Assessing Factors Underpinning PV Degradation through Data Analysis
1. Assessing Factors Underpinning
PV Degradation through Data
Analysis
Xin Chen, Todd Karin, Anubhav Jain
(and soon, Baojie Li) - LBNL
w/collaboration from PV Evolution Labs
DuraMat 2.0 project
2. Overview
Distinguish system and module degradation
Apply PVPRO method to determine if the module is operating off maximum power
point (MPP) because of inverter clipping, temperature derating or MPP errors.
Apply PVPRO methods to large-scale data analysis
Use PVPRO to extract IV parameters to investigate the degradation rate and mode
of 25 large scale field PV systems.
Connecting BOM with degradation
Investigate correlation between bill-of-materials (BOM) data and degradation
results of modules after accelerated stress tests.
3. Task 1: Off-MPP detection
• Data cleaning is a critical part of degradation
analysis
• Existing methods detect inverter clipping
based on power exceeding a certain value
[Rdtools], identifying point masses in
probability density function of power output
[SCSF], or inspecting levels, thresholds or the
shape of the AC power curve [PVanalytics].
• However inverter tracking can fail for a
multitude of reasons
• Use PVPRO analysis to fit a single diode
model to the DC IV data, and comparing the
best-fit model to the data to identify off-MPP
points
DC voltage, current,
POA irradiance,
temperature
Extracted IV parameters
(series resistance, shunt
resistance, photocurrent,
etc.)
Simulated Imp, Vmp
Off-MPP
PVPRO
Single diode
model
Euclidean
distance or
other
criterions
Schematic process
of determine off-
MPP with PVPRO
4. Task 1: Off-MPP detection
[1] D. C. Jordan, B. Sekulic, B. Marion and S. R. Kurtz, "Performance and Aging of a 20-Year-Old Silicon PV System,"
in IEEE Journal of Photovoltaics, vol. 5, no. 3, pp. 744-751, May 2015, doi: 10.1109/JPHOTOV.2015.2396360.
• Right figure is a MPP Vmp-Imp
scatter plot along with a PVPRO
single-diode best-fit model for 10
weeks of data from the NREL SERF
East system[1].
• The figure shows multiple off-MPP
points, for example clustering of
points near 14.5 V is due to inverter
clipping. Vmp and Imp are per-
module values.
5. Task 2: Large-scale PV system analysis
IV parameters
Single diode
model
IV simulated IV observed
Loss
Initial guess
update
PVPRO fit of a single diode model to NIST
Ground data[1] at a single time step.
[1] Boyd, M. (2017), Performance Data from the NIST Photovoltaic (PV) Arrays and Weather Station, Journal of Research (NIST JRES),
National Institute of Standards and Technology, Gaithersburg, MD, [online], https://doi.org/10.6028/jres.122.040 (Accessed July 13, 2021)
MPP data
6. Task 2: Large-scale PV system analysis
• For NIST dataset, the estimated power
degradation rate is -1.7%/yr. Inspecting the
first panel, photocurrent loss is estimated to
cause a -0.8%/yr loss in power, making
photocurrent loss responsible for 47% of the
observed power loss. This system also
appears to show an increase in series
resistance over time.
• 25 large scale PV system datasets will be
collected from PV Fleet[1]. For each system,
we will identify the presence of both overall
module degradation as well as break down
the overall degradation into specific I-V
parameters.
[1] https://www.nrel.gov/pv/fleet-performance-data-initiative.html
7. • Goal: determine how aspects of the bill of materials (BOM)
influence accelerated testing results
• Such an analysis could conceivably be performed using data sets
from testing labs such as PVEL
• Step 1 (current) is getting the data into a structured format, which
is typically as PDF files with embedded graphics
Task 3: BOM analysis
8. Task 3: BOM analysis
• Manual text extraction of BOM data set launched via PVEL collaboration; estimated to be finished in one month.
• Perform basic exploratory data analysis - e.g., the number of distinct materials used for various components, the
range of measured performances, outliers, etc.
• Develop descriptors for the various materials present, i.e., so that materials can be grouped together for analysis
along multiple dimensions rather than treated as individual categorical variables.
• Correlate the materials present with performance of PV modules after accelerated stress tests.
Materials 1
(%)
Materials 2
(%)
……
……
Data
points
Attributes: percentage of different
materials, dimension of modules, etc.
Inputs
Hidden
layers
Outputs:
Imp, Vmp, etc.
Truth:
Imp, Vmp, etc.
Loss
Update
9. Summary
Distinguish system and module degradation
T1-1: Develop the algorithm for determining off-MPP points
T1-2: Compare the output of this algorithm to existing tools like Rdtools [1]
T1-3: Integrate the final data filter with the degradation analysis tools of existing open-source packages
such as RdTools or statistical clear sky fitting
Apply PVPRO methods to large-scale data analysis
T2-1: Refine the methods on synthetic data sets to correct for known convergence issues in the presence
of measurement noise/drift
T2-2: Apply PVPRO on 2-3 additional well-qualified systems and perform a comparison between regular
field IV and PVPRO predictions
T2-3: Deploy PVPRO on a much larger scale, e.g. 25 systems selected from the PV fleets data set
Connecting BOM with degradation
T3-1: Perform a data-driven analysis that attempts to predict aspects of measured performance through
the materials and their descriptors and publish a report on the findings
T3-2: A properly anonymized summary will be disseminated to the data hub
[1] RdTools, version 2.0.5, https://github.com/NREL/rdtools, DOI:10.5281/zenodo.1210316