OCCUPANT BEHAVIOR OF WINDOW OPENING AND CLOSING IN OFFICE BUILDINGS: DATA MINING APPROACHES. Simona D Oca. December 8 th, Washington DC

Transcription

1 OCCUPANT BEHAVIOR OF WINDOW OPENING AND CLOSING IN OFFICE BUILDINGS: DATA MINING APPROACHES December 8 th, Washington DC

2 T H E C R E D I B I L I T Y G A P O F B U I L D I N G E N E R G Y C O N S U M P T I O N The US Energy Big Picture 28% Transportation 32% Industry 40% Building sector 22% Residential 18% Commercial US Energy Information Administration (EIA 2013) the loss of credibility when design expectations of energy efficiency and actual building consumption outcomes differ substantially. Bordass et al., 2004

3 T H E C R E D I B I L I T Y G A P O F B U I L D I N G E N E R G Y C O N S U M P T I O N E N E R G Y P E R F O R M A N C E REAL PREDICTED [ ] credibility gaps arise because the assumptions often used are not well enough informed by what really happens in practice. Bordass et al., 2004 S T O C H A S T I C L a c k o f k n o w l e d g e

4 T H E C R E D I B I L I T Y G A P O F B U I L D I N G E N E R G Y C O N S U M P T I O N 5 times Delivered energy intensities distribution Commercial buildings, USA (LBNL 2014) Same building Different working users styles WIDE RANGE Energy consumption

5 D A T A M I N I N G A P P R O A C H E S Techniques that take advantage of large-scale data sets to detect patterns of behavior Information extraction from data (classification techniques) Provide knowledge about structure and interrelation among data (association techiques) Create models predicting future events (decision trees) Some applications: Market sales: actionable associations between buying diapers and beer on thursdays (product placement) Witten and Frank (2005) Telecomunication companies: identify clients needs and trends relater to household charactericts McCarty (1997) Political campaign operatives: identify potential supporters by datamining of database of existing voters Esdall (2006) Patterns of energy related behavior?

6 D A T A S E T Bank office building (Frankfurt, Germany) 16 offices (W, E oriented) Single/double occupancy Monitoring aspects Parameters Interval Climate Outdoor air temperature, outdoor humidity, wind speed, solar radiance 10 min Operation & Maintenance Monitoring of heating, cooling, lighting and ventilation system, and related energy flows 10 min Indoor environmental quality Indoor (operative) temperature, humidity, (CO2) 10 min Occupants activities and behavior Window state (open/closed) Presence 10 min 2 years data

7 D A T A M I N I N G A P P R O A C H E S K N O W L E D G E D I S C O V E R Y I N D A T A B A S E ( K D D ) Impact users profiles More than 1 million data points for each office

8 D A T A M I N I N G A P P R O A C H E S STEP 1 CLUSTER ANALYSIS K-MEANS ALGORITHM Grouping data objects into clusters objects in the same cluster have high similarity, objects in different clusters have low similarity. STEP 2 ASSOCIATION RULE MINING FP-GROWTH ALGORITHM Associations and correlations (rules) between attributes of the same dataset based on information obtained from cluster analysis Intra cluster distances are minimized Extra cluster distances are maximized THE MARKET BASKET ANALYSIS P A T T E R N S O F B E H A V I O R W O R K I N G U S E R P R O F I L E S

9 D A T A M I N I N G F R A M E W O R K STEP 1 CLUSTER ANALYSIS k-means algorithm P A T T E R N S O F B E H A V I O R Improvement of the notion of behavioral patterns not only as statistical relevant clusters Incorporating the motivational dimension with typical window opening habits, preferences and attitudes. Patterns of behavior Mined parameters Subset data Motivational Window opening/closing drivers Influencing variables Opening duration Window state h window open or close/day Interactivity Window changes n changes/day Position Window degree of opening tilting angle (from 0 to 1)

10 D A T A M I N I N G F R A M E W O R K STEP 1 CLUSTER ANALYSIS k-means algorithm P A T T E R N S O F B E H A V I O R Patterns of behavior Mined parameters Subset data Motivational Window opening/closing drivers Influencing variables Top 4 factors influencing window opening Top 5 factors influencing window closing 7 Indoor air temperature Arriving time Time of the day (early morning) Occupancy presence 6 Indoor air temperature Outdoor air temperature Time of the day (evening) Leaving time Occupancy presence Impact [abs value] Impact [abs value] Thermal driven Thermal&Time driven Time driven Thermal driven Time driven OPENING CLUSTER 1 OPENING CLUSTER 2 OPENING CLUSTER 3 CLOSING CLUSTER 1 CLOSING CLUSTER 2

