The DuraMat Data Hub and Analytics Capability provides a centralized resource for sharing solar PV data. It collects performance, materials properties, meteorological, and other data through a central Data Hub. A data analytics thrust works with partners to provide software, visualization, and data mining capabilities. The goal is to enhance efficiency, reproducibility, and new analyses by combining multiple data sources in one location. Examples of ongoing projects using the hub include clear sky detection modeling to automatically classify sky conditions from irradiance data.
Call Girls in Majnu Ka Tilla Delhi 🔝9711014705🔝 Genuine
DuraMat Data Hub and Analytics Capability Resource for Solar PV Data
1. The DuraMat Data Hub and
Analytics Capability
A Resource for Solar PV Data
Robert White1 and Anubhav Jain2
1National Renewable Energy Laboratory 2Lawrence Berkeley National Laboratory
2. Data and Analy*cs overview - 1
Field deployment
Data Hub Data analytics
Materials
Forensics &
Characterization
Predictive
simulationModule testing
Techno-economic
analysis
DuraMAT capability areas
and partners will share PV
performance, materials
properties, meteorological
data, & other data through
the Data Hub
3. Data and Analy*cs overview - 2
Field deployment
Data Hub Data analytics
Materials
Forensics &
Characterization
Predictive
simulationModule testing
Techno-economic
analysis
Data analytics is a
cross-cutting thrust
that will work with
DuraMat partners to
provide software,
visualization, and data
mining capabilities
8. Examples of data for “the commons”
Footer 4
4 4
Time series data Meteorological data Test chamber data
Materials simulation Materials properties Degradation analysis
The Data Hub will host
and connect to data
from multiple sources
and specializations
These research areas
and data sources are a
small sample of what
the hub plans to host
19. Data Analy*cs capability
Visualize, model, and predict
Release open
source software
• The data analy=cs thrust is cross-
cuang, i.e., we can work with any
thrust that needs help
• Help visualize, model, and predict
based on your data
• Seeking collabora*ons!
22. Clear sky detec*on analy*cs
x1
x2
x3
…
x1
x2
x3
…
x1
x2
x3
…
x1
x2
x3
…
x1
x2
x3
…
x1
x2
x3
…
+ NSRDB clear sky labels
ML classifier,
e.g., random
forest
Features used (similar to
pvlib)
Single point features:
• GHI - GHICS
• GHI’ - GHI’CS
• Time from solar
noon
Window based metrics
calculated every 1 hour
• GHI - GHICS average
and standard
devia=on
• GHI’ - GHI’CS
average and
standard devia=on
• GHILL - GHILL
CS
(difference of line
length)
Step 2: describe each
irradia*on measurement
point by a set of
numerical features
Step 3: train a
machine learning
algorithm to make
predic*ons
Step 1: clean
data set of
unambiguously
mislabeled data
clear points labeled as
“cloudy” in NSRDB – filter
from analysis
23. ML model achieves ~98% accuracy!
No data cleaning Data cleaning applied
Next steps:
• Evaluate sensi=vity of model to different irradiance measurement intervals
• Run analysis across en=re NSRDB data set and test applicability of a single
model across mul=ple geographical loca=ons
• Comparisons of satellite vs ground-based measurements
• Benchmark against exis=ng algorithms as well as expert labeling
• Publish model and accompanying so_ware implementa=on
25. NREL EMN
Development Team
• Kris=n Munch
• Nick Wunder
• Chris Webber
• Dave Evenson
• Courtney
Pailing
DuraMat Data
Team
• Ben Ellis
• Birk Jones
• Pedro Perez
• Stephanie Moffi5
• Kevin Leung
• Jonathan Trinas=c
Acknowledgements
A Coordinated Team Effort
Data Analy@cs
• Ben Ellis
• Mike Deceglie
• NSRDB / PVDAQ
teams