Data Driven Buildings Bharathan Balaji Amazon AI Labs Buildings - PowerPoint PPT Presentation

Data Driven Buildings Bharathan Balaji Amazon AI Labs

Buildings are evolving.. Connectivity Water Security Shelter Sensors Electricity Thermal Comfort Software Fire Safety Data: Interact. Optimize. Innovate. 8 November 2018 Bharathan Balaji, SBIoT 2018 2

Data Generation Bottleneck: Sensors Wired Wireless + Battery Wireless + Harvesting Battery Maintenance Unreliable Reliable Inexpensive Inexpensive Expensive Flexible Flexible Not Flexible 8 November 2018 Bharathan Balaji, SBIoT 2018 3

Pible: Perpetual Indoor BLE Sensor • Sensors • Light • Temperature • Humidity • PIR • Door Event • Bluetooth beacon • Bluetooth Low Energy • Solar harvesting • Super capacitor Limitation: Manual configuration Pible: BuildSys 2018 8 November 2018 Bharathan Balaji, SBIoT 2018 4

Duty Cycle with Reinforcement Learning Reward • Automatically adapts to each lighting condition • 1 sample every 56 seconds on average Days Best Demo Award: BuildSys 2018 Scaling Energy Harvesting Configuration: EnsSys 2018 8 November 2018 Bharathan Balaji, SBIoT 2018 5

Data Collation Bottleneck: Vertical Systems Plug Loads Lighting System Heating, Ventilation and Air Conditioning (HVAC) Enterprise Network Security System November 6, 2018 Bharathan Balaji, SBIoT 2018 6

Integration Platform for Applications Next generation Apps building applications via standardized API Visualize Maintain Analyze Control REST/ Native API Ø Scalable, distributed data storage Data management Ø Metadata and contextual tagging system for sensors and Ø Access control across users actuators Ø REST API for app development Data Connectors Large amount of data generated in modern buildings Building A Building B Building C Building D 6 November 2018 Bharathan Balaji, SBIoT 2018 7

Smart Building Applications Personalized Control Energy Disaggregation BuildSys ’13, Ubicomp ‘16 BuildSys ‘10, BuildSys ‘13 Fault Detection and Diagnosis Occupancy Based Control BuildSys ‘14 IPSN ‘11, SenSys ‘13 11/6/18 Bharathan Balaji, SBIoT 2018 8

App Portability Bottleneck: Naming Semantics 8 November 2018 Bharathan Balaji, SBIoT 2018 9

Each Building is Different v University, Hotels, Hospitals, Shopping Malls Lecture Hall Bio Labs Mixed Use Library • Equipment, Vendor, Institution • Changes with time: Repairs, Retrofits November 6, 2018 Bharathan Balaji, SBIoT 2018 10

Brick: Building Metadata Schema Applications Demand Response Occupant Interaction Fault Detection Management Monitoring Access Control APIs Brick Smoke Detector Thermostat Motion Sensor Infrastructure: Sensors Luminaire HVAC Fire Safety Equipment Brick: BuildSys 2016, Applied Energy 2018 https://brickschema.org 11/8/18 Bharathan Balaji, SBIoT 2018 11

Brick Fundamentals Tags TagSets Relationship Zone Sensor Location RM-3 type hasLocation Temp Room Temp_Sensor type Sensor Zone_Temp_Sensor ZNT-3 6 November 2018 Bharathan Balaji, SBIoT 2018 12

Brick Class Hierarchy Point Location Equipment Floor Command Fire Safety System Room Sensor HVAC Kitchen AHU Temperature Sensor Lab Room Terminal Unit Temperature VAV Sensor 8 November 2018 Bharathan Balaji, SBIoT 2018 13

An Example “Model” Building AHU Power Meter Lighting Controller Return Fan Supply Fan Lighting Zone VAV Damper Supply Air Return Air Room 101 Room 102 Thermostat Temperature HVAC ZONE CO2 Sensor

Relationships in Example Building hasPoint Lighting Power AHU Controller Meter controls feeds isLocationOf Lighting feeds HVAC VAV Zone Zone hasPoint isLocationOf isLocationOf hasPart hasPoint Room Room Temperature Damper 102 101 Sensor Relationship Legend Location Equipment Point 11/6/18 Bharathan Balaji, SBIoT 2018 15

1.Complete Vocabs, Extensible Framework 2.Represent all necessary relationships -> Using RDF 3.Usable query mechanism -> SPARQL over RDF 4.Open Source -> BSD and RFC https://brickschema.org 11/6/18 Bharathan Balaji, SBIoT 2018 16

Need to Map Existing Building Metadata 8 November 2018 Bharathan Balaji, SBIoT 2018 17

Scrabble: Map existing buildings to Brick Source Building Target Building Buildings R3.WtrTemp RM120.Temp Expert Known RM = Room Temp = Temperature Wtr = Water Water Temperature Temperature Sensor Sensor in Room-3 in Room-120 • Learn from prior examples • Ask expert when known examples are not enough 8 November 2018 Bharathan Balaji, SBIoT 2018 18

Scrabble: Multi-Stage Algorithm 8 November 2018 Bharathan Balaji, SBIoT 2018 19

Experts Give Examples at Two Stages 8 November 2018 Bharathan Balaji, SBIoT 2018 20

Dataset • University of California, San Diego – 3 buildings, 3000 data points { ‘vendor_name’: VAV ‘NAE 99 N2 0 VMA999 SA T’, Supply Air Temp Sensor ‘bacnet_desc’: ‘Supply Air Temp’, ‘bacnet_unit’: ‘64’, } Building • Carnegie Mellon University – 1 building, 1000 datapoints Floor ‘CMU/CMPS BLDG/First Floor/VAV Corridor Corridor 9900 Central/ VAV Airflow Setpoint’ Air Flow Setpoint 8 November 2018 Bharathan Balaji, SBIoT 2018 21

Results: Compared to State of the Art Baseline • Faster learning speed than baseline • Capable of accumulating examples given by expert 8 November 2018 Bharathan Balaji, SBIoT 2018 22

Data Driven Buildings Visualize Sensor Maintain Brick Devices Analyze Control Equipment 8 November 2018 Bharathan Balaji, SBIoT 2018 23

“ Alexa, join the meeting ” Alexa for Business https://aws.amazon.com/alexaforbusiness Alexa for Hospitality https://www.amazon.com/alexahospitality

Thank you! Amazon AI Labs – Reinforcement Learning Team LinkedIn: https://www.linkedin.com/in/bharathanbalaji/ Email: bhabalaj@amazon.com Acknowledgements Francesco Fraternali, Jason Koh, Anna Levitt, Gabe Fiero Mani Srivastava, Rajesh Gupta, Yuvraj Agarwal 8 November 2018 Bharathan Balaji, SBIoT 2018 25

Results: Buildings in the Same Campus 8 November 2018 Bharathan Balaji, SBIoT 2018 26