TimePro: Efficient Multivariate Long-term Time Series Forecasting with Variable- and Time-Aware Hyper-state

2025-05-27

AI Summary

Multivariate long-term time series forecasting faces the multi-lag challenge—heterogeneous temporal lags across variables influencing the target. Method: We propose a variable- and time-aware hyperstate mechanism embedded within the Mamba architecture, leveraging learnable joint variable-time attention to adaptively attend to critical lagged timestamps, thereby jointly modeling inter-variable dependencies and dynamic temporal patterns while preserving fine-grained temporal features. The method retains linear computational complexity O(L) without sacrificing expressiveness. Contribution/Results: Our approach achieves state-of-the-art or significantly superior performance on eight real-world long-term forecasting benchmarks. To our knowledge, it is the first method to systematically address multi-lag modeling under strict O(L) complexity constraints, bridging a fundamental gap between efficiency and representational fidelity in multivariate forecasting.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Data Mining & Knowledge Management: Mining of Spatial, Temporal or Spatio-Temporal Data

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
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

In long-term time series forecasting, different variables often influence the target variable over distinct time intervals, a challenge known as the multi-delay issue. Traditional models typically process all variables or time points uniformly, which limits their ability to capture complex variable relationships and obtain non-trivial time representations. To address this issue, we propose TimePro, an innovative Mamba-based model that constructs variate- and time-aware hyper-states. Unlike conventional approaches that merely transfer plain states across variable or time dimensions, TimePro preserves the fine-grained temporal features of each variate token and adaptively selects the focused time points to tune the plain state. The reconstructed hyper-state can perceive both variable relationships and salient temporal information, which helps the model make accurate forecasting. In experiments, TimePro performs competitively on eight real-world long-term forecasting benchmarks with satisfactory linear complexity. Code is available at https://github.com/xwmaxwma/TimePro.

Problem

• Addresses multi-delay issue in multivariate long-term forecasting
• Captures complex variable relationships and temporal features
• Improves accuracy with variate-time-aware hyper-states

Innovation

• Mamba-based model with hyper-states
• Adaptive time point selection
• Linear complexity forecasting