策所述开机数；
将所述开机数、所述当前控制周期到达时采集的状态参数和所述候选温度设定点输入所述回归模型中，预测对应的环境温度预测值；
根据所述环境温度预测值决策所述温度设定点。
方式2：
在方式2中，所述能效样本中样本特征还包括：上一控制周期的温度设定点。
如图4-2所示，将所述候选温度设定点和所述当前控制周期到达时采集的状态参数输入所述回归模型中，预测对应的环境温度预测值；
根据所述环境温度预测值决策所述温度设定点；
将所述候选开机数、所述当前控制周期到达时采集的状态参数和所述温度设定点输入所述UCB模型中，分别预测对应的能效评分预测值；
根据所述能效评分预测值决策所述开机数。
方式3：
如图4-3所示，将所述候选开机数和所述当前控制周期到达时采集的状态参数输入所述UCB模型中，分别预测对应的能效评分预测值；
根据所述能效评分预测值决策所述开机数；
将所述候选温度设定点和所述当前控制周期到达时采集的状态参数输入所述回归模型中，分别预测对应的环境温度预测值；
根据所述环境温度预测值决策所述温度设定点。
在具体实施过程中，决策所述环境温度预测值中最接近所述目标温度对应的温度设定点为当前控制周期的所述温度设定点。如果根据所述能效指标计算的能效评分，能效指标越高则能效评分越高，则决策所述能效评分预测值中的最大值对应的所述候选开机数为当前控制周期的开机数。
这样，方式1和方式2使用两个模型进行耦合决策开机数和温度设定点，减小了推荐空间，极大地提高了模型的收敛速度。方式3使用两个模型分别决策开机数和温度设定点，方案较为简单。
在每到达一个控制周期时，采用上述实施方式对所述环境参数控制设备集群进行开机数和温度设定点的调整外，为了保证设备运行的安全性，如图5所示，还可以设置周期性的状态参数采样时刻进行检测和调整。
可选地，如图6和图7所示，除了与图1、图3所示的方法相同的步骤外，所述的方法还包括：
S160、判断是否到达数据采样时间，采集一次所述数据中心的状态参数；
若所述步骤S160结果为是，执行所述步骤S170；若所述步骤S160结果为否，返回所述步骤S110；
S170、对数据中心的环境温度进行检测，判断是否超出安全温度范围；
若所述步骤S170的结果为是，执行步骤S180；若所述步骤S170结果为否，返回所述步骤S110；
S180、按照设定的调整幅度调整开机数和温度设定点；返回所述步骤S110；
其中，一个所述控制周期的长度为相邻数据采样时间间隔的整数倍。
在具体实施过程中，所述数据采样时间的时长可以根据实际需要进行设置。例如，所述控制周期为1小时，所述数据采样时间为5分钟。
这样，能够避免所述数据中心出现安全隐患。
可选地，所述步骤S170、对数据中心的环境温度进行检测，判断是否超出安全温度范围，包括：
检测数据中心的环境温度TempMeasure；
计算ΔE＝TempMeasure-TempTarget；
判断是否ΔT＜-DeadLine，或ΔT＞DeadLine。
所述步骤S180、按照设定的调整幅度调整开机数和温度设定点，包括：
若ΔT＜-DeadLine，将当前控制周期的温度设定点增大预设温度调整值，当前控制周期的开机数减小预设开机数调整值；
若ΔT＞DeadLine，将当前控制周期的温度设定点减小预设温度调整值，当前控制周期的开机数增大预设开机数调整值；
其中，TempMeasure为当前控制周期到达时采集的温度测量值，TempTarget为所述目标温度，DeadLine为预设温度差值且为正数。
例如，所述DeadLine为预设温度差值为3℃，所述预设温度调整值为1℃， 所述预设开机数调整值为1。
可选地，所述状态参数包括如下至少一种：负载功率、平均送风温度、平均回风温度、热通道侧平均温度、冷通道侧平均温度。
如果在模型决策时使用方式1的技术方案，那么所述UCB模型和所述回归模型的输入与输出的参数将如图8-1和图8-2所示。
在下面给出一个具体的示例，对本发明提供的环境参数控制设备集群控制方法进行说明。
在本示例中，所述温度控制设备为空调，所述控制周期为1小时，相邻数据采样时间间隔5分钟。所述UCB模型为LinUCB模型或GPUCB模型，所述回归模型为xgboost模型。所述目标温度TempTarget＝22℃，所述预设温度差值DeadLine＝3℃，所述空调集群的设备数范围为空调集群设备数的一半至全部，所述温度设定点范围为[15℃，20℃]，所述温度样本集和所述能效样本集的样本集容量为720。从第241个控制周期开始进入推荐阶段，所述第一阈值为-2℃，所述第二阈值为2℃。
如图9所示，所述控制设备集群控制方法包括：
S300、CallTime＝0。
S310、判断是否到达控制周期。
若所述步骤S310的结果为是，执行所述步骤S320；若所述步骤S310的结果为否，执行所述步骤S380。
S320、每到达控制周期，采集温度样本和能效样本。
其中，所述能效样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、开机数n，样本标签为到达当前控制周期时计算的能效评分score，所述温度样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、温度设定点AcTempSet、开机数n，样本标签包括当前控制周期到达时采集的温度测量值TempMeasure。
能效评分score的计算方法为：
若|TempMeasure-Temp|≤DeadLine，
Figure PCTCN2021112714-appb-000003
若|TempMeasure-Temp|＞DeadLine，
Figure PCTCN2021112714-appb-000004
