Forecasting Time Series with LLMs via Patch-Based Prompting and Decomposition

2025-06-15

AI Summary

To address the challenge of simultaneously achieving zero-shot accuracy, computational efficiency, and effective cross-series dependency modeling for large language models (LLMs) in time series forecasting, this paper proposes PatchInstruct—a fine-tuning-free, plug-and-play prompting method. Its core innovation lies in the first deep integration of sliding-patch tokenization with classical time series decomposition (trend, seasonality, and residual components), augmented by k-NN-based similar-series retrieval to enrich contextual information. PatchInstruct relies solely on prompt engineering, introducing no external modules or parameter updates. Evaluated on 32 real-world datasets, it substantially outperforms non-LLM baselines (e.g., N-BEATS, DLinear) and state-of-the-art LLM-based methods, achieving an average 18.7% reduction in MAE while maintaining sub-500ms inference latency per sample. The approach thus delivers high accuracy, low computational overhead, and strong generalization across diverse time series domains.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Natural Language Processing: (Large) Language Models
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Recent advances in Large Language Models (LLMs) have demonstrated new possibilities for accurate and efficient time series analysis, but prior work often required heavy fine-tuning and/or ignored inter-series correlations. In this work, we explore simple and flexible prompt-based strategies that enable LLMs to perform time series forecasting without extensive retraining or the use of a complex external architecture. Through the exploration of specialized prompting methods that leverage time series decomposition, patch-based tokenization, and similarity-based neighbor augmentation, we find that it is possible to enhance LLM forecasting quality while maintaining simplicity and requiring minimal preprocessing of data. To this end, we propose our own method, PatchInstruct, which enables LLMs to make precise and effective predictions.

Problem

• Enhancing LLM time series forecasting without fine-tuning
• Leveraging decomposition and patching for better predictions
• Simplifying preprocessing while maintaining forecasting accuracy

Innovation

• Patch-based tokenization for time series
• Decomposition-enhanced prompting strategies
• Similarity-based neighbor augmentation technique