1. Data Driven Solar Panel Anomaly Detection
Presented by Gao, Xiang

2. Background
The annual installed capacity was about 29.6 Gigawatts (GW) in 2011 and 31 GW in 2012
Solar energy efficiency is vulnerable to external factors(e.g., dust on panels can result in up to a 40% degradation )

3. Related Work
Bo Hu, S. Keshav. Solar panel anomaly detection and classification. 2012
A. Luque and S. Hegedus. Chapter 20. Handbook of Photovoltaic Science and Engineering.
E. Skoplaki, J.A. Palyvos. On the temperature dependence of photovoltaic module electrical performance

4. Our goals Predict the type of particular anomaly
Report the energy loss due to each type of anomaly, including weather factors
Objective anomalies
Shadows: correlation of covering area and decreasing effect
Dust
Snow
wind/ leaves

5. Basic Idea
Derive theoretical power output from irradiance data and compare with real power output
No anomaly
With anomalies

6. Challenge 1
theoretical power output = in-plane irradiance * efficiency
in-plane irradiance: measured by pyranometers
Efficiency: changing with temperature, irradiance,..
What if we do not have in-plane irradiance data but only horizontal?
How to derive current efficiency from standard efficiency in manual?

7. An irradiance model
Theoretical
Measurement

8. Horizontal Irradiance
1367W/m2
Θzs
not a strictly circular
Orbit!

9. Horizontal Irradiance(Non-tracking)
δ :Solar declination
φ: latitude
Fitting constants
Meinel A, Mainel M, Applied Solar Energy, An Introduction, Addison-
Wesley, Reading, MA(1976)

10. Fit constants to local data(TRCA)
Sample selection
Top 10% irradiation from each month
Visual inspection
11 perfect sunny days evenly distributed from spring to fall
Optimal(MMSE)
(0.8,0.41)
(0.7,0.678) obtains % MSE compared with this
Ensure theoretical one is the maximum
(0.831,0.549)
5.7% worser than optimal

11. In-plane Irradiance

12. Evaluation with data from TRCA(2012-3-11)

13. Efficiency Influence factors: temperature, irradiance, materials, etc.
E. Skoplaki, J.A. Palyvos. On the temperature dependence of photovoltaic module electrical performance. Solar Energy 83, 2009.

14. Efficiency
Is it proper to ignore the anomalies with irradiance less than 200 ?

15. Anomaly detection ratio = real power output / theoretical power output
December 2011 :
examining period: 8-18 (daytime)
Single time point
ratio < 0.7: 4658 anomalies (59.7%)
ratio < 0.5: 3529 anomalies (45.2%)
Anomaly sets(ratio < 0.7)
56. group all continuous time point anomalies
13. examine the anomalies with irradiance>200

16. Anomaly detection 2012 (TRCA data) Single time point
Irra. > 200 && ratio < 0.7: 9311 anomalies (4.7%)
Anomaly sets : 896
Distributions:

17. Anomalies time distribution
0.7>Ratio>=0.6
0.1>Ratio>=0

18. Are anomalies at sunset real ?
The anomalies in 17:00 – 18:00
Dynamic examining period

19. An example of snow melting

20. Anomaly classification
Some training sets
More training sets
and test sets are to be
collected

21. Tentative Characteristics
Value
Season
Range
Duration
Speed
Arrays
Correlations:
panels position
whether be affected
Descriptions:
Slope #
Slope order
Slope value

22. Challenge 2 Ground truth Data quality Keep changing
Difficult to derive the truth naturally(e.g. most snow happens at night)
Data quality
Granularity
Alignment
Small values (set a threshold)
Measurement error(e.g. temperature = 850, etc.)

23. Typical measurement error