S321、判断温度样本集和能效样本集中样本数量是否等于样本集容量。
若所述步骤S321结果为是，执行所述步骤S322；若所述步骤S320结果为否，执行所述步骤S323。
S322、删除温度样本集和能耗样本集中最早的控制周期对应的温度样本和能效样本。
S323、将采集的温度样本和能效样本对应更新至温度样本集和能效样本集中。
S330、将所述能效样本集中的样本特征输入UCB模型，以输出所述能效样本集中的样本标签为目标进行模型训练，将所述温度样本集中样本特征输入xgboost模型，以输出所述温度样本集中样本标签为目标进行模型训练。
S340、判断是否CallTime>InitTimeTh；其中InitTimeTh＝240。
若所述步骤S340的结果为是，执行所述步骤S360；若所述步骤S340的结果为否，执行所述步骤S350。
S350、获取数据中心环境的当前控制周期到达时采集的温度测量值TempMeasure，并计算差值ΔT＝TempMeasure-TempTarget。
若ΔT＜-2℃，执行所述步骤S351；若ΔT＞2℃，执行所述步骤S352；若-2℃≤ΔT＜-2℃，执行所述步骤S353。
S351、n＝n-1，AcTempSet＝AcTempSet+1℃。执行所述步骤S370。
S352、n＝n+1，AcTempSet＝AcTempSet-1℃。执行所述步骤S370。
S353、n随机增大或减小1，AcTempSet随机增大或减小1℃。执行所述步骤S370。
S360、根据上一个控制周期的温度设定点AcTempSet分别增加1℃和减小1℃，得到两个备选温度设定点，去除其中不在[15℃，20℃]中的备选温度设定点后，与上一个控制周期的温度设定点AcTempSet共同组成候选温度设定点；根据上一个控制周期的开机数n分别增加1和减小1，得到两个备选开机数，去除其中不在空调集群设备数的一半至全部中的备选开机数后，与上一个控制周期的开机数n共同组成候选开机数。
S361、将所述候选开机数、所述当前控制周期到达时采集的状态参数输入UCB模型中，分别预测对应的能效评分预测值。
S362、决策所述能效评分预测值中最大值对应的候选开机数为当前控制周期的开机数n。
S363、将当前控制周期的开机数、所述当前控制周期到达时采集的状态参数和所述候选温度设定点输入xgboost模型中，预测对应的环境温度预测值；
S364、决策所述环境温度预测值中最接近所述目标温度TempTarget对应的温度设定点为当前控制周期的温度设定点AcTempSet。
S370、CallTime＝CallTime+1。
S380、判断是否到达数据采样时间，采集一次所述数据中心的状态参数。
若所述步骤S380的结果为是，执行所述步骤S381；若所述步骤S380的结果为否，执行所述步骤S310。
S381、对数据中心的环境温度TempMeasure进行检测；计算差值ΔT＝TempMeasure-TempTarget；判断是否ΔT＜-DeadLine，或ΔT＞DeadLine。
若ΔT＜-DeadLine，执行步骤S382；若ΔT＞DeadLine，执行步骤S383；若-DeadLine≤ΔT≤DeadLine，返回所述步骤S310。
S382、AcTempSet＝AcTempSet+1℃，n＝n-1。返回所述步骤S310。
S383、AcTempSet＝AcTempSet-1℃，n＝n+1。返回所述步骤S310。
在上述示例实施过程中，需要对LinUCB模型中的参数Alpha或GPUCB模型中的参数Delta进行预先设置。Alpha/Delta是LinUCB模型/GPUCB模型中调 节预测策略的参数，用于调节模型是倾向于利用还是探索。具体来说，UCB模型从历史样本学习潜在的规律，对于待做推荐决策的工况它能预测出期望值。显然，对于不变的环境来说，模型每次选择期望值最大者是最合适的，收益最大。但是当环境变化时，新的环境状态可能不包含在历史样本中，所以模型没有对之进行学习，模型所做预测很可能是有误差的。所以，当环境变化时，进行适当的探索尝试有利于模型找到新环境下的最大期望。这样，倾向于利用是指模型倾向于选择期望值最大者，倾向于探索则是倾向于选择期望值非最大者。模型越倾向于利用则越稳定，但环境变化时的适应能力就越差一些。反之，模型越倾向于探索则对环境变化的适应能力越强，但过于频繁的探索会导致稳定性较差。
同时，也需要对xgboost模型中的参数max_depth和learning_rate进行预先设置。其中，max_depth表示xgboost模型中树的最大深度，该值越大对样本的拟合能力越强，但太大容易拟合到噪声导致过拟合。所以该值既不能太大也不能太少。在建模时，需要将数据集进行划分，根据测试结果进行选择。优选地，max_depth＝5。learning_rate表示xgboost模型中的学习率，也称之为学习步长。该值越小意味着需要更多的弱学习器的迭代，泛化性越好。但该值若过小可能会降低拟合效果。在建模时，同样需要将数据集进行划分，根据测试结果进行选择。
第二方面：
基于同一发明构思，本发明实施例还提供了一种环境参数控制设备集群控制装置，应用于数据中心的节能调节和环境参数调节，如图10所示，包括：
样本采集模块M101，用于每到达控制周期，采集环境参数样本和能效样本，并对应更新至环境参数样本集和能效样本集中；
模型训练模块M102，用于触发模型训练时，将所述能效样本集中的样本特征输入带上下文的上置信UCB模型，以输出所述能效样本集中的样本标签为目标进行模型训练，将所述环境参数样本集中样本特征输入回归模型，以输出所述环境参数样本集中样本标签为目标进行模型训练；
