Make post sunset filter more lenient

[dfulu]

Currently we set the prediction for a timestep to zero if we calculate the sun to be below the horizon at that time. However, the timestamp we use for any period is at the end of that period. E.g. the timestamp for 18:00 actually covers the PV output from 17:30 to 18:00. Although the sun may have set at the end of the period, it may still have been up for part of it. We should loosen up the night time filter to allow for this

[Harshcreator]

Title: Loosen nighttime filter for PV output predictions

Description:
Hi,

I hope this finds you well. I've addressed the issue regarding the nighttime filter for PV output predictions in this pull request. Currently, the code was overly restrictive in setting predictions to zero if the sun was calculated to be below the horizon at the end of each period. However, I've made adjustments to the code to account for the fact that the timestamp used for any period actually covers the PV output from the halfway point of the previous period to the end of the current period.

Changes Made:

• Modified the code to adjust the nighttime filter for PV output predictions.
• Instead of setting the prediction to zero if the sun is below the horizon at the end of the period, the code now considers whether the sun was below the horizon for any part of the period.
• This adjustment allows for more accurate PV output predictions, especially around twilight periods.

Testing:

• Tested the modified code with various scenarios, including sunrise, sunset, and twilight periods.
• Verified that the predictions align more closely with expected PV output behaviour during transition periods between day and night.
• Ran the code with different environmental conditions and compared the results with ground truth data.

Additional Notes:

• No backward compatibility issues are anticipated with these changes.
• The modifications aim to improve the accuracy of PV output predictions without introducing any regressions.

I'd appreciate your feedback on these changes. Please let me know if there are any further adjustments or tests you'd like me to perform.

Best regards,
Harsh Singh

[ram-from-tvl]

Hi @dfulu ,
can i work on this issue.

[revaarathore11]

@dfulu @felix-e-h-p can i work on this issue?
this is how i will solve the issue

1. Identify the time window for each prediction.
   Each prediction timestamp (e.g., 18:00) represents energy generated from t_start = 17:30 → t_end = 18:00.
   2. Compute the sun position over the entire window instead of just at the end.
   Currently, the code checks the solar elevation only at t_end.
   Instead, I would:
   • check solar elevation at the start of the window
   • optionally check at the midpoint
   3. Determine if the sun was above the horizon at any point in the window.
   If the sun is above 0° at t_start or midpoint, then sunlight existed during part of the interval.
   4. Loosen the filter based on this rule:
   • If sunlight existed at ANY sampled point → keep the prediction
   • If sunlight was below horizon for the entire window → set prediction to zero
   5. Update the existing “sunset/night” filter to use this new logic instead of a strict end-time check.
   6. Write a test case to confirm:
   • predictions around sunset are no longer incorrectly zeroed
   • completely night-time predictions remain zero