Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
KNIME UGM 08/03/2016
Tim WuytsEngineer Data Intelligence
Agenda
• introducing fifthplay
• use cases
• marketing analysis
• PV monitoring
• automating appliance detection
a member of Niko Group
Part of Niko Group
We believe that people deserve a modern
electrical installation that gives them COMFORT,
CONTROL and EFFICIENCY in a pleasurable way.
17070075% °1919
Market share in switches and
socket outlets in Belgium
Family-owned company
founded in 1919
170 mill ion euro turnover
+700 employees
Mission fifthplay
We provide connected solutions, based on a scalable IoT platform, and bring added value for partners and their customers.
Fifthplay platform
AP
I
LAN
Integration of third party hardware
… and many more
Packaged Solutions
Where do we use KNIME?
• Marketing Analysis• IoT Data Intelligence Features
• Data Exploration
• Automation
Marketing Analysis
Marketing Analysis - Workflow Example
Marketing Analysis – PPT ExampleDevice X
Monthsactive
25%
of
dat
a
25% of data
Box50%
TOTAL CUSTOMERS *697
Gw inactive 132
19%
• Out of 697 customers with a device of Type X connected to their gateway, 19% had their gateway inactive on the moment of analysis
• Analysis of these 19% (132 customers) reveals the following:
• 1 in 4 gateways (25%) was active up to 1 month before becoming inactive
• Half of the gateways became inactive within the first 4 months
• 75% was active for at most 8 months before inactivity.
• The longest period recorded before inactivity was 18 months.
• Some gateways were never active.
• Conclusion
• The gateways tend to become inactive very quickly.
• Idea: create triggers to detect customers whose gateway became inactive in the last month & take action (for instance: contacting them to check if there is an issue).
IoT data intelligence features
Alerting Oven, TV, ...Appliance Type Detection
Benchmarking
PV monitoring
Product improvement Self learning thermostat, TRV
. . .
More cases tocome
PV monitoring
The produced electricity from your solar panels is predicted based on historical profile, cloud index, radiation and temperature. When production is lower than expected, the owner is notified.
PV MonitoringPrediction Model Search
PV Monitoring‘Prediction’ Model Search
All PV installations in one single model ?
Time for KNIME
PV MonitoringPrediction Model Search
Loops + Linear/Polynomial Regression + Boxplot
PV MonitoringAnomaly Detection
• problem: training data !!
• only 1 known case of malfunction
• approach:
• find model that would have predicted the
known malfunction
• apply model to other installations
• investigate detected anomalies
Q: Is my PV installation still performingat the same performance level as before?
PV MonitoringAnomaly Detection
PV degradation started 01/2013
• failure started on 27/12• 40% of panels offline on 15/01• 60% of panels offline on 12/03• failure solved on 28/05
PV MonitoringAnomaly Detection Models
Day below prediction
PV MonitoringAnomaly Detection Models
Day below prediction – moving average
PV MonitoringAnomaly Detection Models
% Underperformance – moving average
Time for KNIME
PV MonitoringAnomaly Detection Models
• Applying the final model to all installations reveals
• 20 installations showed anomalies
• 4 installations had miscellaneous inconsistencies in the historic data
• 9 triggered an alert in the winter months
but production values were very low (< 100 Wh per day)
• 7 installations had an alert on the same day (28/12/2014)!
PV MonitoringSo what happened?
PV MonitoringSo what happened?
DEZE GEGEVENS MOGEN VRIJ WORDEN GEBRUIKT MITS DE VOLGENDE
BRONVERMELDING WORDT GEGEVEN:
KONINKLIJK NEDERLANDS METEOROLOGISCH INSTITUUT (KNMI)
STN = stationsnummer
YYYYMMDD = datum (YYYY=jaar MM=maand DD=dag)
RD = 24-uur som van de neerslag in tiende millimeters
van 08.00 voorafgaande dag- 08.00 UTC huidige dag
SX = codecijfer sneeuwdek om 08.00 uur UTC:
codecijfer sneeuwdikte
1 1 cm
... ...
996 996 cm
997 gebroken sneeuwdek < 1 cm
998 gebroken sneeuwdek >=1 cm
999 sneeuwhopen
751,20141223, 0, 0,
751,20141224, 10, 0,
751,20141225, 37, 0,
751,20141226, 70, 0,
751,20141227, 350, 998,
751,20141228, 46, 998,
751,20141229, 18, 0,
751,20141230, 4, 0,
Data from the ‘Neerslagstation’ Vrouwepolder
AutomatingAppliance Detection
• Appliance What???
• KNIME server
Appliance Type Detection
?
?
AutomationPlatform Architecture
AutomationAppliance Detection
Time for KNIME
KNIME tips
• loops
• multiple axes -> manipulate data
• email reports
• wrapped nodes + quickform
• read variables from file
• knime URI
KNIME tipswrapped nodes + quickform
KNIME tipsread variables from file
KNIME URIs
knime://knime.workflow/../../environment.variables
Thank you