In this paper we develop and verify a predictor-corrector method for a one-day-ahead photovoltaic array power production prediction. The most critical inputs to the prediction model are predictions of meteorological variables, such as solar irradiance components and the air temperature, which are the main sources of the power prediction uncertainty. Through a straightforward application of the weather forecast data sequence, photovoltaic array power production prediction is refreshed with the frequency of new forecasts generation by the meteorological service. We show that the prediction sequence quality can be significantly improved by using a neural-network-based corrector which takes into account near-history realizations of the prediction error. In this way it is possible to refresh the prediction sequence as soon as new local measurements become available. Except for predictions of meteorological variables, the prediction model itself is also a source of the prediction uncertainty, which is also taken into account by the proposed approach. The proposed predictor-corrector method is verified on real data over a 2-year time period. It is shown that the proposed approach can reduce the standard deviation of the power production prediction error up to 50%, but only for the first severalinstances of the prediction sequence (up to 6-8 h ahead) which are in turn the most relevant for real-time operation of predictive control systems that use the photovoltaic array power production prediction, like microgrid energy flows control or distribution network regulation. (C) 2016 Elsevier Ltd. All rights reserved.