New version 8.13.1 is released as stable: Power by the Hour | Homey
- Fix solar yesterday graph
- Re-order graphs
- Tweak solar forecaster low light hours
- Fix DAP milliseconds bug
App version 8.13.2 was just released as stable to the app store: Power by the Hour | Homey
- Added solar eclipse awareness to solar forecaster
When this app gets installed before sunset it will try to pull the moon partly or full (depending where you live) in front of the sun. It will only do that very occasionally by the way.
Impact of eclipse aware solar forecaster is visible after 19:30 (NL time)
Compared to solar forecast tomorrow (no eclipse just before sunset)
Dutch interview with Stefan (co-founder) about the future of Homey and energy management, and the role Power by the Hour plays in that:
Yes that is interesting. It also looks like in the afternoon it is the other way around. Looks like a time shift somehow.
Is your Homey (and thus the Solar lat/lon) set to the correct location?
Is your Homey set to the correct time zone?
You are using the latest version 8.13.2? (a timezone fix was implemented about a week ago)
When you do a retrain (from device settings), does the graph of yesterday/today improve?
Yes, this is how it is supposed to work! Thanks for your compliments 
That could be causing it. I was only able to optimize my algorithm on my own panel data, which face south. I sent you a PM. If you could provide the data for both your arrays I can run it through the algorithm and improve it for east-west situations I hope.
New version 8.13.3 is ready for testing: https://homey.app/a/com.gruijter.powerhour/test/
- Fix solar data interpolation and capping
Thanks to the raw data @Leo_Verwaal provided on his east-west oriented solar panels I was able to find a bug in data interpretation for the solar forecaster. This bug was most visible for solar arrays facing east or west, but south facing arrays also have a slightly better working forecast now.
Please test @Leo_Verwaal (do a retrain after installing), and share the new forecast image of yesterday and today.
Did you do a retrain on both solar forecasters? And did you try to fully close and reopen the Homey app after that?
Edit: found another location in the code where the same bug needs to be fixed. Working on that now.
Yes, I installed 8.13.3, retrained, re-started the app …
I made another fix (not released yet). This is the yesterday image I now get based on your raw data. I am still trying to figure out what is causing the over-estimation of the forecaster after 15:00. But any theory I have (e.g. coudlensing on cloudy days, or shadow popping in because of seasonal changes) fails to explain it.
The forecaster bases its estimation on the past 14 days. So in the past 2 weeks your east panels must actually have produced so much power in the afternoon, but not anymore.
Do you see anything that might cause the forecaster to overestimate the power in the afternoon?
New version 8.13.4 has been released to the app store: Power by the Hour | Homey
- Fix solar data interpolation and capping more
- Add era5 model high weight solar learning
This definitely fixes the solar forecaster due to data misinterpretation.
Secondly I added another weather model to better learn from data on days where the meteo forecast was very wrong (that happens sometimes 
).
Solar forecaster now should be improved, and is especially visible for east-west facing panels.
@Leo_Verwaal Your afternoon over-estimation for your east panels were caused by wrong meteo forecasts from the past 2 weeks (where clouds were predicted, but it was actually sunny). I added another meteo source that partly corrects this, so your graph should be better now.
New app version 8.14.0 was pushed as stable to the app store: Power by the Hour | Homey
- Reduce memory peaks
- Fix water meter bug
- Fix several potential racing issues
This reduces the chances of Homey killing the app due to memory usage peaks. Also it fixes several potential racing issues making the use of complex flows more reliable.
Version 8.15.0 is ready for testing: https://homey.app/a/com.gruijter.powerhour/test/
- Further reduce memory peaks
- Improve ROI strategy
- Tweak solar forecaster algorithm for cloudy days
- Extend solar forecast local weather models
- Add action flow ‘time_until_cheapest’
- Fix anomalous meter jumps at period boundaries
- Refactoring and code cleanup
Huge code changes with multiple fixes for extremely hard to find long standing issues. Furthermore the memory peak load has again been significantly reduced.
The solar forecaster now uses more local weather forecast models. This benefits worldwide regions, and especially US/Canada, Benelux, France, UK and the Nordic countries.
Thanks for moving the limits of this app, it’s really fantastic to observe where it goes, it’s almost magic 
What is the risk that some API used, will be blocked again ? (I’m asking because I saw previous restrictions on other prediction models and I’m now considering to switch actually to PbtH and it was also once the case of PbtH).
Also, seems I’m getting some peaks in the graphs, do I have to retrain the model to get rid of that ?
