LangTime: A Language-Guided Unified Model for Time Series Forecasting with Proximal Policy Optimization

Recent research has shown an increasing interest in utilizing pre-trained large language models (LLMs) for a variety of time series applications. However, there are three main challenges when using LLMs as foundational models for time series forecasting: (1) Cross-domain generalization. (2) Cross-modality alignment. (3) Error accumulation in autoregressive frameworks. To address these challenges, we proposed LangTime, a language-guided unified model for time series forecasting that incorporates cross-domain pre-training with reinforcement learning-based fine-tuning. Specifically, LangTime constructs Temporal Comprehension Prompts (TCPs), which include dataset-wise and channel-wise instructions, to facilitate domain adaptation and condense time series into a single token, enabling LLMs to understand better and align temporal data. To improve autoregressive forecasting, we introduce TimePPO, a reinforcement learning-based fine-tuning algorithm. TimePPO mitigates error accumulation by leveraging a multidimensional rewards function tailored for time series and a repeat-based value estimation strategy. Extensive experiments demonstrate that LangTime achieves state-of-the-art cross-domain forecasting performance, while TimePPO fine-tuning effectively enhances the stability and accuracy of autoregressive forecasting.

From traffic flows to stock markets, our world runs on time-based data ,but forecasting these patterns remains challenging, especially across different domains like energy and retail.While powerful language models (like GPT) show promise, they struggle to understand numerical time series. Our solution, LangTime, bridges this gap by giving these models:（1) Data instruction manuals that explain unique characteristics of each dataset. (2) Domain guides highlighting differences between fields (e.g., healthcare vs. transportation).For long-term predictions where small errors can snowball, we adapted the AI training method behind ChatGPT (called PPO) specifically for time series. This acts as a coach providing continuous feedback to correct mistakes and improve prediction.Tested across energy, traffic and finance domains, LangTime outperforms existing methods. Our framework unlocks language AI's potential for real-world numerical forecasting challenges.