推荐模块M104，用于确定当前处于推荐阶段时，每到达控制周期，确定多个候选环境参数设定点和多个候选开机数，基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数、候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点并运行；
其中，所述能效样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、开机数，样本标签为到达当前控制周期时计算的能效评分，所述环境参数样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、环境参数设定点，样本标签包括当前控制周期到达时采集的环境参数测量值。
可选地，所述能效评分为根据所述数据中心的能效指标确定的评分；
所述能效指标包括所述数据中心的总功率或所述环境参数控制设备集群的 总功率或所述数据中心的电源使用效率PUE；其中：
若当前控制周期到达时采集的环境参数测量值与上一个控制周期的环境参数设定点小于等于预设环境参数偏差，所述能效评分根据所述数据中心的能效指标利用第一公式计算得到；
若当前控制周期到达时采集的环境参数测量值与上一个控制周期的环境参数设定点大于预设环境参数偏差，所述能效评分根据所述数据中心的能效指标利用第二公式计算得到；
其中，当所述能效指标相同时，利用所述第一公式计算得到的所述能效评分大于所述第二公式计算得到的所述能效评分。
可选地，所述环境参数样本中样本特征还包括：上一控制周期的开机数；基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分和环境参数预测值决策开机数和环境参数设定点，包括：
将所述候选开机数、所述当前控制周期到达时采集的状态参数输入所述UCB模型中，分别预测对应的能效评分预测值；
根据所述能效评分预测值决策所述开机数；
将所述开机数、所述当前控制周期到达时采集的状态参数和所述候选环境参数设定点输入所述回归模型中，预测对应的环境参数预测值；
根据所述环境参数预测值决策所述环境参数设定点。
可选地，所述能效样本中样本特征还包括：上一控制周期的环境参数设定点；
基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分和环境参数预测值决策开机数和环境参数设定点，包括：
将所述候选环境参数设定点和所述当前控制周期到达时采集的状态参数输入所述回归模型中，预测对应的环境参数预测值；
根据所述环境参数预测值决策所述环境参数设定点；
将所述候选开机数、当前控制周期到达时采集的状态参数和所述环境参数设定点输入所述UCB模型中，分别预测对应的能效评分预测值；
根据所述能效评分预测值决策所述开机数。
可选地，基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分和环境参数预测值决策开机数和环境参数设定点，包括：
将所述候选开机数、所述当前控制周期到达时采集的状态参数输入所述UCB模型中，分别预测对应的能效评分预测值；
根据所述能效评分预测值决策所述开机数；
将所述候选环境参数设定点和所述当前控制周期到达时采集的状态参数输 入所述回归模型中，分别预测对应的环境参数预测值；
根据所述环境参数预测值决策所述环境参数设定点。
可选地，若所述能效评分越大，所述环境参数控制设备集群的能耗效率越高，则根据所述能效评分预测值决策所述开机数，包括：
将最大的所述能效评分预测值对应的候选开机数决策为所述开机数；
根据所述环境参数预测值决策所述环境参数设定点，包括：
将最接近目标环境参数的所述环境参数预测值对应的候选环境参数设定点决策为所述环境参数设定点。
可选地，当到达控制周期时，触发模型训练；
将所述能效指标样本集中的样本特征输入带上下文的上置信UCB模型，以输出能效样本集中的样本标签为目标进行模型训练，包括：
将所述能效样本集中当前控制周期的能效样本中的样本特征输入UCB模型，以输出所述当前控制周期的能效样本中的样本标签为目标进行模型训练；或
将所述能效样本集中所有能效样本的样本特征依次输入上置信UCB模型，以输出对应的样本标签为目标进行模型训练；
将所述环境参数样本集中样本特征输入回归模型，以输出所述环境参数样本集中标签为目标进行模型训练，包括：
将所述环境参数样本集中所有环境参数样本的样本特征依次输入回归模型，以输出对应的样本标签为目标进行模型训练。
可选地，所述的装置还包括：
初始控制模块M103，用于确定不满足模型决策条件时，每达到一个控制周期，获取数据中心环境的当前控制周期到达时采集的环境参数测量值；
根据环境参数测量值与目标环境参数的差值调整开机数和环境参数设定点并运行；
确定满足模型决策条件时，确定进入推荐阶段。
可选地，确定不满足模型决策条件时，每达到一个控制周期，获取数据中心环境的环境参数测量值，根据环境参数测量值与目标环境参数的差值调整开机数和环境参数设定点，包括：
确定不满足模型决策条件时，每到达一个控制周期时，计算差值ΔE＝EnvirMeasure-EnvirTarget；
若所述差值ΔE小于第一阈值，确定当前控制周期的环境参数设定点为上一个控制周期的环境参数设定点增大预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数减少开机数调整值；
若所述差值ΔE大于第二阈值，确定当前控制周期的环境参数设定点为上一个控制周期的环境参数设定点减小预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数增大预设开机数调整值；
若所述差值ΔE大于等于第一阈值且小于等于第二阈值，确定当前控制周期 的环境参数设定点为上一个控制周期的环境参数设定点随机增大或减小预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数随机增大或减小预设开机数调整值；
