Methods and systems for personalized heating, ventilation, and air conditioning

A method for controlling an operation of a heating ventilation and air conditioning (HVAC) system for an occupant in an environment, the occupant is a user of the HVAC system, the method comprising:

• using a memory having stored data including a personalized database, and a historical database, the historical database including a generic thermal comfort model learned from labeled data of other users, wherein each instance of the labeled data is indicative of a thermal comfort level according to a thermal comfort scale of at least one other user in different conditions of at least one environment;
• using a hardware processor is configured to
• initialize a pre-train personalized thermal comfort (PPTC) model with the labeled data from the other users and a transfer active learning framework that utilizes the generic thermal comfort model for transfer learning while all labeled data from the occupant is assumed inaccessible or unknown, and store the PPTC model in the memory;
• receiving, via an input interface, a sequence of unlabeled real-time data including measurements of biometric data of the occupant and measurements of environmental data in the environment the occupant is located;
• calculate a disagreement score by identifying a distance between each unlabeled data point of the sequence of unlabeled real-time data and all data points in the labeled data from the occupant, storing the distances in a vector and sort from a smallest distance to a largest distance, choosing a K smallest distances for each unlabeled data point and obtain their labels from the historical database, predict a label for the unlabeled data point and then calculate the disagreement score using the labels from the historical database and predicted label for the unlabeled data point, using the PPTC model;
• requesting, via a transmitter, the occupant to label an instance of unlabeled data, when there is a disagreement between the labels of stored labeled data from the other users similar to the unlabeled real-time data and a predicted label for the unlabeled real-time data instance which exceeds a predetermined threshold;
• receive, a response back from the occupant including labeling the instance of unlabeled data such that the labeling is indicative of a thermal comfort level of the occupant, and store the labeled instance of unlabeled data as personalized labeled data in the personalized database;
• initiate active learning of the transfer active learning framework by training PPCT model using different models of the stored labeled data from the occupant compared to the stored labeled data from the other users, and for each iteration during the initialization period, updates the personalized database with the personalized labeled data, and retrains the PPCT model using the historical database and the updated personalized database;
• and
• controlling, via a controller, the HVAC system based on the retrained PPTC model.

A system for controlling an operation of a heating ventilation and air conditioning (HVAC) system for an occupant in an environment, such that the occupant is a user of the HVAC system and has control of the HVAC system, and the system including an input interface to receive data, and a memory having stored a personalized database, and a historical database, the historical database including a generic thermal comfort model learned from labeled data of other users, each instance of the labeled data is indicative of a thermal comfort level according to a thermal comfort scale of at least one other user in different conditions of at least one environment, the system comprising:

• a hardware processor to
• initialize a pre-train personalized thermal comfort (PPTC) model with the labeled data from the other users and a transfer active learning framework that utilizes the generic thermal comfort model for transfer learning while all labeled data from the occupant is assumed inaccessible or unknown, and store the PPTC model in the memory;
• receive a sequence of unlabeled real-time data including measurements of biometric data of the occupant and measurements of environmental data in the environment the occupant is located;
• calculate a disagreement score by identifying a distance between each unlabeled data point of the sequence of unlabeled real-time data and all data points in the labeled data from the occupant, storing the distances in a vector and sort from a smallest distance to a largest distance, choosing a K smallest distances for each unlabeled data point and obtain their labels from the historical database, predict a label for the unlabeled data point and then calculate the disagreement score using the labels from the historical database and predicted label for the unlabeled data point, using the PPTC model;
• request the occupant to label an instance of unlabeled data, via a transmitter, when there is a disagreement between the labels of stored labeled data from the other users similar to the unlabeled real-time data and a predicted label for the unlabeled real-time data instance that exceeds a predetermined disagreement threshold;
• initiate active learning of the transfer active learning framework by training the PPCT model using different models of the stored labeled data from the occupant compared to the stored labeled data from the other users, and for each iteration during the initialization period, updates the personalized database with the personalized labeled data, and retrains the PPCT model using the historical database and the updated personalized database;
• and
• a controller to control the HVAC system based on the retrained PPTC model.