BehaveLearn

Experimental implementation in the environmental chamber of the ICE lab
Experimental implementation in the environmental chamber of the ICE lab

Occupant behavior is known as a key driver of building energy use. It is a highly stochastic and complicated phenomenon, driven by a wide variety of factors, and is unique in each building. Therefore, it cannot be addressed using analytical approaches traditionally used to describe physics-based aspects of buildings. Consequently, current controls which rely on hard-coded expert knowledge have a limited potential for integrating occupant behavior. An alternative approach is to program a human-like learning mechanism and develop a controller that is capable of learning the control policy by itself through interacting with the environment and learning from experience. Reinforcement Learning (RL), a machine learning algorithm inspired by neuroscience, can be used to develop such a self-learning controller. Given the learning ability, these controllers are able to learn optimal control policy without prior knowledge or a detailed system model, and can continuously adapt to the stochastic variations in the environment to ensure an optimal operation.

The main question that the study “BehaveLearn: Reinforcement Learning for occupant-centric operation of building energy systems” lead by the Ph.D. student Amirreza Heidari  deals with is how to  develop  a controller that can perceive and adapt to the occupant behavior to minimize energy use without compromising user needs?

In this context, three occupant-centric control frameworks are developed in this study:

  • DeepHot: focused on hot water production in residential buildings
  • DeepSolar: focused on solar-assisted space heating and hot water production in
    residential buildings;
  • DeepValve: focused on zone-level space heating in offices

DeepHot and DeepSolar frameworks are evaluated using real-world hot water use behavior of occupants, and DeepValve is experimentally implemented in the environmental chamber of the ICE lab. Results show promising energy saving potentials of each control framework.

Background

  • Occupant behavior is highly stochastic, and is unique in each building
  • Current controls rely on expert knowledge and cannot easily integrate occupant behavior
  • Learning controls can autonomously learn and adapt to occupant behavior to reduce energy use in buildings

Methodology

  • Collecting data on hot water use behavior in 3 Swiss houses
  • Develop and evaluate DeepHot and DeepSolar frameworks in simulation using real-world data
  • Develop and evaluate DeepValve framework in simulation using real-world data from other studies
  • Experimentally implement DeepValve in the ICE environmental chamber

Goals

  • Develop self-learning control frameworks to integrate occupant behavior in building controls
  • Test the frameworks in simulation using real-world data of occupant behavior
  • Test the frameworks experimentally in the environmental chamber of the ICE lab

Results

  • A significant energy reduction can be achieved by integrating occupant behavior
  • Energy reduction is achieved without compromising health and comfort of occupants
  • Self-learning control can also coordinate stochastic renewable energy use and occupant behavior

Journal Publications

Conference Publications