Researchers recover rooftop PV tilt and azimuth from power data, tested on 17 Australian systems

An international research team has developed a method to determine the tilt and azimuth of rooftop PV arrays from operational power data, according to pv-magazine.com. pv-magazine.compv-magazine.compv-magazine.comOpen source ↗ The authors tested it on 17 commercial rooftop systems in Australia with independently verified orientations. pv-magazine.compv-magazine.compv-magazine.comOpen source ↗ Orientation data is needed to estimate electricity generation and detect faults, but installation records are often missing or incorrect, and site inspections are impractical for large PV portfolios.

Data and method

The researchers used operational data collected in 2025 from commercial rooftop PV systems in Australia. The dataset included DC power measurements recorded at five-minute intervals for each maximum power point tracker (MPPT). Clear-sky modeling and ERA5 weather data were used to estimate irradiance on panels at different tilt and azimuth angles. pv-magazine.compv-magazine.compv-magazine.comOpen source ↗

After selecting periods with high direct irradiance and enough morning and afternoon measurements, the team applied an optimization algorithm to find the orientation that best matched the measured power output. The algorithm also adjusted effective system capacity to account for power losses. The researchers compared two fitting metrics, R² and Huber loss. R² measures how well modeled power matches observed data, while Huber loss reduces the influence of outliers, such as deviations caused by shading. They also assessed the effects of shading, low tilt angles, data resolution, module temperature and inverter clipping.

Accuracy and limits

On unshaded systems the method recovered tilt with a median error of about 2° and azimuth with about 4°, using only DC power data and site coordinates, said corresponding author Hanrong Huang. pv-magazine.compv-magazine.compv-magazine.comOpen source ↗ Because it works on DC data, it can also find the orientation of each array separately at the MPPT level.

Clear-sky modeling consistently outperformed ERA5, which gave median tilt errors of 10° to 12° and azimuth errors of 48° to 62° for unshaded systems. pv-magazine.compv-magazine.compv-magazine.comOpen source ↗ Shading raised clear-sky azimuth errors from 4° to 35° with Huber loss and from 15° to 27° with R². At tilt angles below 5°, azimuth was difficult to determine because panel orientation has little effect on incident irradiance, which limits the impact of such errors on estimated generation.

The authors named partial shading as the main limitation. Their next steps are to reduce its effect on accuracy and to make the method robust to inverter clipping and tripping. The work was published in Solar Energy under the title "Recovering rooftop photovoltaic array tilt and azimuth from operational power data." The team included scientists from the Chinese Academy of Sciences, the University of New South Wales, Diagno Energy and Global Sustainable Energy Solutions.