Power-only estimation of tilt and azimuth angles in distributed photovoltaic systems

Distributed photovoltaic (PV) systems increasingly occupy mountainous terrain, where modules at one site often differ in tilt and azimuth. Such installation-angle differences can be misidentified as faults and trigger false alarms. With module-level plane-of-array irradiance and back-of-module temperature sensors rarely installed, this paper estimates both angles from the power time series alone. Clear-sky intervals are detected from the power signal, and a variance-of-ratio objective between modeled and measured power is introduced. Since the ratio absorbs slowly varying losses such as soiling and degradation, the fit follows temporal profile shape, not amplitude. The objective is minimized by an improved sparrow search algorithm (SSA). The method is validated on a rooftop dataset from Nanjing, China, and the National Renewable Energy Laboratory dataset, covering three climates, six PV module technologies, sampling intervals from one to fifteen minutes, and observation horizons from hours to years. It estimates tilt and azimuth more accurately than PVWatts 5, PV-Peak, and Solar Data Tools, at comparable runtime. An ablation study attributes most of this accuracy to the clear-sky filtering and the variance-of-ratio objective, while the improved SSA mainly speeds convergence and stabilizes repeated runs. The estimates are insensitive to temperature-model error: a ±20 °C input bias shifts the estimated tilt by at most 0.15°. A Fisher–Cramér–Rao lower bound analysis quantifies window-level identifiability and guides window selection, showing that longer clear-sky windows spanning solar noon are the most informative. This offers a low-cost route to improved monitoring and forecasting for distributed PV plants.