其中，EnvirMeasure为当前控制周期到达时采集的环境参数测量值，EnvirTarget为所述目标环境参数。
可选地，确定多个候选环境参数设定点和多个候选开机数，包括：
确定以上一个控制周期的环境参数设定点为中心的多个备选环境参数设定点；
确定所述多个备选环境参数设定点中符合环境参数设定点范围的备选环境参数设定点，及所述上一个控制周期的环境参数设定点为候选环境参数设定点；
确定以上一个控制周期的开机数为中心的多个备选开机数；
确定所述多个备选开机数中符合开机数范围的备选开机数，及所述上一个控制周期的开机数为候选开机数。
可选地，所述的装置还包括：
安全维护模块M105，用于每到达数据采样时间，采集所述数据中心的状态参数，并触发对数据中心的环境参数进行检测，在超出安全环境参数范围时，按照设定的调整幅度调整开机数和环境参数设定点；
其中，一个所述控制周期的长度为相邻数据采样时间间隔的整数倍。
可选地，每到达数据采样时间，采集所述数据中心的状态参数，并触发对数据中心的环境参数进行检测，在超出安全环境参数范围时，按照设定的调整幅度调整开机数和环境参数设定点，包括：
每到达数据采样时间，采集所述数据中心的状态参数，并触发检测数据中心的环境参数EnvirMeasure；
计算ΔE＝EnvirMeasure-EnvirTarget；
若ΔE＜-DeadLine，将当前控制周期的环境参数设定点增大预设环境参数调整值，当前控制周期的开机数减小预设开机数调整值；
若ΔE＞DeadLine，将当前控制周期的环境参数设定点减小预设环境参数调整值，当前控制周期的开机数增大预设开机数调整值；
其中，EnvirMeasure为当前控制周期到达时采集的环境参数测量值，EnvirTarget为所述目标环境参数，DeadLine为预设环境参数差值且为正数。
可选地，所述UCB模型为线性上置信LinUCB模型或高斯UCB模型。
可选地，所述回归模型为如下任一种：
xgboost模型、随机森林RF模型、支持向量机SVM模型、神经网络模型。
可选地，所述状态参数包括如下至少一种：负载功率、平均送风环境参数、平均回风环境参数、热通道侧平均环境参数、冷通道侧平均环境参数。
可选地，所述环境参数为温度或者湿度。
在具体实施过程中，所述环境参数控制设备集群控制装置与所述环境参数控制设备集群控制方法的具体工作原理相似，故可以参考所述环境参数控制设 备集群控制方法的具体实施方式对应实施，此处不再赘述。
第三方面：
基于同一发明构思，本发明实施例还提供了一种电子设备100，如图11所示，包括：处理器110和用于存储所述处理器110可执行指令的存储器120；其中，所述处理器110被配置为执行所述指令，以实现所述环境参数控制设备集群控制方法。
在具体实施过程中，所述设备100可因配置或性能不同而产生比较大的差异，可以包括一个或一个以上处理器110和存储器120，一个或一个以上存储应用程序131或数据132的存储介质130。其中，存储器120和存储介质130可以是短暂存储或持久存储。存储在存储介质130的应用程序131可以包括一个或一个以上所述单元(图11中未示出)，每个模块可以包括对环境参数控制设备集群控制装置中的一系列指令操作。更进一步地，处理器110可以设置为与存储介质130通信，在所述设备100上执行存储介质130中的一系列指令操作。所述设备100还可以包括一个或一个以上电源(图11中未示出)；一个或一个以上网络接口140，所述网络接口140包括有线网络接口141或无线网络接口142；一个或一个以上输入输出接口143；和/或，一个或一个以上操作系统133，例如Windows、Mac OS、Linux、IOS、Android、Unix、FreeBSD等。
图12示意了应用了本发明实施例提供的电子设备100所组成的数据中心控制系统。如图12所示，所述数据中心控制系统包括所述电子设备100、数据中心设备200和监控系统设备300。其中所述数据中心设备200包括温度控制设备集群和/或湿度控制设备集群、安装有数据中心的服务器设备的机柜、温度传感器和/或湿度传感器等。所述监控系统设备300为至少一个，用于控制所述数据中心设备的运行状态。所述电子设备100接收由所述监控系统设备300处理并转发的数据中心采集的状态参数，并根据所述状态参数决策环境参数设定点及开机数，所述监控系统设备300根据所述电子设备100的决策结果控制对应的数据中心设备200。
第四方面：
基于同一发明构思，本发明实施例还提供了一种计算机存储介质，所述计算机存储介质存储有计算机程序，所述计算机程序被用于实现所述环境参数控制设备集群控制方法。
本发明实施例提供的温湿度控制设备集群控制方法、装置、设备及存储介质，将带上下文的UCB模型算法与回归模型算法相结合进行数据中心配置推荐，具有较高的准确性和收敛速度。以一定的控制周期对空调的开机数和温度设定点进行调节，从而改良数据中心的温度并降低PUE，且安全可靠。
本领域内的技术人员应明白，本申请的实施例可提供为方法、系统、或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本申请是参照根据本申请的方法、设备(系统)、和计算机程序产品的流程 图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
显然，本领域的技术人员可以对本申请进行各种改动和变型而不脱离本申请的精神和范围。这样，倘若本申请的这些修改和变型属于本申请权利要求及其等同技术的范围之内，则本申请也意图包含这些改动和变型在内。

