Quby is a leading company offering data driven home services technology across European markets, known for creating the in-home display and smart thermostat Toon. We enable our partners to take on a leading role in the home services domain, by offering data driven home services. Our services enable users to control and monitor their homes using both an in-home display and app.
As a data driven company, we use AI and machine learning, backed by Apache Spark, to generate actionable insights for all our end users. Via our IoT devices we have access to Europe’s largest energy dataset, petabytes in scale and growing exponentially. This unique dataset enables us to introduce new data driven services, with a particular focus on homes with smart meter installations.
In this talk, Ellissa will describe how machine learning is implemented on the Quby platform and will show multiple use cases backed by high-resolution IoT data. We’ll take a look at super resolution techniques for time series data, where using detailed high-resolution energy data is used to show personalized energy insights for users where only limited low-resolution energy data is available. We’ll show how ML algorithms offer the possibility for non-intrusive monitoring of elderly patients.
Ellissa will share the experiences from the Data Science and Data Engineering teams at Quby with bringing these data science algorithms from R&D to production using Databricks and the lessons learned in offering these services to hundreds of thousands of users on a daily basis.
1.WIFI SSID: Spark+AISummit | Password: UnifiedDataAnalytics
2.Making Homes Efficient and Comfortable Using AI and IoT Data Ellissa Verseput, Quby #UnifiedDataAnalytics #SparkAISummit
3.Outline • Why, What, How Quby • 2 Example Use Cases – Bill Breakdown (Efficient) – Thermostat Program Advice (Comfortable) 3
4.We believe the future can be better. Easier, more comfortable, and more sustainable. We help businesses and their customers to make this change without compromising on the important things in life. 4
6. Solar panels and power Central heating Boiler Water sensor storage system adapters Water Meter adaptors Gas sensor Gas Smart plugs & smoke detectors Z-Wave Z-Wave Electricity sensor Electricity Olisto Drebble Philips Hue Athom Homey Amazon Alexa Google Home
7.Over 400,000 connected homes across Europe Our partners: #UnifiedDataAnalytics #SparkAISummit 7
8.Outline ü Why, What, How Quby • 2 Example Use Cases – Bill Breakdown (Efficient) – Thermostat Program Advice (Comfortable) 8
9.Current Quby Portfolio Home services Efficient home Comfortable home Trusted home Energy insights gas & electricity Smart thermostat functionality Monitoring the home (high frequency) Energy insights with low frequency Smart home integration Smart security solution Waste checker (~7 use cases) Air quality measurement & insights Assisted living Solar generation integration Water insights & saving tips Dedicated app development for utilities with superior user experience & personal relevance 9
10.Use Case #1 Bill Breakdown Show how different appliances and activities in the home contribute to the energy bills 10
11.Key Technology: Load Disaggregation 11
12.Quby’s disaggregation algorithms Patented algorithms can detect appliances from 10 second resolution electricity meter data 12
13.Bill Breakdown Categories for 10sec-data Elec bill Elec data 13
14.10sec-data VS 15min-data High Resolution Low Resolution 14
15.Bill Breakdown Categories for 15min-data Elec bill Elec data 15
16.Super Resolution 16
17.Utilizing our large database of high-resolution data, we apply advanced techniques to offer a more personalized, more dynamic and more accurate bill breakdown service. Appliance User Low resolution data detections Enhancement Bill breakdown Most similar Similar users High resolution data users with high resolution data 17
18.Bill Breakdown Architecture API Toon data collector P4 data collector 18
19.Bill Breakdown for low resolution data 19
20.Outline ü Why, What, How Quby Ø 2 Example Use Cases • Bill Breakdown (Efficient) • Thermostat Program Advice (Comfortable) 20
21.Outline ü Why, What, How Quby Ø 2 Example Use Cases 🤯 Bill Breakdown (Efficient) • Thermostat Program Advice (Comfortable) 21
22.Use Case #2 Thermostat Program Advice Suggest updates to the Toon users’ thermostat program, such that the program better reflects their behavioural patterns 22
23.Thermostat Program Advice Key Technology Presence Detection Somebody home? Cooling down and Humidity sensor warming up rate 23
24.Non-intrusive Monitoring Proof of Life – Toon can detect when a person is present Safety – Toon can indicate an active/inactive (elderly) resident Heating/Lighting – Toon can detect people are active to optimize heating 24
25. Training a Machine Learning Model Hour Mon Tues Wed 08:00 Home Away Home 09:00 Away Away Home 10:00 Away Away Home Hour Mon Tues Wed 08:00 Home Home Home 09:00 Home Away Away Cooling down and Humidity sensor warming up rate 10:00 Away Away Away 25
26. Challenge We want to track model improvements and reproduce models
28.This is how we did it 28
29. Challenge • We have saved the best, trained machine learning model in Python, • but for our production data pipeline we want to use Spark Scala