Statistics notes ,it includes mean to index numbers
GrowSmarter Webinar : Generating Insights from Energy Household by AGT
1. Generating Insights from Energy Household
Data
Stefaniia Legostaieva - Data Scientist at AGT
Manuel Görtz – Project Manager at AGT
18th September 2018, Darmstadt
Stefaniia Legostaieva - Data Scientist at AGT
Manuel Görtz – Project Manager at AGT
18th September 2018, Darmstadt
This project has received funding from the European Union’s Horizon 2020 research and innovation
programme under grant agreement no 646456. The sole responsibility for the content of this
presentation lies with the GrowSmarter project and in no way reflects the views of the European Union.
2. AGT is a pioneer in IoT and Social Data management,
Big Data integration and advanced analytics.
MANUFACTURING
& ENERGY
SMART TRAFFIC
MANAGEMENT
SMART
CITIES
NEW MEDIA
CONTENT
SPORTS &
ENTERTAINMENT
EVENTS
Private organization
FOUNDED in 2007
Headquartered in
SWITZERLAND
>$8B in
IoT projects
R&D CENTERS in
IL, DE, U.S. > 250 engineers
~$1B ANNUAL REVENUE
Profitable, CFP
3. INTRODUCTION
Project goal
To reduce energy consumption in private sector
Solution
with the help of smart plugs and insights about
energy consumption of individual devices
AGT proprietary and confidential
4. AGENDA
System
Deployment with Fibaro Smart plugs and Homee Gateway in Smart Homes
Dashboard
Web-based application (Dashboard) for device-level energy awareness in Smart Homes
Services / analytics
Awareness
Activity recognition and behaviour analytics
Demo
AGT proprietary and confidential
6. ENERGY AWARENESS DASHBOARD
GrowSmarter Insight Dashboard
Web-based application for device-level energy awareness in Smart Homes
Productized and deployed on AWS
Real-time data collection, processing and analytics
Scalable up to 10 000 households
AGT proprietary and confidential
7. ANALYTICS / SERVICES
Smart Home Energy Analytics Services
Awareness
Device usages statistics
Connected device type recognition
Activity recognition and behaviour analytics
Device usage mode recognition
Device abnormal behavior detection
Recommendations
Device replacement
AGT proprietary and confidential
8. ENERGY AWARENESS DASHBOARD
Device energy consumption costs
Value
- how much and electrical energy do I spend on each home appliance?
AGT proprietary and confidential
9. ENERGY AWARENESS DASHBOARD
Device usages statistics
Value
- how many times do I use a device?
- for how long do I use a device?
- how much it costs (based on provided price per kWh)?
AGT proprietary and confidential
10. Device Mode Recognition
Value
- how much the fast program of my dishwasher
consumes comparing to the eco program, how much
does it cost?
Analytics
Analyzing consumption behavior of a user
Comparison of different modes of usages within a
device
Recommendations of using energy efficient mode
based on consumption behavior
ENERGY AWARENESS DASHBOARD
RegularEcoFast
Dishwasher
AGT proprietary and confidential
11. Device Type Recognition
Value
- did I plug a new device, shall my devices be re-arrange
based on a device type?
Analytics
Detect when and what kind of new device was
connected
Re-group devices in the dashboard based on the
device and not the smart plug
ENERGY AWARENESS DASHBOARD
KettleMicrowave
Washing
machineTVAGT proprietary and confidential
12. Device Abnormal Behavior Recognition
Value
- does my device work normally?
- do I use my device as usually?
Analytics
Analyzing consumption patterns of a device
Detect abnormal states of the device
Detect abnormal use of the device
ENERGY AWARENESS DASHBOARD
normalanomalous
Abnormality: door is open for too long
AGT proprietary and confidential
13. Data
Public dataset
Office data
ANALYTICS - PIPELINE
AGT proprietary and confidential
Semantic data
type matching
Zero consumption
removal
Data fusion
Data cleaning
Data
preprocessing
Semantic data
type matching
Zero consumption
removal
Transforming into
windows of pre-
defined size: 1
hour, 30 or 15
minutes
Feature
extraction
Statistical features
extraction: mean,
max, min,
variance, peaks,
time to max
DTW (Dynamic
time wrapping) as
similarity measure
Exploratory
analysis
Model building
Random Forest
classifier
KNN (K-nearest
neighbors)
Evaluation
Visualization
Confusion matrix
F1 measure
Plots
Make decision
Build data
product
Device type recognition
⁻ approach 1
⁻ approach 2
feedback loop
User’s feedback