Claims (26)

  1. 一种环境参数控制设备集群控制方法，应用于数据中心的节能调节和环境参数调节，其特征在于，包括：
    每到达控制周期，采集环境参数样本和能效样本，并对应更新至环境参数样本集和能效样本集中；
    触发模型训练时，将所述能效样本集中的样本特征输入带上下文的上置信UCB模型，以输出所述能效样本集中的样本标签为目标进行模型训练，将所述环境参数样本集中样本特征输入回归模型，以输出所述环境参数样本集中样本标签为目标进行模型训练；
    确定当前处于推荐阶段时，每到达控制周期，确定多个候选环境参数设定点和多个候选开机数，基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点并运行；
    其中，所述能效样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、开机数，样本标签为到达当前控制周期时计算的能效评分，所述环境参数样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、环境参数设定点，样本标签包括当前控制周期到达时采集的环境参数测量值。
  2. 如权利要求1所述的方法，其特征在于，所述能效评分为根据所述数据中心的能效指标确定的评分；
    所述能效指标包括所述数据中心的总功率或所述环境参数控制设备集群的总功率或所述数据中心的电源使用效率PUE；其中：
    若当前控制周期到达时采集的环境参数测量值与上一个控制周期的环境参数设定点小于等于预设环境参数偏差，所述能效评分根据所述数据中心的能效 指标利用第一公式计算得到；
    若当前控制周期到达时采集的环境参数测量值与上一个控制周期的环境参数设定点大于预设环境参数偏差，所述能效评分根据所述数据中心的能效指标利用第二公式计算得到；
    其中，当所述能效指标相同时，利用所述第一公式计算得到的所述能效评分大于所述第二公式计算得到的所述能效评分。
  3. 如权利要求1所述的方法，其特征在于，所述环境参数样本中样本特征还包括：上一控制周期的开机数；
    基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点，包括：
    将所述候选开机数、所述当前控制周期到达时采集的状态参数输入所述UCB模型中，分别预测对应的能效评分预测值；
    根据所述能效评分预测值决策所述开机数；
    将所述开机数、所述当前控制周期到达时采集的状态参数和所述候选环境参数设定点输入所述回归模型中，预测对应的环境参数预测值；
    根据所述环境参数预测值决策所述环境参数设定点。
  4. 如权利要求1所述的方法，其特征在于，所述能效样本中样本特征还包括：上一控制周期的环境参数设定点；
    基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点，包括：
    将所述候选环境参数设定点和所述当前控制周期到达时采集的状态参数输入所述回归模型中，预测对应的环境参数预测值；
    根据所述环境参数预测值决策所述环境参数设定点；
    将所述候选开机数、所述当前控制周期到达时采集的状态参数和所述环境参数设定点输入所述UCB模型中，分别预测对应的能效评分预测值；
    根据所述能效评分预测值决策所述开机数。
  5. 如权利要求1所述的方法，其特征在于，基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点，包括：
    将所述候选开机数和所述当前控制周期到达时采集的状态参数输入所述UCB模型中，分别预测对应的能效评分预测值；
    根据所述能效评分预测值决策所述开机数；
    将所述候选环境参数设定点和所述当前控制周期到达时采集的状态参数输入所述回归模型中，分别预测对应的环境参数预测值；
    根据所述环境参数预测值决策所述环境参数设定点。
  6. 如权利要求3-5任一项所述的方法，其特征在于，若所述能效评分越大，所述环境参数控制设备集群的能耗效率越高，则根据所述能效评分预测值决策所述开机数，包括：
    将最大的所述能效评分预测值对应的候选开机数决策为所述开机数；
    根据所述环境参数预测值决策所述环境参数设定点，包括：
    将最接近目标环境参数的所述环境参数预测值对应的候选环境参数设定点决策为所述环境参数设定点。
  7. 如权利要求1所述的方法，其特征在于，当到达控制周期时，触发模型训练；
    将所述能效指标样本集中的样本特征输入带上下文的UCB模型，以输出能效样本集中的样本标签为目标进行模型训练，包括：
    将所述能效样本集中当前控制周期的能效样本中的样本特征输入UCB模型，以输出所述当前控制周期的能效样本中的样本标签为目标进行模型训练； 或
    将所述能效样本集中所有能效样本的样本特征依次输入UCB模型，以输出对应的样本标签为目标进行模型训练；
    将所述环境参数样本集中样本特征输入回归模型，以输出所述环境参数样本集中标签为目标进行模型训练，包括：
    将所述环境参数样本集中所有环境参数样本的样本特征依次输入回归模型，以输出对应的样本标签为目标进行模型训练。
  8. 如权利要求1所述的方法，其特征在于，还包括：
    确定不满足模型决策条件时，每达到一个控制周期，获取数据中心环境的当前控制周期到达时采集的环境参数测量值；
    根据环境参数测量值与目标环境参数的差值调整开机数和环境参数设定点并运行；
    确定满足模型决策条件时，确定进入推荐阶段。
