Rapid growing of photovoltaic (PV) power puts forward higher requirements for PV power generation forecasting. However, the volatility and fluctuation of meteorological factors bring severe forecasting uncertainties. This paper proposes a hybrid approach integrating extreme learning machine-based quantile regression and hidden Markov model (HMEQR) for probabilistic forecasting of PV power to effectively quantify the forecasting uncertainties. The constructed hidden Markov model takes the numerical weather prediction (NWP) data as observation series and succeeds to extract the latent meteorological states for every moment, which benefits the conditional modeling of probabilistic forecasting. The extreme learning machine-based quantile regression (EQR) model is constructed under various weather conditions to produce conditional predictive quantiles for forecasting distribution description. Trained with efficient linear programming, the EQR model lowers the computational complexity and maintains satisfactory overall performances. Comprehensive case studies validate the effectiveness of the proposed HMEQR model compared with mature approaches.