Abstract: With the rapid development of the Internet of Things technology and industrial infrastructure, the digitalization and networking of manufacturing sites are continuously improving, providing technical foundation and guarantee for intelligent scheduling and control of production tasks in discrete manufacturing job shops. At present, the growing demand for multi-variety, small-batch, and customized products is making the workshop environment increasingly complex and variable, thereby amplifying the uncertainty of order delivery deadlines. However, the remaining completion time of orders is a key factor affecting delivery deadlines. Based on the strong data perception and acquisition ability of shop sites, an active scheduling method for flexible job shops considering order delivery deadlines is proposed. Firstly, a scheduling decision model based on an improved deep Q-network is established, where the predicted remaining completion time of orders is incorporated as one of the state features of the decision-making model to enhance the proactivity of scheduling. A composite scheduling rule action set is designed for the problem of job assignment to machines and job selection in machine buffers. The optimization objectives are to minimize the maximum completion time, total delay time, and average delay time. The decision-making model is trained to select the optimal action according to the real-time data through the prediction network and the target network, so as to realize active scheduling of the production process while ensuring multi-objective global optimization. Finally, the effectiveness and superiority of the proposed scheduling method are verified by application.