Retrieval Augmented Time Series Forecasting

2025-05-07

AI Summary

Time series forecasting suffers from insufficient exploitation of historical patterns and weak inductive biases. To address this, we propose a retrieval-augmented forecasting paradigm: during inference, the most similar historical time segments are retrieved from the training set via similarity matching; their ground-truth future values are then jointly modeled with the current input to dynamically inject strong, explicit prior knowledge. This is the first work to explicitly incorporate historical pattern retrieval into time series forecasting—requiring no architectural modifications and overcoming the inductive bias limitations inherent in end-to-end learning. Leveraging time-series slice embeddings and multi-source fusion modeling, our method achieves an average win rate of 86% across ten benchmark datasets, significantly outperforming state-of-the-art models. It demonstrates both high computational efficiency and strong robustness under diverse distributional shifts.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs

Abstract

Time series forecasting uses historical data to predict future trends, leveraging the relationships between past observations and available features. In this paper, we propose RAFT, a retrieval-augmented time series forecasting method to provide sufficient inductive biases and complement the model's learning capacity. When forecasting the subsequent time frames, we directly retrieve historical data candidates from the training dataset with patterns most similar to the input, and utilize the future values of these candidates alongside the inputs to obtain predictions. This simple approach augments the model's capacity by externally providing information about past patterns via retrieval modules. Our empirical evaluations on ten benchmark datasets show that RAFT consistently outperforms contemporary baselines with an average win ratio of 86%.

Problem

• Enhancing time series forecasting with retrieval-augmented methods
• Leveraging historical data patterns for improved future predictions
• Outperforming baselines via external retrieval of similar past sequences

Innovation

• Retrieval-augmented method enhances forecasting accuracy
• Retrieves similar historical patterns for predictions
• Outperforms baselines with 86% average win ratio