Nowadays, the rapidly growing energy consumption in the world along with the shortage of fossil fuels and their high environmental pollution has led to increased attention to renewable energies such as solar, wind, and geothermal energies. Among these, solar energy has many advantages such as no ecological and noise pollution and free use. However, photovoltaic power plant output, due to its dependency on solar irradiance and other weather conditions, has uncontrollable uncertainty. Therefore, for providing high-quality electric energy for end-consumers and enhancing the reliability of the system, photovoltaic power output needs to be predicted accurately. The aim of this paper is to address this issue by proposing three short-term photovoltaic power-forecasting models based on deep-learning neural networks, which differ in terms of input types and network structures. The proposed models use long short-term memory (LSTM) in their structures and historical power outputs and weather conditions as their inputs to forecast one-hour-ahead PV power. Conducted experiments show that employing weather conditions, in addition to the historical output powers, increases the prediction accuracy. Moreover, utilizing more complicated network structures leads to performance improvements.