  9. 如权利要求8所述的方法，其特征在于，确定不满足模型决策条件时，每达到一个控制周期，获取数据中心环境的环境参数测量值，根据环境参数测量值与目标环境参数的差值调整开机数和环境参数设定点，包括：
    确定不满足模型决策条件时，每到达一个控制周期时，计算差值ΔE＝EnvirMeasure-EnvirTarget；
    若所述差值ΔE小于第一阈值，确定当前控制周期的环境参数设定点为上一个控制周期的环境参数设定点增大预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数减少开机数调整值；
    若所述差值ΔE大于第二阈值，确定当前控制周期的环境参数设定点为上一个控制周期的环境参数设定点减小预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数增大预设开机数调整值；
    若所述差值ΔE大于等于第一阈值且小于等于第二阈值，确定当前控制周期的环境参数设定点为上一个控制周期的环境参数设定点随机增大或减小预设环 境参数调整值，当前控制周期的开机数为上一个控制周期的开机数随机增大或减小预设开机数调整值；
    其中，EnvirMeasure为当前控制周期到达时采集的环境参数测量值，EnvirTarget为所述目标环境参数。
  10. 如权利要求1所述的方法，其特征在于，确定多个候选环境参数设定点和多个候选开机数，包括：
    确定以上一个控制周期的环境参数设定点为中心的多个备选环境参数设定点；
    确定所述多个备选环境参数设定点中符合环境参数设定点范围的备选环境参数设定点，及所述上一个控制周期的环境参数设定点为候选环境参数设定点；
    确定以上一个控制周期的开机数为中心的多个备选开机数；
    确定所述多个备选开机数中符合开机数范围的备选开机数，及所述上一个控制周期的开机数为候选开机数。
  11. 如权利要求1所述的方法，其特征在于，还包括：
    每到达数据采样时间，采集所述数据中心的状态参数，并触发对数据中心的环境参数进行检测，在超出安全环境参数范围时，按照设定的调整幅度调整开机数和环境参数设定点；
    其中，一个所述控制周期的长度为相邻数据采样时间间隔的整数倍。
  12. 如权利要求11所述的方法，其特征在于，每到达数据采样时间，采集所述数据中心的状态参数，并触发对数据中心的环境参数进行检测，在超出安全环境参数范围时，按照设定的调整幅度调整开机数和环境参数设定点，包括：
    每到达数据采样时间，采集所述数据中心的状态参数，并触发检测数据中心的环境参数EnvirMeasure；
    计算ΔE＝EnvirMeasure-EnvirTarget；
    若ΔE＜-DeadLine，将当前控制周期的环境参数设定点增大预设环境参数调整值，当前控制周期的开机数减小预设开机数调整值；
    若ΔE＞DeadLine，将当前控制周期的环境参数设定点减小预设环境参数调整值，当前控制周期的开机数增大预设开机数调整值；
    其中，EnvirMeasure为当前控制周期到达时采集的环境参数测量值，EnvirTarget为所述目标环境参数，DeadLine为预设环境参数差值且为正数。
  13. 如权利要求1所述的方法，其特征在于，所述UCB模型为线性上置信LinUCB模型或高斯上置信GPUCB模型。
  14. 如权利要求1所述的方法，其特征在于，所述回归模型为如下任一种：
    xgboost模型、随机森林RF模型、支持向量机SVM模型、神经网络模型。
  15. 如权利要求1所述的方法，其特征在于，所述状态参数包括如下至少一种：
    负载功率、平均送风环境参数、平均回风环境参数、热通道侧平均环境参数、冷通道侧平均环境参数。
  16. 如权利要求1所述的方法，其特征在于，所述环境参数为温度或者湿度。
  17. 一种环境参数控制设备集群控制装置，应用于数据中心的节能调节和环境参数调节，其特征在于，包括：
    样本采集模块，用于每到达控制周期，采集环境参数样本和能效样本，并对应更新至环境参数样本集和能效样本集中；
    模型训练模块，用于触发模型训练时，将所述能效样本集中的样本特征输入带上下文的UCB模型，以输出所述能效样本集中的样本标签为目标进行模型训练，将所述环境参数样本集中样本特征输入回归模型，以输出所述环境参数样本集中样本标签为目标进行模型训练；
    推荐模块，用于确定当前处于推荐阶段时，每到达控制周期，确定多个候选环境参数设定点和多个候选开机数，基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于 所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点并运行；
    其中，所述能效样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、开机数，样本标签为到达当前控制周期时计算的能效评分，所述环境参数样本中样本特征包括上一控制周期到达时获取的数据中心的状态参数、环境参数设定点，样本标签包括当前控制周期到达时采集的环境参数测量值。
  18. 如权利要求17所述的装置，其特征在于，所述环境参数样本中样本特征还包括：上一控制周期的开机数；
    基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点，包括：
    将所述候选开机数、所述当前控制周期到达时采集的状态参数输入所述UCB模型中，分别预测对应的能效评分预测值；
    根据所述能效评分预测值决策所述开机数；
