The smart manufacturing system employs a large number of advanced information technologies, which makes it possible to collect real-time data in production systems. The wide application of various types of information technology in the manufacturing process has enabled the manufacturing system to accumulate a large amount of data relating to production scheduling. Therefore, the historical production scheduling data and the real-time production data collected by smart equipments are used to establish a data-driven production scheduling method. Focusing on real-time hybrid flow shop scheduling problems, a real-time data-driven scheduling method based on the BP neural network is proposed. Firstly, the sample data for scheduling knowledge mining is extracted from the historical optimal and near-optimal scheduling scenarios. Through the BP neural network, the mapping relationship network between the production system state and the dispatching rules is obtained, which is then applied to production online real-time scheduling. Finally, numerical experiments verify that the proposed method outperforms the fixed single dispatching rule, and is stable under different scheduling objectives.