eduzhai > Applied Sciences > Engineering >

One for Many Transfer Learning for Building HVAC Control

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: The design of building heating, ventilation, and air conditioning (HVAC)system is critically important, as it accounts for around half of buildingenergy consumption and directly affects occupant comfort, productivity, andhealth. Traditional HVAC control methods are typically based on creatingexplicit physical models for building thermal dynamics, which often requiresignificant effort to develop and are difficult to achieve sufficient accuracyand efficiency for runtime building control and scalability for fieldimplementations. Recently, deep reinforcement learning (DRL) has emerged as apromising data-driven method that provides good control performance withoutanalyzing physical models at runtime. However, a major challenge to DRL (andmany other data-driven learning methods) is the long training time it takes toreach the desired performance. In this work, we present a novel transferlearning based approach to overcome this challenge. Our approach caneffectively transfer a DRL-based HVAC controller trained for the sourcebuilding to a controller for the target building with minimal effort andimproved performance, by decomposing the design of neural network controllerinto a transferable front-end network that captures building-agnostic behaviorand a back-end network that can be efficiently trained for each specificbuilding. We conducted experiments on a variety of transfer scenarios betweenbuildings with different sizes, numbers of thermal zones, materials andlayouts, air conditioner types, and ambient weather conditions. Theexperimental results demonstrated the effectiveness of our approach insignificantly reducing the training time, energy cost, and temperatureviolations.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×