Diffusion Models for Time Series Forecasting: A Survey

2025-07-19

AI Summary

Diffusion models, originally developed for image generation, face challenges in time-series forecasting (TSF) due to fundamental differences in data structure, temporal dependencies, and evaluation criteria. Method: This work presents the first systematic survey of diffusion-based TSF, introducing a novel taxonomy grounded in conditional information sources (e.g., historical observations, covariates, timestamps) and fusion mechanisms (early/late/mid-layer injection). It unifies the analysis of conditional control paradigms—including denoising objectives, feature encoding strategies, and reverse sampling procedures—and conducts a comprehensive empirical assessment across standard datasets, baselines, and metrics. Contribution/Results: The survey identifies critical bottlenecks in scalability, long-horizon dependency modeling, and computational efficiency. It establishes a principled framework for theoretical understanding, methodological design, and reproducible benchmarking of diffusion models in TSF, while outlining concrete directions for future research.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Time-Series/Data Streams
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• Web Mining and Content Analysis: Content-based information diffusion
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization

Abstract

Diffusion models, initially developed for image synthesis, demonstrate remarkable generative capabilities. Recently, their application has expanded to time series forecasting (TSF), yielding promising results. In this survey, we firstly introduce the standard diffusion models and their prevalent variants, explaining their adaptation to TSF tasks. We then provide a comprehensive review of diffusion models for TSF, paying special attention to the sources of conditional information and the mechanisms for integrating this conditioning within the models. In analyzing existing approaches using diffusion models for TSF, we provide a systematic categorization and a comprehensive summary of them in this survey. Furthermore, we examine several foundational diffusion models applied to TSF, alongside commonly used datasets and evaluation metrics. Finally, we discuss current limitations in these approaches and potential future research directions. Overall, this survey details recent progress and future prospects for diffusion models in TSF, serving as a reference for researchers in the field.

Problem

• Adapting diffusion models for time series forecasting tasks
• Reviewing conditional information integration in diffusion models
• Analyzing limitations and future directions in TSF applications

Innovation

• Adapt diffusion models to time series forecasting
• Integrate conditional information into diffusion models
• Systematically categorize diffusion-based TSF approaches