Optimizing data center controls using neural networks

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for improving operational efficiency within a data center by modeling data center performance and predicting power usage efficiency. An example method receives a state input characterizing a current state of a data center. For each data center setting slate, the state input and the data center setting slate are processed through an ensemble of machine learning models. Each machine learning model is configured to receive and process the state input and the data center setting slate to generate an efficiency score that characterizes a predicted resource efficiency of the data center if the data center settings defined by the data center setting slate are adopted t. The method selects, based on the efficiency scores for the data center setting slates, new values for the data center settings.

What is claimed is: 
     
         1 . (canceled) 
     
     
         2 . A method comprising:
 receiving a state input characterizing a state of a facility and data defining a plurality of facility setting slates, each slate defining a respective combination of facility settings for the facility;   for each facility slate in the plurality of facility setting slates and for a first operating constraint of a plurality of operating constraints for the facility, processing the state input and the facility slate through one or more first machine learning models that are each specific to the first operating constraint to generate a first constraint score, wherein facility the first constraint score characterizes a predicted value of an operating property of the facility if facility settings defined by the facility slate are adopted in response to receiving the state input;   generating a filtered plurality of setting slates, comprising filtering slates from the plurality of facility setting slates having a respective first constraint score that does not satisfy a predetermined threshold value for the first operating constraint;   processing each slate of the filtered plurality of facility setting slates through one or more second machine learning models to generate an efficiency score for the slate, the efficiency score characterizing facility if the facility settings defined by the facility setting slate are adopted in response to receiving the state input; and   selecting, based on the respective efficiency scores for the facility setting slates in the second set of facility setting slates, new settings for the facility.   
     
     
         3 . The method of  claim 2 , wherein processing the state input and the slate through the one or more first machine learning models comprises:
 processing the state input and the slate through each of the one or more first machine learning models to obtain a respective candidate constraint score, the respective candidate constraint score characterizing a predicted value of the operating property of the facility according to the first machine learning model; and   generating the respective first constraint score based on an average of the respective candidate constraint scores.   
     
     
         4 . The method of  claim 2 , wherein processing each slate of the filtered plurality of facility settings through the one or more second machine learning models to generate the efficiency score for the slate comprises:
 processing the slate through each of the one or more second machine learning models to generate a respective candidate efficiency score, the respective candidate efficiency score characterizing a predicted resource efficiency of the facility according to the second machine learning model; and   generating the efficiency score for the slate based on an average of the respective candidate efficiency scores.   
     
     
         5 . The method of  claim 4 , wherein selecting the new settings based on the respective final efficiency scores comprises:
 ranking the facility setting slates based on the efficiency scores; and   selecting settings from a highest-ranked slate as the new settings for the facility.   
     
     
         6 . The method of  claim 4 , wherein each second machine learning model of the one or more second machine learning models:
 has been trained on a different sample of training data from each other second machine learning model of the one or more second machine learning models, or   has a different model architecture from each other second machine learning model of the one or more second machine learning models.   
     
     
         7 . The method of  claim 2 , further comprising:
 for each slate in the plurality of facility setting slates and for a second operating constraint of the plurality of operating constraints for the facility that is different than the first operating constraint, processing the state input and the slate through one or more machine learning models specific to the second operating constraint to generate a respective second constraint score; and   filtering slates from the plurality of facility setting slates having a respective second constraint score that does not satisfy a predetermined threshold value for the second operating constraint.   
     
     
         8 . The method of  claim 2 , further comprising:
 receiving a second state input characterizing a state of a second facility different from the facility;   receiving data defining a second plurality of facility setting slates for the second facility;   for each slate in the second plurality of facility setting slates and for a second operating constraint of the plurality of operating constraints, processing the state input and the slate through the one or more first machine learning model specific to the first operating constraint to generate a respective first constraint score for the slate; and   filtering slates from the second plurality of facility setting slates having a respective first constraint score that does not satisfy the predetermined threshold value for the first operating constraint.   
     
     
         9 . The method of  claim 2 , wherein the respective efficiency score is a predicted long-term power usage effectiveness of the facility if the facility settings defined by the slate are adopted in response to receiving the state input. 
     
     
         10 . A system comprising:
 one or more computers and one or more storage devices on which are stored instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:   receiving a state input characterizing a state of a facility and data defining a plurality of facility setting slates, each slate defining a respective combination of facility settings for the facility;   for each slate in the plurality of facility setting slates and for a first operating constraint of a plurality of operating constraints for the facility, processing the state input and the slate through one or more first machine learning models that are each specific to the first operating constraint to generate a first constraint score, wherein the first constraint score characterizes a predicted value of an operating property of the facility if settings defined by the slate are adopted in response to receiving the state input;   generating a filtered plurality of setting slates, comprising filtering slates from the plurality of facility setting slates having a respective first constraint score that does not satisfy a predetermined threshold value for the first operating constraint;   processing each slate of the filtered plurality of facility setting slates through one or more second machine learning models to generate an efficiency score for the slate, the efficiency score characterizing a predicted resource efficiency of the facility if the facility settings defined by the facility setting slate are adopted in response to receiving the state input; and   selecting, based on the respective efficiency scores for the facility setting slates in the second set of facility setting slates, new settings for the facility.   
     
