Building flexible energy systems (BFES) can be enhanced by introducing storage batteries. Providing timely scheduling strategies for flexible resources can improve the system’s energy utilization. BFES’s scheduling strategies are often adjusted based on Intra-hour photovoltaic(PV) output. Intra-hour PV power generation can be predicted by analyzing cloud imagery data; however, this method does not meet the economic requirements of BFES due to its cost and instrumentation. Therefore, this study proposes a low-cost method for intra-hour PV power generation prediction (IHP) for BFES and explores the impact of integrating this approach into BFES on the rate of renewable energy consumption. This method combined low-quality sky images captured using fisheye cameras installed above buildings with historical electricity generation data and employed convolutional neural networks. The feasibility of the IHP method and the advantages of incorporating it into BFES were verified by applying it to a building equipped with PV devices in Changping, Beijing. The performance of the proposed model algorithm was compared with those of existing models. The proposed method achieved average prediction accuracy improvements of 25.1 and 12.5 % compared with existing models under sunny and cloudy conditions, respectively. Under clear conditions, the model could predict the PV power generation within the next 25 min, whereas under cloudy conditions, the model could predict the power generation within 10 min. In addition, integrating IHP into the scheduling strategy of BFES can improve the renewable energy consumption rate by 44.4 % on the original basis.