Abstract: Under the background of the high percentage renewable energy access and the rapid development of multiple energy coupling networks, it is difficult for a single model-driven methodology to satisfy the speed requirements of the real-time optimal scheduling decisions for the integrated regional energy systems. Therefore, it is significant to study the scheduling decision methods with high intelligence and fast decision-making capabilities. In this paper, a data- and model-driven method for a bi-level optimal scheduling of the regional integrated energy systems is proposed. On the upper layer, the mixed integer linear programming (MILP) is used to obtain a day-ahead scheduling plan which provides reference values for the within-day rolling optimization. The lower layer combines the Convolutional Neural Network (CNN) with the Gated Recurrent Unit (GRU) for the within-day rolling optimization decision making. Using the adaptive power correction model it finely tunes its outputs to obtain the accurate solution. Finally, the effectiveness of the method proposed in this paper is demonstrated by means of example analysis.