Geolocating Photovoltaic Systems using Geometry and Weather Data Correlation

With the increasing deployments of residential photovoltaic systems, successful attacks on photovoltaic inverters pose a threat to the stability of the power grid. If attackers with remote control over many inverters want to target segments of a power grid, they require knowledge of their location. To enable the development of countermeasures, we analyze how precisely inverters can be localized. We present two approaches to determine the location of an inverter, requiring only historical power generation data from each photovoltaic system. The first approach estimates times of equal solar elevation over multiple days, which uniquely identify a location through some astronomic and geometric observations. The second approach uses the correlation of daily differences in power generation with changes in local weather by measuring similarity to publicly available irradiation data. Both approaches are evaluated against real generation data from 1.428 photovoltaic systems, obtained from \url{pvoutput.org}, in a diverse set of scenarios. The first approach achieves a median spatial error of $108\,\text{km}$, while the second attains a median error of only $6\,\text{km}$. Given the broad range and rather low quality of the input data used, we achieve an accuracy that significantly exceeds existing research, even outperforming most previous results that use much higher quality input data. Multiple limiting factors are also identified, leaving opportunities for future work to further improve both approaches. The algorithms described in the present work are open-source.