     
         11 . The system of  claim 10 , wherein processing the state input and the slate through the one or more first machine learning models comprises:
 processing the state input and the slate through each of the one or more first machine learning models to obtain a respective candidate constraint score, the respective candidate constraint score characterizing a predicted value of the operating property of the facility according to the first machine learning model; and   generating the respective first constraint score based on an average of the respective candidate constraint scores.   
     
     
         12 . The system of  claim 10 , wherein processing each slate of the filtered plurality of facility settings through the one or more second machine learning models to generate the efficiency score for the slate comprises:
 processing the slate through each of the one or more second machine learning models to generate a respective candidate efficiency score, the respective candidate efficiency score characterizing a predicted resource efficiency of the facility according to the second machine learning model; and   generating the efficiency score for the slate based on an average of the respective candidate efficiency scores.   
     
     
         13 . The system of  claim 12 , wherein selecting the new settings based on the respective final efficiency scores comprises:
 ranking the facility setting slates based on the efficiency scores; and   selecting settings from a highest-ranked slate as the new settings for the facility.   
     
     
         14 . The system of  claim 12 , wherein each second machine learning model of the one or more second machine learning models:
 has been trained on a different sample of training data from each other second machine learning model of the one or more second machine learning models, or   has a different model architecture from each other second machine learning model of the one or more second machine learning models.   
     
     
         15 . The system of  claim 10 , wherein the operations further comprise:
 for each slate in the plurality of facility setting slates and for a second operating constraint of the plurality of operating constraints for the facility that is different than the first operating constraint, processing the state input and the slate through one or more machine learning models specific to the second operating constraint to generate a respective second constraint score; and   filtering slates from the plurality of facility setting slates having a respective second constraint score that does not satisfy a predetermined threshold value for the second operating constraint.   
     
     
         16 . The system of  claim 10 , wherein the operations further comprise:
 receiving a second state input characterizing a state of a second facility different from the facility;   receiving data defining a second plurality of facility setting slates for the second facility;   for each slate in the second plurality of facility setting slates and for a second operating constraint of the plurality of operating constraints, processing the state input and the slate through the one or more first machine learning model specific to the first operating constraint to generate a respective first constraint score for the slate; and   filtering slates from the second plurality of facility setting slates having a respective first constraint score that does not satisfy the predetermined threshold value for the first operating constraint.   
     
     
         17 . The system of  claim 10 , wherein the respective efficiency score is a predicted long-term power usage effectiveness of the facility if the facility settings defined by the slate are adopted in response to receiving the state input. 
     
     
         18 . One or more non-transitory computer-readable storage media encoded with instructions that, when executed by one or more computers, cause the one or more computers to perform operations comprising:
 receiving a state input characterizing a state of a facility and data defining a plurality of facility setting slates, each slate defining a respective combination of facility settings for the facility;   for each slate in the plurality of facility setting slates and for a first operating constraint of a plurality of operating constraints for the facility, processing the state input and the slate through one or more first machine learning models that are each specific to the first operating constraint to generate a first constraint score, wherein the first constraint score characterizes a predicted value of an operating property of the facility if settings defined by the slate are adopted in response to receiving the state input;   generating a filtered plurality of setting slates, comprising filtering slates from the plurality of facility setting slates having a respective first constraint score that does not satisfy a predetermined threshold value for the first operating constraint;   processing each slate of the filtered plurality of facility setting slates through one or more second machine learning models to generate an efficiency score for the slate, the efficiency score characterizing a predicted resource efficiency of the facility if the facility settings defined by the facility setting slate are adopted in response to receiving the state input; and   selecting, based on the respective efficiency scores for the facility setting slates in the second set of facility setting slates, new settings for the facility.   
     
     
         19 . The computer-readable media of  claim 18 , wherein processing the state input and the slate through the one or more first machine learning models comprises:
 processing the state input and the slate through each of the one or more first machine learning models to obtain a respective candidate constraint score, the respective candidate constraint score characterizing a predicted value of the operating property of the facility according to the first machine learning model; and   generating the respective first constraint score based on an average of the respective candidate constraint scores.   
     
     
         20 . The computer-readable media of  claim 18 , wherein processing each slate of the filtered plurality of facility settings through the one or more second machine learning models to generate the efficiency score for the slate comprises:
 processing the slate through each of the one or more second machine learning models to generate a respective candidate efficiency score, the respective candidate efficiency score characterizing a predicted resource efficiency of the facility according to the second machine learning model; and   generating the efficiency score for the slate based on an average of the respective candidate efficiency scores.   
     
     
         21 . The computer-readable media of  claim 18 , wherein the operations further comprise:
 for each slate in the plurality of facility setting slates and for a second operating constraint of the plurality of operating constraints for the facility that is different than the first operating constraint, processing the state input and the slate through one or more machine learning models specific to the second operating constraint to generate a respective second constraint score; and   filtering slates from the plurality of facility setting slates having a respective second constraint score that does not satisfy a predetermined threshold value for the second operating constraint.