The microgrid is a promote solution for renewable energy systems integration in the grid; it can reduce the penetration limits in the electrical grid due to the intermittency nature of such systems like photovoltaic (PV) power. The ability to forecast the PV power availability for a short-term horizon affects the reliability of the microgrid. Proper forecasting significantly helps to improve the operation of the microgrid. In this work, short-term PV power generation forecasting using long short-term memory (LSTM) recurrent neural network is developed. The impact of changing the LSTM model on forecasting performance was studied. Analysis of forecasting results indicates that when the variation of PV power is higher, it is more complex to deal with the nonlinearity of the forecasting model to provide accurate forecasting.