Nowcasting of Solar Irradiance and Photovoltaic Production Using a Network of All-Sky Imagers

Kurzfassung

To mitigate the ongoing climate change, carbon-neutral sources of energy are required. Photovoltaics (PV) is seen as one of the most promising technologies in this regard. With the rapid extension of PV capacities and the increasing PV penetration in power grids, the intermittency of the solar resource can affect the stability of these grids. Forecasts of the solar resource for horizons from minutes to weeks ahead can help to plan the PV production and to compensate fluctuations of the same. In an urban distribution grid, this can create the challenge to forecast the production of a large number of distributed PV installations with a capacity of around 20 kW or less. So far, all-sky-imager- (ASI-) based nowcasts (i.e. very short-term forecasts) have a spatial resolution of down to 5 m × 5 m and can cover lead times 0-15 min ahead in 1-min steps. The setup of state-of-the-art nowcasting systems allows accurate cloud observations and nowcasts only in a limited area close to the ASIs. Further, the accuracy of these systems is only satisfying for certain applications. To cope with these shortcomings, I develop and investigate an ASI network, i.e. a nowcast based on numerous distributed ASIs and a combined evaluation of the individual ASIs’ data. Such a nowcast has not been investigated yet based on real camera images. The Eye2Sky ASI network provides the experimental basis for this work. The ASIs of Eye2Sky are combined in the ASI network for two main purposes. Around Oldenburg, a high density of 12 ASIs was placed in an area of 13 km × 12 km. This dense setup aims to reduce nowcast errors in particular at lead times up to 30 min. In the whole region of Eye2Sky, the ASI network uses up to 23 ASIs at distances of up to 60 km to achieve an increased spatial coverage compared to an ASI pair and to cover lead times of up to multiple hours. To achieve these goals, this thesis develops dedicated approaches to estimate cloud height using multiple ASI pairs, to model clouds based on an ASI image-based cloud classification and a subsequent accuracyweighted merger of the multiple ASIs observations, to measure diffuse irradiance by the ASIs, to estimate the cloud motion vector field based on the numerous ASIs and to track clouds along their trajectories through the cloud motion vector field. The ASI network’s nowcast of GHI is validated using three meteorological stations in Oldenburg. For the validation dataset of 92 days, at the temporal resolution of 1 min and spatial resolution of 50 m × 50 m, the ASI network brings a clear reduction of nowcast errors over state-of-the art nowcasts. E.g. at lead times of 5, 10 and 20 min, the ASI network provides an RMSD which is 23 − 30% (35 − 50 W/m2), 15 − 21% (25 − 36 W/m2) and 8 − 12% (14 − 21 W/m2) smaller than the RMSD of a persistence nowcast based on a single meteorological station and 21 − 25% (27 − 38 W/m2), 17 − 19% (27 − 31 W/m2) and 11 − 13% (21 − 24 W/m2) smaller than the RMSD of a state-of-the-art ASI pair, depending on the evaluated site. The ASI network and a PV model chain are applied to nowcast the production of distributed PV installations in Oldenburg. Also in this application, the ASI network provides a notably more accurate nowcast with an nRMSD up to 41% smaller compared to the reference nowcasts. Overall, these validations show that the ASI network with its extended setup and enhanced methodology can augment the scope of ASI-based nowcasts over the stateof-the-art: The spatial coverage, the achievable forecast horizon and the accuracy of the ASI-based nowcast are all increased notably.