Abstract: The two-way empowerment of industrial Internet and artificial intelligence has been elevated to a national strategy, which will inevitably drive the generational evolution of industrial Internet and accelerate the realization of the vision of intelligent interconnection of everything. To this end, we propose the concept of industrial intelligent Internet, establishes an integrated collaborative architecture of “sensing-communication-computing-control-intelligence”, and systematically elaborates the core roles, technical connotations, and cross-domain coordination mechanisms of these five functional domains. On this basis, we analyze and formulate the fundamental mathematical model for “sensing-communication-computing-control-intelligence” coordination, where the objectives and constraints of the collaborative scheduling problem are characterized and typical problem models are given. Then, we review intelligent optimization scheduling methods based on single-agent and multi-agent deep reinforcement learning, with a detailed comparison of related works. Furthermore, we explore the large-model-driven paradigm for intelligent optimization scheduling, review the latest research progress in large-model-assisted reinforcement learning and specialized networking large models, and discuss the potential application paths of large models in intelligent optimization scheduling. Finally, we analyze the key technical challenges and future research directions from five aspects: multimodal fusion sensing, end-to-end deterministic communication, cloud-edge-end collaborative computing, universal virtualized control, and large-model-based intelligent decision-making.