The participation of flexible load in the optimal scheduling of new power system has a significant effect on improving the consumption capacity of new energy, but the potential of flexible load has not been fully explored. To solve this problem, this paper presents a day-ahead and intra-day optimal scheduling method based on source-load prediction. Firstly, the sparrow search algorithm is used to optimize the convolutional long-term and short-term memory neural network ( SSA-CNN-LSTM ) for day-ahead and intra-day power prediction of new energy and load. Secondly, according to the characteristics of flexible load and the flexibility of demand response, the load is divided into different types, such as shiftable, transferable and reducible load. The day-ahead and intra-day two-stage low-carbon environmental economic dispatch model of source-load interaction is constructed with the goal of optimizing the system operation cost and pollutant gas emission considering the staged carbon transaction cost. Finally, the improved multi-objective grey wolf algorithm (MOGWO) is used to solve the model. The example analysis shows that the total cost of the flexible load classification is reduced by 8.6 %, the pollutant emission is reduced by 4.1 %, and the new energy consumption capacity is increased by 4.2 % compared with the traditional method. The uncertainty of new energy and load response is significantly reduced in multiple time scales and the low-carbon environmental and economic comprehensive benefits of the new power system are improved.