Abstract: The representation and modeling of complex system dynamics face multiple challenges, including intricate structural forms, highly diverse internal mechanisms, and fragmented prior knowledge. However, the rapid advancement of artificial intelligence technologies has provided transformative solutions to these difficulties. This article presents a systematic review of the latest progress in intelligent modeling techniques for complex system dynamics, with a particular focus on proposing and elaborating a unified representation framework aimed at overcoming modeling barriers across diverse complex scenarios. Centered around this framework, the article explores a range of intelligent modeling approaches, including stochastic skeletons, topological decoupling strategies, data-adaptive neural ordinary differential processes, stochastic skeleton operators for network dynamics, and foundational models based on symbolic regression for networks. Looking ahead, by integrating the unified representation framework with multimodal data and knowledge, and further combining it with control theory and practical tools for scientific discovery, intelligent models will be capable of achieving deeper, more precise understanding, prediction, and regulation of complex systems. This will not only provide robust support for scientific advancement but also demonstrate broad application potential in key areas such as epidemic control, online public opinion governance, and complexity research.