Question 1

How is the expected power computed?

Accepted Answer

Expected power for each string uses P_expected = P_stc × (irradiance/1000) × (1 + tempCoeff × (panelTemp − 25)), where P_stc is the rated string power at standard test conditions, irradiance is the simulated sensor reading in W/m², panelTemp is derived from ambient temperature plus irradiance-driven cell heating, and tempCoeff is a negative coefficient (about −0.4%/°C) capturing why panels lose output as they get hot.

Question 2

How does the anomaly detector decide a string is faulted?

Accepted Answer

Each simulated timestep the residual (actual minus expected power) is fed into a rolling exponentially-weighted mean and variance for that string. The residual is converted into a z-score, and once the z-score drops below the negative anomaly threshold for long enough the string is flagged. This is genuine statistical process control, not a scripted trigger — it responds directly to whatever fault is injected.

Question 3

What is the difference between shading, soiling and degradation faults?

Accepted Answer

Partial shading derates a fraction of a string's irradiance instantly and non-linearly because series-connected cells are limited by the weakest sub-string. Soiling is a slow multiplicative dust build-up that accumulates day by day. Degradation is a gradual efficiency decline simulating accelerated cell ageing, again compounding day over day. All three reduce actual output relative to the expected-power model, but with distinct time signatures.

☀️ Solar Panel Fault Detector — Power-Curve Anomaly Live

What did you find?