Significant uncertainty is introduced into the power system by the intermittent and volatile nature of new energy power generation, significantly raising the risk of the power system operational safely and steadily. To show a more precise and understandable Photovoltaic (PV) power output interval and to lessen the influence of PV power generation uncertainty on the electricity grid. Therefore. This research presents a deep learning and similar day clustering based short-term PV power interval prediction approach. Firstly, similar weather is clustered using the improved OPTICS algorithm to get the dataset under sunny and rainy day conditions. Then a Bayesian neural network is constructed by adding dropout layer based on LSTM to build a PV output prediction model for short-term PV output interval prediction. Finally, the actual PV dataset from the Australian Solar Energy Center is used for example analysis to confirm the suggested method’s dependability.