Difference between PVFactors and ModelChain effective irradiance

[pasquierjb]

When I calculate the effective irradiance with PVFactors and only consider the front surface, I get (slightly) different results than when calculating the effective irradiance with ModelChain.

Here is an example with a South Facing Fixed Mount with no losses.

I understand that the gcr has an influence on PVFactors calculation (and not on ModelChain)

How can I configure PVFactors to match the results of ModelChain ?

import pandas as pd
from pvlib import location
from pvlib import tracking
from pvlib.bifacial.pvfactors import pvfactors_timeseries
from pvlib import temperature
from pvlib import pvsystem
from pvlib.temperature import TEMPERATURE_MODEL_PARAMETERS as PARAMS
from pvlib import modelchain
import matplotlib.pyplot as plt
import warnings


# supressing shapely warnings that occur on import of pvfactors
warnings.filterwarnings(action='ignore', module='pvfactors')

# using Greensboro, NC for this example
lat, lon = 36.084, -79.817
tz = 'Etc/GMT+5'
times = pd.date_range('2021-06-21', '2021-06-22', freq='1T', tz=tz)

# create location object and get clearsky data
site_location = location.Location(lat, lon, tz=tz, name='Greensboro, NC')

cs = site_location.get_clearsky(times)

# get solar position data
solar_position = site_location.get_solarposition(times)

# example array geometry
pvrow_height = 1
pvrow_width = 4
pitch = 10
gcr = pvrow_width / pitch
surface_azimuth = 180 # south-facing array
axis_azimuth = surface_azimuth - 90
albedo = 0.25
surface_tilt = 25


## 1. Calculate the effective Irradiance with ModelChain
mount = pvsystem.FixedMount(surface_tilt=surface_tilt, surface_azimuth=surface_azimuth)
temp_model_parameters = PARAMS['sapm']['open_rack_glass_glass']
cec_modules = pvsystem.retrieve_sam('CECMod')
cec_module = cec_modules['Trina_Solar_TSM_300DEG5C_07_II_']
cec_inverters = pvsystem.retrieve_sam('cecinverter')
cec_inverter = cec_inverters['ABB__MICRO_0_25_I_OUTD_US_208__208V_']
array = pvsystem.Array(mount=mount,
                       module_parameters=cec_module,
                       temperature_model_parameters=temp_model_parameters)
system = pvsystem.PVSystem(arrays=[array],
                           inverter_parameters=cec_inverter)
mc = modelchain.ModelChain(system, site_location, spectral_model='no_loss', aoi_model='no_loss')
_ = mc.run_model(cs)

## 2. Calculate the effective Irradiance with PVFactors
irrad = pvfactors_timeseries(solar_azimuth=solar_position['azimuth'],
    solar_zenith=solar_position['apparent_zenith'],
    surface_azimuth=surface_azimuth,  # south-facing array
    surface_tilt=surface_tilt,
    axis_azimuth = axis_azimuth, 
    timestamps=times,
    dni=cs['dni'],
    dhi=cs['dhi'],
    gcr=gcr,
    pvrow_height=pvrow_height,
    pvrow_width=pvrow_width,
    albedo=albedo,
    n_pvrows=3,
    index_observed_pvrow=1)
irrad = pd.concat(irrad, axis=1)


#Compare both Effective Irradiance
effec_irr_MC = mc.results.effective_irradiance.sum()
effec_irr_PVFactors = irrad['total_inc_front'].sum() #We consider the Incident irradiance not to include AOI losses

print("MC: ", effec_irr_MC, "PVFactors: ", effec_irr_PVFactors)
>> MC:  437953.13574622635 PVFactors:  441780.10564860125

SImilarly with a Single Axis Tracker facing East to West

import pandas as pd
from pvlib import location
from pvlib import tracking
from pvlib.bifacial.pvfactors import pvfactors_timeseries
from pvlib import temperature
from pvlib import pvsystem
from pvlib.temperature import TEMPERATURE_MODEL_PARAMETERS as PARAMS
from pvlib import modelchain
import matplotlib.pyplot as plt
import warnings


# supressing shapely warnings that occur on import of pvfactors
warnings.filterwarnings(action='ignore', module='pvfactors')

# using Greensboro, NC for this example
lat, lon = 36.084, -79.817
tz = 'Etc/GMT+5'
times = pd.date_range('2021-06-21', '2021-06-22', freq='1T', tz=tz)

# create location object and get clearsky data
site_location = location.Location(lat, lon, tz=tz, name='Greensboro, NC')

cs = site_location.get_clearsky(times)

# get solar position data
solar_position = site_location.get_solarposition(times)

# example array geometry
pvrow_height = 5 # height must be higher than width for trackers
pvrow_width = 4
pitch = 10
gcr = pvrow_width / pitch
surface_azimuth = 90 # Facing East/West
axis_azimuth = surface_azimuth - 90
albedo = 0.25
surface_tilt = 25

