Short-term photovoltaics power forecasting using Jordan recurrent neural network in Surabaya

Abstract

Photovoltaic (PV) is a renewable electric energy generator that utilizes solar energy. PV is very suitable to be developed in Surabaya, Indonesia. Because Indonesia is located around the equator which has 2 seasons, namely the rainy season and the dry season. The dry season in Indonesia occurs in April to September. The power generated by PV is highly dependent on temperature and solar radiation. Therefore, accurate forecasting of short-term PV power is important for system reliability and large-scale PV development to overcome the power generated by intermittent PV. This paper proposes the Jordan recurrent neural network (JRNN) to predict short-term PV power based on temperature and solar radiation. JRNN is the development of artificial neural networks (ANN) that have feedback at each output of each layer. The samples of temperature and solar radiation were obtained from April until September in Surabaya. From the results of the training simulation, the mean square error (MSE) and mean absolute percentage error (MAPE) values were obtained at 1.3311 and 34.8820, respectively. The results of testing simulation, MSE and MAPE values were obtained at 0.9858 and 1.3311, with a time of 4.591204. The forecasting has minimized significant errors and short processing times.