11 D A T A M I N I N G F R A M E W O R K STEP 1 CLUSTER ANALYSIS k-means algorithm P A T T E R N S O F B E H A V I O R Patterns of behavior Mined parameters Subset data Opening duration Window state h window open or close/day h/day h/day h/day h/day E10 E04 W05 E07 W01 E08 E05 E09 W04 E02 W03 E06 E01 E11 W02 E03 LONG OPENING winter summer spring autumn MEDIUM OPENING SHORT OPENING

12 D A T A M I N I N G F R A M E W O R K STEP 1 CLUSTER ANALYSIS k-means algorithm P A T T E R N S O F B E H A V I O R Patterns of behavior Mined parameters Subset data Interactivity Window changes n changes/day n changes/day changes/day changes/day changes/day E08 E04 E07 E02 W01 E09 E05 W05 E06 W04 E01 W03 E11 W02 E10 E03 winter summer spring autumn ACTIVE OPERATOR NEUTRAL OPERATOR PASSIVE OPERATOR

13 D A T A M I N I N G F R A M E W O R K STEP 1 CLUSTER ANALYSIS k-means algorithm P A T T E R N S O F B E H A V I O R Tilting Angle :00:00 AM 1:00:00 AM 2:00:00 AM 3:00:00 AM 4:00:00 AM 5:00:00 AM 6:00:00 AM 7:00:00 AM 8:00:00 AM 9:00:00 AM 10:00:00 AM 11:00:00 AM 12:00:00 PM 1:00:00 PM 2:00:00 PM 3:00:00 PM 4:00:00 PM 5:00:00 PM 6:00:00 PM 7:00:00 PM 8:00:00 PM 9:00:00 PM 10:00:00 PM 11:00:00 PM BIG OPENING INTERMEDIATE OPENING SMALL OPENING

14 D A T A M I N I N G F R A M E W O R K A S S O C I A T I O N R U L E S M I N I N G STEP 2 ASSOCIATION RULES MINING W O R K I N G U S E R P R O F I L E S Find unsuspected relationships and summarize the data in novel ways Extract frequent correlations from patterns of behavior Motivational Duration Interactivity Position Office window opening window closing window state window change window tilting angle E01 thermal driven thermal driven short openings passive operation small openings E02 thermal/time driven thermal driven short openings active operation small openings E03 thermal driven thermal driven short openings passive operation small openings E04 thermal driven thermal driven long openings active operation big openings E05 thermal/time driven time driven medium openings neutral operation intermediate openings E06 time driven time driven short openings neutral operation small openings E07 time driven time driven medium openings active operation intermediate openings E08 time driven time driven medium openings active operation intermediate openings E09 thermal/time driven thermal driven medium openings neutral operation small openings E10 time driven time driven long openings* passive operation big openings* E11 thermal driven thermal driven short openings passive operation small openings W01 time driven time driven medium openings active operation intermediate openings W02 thermal driven thermal driven short openings passive operation small openings W03 time driven time driven short openings passive operation small openings W04 thermal/time driven time driven short openings passive operation intermediate openings W05 thermal/time driven time driven long openings neutral operation big openings

15 D A T A M I N I N G F R A M E W O R K A S S O C I A T I O N R U L E S M I N I N G STEP 2 ASSOCIATION RULES MINING W O R K I N G U S E R P R O F I L E S Find unsuspected relationships and summarize the data in novel ways Extract frequent correlations from patterns of behavior 80% 20% Patterns of behavior User α User β Motivational physical environmental driven contextual driven Opening duration short periods ( hours/day) long periods ( hours per day) Interactivity infrequently (0-0.7 times per day) frequently (1-1.7 times per day) Position small openings (< 0.3 degree of tilting angle) intermediate openings (< 0.6 degree of tilting angle)

16 C O N C L U S I O N S 1. automatically extrapolate fast legible, valid, novel occupancy patterns from big data streams 2. key intermediate to visualize behavioral patterns in (big) energy data 3. provide accurate assumption of actual natural ventilation scenarios in office buildings 4. quantify the energy/economic impacts of diverse ventilation scenarios on energy use in a building 5. quantify motivational drivers, habits, preferences and attitudes on office building 6. deliver a set of behavioral rules at the office level to direct specific energy saving strategies 7. allow building designers, operator, manager to tailor efficient and robust system and building envelope control strategies and design

17 I E A E B C A N N E X 6 6

18