Reducing source-load uncertainty while balancing economic efficiency and low carbon emissions has become the focus of optimizing the scheduling of park-level integrated energy systems (PIES). To this end, an integrated framework of prediction, regulation, and decision-making is proposed. Firstly, a PIES incorporating combined heat and power (CHP), power to gas (P2G), and carbon capture and storage (CCS) is constructed. Secondly, a data prediction method based on the rime algorithm optimized convolutional neural network-support vector machine (RIME-CNN-SVM) is proposed, and the information gap decision theory (IGDT) is used to account for severe source-load uncertainties with unknown probability distribution. Finally, a low-carbon optimization scheduling strategy for PIES is established, considering source-load uncertainties, a tiered carbon trading mechanism, and penalties for abandoning wind and solar power. Through numerical analysis, the rationality and effectiveness of the proposed model are verified, demonstrating that the proposed method improves the accuracy of PIES scheduling while balancing economic efficiency and low-carbon emissions.