Problem about user-defined time-dependent tilt

[smallduan]

I want to input custom tilt and azimuth values, so I attempted to read this data from a CSV file. I used part of the code from Tutorial 5 in PV-tutorials.

tilt = tracker_data['surface_tilt'].fillna(0)
tilt.to_csv('tilt.csv', header=True)
data2 = pd.read_csv('tilt.csv', index_col=0).squeeze()
assert (data2.index == tilt.index).all()
azimuth = tracker_data['surface_azimuth'].fillna(0)
df_poa_tracker = pvlib.irradiance.get_total_irradiance(
                                                        surface_tilt=data2,
                                                       surface_azimuth=azimuth,
                                                       dni=df_tmy['DNI'],
                                                       ghi=df_tmy['GHI'],
                                                       dhi=df_tmy['DHI'],
                                                       solar_zenith=solar_position['apparent_zenith'],
                                                       solar_azimuth=solar_position['azimuth'])

The calculation failed to produce correct results. How should I properly implement tilt/azimuth inputs from external files like CSV?

[kandersolar]

Notice how df_poa_tracker has 17520 elements per column. I bet that means there is something going wrong with the index of one of the timeseries variables, and the result is the combination of two incompatible indexes (17520 = 8760*2).

One thing to be sure of is that the dtype of all indexes is a datetime, and not a string that looks like a datetime. You might need to be using parse_dates=True in the call to pd.read_csv, for example.

[smallduan]

Dear @kandersolar

Thank you very much. It was indeed an issue with the index. I was able to make the calculation work properly by directly replacing the index of data2 with the original tilt's index. However, I still haven't found out why the indexes were different in the first place.

tilt = tracker_data['surface_tilt'].fillna(0)
tilt.to_csv('tilt.csv', header=True)
data2 = pd.read_csv('tilt.csv', index_col=0).squeeze()
data2.index = tilt.index
azimuth = tracker_data['surface_azimuth'].fillna(0)

df_poa_tracker = pvlib.irradiance.get_total_irradiance(surface_tilt=data2,
                                                       surface_azimuth=azimuth,
                                                       dni=df_tmy['DNI'],
                                                       ghi=df_tmy['GHI'],
                                                       dhi=df_tmy['DHI'],
                                                       solar_zenith=solar_position['apparent_zenith'],
                                                       solar_azimuth=solar_position['azimuth'])

[kandersolar]

Glad it's working now!

However, I still haven't found out why the indexes were different in the first place.

Did you check the datetime/string thing I mentioned? In my experience, df.to_csv() followed by pd.read_csv() will convert a datetime index into a string index (where the string values look identical to datetimes). Using parse_dates=True in the call to pd.read_csv would fix that.

If that's not the problem, you'll have to investigate yourself why the new index and the original index are different. Checking the value and type of .index[0] for the two dataframes is where I would start.

[smallduan]

Dear @kandersolar

Thank you very much! You are absolutely right. I didn't notice this issue earlier. Following your suggestion, the problem has now been fully resolved.