PHAT: Modeling Period Heterogeneity for Multivariate Time Series Forecasting

2026-01-31

AI Summary

This work addresses the widespread neglect of periodic heterogeneity—characterized by diverse and dynamically varying cycle lengths across variables—in multivariate time series modeling. To tackle this challenge, the authors propose a period-aware three-dimensional “period bucket” tensor structure that organizes inputs through cycle grouping and phase alignment. They further design intra-bucket interaction and inter-bucket masking mechanisms, complemented by a decomposition of attention into positive and negative components and a period-prior modulation strategy to explicitly model both period alignment and deviation. Built upon a Transformer architecture, the proposed method significantly outperforms 18 baseline models across 14 real-world datasets, offering the first systematic solution to modeling periodic heterogeneity in multivariate time series and achieving state-of-the-art forecasting performance.

Technology Category

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

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics

Abstract

While existing multivariate time series forecasting models have advanced significantly in modeling periodicity, they largely neglect the periodic heterogeneity common in real-world data, where variables exhibit distinct and dynamically changing periods. To effectively capture this periodic heterogeneity, we propose PHAT (Period Heterogeneity-Aware Transformer). Specifically, PHAT arranges multivariate inputs into a three-dimensional"periodic bucket"tensor, where the dimensions correspond to variable group characteristics with similar periodicity, time steps aligned by phase, and offsets within the period. By restricting interactions within buckets and masking cross-bucket connections, PHAT effectively avoids interference from inconsistent periods. We also propose a positive-negative attention mechanism, which captures periodic dependencies from two perspectives: periodic alignment and periodic deviation. Additionally, the periodic alignment attention scores are decomposed into positive and negative components, with a modulation term encoding periodic priors. This modulation constrains the attention mechanism to more faithfully reflect the underlying periodic trends. A mathematical explanation is provided to support this property. We evaluate PHAT comprehensively on 14 real-world datasets against 18 baselines, and the results show that it significantly outperforms existing methods, achieving highly competitive forecasting performance. Our sources is available at GitHub.

Problem

• period heterogeneity
• multivariate time series forecasting
• periodicity
• time series

Innovation

• Period Heterogeneity
• Multivariate Time Series Forecasting
• Periodic Bucket Tensor
• Positive-Negative Attention
• Periodic Prior Modulation