Very short-term temperature forecaster using MLP and N-nearest stations for calculating key control parameters in solar photovoltaic generation

Although photovoltaic generation has been proposed as a solution for the world's energy challenges, it depends to a large extent on solar irradiation and air temperature. Therefore, small variations in these meteorological parameters produce sudden changes in power generation, which makes it difficult to integrate photovoltaic generators into the electrical grid. The aim of this study is to develop a very short-term temperature forecaster that makes photovoltaic generation more reliable in order to provide not only power but also ancillary services. To predict ambient temperature in a specific area (Vitoria-Gasteiz, Basque Country) in the next 10 min, this forecaster combines a multilayer perceptron and the optimal nearest number of meteorological. In addition, the distance and relative location between each station and the target station were taken into account. The accumulated deviation between actual and forecasted temperature was lower than 1% in 96.60% of the examined days from the validation database. Moreover, the root mean square error was 0.2557 °C, which represents an improvement of 13.20% as compared with the benchmark result. The results indicated that the forecaster can be considered for implementation in photovoltaic generators to compute key control parameters and improve their integration into the electrical grid.