To solve the challenges of accommodating large-scale renewable energy integration into the power grid, a day-ahead dispatching strategy utilizing flexible resources on both source-side and load-side is proposed. Firstly, the flexible regulation capabilities of deep peak regulation, shiftable loads and reducible loads are analyzed, and a power system dispatching model includes these flexible resources is established. Subsequently, a day-ahead scheduling method based on convolutional sequence to sequence model is proposed. Load forecasting data and other relevant information are extracted through the encoder using multi-layer Convolutional Neural Networks, which enhances the capability and speed of information extraction in deep learning network. The decoder employing a Gated Recurrent Unit decodes the information extracted by the encoder and outputs the scheduling plan. Finally, auxiliary decision correction is applied to refine the outputted dispatching plan to ensure security. The effectiveness and correctness of the proposed method are verified using an improved IEEE 39-bus system.