You are on page 1of 1

Adaptive occupant and building learning controls for zone level building systems

Project 4.2: Ongoing commissioning for commercial buildings


Burak Gunay1, Liam OBrien1, Ian Beausoleil-Morrison2
Civil and Environmental Engineering Department, Carleton University
Mechanical and Aerospace Engineering Department, Carleton University
Problem Definition:: Learning from the heat transfer and storage characteristics:
Terminal heating and cooling systems, lights and blinds in offices have been either controlled by
Simple control-oriented models (see Fig 4) were
occupants or they have been automated based on fixed setpoints and schedules.
developed and implemented in an academic building.
Occupants' control targets maintaining a comfortable indoor climate with minimum number of
The model parameters were identified in recursion by
control actions with little intention to save energy [1].
using the Extended Kalman Filter (EKF) (see Fig 5).
Operators tend to choose conservative setpoints and schedules that maintain the comfort of the
majority and minimize the frequency of complaints with little consideration to save energy again [1].

Research objectives and motivation:


Learn occupancy patterns and zone heat transfer characteristics based upon inexpensive sensors.
Dynamically adapt heating/cooling setpoints based upon learned occupancy patterns and learned Figure 4: The control oriented models.
zone heat transfer characteristics.
The models could make predictions over a five day horizon.
Learning from occupants adaptive behaviours: The models were used to adapt the nighttime setback to setpoint
transition period (see Fig 7).
Occupant models provide the likelihood of occupants interactions with electric lighting, blinds, and
thermostats (see Fig 4) [2].
An adaptive algorithm was developed and implemented in an academic building to build these
behaviour models in real-time as occupants interact with their electric lighting, blinds, and
thermostats.
Dynamically evolving and individual heating, cooling, lighting, and blinds automation setpoints were
employed.

Figure 6: The performance of the one-state model in making offline


and time-step (15 min) ahead predictions for the indoor temperature.

Figure 1: Recursively developed occupant behaviour models in a shared-office space.

Learning recurring occupancy patterns: Figure 5: The model inputs and the measured
temperature (output). The evolution of the one-state Figure 7: An anecdote of the setpoint transition time period (t) time
More than a quarter of the North American workforce reports having flexible work schedules [3]. model's parameters in time when the EKF algorithm was to bring a room to the intended temperature range upon a setback
Individual occupancy profiles become diverse (see Fig 2). employed. period computed dynamically by the one-state model.
An algorithm which learns the recurring occupancy patterns was developed and implemented in
an academic office building (see Fig 3). Future Work:

Lab experimentation (see Fig 8) and field implementation


(see Fig 9) to quantify energy savings potential and comfort
implications.
Learning from occupants movements to adapt motion
sensor delay time for electric lighting.
An energy management system interface that reads sensory
data from a BACnet controller and adapts the setpoints and
schedules autonomously.
Detecting envelope degradations and operational faults
using simple control-oriented models.
Diversity of the individual heating and cooling load profiles
will be studied with the adaptive control algorithms. Peak
load reduction potential will be analyzed.
Influence of the terminal heating and cooling equipment
granularity will be studied.

Figure 8: Lab experimentation. Each


week the traditional control algorithm
alternating with the adaptive control
algorithm.

Figure 2: The individual mean weekday occupancy profiles for the seven private offices

Figure 9: Field implementation of the adaptive control algorithms in 8 offices. Annual space conditioning and lighting loads will be
contrasted with 8 identical offices.

References:
[1] Bordass, B., Cohen, R., Standeven, M., & Leaman, A. (2001). Assessing building performance in use 3: energy performance of the Probe buildings.Building Research &
Information, 29(2), 114-128.
Figure 3: Evolution of the learned arrival and departure times during a simulation; and a comparison [2] Gunay, H. Burak, William O'Brien, and Ian Beausoleil-Morrison. "A critical review of observation studies, modeling, and simulation of adaptive occupant behaviors in
between the recursively learned and observed distributions for the arrival and departure times. offices." Building and Environment 70 (2013): 31-47.
[3] McMenamin, T. M. (2007). Time to work: recent trends in shift work and flexible schedules, A. Monthly Lab. Rev., 130, 3.