Outboard Pyranometer Bifacial Capacity Testing

Jeff Newmiller (DNV)

2023-05-09

Overview

Scope of work
Foundations of ASTM2848
Possible adaptations for bifaciality
- Proportionality
- In-plane measurement
- Displaced measurement
- Outboard measurement
Preliminary stability review
Conclusions

Characterization of impact of bifaciality
Not a discussion of a complete test
Commenting on approaches to aligning measurements with intermediate model results
Goal is to evaluate whether observed power generation under reference conditions agrees with energy model associated with contract (PV system is functional)
- Traditionally supports financing energy model
- May also apply (differently) to developer-EPC contract

Published: \(P = E \cdot (a_1 + a_2 \cdot E + a_3 \cdot T_a + a_4 \cdot v)\)
Derived from:
- \(P = E \cdot (c_1 + c_2 \cdot (T_c - 25))\)
- \(T_c = T_a + E \cdot d_1 \cdot (v_{max} - v)\)
- The \(E^2 \cdot v\) second degree was dropped to first degree \(E \cdot v\)
Ambient temperature was chosen to capture thermal modeling disagreement
Wind speed \(v\) was originally specified to be at 10m height because free-stream correlations are much clearer than turbulent 2m correlations, but this requirement was removed during standardization
GHI was not chosen for power capacity testing because it needs long evaluation periods to average out the typically-high random model error
Direct shade on modules or sensors represents a dramatic discrepancy, but diffuse shading was intentionally not excluded as the model was expected to address that if needed.
Key strategy: set the evaluation boundary at the simplest feasible measurement points

Cell
- Bifaciality factor: \(E_{net} = \left(1 + \varphi \cdot \frac{G_{RPOA,eff}}{G_{POA,eff}}\right) \cdot G_{POA,eff}\)
- Evaluation boundary is distinctly narrower than ASTM system boundary
Module
- Glass-surface reflection losses
System
- Rear-side structural shading, mismatch, reflection, and spectrum
- Front-side near shading, mismatch, reflection, and spectrum
- PVsyst does not currently model spectral impacts on rear irradiance at all, but the correct adjustments are still a research topic

Evaluating the power plant as if it were monofacial is quite undesirable from an investor point of view, but disentangling the contributions of site albedo, terrain, and cloud conditions from the performance of the equipment is non-trivial
As it is not feasible to disable the bifacial power conversion, this typically leads to high apparent performance… only excessively-large shortfalls will be identified if a monofacial standard of evaluation is used.
Some developers have been forced to agree to evaluate based only on monofacial expectation and test procedures by EPCs
- Developer carries risk of bifacial underperformance in investor energy estimates
In some cases, expected rear-side power contribution may be quite low, e.g. when bifaciality is low, or when GCR is quite high
- Bifacial contribution may be within the uncertainty of a monofacial test

This module-based equation:
- \(E_{net} = \left(1 + \varphi \cdot \frac{G_{RPOA,eff}}{G_{POA,eff}}\right) \cdot G_{POA,eff}\)
tempts some practitioners to estimate a “constant” bifacial irradiance factor
- \(BIF = \left(1 + \varphi \cdot \frac{G_{RPOA,eff}}{G_{POA,eff}}\right)\)
and modify all readings from the front-side \(G_{POA}\) with it, but this value is very much not a constant… \(G_{RPOA,eff}\) is quite uncorrelated with \(G_{POA,eff}\)
Results from this approach can be sensitive to short-term variations in \(BIF\).

One approach is to create a space between two modules and mount one or more rear-facing sensors in the plane of the module laminate.
- This is usually associated with the sensor field-of-view being near the module frames or support purlins, creating a bias toward low irradiance relative to the average rear-surface irradiance.
- NREL has recommended that a pair of sensors be mounted one halfway between the torque tube and the east edge, and the other halfway between the torque dube and the west edge, and average these. However, this does not capture the lengthwise variation in irradiance.
- It also fails to demonstrate low capacity when purlins or torque-tubes are obstructing excessively relative to the model (too large).
All sensors exposed directly to the shade under the modules are also exposed to the non-uniform time-and-spatial varying irradiance described by McIntosh ¹

A variation on the in-plane measurement approach is to use a calibrated bifacial module to measure a combined front+rear irradiance.
- This approach assumes the module bifaciality is the same, but bifaciality can easily vary by +/-10% between modules.
- Like the in-plane measurements, this also is blind to structural shading deviations and irradiance positional variations.

In order to address the structural shading blind spot, the rear sensors could be moved backward to clear the frames/purlins, but the torque tube would still be a factor.
Some displaced sensors are mounted on the bottom of the torque tube, but this position underestimates the average irradiance available to the rear side and is still subject to moving speckled illumination.

To avoid trying to measure \(RPOA_{avg}\), measure in a more stable location \(RPOA_{outboard}\):
- along the torque-tube axis, roughly 0.5-1m extended from the south edge of the row
- South end is dominated by un-shaded ground
- Measurement is biased compared to \(RPOA_{avg}\), but unaffected by structural shading issues or moving shade patterns
- Regression will correct for bias
- Measurement is straightforward, modeling is a math exercise not covered here
Proposed: \(P = E_{tot} \cdot (a_1 + a_{2a} \cdot E_{POA} + a_{2b} \cdot E_{RPOA,outboard} + a_3 \cdot T_a + a_4 \cdot v)\)
where:
- \(E_{tot} = E_{POA} + E_{RPOA,outboard}\)

	estimate	std.error	statistic	p.value
term
(Intercept)	0.086425	0.019896	4.343906	1.536331e-05
row7Gfront	0.006268	0.000024	256.971475	0.000000e+00
row7Grear	0.005031	0.000131	38.495146	1.884400e-202

	estimate	std.error	statistic	p.value
term
(Intercept)	0.224668	0.020826	10.788037	8.499398e-26
row7RotatingAlbedometer_CM11_Up	0.006009	0.000027	224.415548	0.000000e+00
row7RotatingAlbedometer_CM11_Down	0.001014	0.000097	10.429101	2.722766e-24

Residuals are higher in this outboard data
Down instrument sees more irradiance than the rear side, so coefficient is smaller
Instrumentation is on a different row (7) than available dc power (2)

Many proposed methods move the evaluation boundary, shifting risks
Few proposed methods address ground albedo variation at test time
Proposed outboard method
- Keeps more traditional evaluation boundaries, less shifting of risk
- Requires an auxiliary estimate of outboard sensor readings in model (not discussed here)
- Preliminary data suggests larger residuals than inboard measurements