Fix bug in previous days gap filling, fix crash with tried_list

[springfall2008]

No description provided.

[Copilot]

Pull request overview

This PR fixes a bug in the historical load data gap filling logic where days beyond the available data age were incorrectly mapped to earlier days due to clamping in the use_days calculation.

Changes:

• Fixed the loop in previous_days_modal_filter to compute sums for all days up to the maximum requested, rather than clamping to load_minutes_age
• Changed use_days calculation from max(min(days, self.load_minutes_age), 1) to max(days, 1) to allow proper detection of days with no data
• Added calculation of average non-zero day consumption to use as fallback instead of hardcoded 24 kWh
• Added standalone test file to verify the bug fix
• Reset tried_list cache between optimization passes in plan.py

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

File	Description
apps/predbat/fetch.py	Core bug fix - Changed loop to iterate through all days and compute average non-zero consumption for gap filling
coverage/test_modal_filter.py	Added standalone test to verify the fix (references missing YAML file)
apps/predbat/plan.py	Reset optimization cache between passes (appears unrelated to gap filling)

[Copilot]

The loop range has two issues: (1) it starts from 0, but days=0 and days=1 both result in use_days=1 and full_days=0, causing redundant computation; (2) if days_list is empty, max(days_list) will raise ValueError. The loop should start from 1 and handle empty lists: for days in range(1, max(max(days_list, default=1), self.load_minutes_age) + 1). This ensures historical days are properly 1-indexed and handles edge cases safely.

[Copilot]

The loop iterates from 0, but days=0 and days=1 both result in use_days=1 and full_days=0, causing the same data to be computed twice. Historical days should be 1-indexed (day 1 = yesterday). The loop should start from 1 instead: for days in range(1, max(max(days_list), self.load_minutes_age) + 1). This will avoid redundant computation and ensure sum_all_days is calculated correctly.

[Copilot]

The comment says "Work out the highest non-zero day load" but the code actually computes the average of all non-zero days, not the highest value. The comment should be corrected to "Work out the average non-zero day load" to accurately describe what the code does.

[Copilot]

This change appears unrelated to the "previous days gap filling" bug fix described in the PR title. Resetting tried_list here clears the optimization cache between the initial optimization pass and the region-based optimization passes. While this may be a valid optimization improvement, bundling unrelated changes in a single PR makes it harder to review and understand the purpose of each change. Consider either documenting why this is related to gap filling, or moving it to a separate PR.

[Copilot]

The threshold of 2.5 kWh for determining zero days seems arbitrary and could be problematic for very low consumption households or edge cases. Consider making this threshold configurable or basing it on a percentage of the average consumption rather than a fixed value. For example, a household with naturally low consumption (e.g., 5 kWh/day) would still be counted, but a household with a partial day of 3 kWh due to data issues would also be counted.