Abstract: Time series record state changes in complex systems across transportation, energy, finance, healthcare, industry, and cloud services, providing a foundation for studying system dynamics. Traditional time series analysis focuses on modeling numerical patterns for domain-specific tasks such as forecasting, classification, and anomaly detection. As applications expand, demand is growing for cross-domain transfer, semantic interpretation, knowledge integration, contextual reasoning, and automated decision-making. Advances in foundation models, large language models, multimodal learning, and agent technologies are extending time series analysis from numerical modeling to intelligent analysis. Time series foundation models learn transferable temporal representations through large-scale pretraining, and large language models link numerical patterns with textual semantics. Multimodal and knowledge-enhanced methods integrate text, images, events, and domain knowledge to support contextual understanding. Agents organize closed-loop analytical workflows through task planning, tool use, and feedback-based verification. Organized around these technical directions, this survey systematically reviews the data forms, task taxonomy, key methods, evaluation resources, and industry applications of intelligent time series analysis driven by large models. It also examines current challenges and future research directions, providing a reference for developing intelligent time series analysis systems for complex real-world settings.