Optimal Personal Comfort Management Using SPOT Peter Xiang

Optimal Personal Comfort Management Using SPOT+ Peter Xiang Gao, S. Keshav University of Waterloo

HVAC Energy use �Buildings use 1/3 of all energy � 30 -50% of building energy is for HVAC �Can save energy by changing temperature setpoint: � 1 o. C higher when cooling ≈ 10% saving � 1 o. C lower when heating ≈ 2 -3% saving

Focus of this work Consider a single office heating system in winter Assume �Thermal isolation �Personal thermal control (heater)

Personal Office Thermal Comfort Management Office Corridor Office

SPOT+: A Smart Personalized Office Thermal Control System Personal Thermal Comfort Evaluation Learning-Based Modeling Occupancy Arrive office Lunch Leave office 6 7 8 9 10 11 12 13 14 15 16 17 18 Time of a day Occupancy Prediction Temperature 500 W -> f ( • ) -> + 1 o. C 26. 5 24. 5 22. 5 20. 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Time of a day Setpoint Scheduling

SPOT+: A Smart Personalized Office Thermal Control System Personal Thermal Comfort Evaluation Learning-Based Modeling Occupancy Arrive office Lunch Leave office 6 7 8 9 10 11 12 13 14 15 16 17 18 Time of a day Occupancy Prediction Temperature 500 W -> f ( • ) -> + 1 o. C 26. 5 24. 5 22. 5 20. 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Time of a day Setpoint Scheduling

Predicted Mean Vote (PMV) model �Air Temperature, Background Radiation, Air Velocity, Humidity, Metabolic Rate, Clothing Level Cold Cool Slightly Cool Neutral Slightly Warm Hot -3 -2 -1 0 1 2 3 ASHRAE Scale

[1] SPOT Clothing Sensing 5° infrared sensor: • Detects users’ clothing surface temperature Microsoft Kinect: • • Detects occupancy Detects location of the user Weather. Duck: • Senses other environmental variables ________________ [1] P. X. Gao, S. Keshav, SPOT: A Smart Personalized Office Thermal Control System, e-Energy 2013

Clothing level estimation �Estimate clothing by measuring emitted infrared �More clothing => lower infrared reading Clo = k * (tclothing – tbackground) + b �tclothing is the infrared measured from clothes on human body �tbackground is the background infrared radiation �k and b are parameters to be estimated by regression

Personalization �PMV model represents the average �for a single office, only the occupant’s vote matters �Predicted Personal Vote (PPV) Model ppv = fppv (pmv) where fppv( • ) is a linear function

SPOT+: A Smart Personalized Office Thermal Control System Personal Thermal Comfort Evaluation Learning-Based Modeling Occupancy Arrive office Lunch Leave office 6 7 8 9 10 11 12 13 14 15 16 17 18 Time of a day Occupancy Prediction Temperature 500 W -> f ( • ) -> + 1 o. C 26. 5 24. 5 22. 5 20. 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Time of a day Setpoint Scheduling

Learning-Based Model Predictive Control �We model thermal characteristics of a room using LBMPC �The model can predict future temperature = flbmpc (current temperature, heater power)

Learning-Based Model Predictive Control �

Learning-Based Model Predictive Control �

SPOT+: A Smart Personalized Office Thermal Control System Personal Thermal Comfort Evaluation Learning-Based Modeling Occupancy Arrive office Lunch Leave office 6 7 8 9 10 11 12 13 14 15 16 17 18 Time of a day Occupancy Prediction Temperature 500 W -> f ( • ) -> + 1 o. C 26. 5 24. 5 22. 5 20. 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Time of a day Setpoint Scheduling

Occupancy Prediction �We predict occupancy using historical data. 0. 3 1 1 1. 3 0 Match Previous similar history Predict using matched records ________________ [1] James Scott et. al. , Pre. Heat: Controlling Home Heating With Occupancy Prediction, Ubi. Comp 2011

SPOT+: A Smart Personalized Office Thermal Control System Personal Thermal Comfort Evaluation Learning-Based Modeling Occupancy Arrive office Lunch Leave office 6 7 8 9 10 11 12 13 14 15 16 17 18 Time of a day Occupancy Prediction Temperature 500 W -> f ( • ) -> + 1 o. C 26. 5 24. 5 22. 5 20. 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Time of a day Setpoint Scheduling

Optimal Control �We use the optimal control strategy to schedule the setpoint over a day. �The control objective is to reduce energy consumption and still maintain thermal comfort Overall energy consumption in the optimization horizon S Weight of comfort, set to large value to guarantee comfort first Predicted occupancy, we only guarantee comfort when occupied. aka m(s) = 1 Thermal comfort penalty. Both term equal zero when the user feels comfortable

Optimal Control - Constraints �ε is the tolerance of predicted personal vote (PPV) �So when | ppv(x(s)) | is smaller than ε, there is no penalty �Otherwise, either βc(s) or βh(s) will be positive to penalize the discomfort thermal environment

Evaluation

Evaluation of clothing level estimation �Root mean square error (RMSE) = 0. 0918 �Linear correlation = 0. 9201

Predicted Personal Vote Estimation �Root mean square error (RMSE) = 0. 5377 �Linear correlation = 0. 8182

Accuracy of LBMPC �The RMSE over a day is 0. 17 C.

Accuracy of Occupancy Prediction �The result of optimal prediction is affected by occupancy prediction. �False negative 10. 4% (From 6 am. to 8 pm. ) �False positive 8. 0% (From 6 am. to 8 pm. ) �Still an open problem

Comparison of schemes

Limitations �SPOT+ requires thermal Insulation for personal thermal control �Current SPOT+ costs about $1000 �PPV requires some initial calibration �State of window/door is not modelled in the current LBMPC �Accuracy of clothing level estimation is affected by �Accuracy of Kinect �Distance effect of the infrared sensor

Conclusion �We extended PMV model for personalized thermal control �We design and implement SPOT+ �We use LBMPC and optimal control for personalized thermal control �SPOT+ can accurately maintain personal comfort despite environmental fluctuations �allows a worker to balance personal comfort with energy use.

Relationship between PPV and Energy cost � Maintaining a PPV of 0 consumes about 6 k. Wh electricity daily. By setting the target PPV to -0. 5, we can save about 3 k. Wh electricity per day.

Average Discomfort vs Energy Consumption