Original paper

Urban solar irradiance and power prediction from nearby stations

Chen, Zihao; Troccoli, Alberto

Meteorologische Zeitschrift Vol. 26 No. 3 (2017), p. 277 - 290

21 references

published: Jun 14, 2017
published online: Oct 27, 2016
manuscript accepted: Feb 11, 2016
manuscript revision received: Jan 13, 2016
manuscript revision requested: Nov 24, 2015
manuscript received: Jul 15, 2015

DOI: 10.1127/metz/2016/0725

BibTeX file

O

Open Access (paper can be downloaded for free)

Download paper for free

Abstract

With the proliferation of small-scale solar PV installations, global horizontal irradiance (GHI) and power predictions are becoming critical elements in the integration of PV generation into the grid. This paper considers short-term prediction, from 5 minutes to a few hours, based on historical meteorological measurement data from weather and power monitoring stations located in the Canberra (Australia) region. The specific objective of this study is to produce skilful forecasts for (a generic) target station using a minimal amount of observations from nearby stations. Thus, although a large number of weather and power variables are collected and used for developing and testing the prediction algorithms, the ultimate aim is to rely on a few predictors, mainly meteorologically based. This will allow the identification of critical instruments which would need to be installed in order to provide satisfactory PV power predictions while limiting capital and operating costs of monitoring. Relative mean absolute error (rMAE) is used here to indicate prediction performance. Three statistical methods are tested for two different seasons, a winter and a summer. The relative importance of predictors and stations is assessed. A conversion from GHI to global irradiance on tilted surfaces, by means of simple geometry arguments and notion of irradiance components at a nearby site, is also introduced and tested. Finally, the prediction accuracy is categorised according to different clear-sky indices. Results show that when the clear-sky index exceeds 0.9 (near-to-cloudless conditions), the prediction performance is distinctly better than at lower clear sky indices, by at least 0.05 and 0.2 in terms of rMAE in summer and winter, respectively.

Keywords

Solar irradiance forecastingsolar power forecastingstatistical methodsmeteorological and power observationsrooftop photovoltaic