Photovoltaic (PV) power forecasting is important for the safety and economic efficiency of power system operation. Existing PV power forecasting studies usually focus on a single PV plant, ignoring the spatial and temporal coupling between neighboring PV plants. This may limit the improvement of PV prediction accuracy. In this paper, a multi-task learning-based PV power forecasting framework is proposed, which exploits the coupling relationship between neighboring PV plants with the help of the information sharing mechanism of multi-task learning to improve the forecasting accuracy. In the proposed framework, long short-term memory (LSTM) network is used as the information sharing layer to fully exploit the temporal characteristics in PV power data. Finally, real PV power datasets are used for the validation of the algorithm, and the results show that the proposed multi-task learning based method outperforms the current state-of-the-art single-task forecasting methods.