The energy produced by photovoltaic farms has a variable nature depending on astronomical and meteorological factors. The former are the solar elevation and the solar azimuth, which are easily predictable without any uncertainty. The amount of liquid water met by the solar radiation within the troposphere is the main meteorological factor influencing the solar power production, as a fraction of short wave solar radiation is reflected by the water particles and cannot reach the earth surface. The total cloud cover is a meteorological variable often used to indicate the presence of liquid water in the troposphere and has a limited predictability, which is also reflected on the global horizontal irradiance and, as a consequence, on solar photovoltaic power prediction. This lack of predictability makes the solar energy integration into the grid challenging. A cost-effective utilization of solar energy over a grid strongly depends on the accuracy and reliability of the power forecasts available to the Transmission System Operators (TSOs). Furthermore, several countries have in place legislation requiring solar power producers to pay penalties proportional to the errors of day-ahead energy forecasts, which makes the accuracy of such predictions a determining factor for producers to reduce their economic losses. Probabilistic predictions can provide accurate deterministic forecasts along with a quantification of their uncertainty, as well as a reliable estimate of the probability to overcome a certain production threshold. In this paper we propose the application of an analog ensemble (AnEn) method to generate probabilistic solar power forecasts (SPF). The AnEn is based on an historical set of deterministic numerical weather prediction (NWP) model forecasts and observations of the solar power. For each forecast lead time and location, the ensemble prediction of solar power is constituted by a set of past production data. These measurements are those concurrent to past deterministic NWP forecasts for the same lead time and location, chosen based on their similarity to the current forecast and, in the current application, are represented by the one-hour average produced solar power. The AnEn performance for SPF is compared to a quantile regression (QR) technique and a persistence ensemble (PeEn) over three solar farms in Italy spanning different climatic conditions. The QR is a state-of-the-science method for probabilistic predictions that, similarly to AnEn, is based on a historical data set. The PeEn is a persistence model for probabilistic predictions, where the most recent 20 power measurements available at the same lead-time are used to form an ensemble. The performance assessment has been carried out evaluating important attributes of a probabilistic system such as statistical consistency, reliability, resolution and skill. The AnEn performs as well as QR for common events, by providing predictions with similar reliability, resolution and sharpness, while it exhibits more skill for rare events and during hours with a low solar elevation. (C) 2015 Elsevier Ltd. All rights reserved.
