Daily Solar Radiation Forecasting based on a Hybrid NARX-GRU Network in Dumaguete, Philippines

In recent years, solar radiation forecasting has become highly important worldwide as energy increases its contribution to electricity grids. However, due the intermittent nature of caused by meteorological parameters, errors arise, and fluctuations in power output photovoltaic (PV) systems a severe issue. This paper aims introduce hybrid model daily global time series. Meteorological data samples from Dumaguete, Philippines, are used assess accuracy proposed nonlinear autoregressive network with exogenous inputs (NARX) – gated recurrent unit (GRU) model. Four different models were trained using data, which Optimizable Gaussian Process Regression (GPR), Nonlinear Autoregressive Network (NAR), NARX, Hybrid NARX-GRU Network. Results show that root mean square error (RMSE) ~0.05 training 33 seconds. The better performance compared three obtained RMSE values 27.741, 39.82, 28.92, for GPR, NAR, respectively. simulation results demonstrate significantly outperforms regression single terms statistical metrics efficiency. Furthermore, this study shows hybridized is able provide an effective estimation radiation, operation PV plants country, specifically commitment purposes