How Connected Cars can Improve Solar Forecasting - Expanding the Scale of Local Sensor Networks

Solar forecasting is an important research branch that aims to retain grid stability with the rapidly increasing share of fluctuating electricity from photovoltaics. While approaches based on sensor networks are prominent in targeting forecast horizons too short for image-based methods, they are not considered capable of capturing cloud dynamics in more extensive temporal and spatial scales. This work presents an approach that may provide both simultaneously: An expansive spatial scale with the spatial and temporal resolution to sense cloud shadow dynamics. The approach uses automotive light sensors to measure a correlated irradiance proxy. The spatial resolution and dynamics are made static by utilizing a “virtual” sensor grid, targeted by a 5-nearest neighbor inverse distance weighted interpolation. Subsequently, the cumulative mean absolute error method extracts cloud shadow movement direction and velocity from subsequent simulated clear-sky index snapshots. This first work investigating the novel approach assumes perfect sensor irradiance measurements to examine theoretical limitations. The results for an area of 600 m × 900 m indicate that, under this assumption, cloud shadow speeds could be detected with a root mean squared error (RMSE) of 2.4 m/s and directions with an RMSE of 9.5°. The results deteriorate but appear robust to some extent when reducing the percentage of cars sharing their sensor measurements.