# load solar position and tracker orientation for use in pvsystem object
mount = pvsystem.SingleAxisTrackerMount(axis_azimuth=axis_azimuth,
                                            backtrack=False)

# created for use in pvfactors timeseries
orientation = mount.get_orientation(solar_position['apparent_zenith'],
                                        solar_position['azimuth'])


## 1. Calculate the effective Irradiance with ModelChain
temp_model_parameters = PARAMS['sapm']['open_rack_glass_glass']
cec_modules = pvsystem.retrieve_sam('CECMod')
cec_module = cec_modules['Trina_Solar_TSM_300DEG5C_07_II_']
cec_inverters = pvsystem.retrieve_sam('cecinverter')
cec_inverter = cec_inverters['ABB__MICRO_0_25_I_OUTD_US_208__208V_']
array = pvsystem.Array(mount=mount,
                       module_parameters=cec_module,
                       temperature_model_parameters=temp_model_parameters)
system = pvsystem.PVSystem(arrays=[array],
                           inverter_parameters=cec_inverter)
mc = modelchain.ModelChain(system, site_location, spectral_model='no_loss', aoi_model='no_loss')
_ = mc.run_model(cs)

## 2. Calculate the effective Irradiance with PVFactors
irrad = pvfactors_timeseries(solar_azimuth=solar_position['azimuth'],
    solar_zenith=solar_position['apparent_zenith'],
    surface_azimuth=orientation['surface_azimuth'],  # south-facing array 
    surface_tilt=orientation['surface_tilt'],
    axis_azimuth = axis_azimuth,  
    timestamps=times,
    dni=cs['dni'],
    dhi=cs['dhi'],
    gcr=gcr,
    pvrow_height=pvrow_height,
    pvrow_width=pvrow_width,
    albedo=albedo,
    n_pvrows=3,
    index_observed_pvrow=1)
irrad = pd.concat(irrad, axis=1)


#Compare both Effective Irradiance
effec_irr_MC = mc.results.effective_irradiance.sum()
effec_irr_PVFactors = irrad['total_inc_front'].sum() #We consider the Incident irradiance not to include AOI losses

print("MC: ", effec_irr_MC, "PVFactors: ", effec_irr_PVFactors)

[kandersolar]

How can I configure PVFactors to match the results of ModelChain ?

Can you say a bit about why you'd want to do this? If you want the results from ModelChain, why not just use ModelChain (or pvlib.irradiance.get_total_irradiance)? And at some point the difference in results between pvfactors and ModelChain will be due to pvfactors' increased level of simulation detail, so configuring it to match ModelChain seems like a step backwards anyway.

There are two main sources of difference in your results:

• ModelChain defaults to the Hay-Davies transposition model (try print(mc) to see what yours is using), while pvfactors uses the Perez model. Specify transposition_model='perez' in the ModelChain constructor to use Perez in ModelChain as well.
• pvfactors considers the effect of neighboring rows (ground shading, row-row reflection, diffuse sky shading) while ModelChain simulates an isolated row. Specify n_pvrows=1, index_observed_pvrow=0 in the pvfactors call to simulate an isolated row in pvfactors.

When I reconfigure the fixed tilt script to account for those two differences, the difference in summed irradiance goes nearly to zero (<0.04%). But again it's not clear to me why you would want to do this; pvfactors will be slower for the same answer.

P.S., "effective irradiance" usually includes AOI loss (and other effects like soiling and spectrum adjustment); what is being compared here is the incident irradiance aka POA irradiance or GTI.

[pasquierjb]

Thanks @kanderso-nrel.

In fact, when isolating the row and specifying the same transposition model, I get very close results. On the Single-Axis Tracker, I get 0.1% difference with an albedo=0 (that I also specify in my pvsystem.Array) and 0.7% with an albedo=1. The ground diffuse is probably considered differently in both tools.

Agreed that I should have rather used mc.results.total_irrad['poa_gloal'] than effective_irradiance to compare POA irradiances.

I want to compare the power ouput of results calculated with PVLIB ModelChain with a monofacial panel with the output of the same array adding bifaciality. I agree that it is probably simpler to re-calculate everything with PVFactors but out of curiosity I wanted to understand the level of differences between the 2 modeling approaches.

Side question, on a field with important gcr, is it good practice to add the self-shading diffuse losses calculated with pvlib.shading.sky_diffuse_passias to my Model Chain losses ?

[cwhanse]

is it good practice to add the self-shading diffuse losses calculated with pvlib.shading.sky_diffuse_passias to my Model Chain losses ?

If you are comparing between the monofacial and bifacial irradiances as described above, then I would say yes, the monofacial simulation should account for self-shading of diffuse (because the bifacial calculation in PVFactors is accounting for that effect).