    将所述开机数、所述当前控制周期到达时采集的状态参数和所述候选环境参数设定点输入所述回归模型中，预测对应的环境参数预测值；
    根据所述环境参数预测值决策所述环境参数设定点。
  19. 如权利要求17所述的装置，其特征在于，所述能效样本中样本特征还包括：上一控制周期的环境参数设定点；
    基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点，包括：
    将所述候选环境参数设定点和所述当前控制周期到达时采集的状态参数输入所述回归模型中，预测对应的环境参数预测值；
    根据所述环境参数预测值决策所述环境参数设定点；
    将所述候选开机数、所述当前控制周期到达时采集的状态参数和所述环境参数设定点输入所述UCB模型中，分别预测对应的能效评分预测值；
    根据所述能效评分预测值决策所述开机数。
  20. 如权利要求17所述的装置，其特征在于，基于当前控制周期到达时采集的状态参数和候选开机数，利用所述UCB模型预测对应的能效评分预测值，基于所述状态参数和候选环境参数设定点，利用回归模型预测对应环境参数预测值，基于所述能效评分预测值和环境参数预测值决策开机数和环境参数设定点，包括：
    将所述候选开机数和所述当前控制周期到达时采集的状态参数输入所述UCB模型中，分别预测对应的能效评分预测值；
    根据所述能效评分预测值决策所述开机数；
    将所述候选环境参数设定点和所述当前控制周期到达时采集的状态参数输入所述回归模型中，分别预测对应的环境参数预测值；
    根据所述环境参数预测值决策所述环境参数设定点。
  21. 如权利要求17所述的装置，其特征在于，还包括：
    初始控制模块，用于确定不满足模型决策条件时，每达到一个控制周期，获取数据中心环境的当前控制周期到达时采集的环境参数测量值；
    根据环境参数测量值与目标环境参数的差值调整开机数和环境参数设定点并运行；
    确定满足模型决策条件时，确定进入推荐阶段。
  22. 如权利要求21所述的装置，其特征在于，确定不满足模型决策条件时，每达到一个控制周期，获取数据中心环境的环境参数测量值，根据环境参数测量值与目标环境参数的差值调整开机数和环境参数设定点，包括：
    确定不满足模型决策条件时，每到达一个控制周期时，计算差值ΔE＝EnvirMeasure-EnvirTarget；
    若所述差值ΔE小于第一阈值，确定当前控制周期的环境参数设定点为上一 个控制周期的环境参数设定点增大预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数减少开机数调整值；
    若所述差值ΔE大于第二阈值，确定当前控制周期的环境参数设定点为上一个控制周期的环境参数设定点减小预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数增大预设开机数调整值；
    若所述差值ΔE大于等于第一阈值且小于等于第二阈值，确定当前控制周期的环境参数设定点为上一个控制周期的环境参数设定点随机增大或减小预设环境参数调整值，当前控制周期的开机数为上一个控制周期的开机数随机增大或减小预设开机数调整值；
    其中，EnvirMeasure为当前控制周期到达时采集的环境参数测量值，EnvirTarget为所述目标环境参数。
  23. 如权利要求17所述的装置，其特征在于，还包括：
    安全维护模块，用于每到达数据采样时间，采集所述数据中心的状态参数，并触发对数据中心的环境参数进行检测，在超出安全环境参数范围时，按照设定的调整幅度调整开机数和环境参数设定点；
    其中，一个所述控制周期的长度为相邻数据采样时间间隔的整数倍。
  24. 如权利要求23所述的装置，其特征在于，每到达数据采样时间，采集所述数据中心的状态参数，并触发对数据中心的环境参数进行检测，在超出安全环境参数范围时，按照设定的调整幅度调整开机数和环境参数设定点，包括：
    每到达数据采样时间，采集所述数据中心的状态参数，并触发检测数据中心的环境参数EnvirMeasure；
    计算ΔE＝EnvirMeasure-EnvirTarget；
    若ΔE＜-DeadLine，将当前控制周期的环境参数设定点增大预设环境参数调整值，当前控制周期的开机数减小预设开机数调整值；
    若ΔE＞DeadLine，将当前控制周期的环境参数设定点减小预设环境参数调 整值，当前控制周期的开机数增大预设开机数调整值；
    其中，EnvirMeasure为当前控制周期到达时采集的环境参数测量值，EnvirTarget为所述目标环境参数，DeadLine为预设环境参数差值且为正数。
  25. 一种电子设备，其特征在于，包括：处理器和用于存储所述处理器可执行指令的存储器；
    其中，所述处理器被配置为执行所述指令，以实现如权利要求1-16任一项所述的环境参数控制设备集群控制方法。
  26. 一种存储介质，其特征在于，所述计算机存储介质存储有计算机程序，所述计算机程序被用于实现如权利要求1-16任一项所述的环境参数控制设备集群控制方法。
PCT/CN2021/112714 2021-06-07 2021-08-16 环境参数控制设备集群控制方法、装置、设备及存储介质 WO2022257267A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110632624.5A CN115453860A (zh) 2021-06-07 2021-06-07 环境参数控制设备集群控制方法、装置、设备及存储介质
CN202110632624.5 2021-06-07

