Optimization of Solar PV Power Forecasting Using Bootstrap Techniques and the Feed-Forward Neural Network Model

Authors

Zarate-Perez, Eliseo

Palumbo, Mariana

da Motta, Ana

Grados, Juan

Abstract

The outbreak of the COVID-19 disease has exerted a deep and extensive influence on the energy sector. The work modality and lifestyle caused by the confinement policy have increased electricity consumption in the residential sector. In such a way that the application of photovoltaic solar energy (PV) is rapidly evolving to mitigate the problems caused. However, due to the variability and uncertainty of solar irradiance, several technical challenges are created to produce PV energy. To reduce these adverse effects, forecasting of energy production at multiple scales is used. In this sense, the objective of this study is to determine the forecast performance of a hybrid model through the application of a Feed-Forward Neural Network (FFNN), together with the application of the moving block bootstrap technique (MBB), using the real data of the production of a PV system. The results show that the FFNN method combined with MBB techniques consistently outperform the original FFNN method in terms of forecast accuracy. That is, the original model presents a performance of 4.48% percentage forecast error (MAPE), compared to 3.14% for the proposed hybrid model. Finally, through the Ljung-Box test it is shown that the results are not correlated; therefore, the recommended model is validated.