PV power forecast could mitigate the effects of high solar power injection into the Electricity Grid. Two main methods are currently used for PV power generation forecast. A deterministic approach that uses physical based models requiring detailed PV plant information. A data-driven approach based on statistical or stochastic machine learning techniques needing historical power measurements. The main goal of this work is to analyze the accuracy of these different approaches. Two original deterministic and a stochastic model for day ahead PV generation forecast were developed and a detailed error analysis was performed. Four years of site measurements were used to train and test the models. NWP data generated by the WRF models were used as input . Additionally a new parameter, called Clear Sky Performance index, is defined. It is the equivalent of the Clear Sky index for PV power generation forecast and it is here used to develop the stochastic model and an outperforming persistence model. The stochastic model not only was able to correct NWP bias errors, but also provides a better transposition. The deterministic approach leads to a skill score of 35% with respect to the persistence model, while the stochastic approach provides a skill score of 39%.