Publications (1)

Publication Number Publication Date
WO2022257267A1 true WO2022257267A1 (zh) 2022-12-15

Family

ID=84294700

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/112714 WO2022257267A1 (zh) 2021-06-07 2021-08-16 环境参数控制设备集群控制方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN115453860A (zh)
WO (1) WO2022257267A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118329802A (zh) * 2024-04-10 2024-07-12 江苏邦腾环保技术开发有限公司 环境监测设备校准方法和系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100100337A1 (en) * 2008-10-16 2010-04-22 Oxfordian, Llc Method to extract parameters from in-situ monitored signals for prognostices
CN102213475A (zh) * 2011-03-22 2011-10-12 曙光信息产业(北京)有限公司 一种数据中心功耗自适应管理方法
CN104930668A (zh) * 2015-07-08 2015-09-23 中国电信股份有限公司南京分公司 一种机房精密空调集群控制方法
CN109405202A (zh) * 2018-11-12 2019-03-01 国网江苏省电力有限公司南通供电分公司 一种基于用户需求的中央空调集群优化控制方法
CN111783902A (zh) * 2020-07-30 2020-10-16 腾讯科技（深圳）有限公司 数据增广、业务处理方法、装置、计算机设备和存储介质
CN112577161A (zh) * 2019-09-30 2021-03-30 北京国双科技有限公司 空调能耗模型训练方法与空调系统控制方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100100337A1 (en) * 2008-10-16 2010-04-22 Oxfordian, Llc Method to extract parameters from in-situ monitored signals for prognostices
CN102213475A (zh) * 2011-03-22 2011-10-12 曙光信息产业(北京)有限公司 一种数据中心功耗自适应管理方法
CN104930668A (zh) * 2015-07-08 2015-09-23 中国电信股份有限公司南京分公司 一种机房精密空调集群控制方法
CN109405202A (zh) * 2018-11-12 2019-03-01 国网江苏省电力有限公司南通供电分公司 一种基于用户需求的中央空调集群优化控制方法
CN112577161A (zh) * 2019-09-30 2021-03-30 北京国双科技有限公司 空调能耗模型训练方法与空调系统控制方法
CN111783902A (zh) * 2020-07-30 2020-10-16 腾讯科技（深圳）有限公司 数据增广、业务处理方法、装置、计算机设备和存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
FENG LUHAN : "Research on Anomaly Localization of Multi-Dimensional Monitoring Indicators in Artificial Intelligence IT Operations", MASTER'S E-JOURNAL, 15 February 2020 (2020-02-15), pages 1 - 82, XP093014256, [retrieved on 20230116] *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118329802A (zh) * 2024-04-10 2024-07-12 江苏邦腾环保技术开发有限公司 环境监测设备校准方法和系统

Also Published As

Publication number Publication date
CN115453860A (zh) 2022-12-09

Similar Documents

Publication Publication Date Title
WO2023093820A1 (zh) 设备的控制优化方法、展示平台、云服务器及存储介质
US11513480B2 (en) Method and device for automatically diagnosing and controlling apparatus in intelligent building
WO2022012542A1 (zh) 基于数据分析的云计算数据中心精密空调节能控制方法
JP5511698B2 (ja) 空調機連係制御システム、空調機連係制御方法および空調機連係制御プログラム
US20120197828A1 (en) Energy Saving Control for Data Center
CN110726218B (zh) 空调器及其控制方法、装置、存储介质和处理器
WO2023116742A1 (zh) 一体化数据中心柜末端空调系统节能优化方法及装置
CN110866592B (zh) 模型训练方法、装置、能效预测方法、装置和存储介质
CN111649457A (zh) 一种动态预测性机器学习型空调节能控制方法
CN116266253A (zh) 空调参数的优化控制方法、系统和计算机可读存储介质
CN113821903B (zh) 温度控制方法和设备、模块化数据中心及存储介质
WO2022257267A1 (zh) 环境参数控制设备集群控制方法、装置、设备及存储介质
CN114046593A (zh) 一种动态预测性机器学习型空调节能控制方法及系统
KR101727434B1 (ko) 랜덤 포레스트 모델을 이용한 냉동기의 성능 판단 방법
CN114556027B (zh) 空调控制装置、空调系统、空调控制方法和记录介质
CN113791538B (zh) 一种机房设备的控制方法、控制设备及控制系统
CN108446783A (zh) 一种新风机运行功率的预测及监控方法
US20150127164A1 (en) Performance management system, method and non-transitory computer readable storage medium thereof
CN112325447B (zh) 一种基于强化学习的制冷机组控制装置及控制方法
US11662696B2 (en) Automatic control artificial intelligence device and method for update control function
CN116954329A (zh) 制冷系统的状态调节方法、装置、设备、介质及程序产品
CN115455797A (zh) 温度预测模型训练和温度决策方法及装置、电子设备
KR20160063839A (ko) 지능형 축사 환풍구 제어방법
CN117408170B (zh) 一种适用于数据中心水冷系统的节能预测控制方法
CN117236181A (zh) 一种基于dqn的数据中心水冷机组能耗优化方法与系统

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21944769

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21944769

Country of ref document: EP

Kind code of ref document: A1