Connecting to the API
Waiting for the first readings…
Next 7 days
Energy forecast for all panels, from the weather feed's sunshine and temperature. The gap between a solid bar and its dashed outline is what today's dirt costs.
What changed lately
One line each time a panel's verdict or wash decision changes.
How a verdict is made
- Read the panel: voltage, current, four corner light sensors, eight temperatures and the tank level.
- Turn that into features: the heat pattern against the panel's median, how unevenly the light falls, and power against light.
- Classify the panel as healthy, dusty, shaded, hot cell or cracked cell. A second model judges it again after a wash, which separates dirt from a real fault.
- Apply the rules: a shadow is never washed, no wash if rain is forecast, refill a low tank first, and wash only when the loss per day is more than the cost of a wash.
- Send the verdict to the API in one request. The panel's node does the washing.
About the accuracy. The classifier was trained and tested on data from the physics simulator in this project (about 99.8% on that data). It has not been measured on real panels yet, and real data will be used to retrain it.