Personalized Shading Control Strategy to
Maximize Occupant Satisfaction while
Minimizing Lighting Energy Use by
Multi-Objective Optimization
Jie Xiong, Seungjae Lee,
Athanasios Tzempelikos, Panagiota Karava
July 11 -14, 2016
OUTLINE
INTRODUCTION
MODELING
MULTI-OBJECTIVE OPTIMIZATION
CONCLUSION AND FUTURE WORK
July 11-14, 2016
Purdue Conferences
2
INTRODUCTION
Previous studies have proven the energy benefits of automated
shading control strategies, while occupant dissatisfaction issues
with respect to visual comfort could arise with such systems
Three main challenges related to “optimal” shading and lighting
controls, and considerations of visual comfort, occupant satisfaction,
and energy use
» The development of personalized visual comfort profiles related
to shading operation are required
» Existing models were developed for the case of manual shading
controls (no automation or motorization)
» The common approach has been to consider visual comfort and
energy use in optimization schemes, with different levels of
complexity and formulation types
July 11-14, 2016
Purdue Conferences
3
INTRODUCTION
In this study, a personalized shading control
framework is developed to maximize occupant
satisfaction while minimizing lighting energy use using
such a multi-objective optimization scheme.
July 11-14, 2016
Purdue Conferences
4
MODELING
Experiment and Data
» A set of experiments were conducted to gather data for
developing personal satisfaction model
» Experiment facilities
– Four private offices of the Herrick Laboratories, Purdue
University, 09/29/2015 – 10/23/2015
– 3.17m × 4.03m × 3.1m, 60% window-to-wall ratio, 70% normal
visible transmittance glazing
MODELING
Experiment and Data
» Indoor sensing system
– Environmental conditions
– States of the shading system
– Occupancy status
» Experiment procedure
– Programmed shading control intentionally raises the shades
gradually
– The test subject could override the system through a webbased control interface
– The data used in modeling study is the data of 14-days (nonconsecutive) in one office (1735 training data points and 2601
test observations)
MODELING
Integrated Daylight, Electric-light and Lighting Energy Model
» A validated hybrid ray-tracing and radiosity daylight model
(Chan and Tzempelikos, 2012)
Personal Satisfaction (Override) Model
» Separate lowering and raising models
» Logistic regression
» Probability theory
MODELING
Personal Satisfaction (Override) Model
» Global sensitivity analysis
» Input variable selection
– SP (shade position)
– VI (vertical illuminance)
» Training results
MODELING
Model Visualization
Lowering
vs.
Raising
MULTI-OBJECTIVE OPTIMIZATION
Formulation of the Optimization Problem
» Variables and Objectives
– Energy objective
– Satisfaction objective
– Optimization problem
» Optimization Strategy and Algorithm
– Straightforward search strategy
– Controlled variable (shade position) is pre-defined with
discrete options as a feasible set
MULTI-OBJECTIVE OPTIMIZATION
Results and Discussion
» Developed optimization algorithm implemented on the
experiment training data (7 days)
» 15-minute interval
» The outdoor conditions (measured transmitted
illuminance values)
MULTI-OBJECTIVE OPTIMIZATION
Results and Discussion
» Representative Pareto Fronts
MULTI-OBJECTIVE OPTIMIZATION
Results and Discussion
» Tolerance in Objective
» 0.1 W/m2 tolerance in power objective
MULTI-OBJECTIVE OPTIMIZATION
Results and Discussion
» Analysis of Pareto
Solutions
– Numbers of Pareto
solutions
– Pareto solution range
MULTI-OBJECTIVE OPTIMIZATION
Results and Discussion
» Analysis of Pareto Optimal
– Without tolerance
– With tolerance
MULTI-OBJECTIVE OPTIMIZATION
Framework of Multi-Objective Optimization Application
in Shading Control
» Instead of assigning “weights”
» Provide the “option pool” to occupant
Application
Flowchart
CONCLUSION AND FUTURE WORK
This study presents
» The development of a personalized shading control
strategy, using a multi-objective optimization algorithm
» The application framework aiming at maximizing
personal occupant satisfaction while minimizing lighting
energy consumption with automated shading systems
Modeling
» Experiments
» Combining separately-built lowering and raising models
» Multivariate logistic regression
» Quantitative evaluation method for probability function is
required, such as Jensen–Shannon divergence method
CONCLUSION AND FUTURE WORK
Multi-objective optimization
» Enlarged time horizon based on serial-independency and
fixed condition assumptions
» Pre-defined finite controlled variables and straight-forward
search strategies
» Effectiveness of tolerance
» Cloudy conditions lead to more Pareto solutions as well as
larger Pareto optimal override probability ranges
CONCLUSION AND FUTURE WORK
The application framework of developed personal
shading control strategy using multi-objective
optimization strategy
» Complete Pareto solutions – pool of options and decision
» Occupants – decision makers balancing between their
personalized comfort limits and energy use considerations
» prototype study on adaptive shading control strategies with
learned personalized comfort profiles and parallel energy use
considerations
» Preliminary step for application of MOO strategy in integrated
building control
CONCLUSION AND FUTURE WORK
Future Work
» Developing adaptive learning satisfaction or preference
models
» New strategies for constructing pools of options given
varying numbers of Pareto solutions
» Investigation and evaluation of developed strategy by
application in buildings and feedback from occupants
» Extension of objectives and controlled variables to
energy consumed in building, overall comfort and indoor
environment control
Thank you!
Questions?