Short-term nonlinear autoregressive photovoltaic power forecasting using statistical learning approaches and in-situ observations