Abstract: The heterogeneity of time series data in different scenarios significantly impacts the generalization ability and effectiveness of time series forecasting algorithms in intelligent decision-making, posing a major obstacle to their application. Large time series forecasting models are essential techniques to address this challenge. The latest research trends in the field of time series forecasting are integrated and four im-plementation approaches of large time series forecasting models are explored from a modal perspective from top to bottom: prompt-based methods, fine-tuning-based methods, alignment-based methods, and basic models for time series forecasting. Additionally, the core elements and available technologies in the construction process of large time series forecasting models are sorted out  Furthermore, the important challenges and research directions in the future are explored.