DeepHGNN: Study of Graph Neural Network based Forecasting Methods for Hierarchically Related Multivariate Time Series

Abishek Sriramulu,Nicolas Fourrier,Christoph Bergmeir

2024-05-29

Abstract:Graph Neural Networks (GNN) have gained significant traction in the forecasting domain, especially for their capacity to simultaneously account for intra-series temporal correlations and inter-series relationships. This paper introduces a novel Hierarchical GNN (DeepHGNN) framework, explicitly designed for forecasting in complex hierarchical structures. The uniqueness of DeepHGNN lies in its innovative graph-based hierarchical interpolation and an end-to-end reconciliation mechanism. This approach ensures forecast accuracy and coherence across various hierarchical levels while sharing signals across them, addressing a key challenge in hierarchical forecasting. A critical insight in hierarchical time series is the variance in forecastability across levels, with upper levels typically presenting more predictable components. DeepHGNN capitalizes on this insight by pooling and leveraging knowledge from all hierarchy levels, thereby enhancing the overall forecast accuracy. Our comprehensive evaluation set against several state-of-the-art models confirm the superior performance of DeepHGNN. This research not only demonstrates DeepHGNN's effectiveness in achieving significantly improved forecast accuracy but also contributes to the understanding of graph-based methods in hierarchical time series forecasting.

Machine Learning,Artificial Intelligence

What problem does this paper attempt to address?

The paper focuses on the use of Graph Neural Networks (GNN) in hierarchical multivariate time series forecasting. Existing hierarchical forecasting methods typically use a two-step approach: first predicting each time series in the hierarchy individually, and then coordinating them to ensure consistency. However, these methods may not guarantee both prediction accuracy and consistency between hierarchies. The paper proposes a new hierarchical GNN framework called DeepHGNN, which is specifically designed for complex hierarchical structures. The uniqueness of DeepHGNN lies in its innovative graph-based hierarchical interpolation and end-to-end coordination mechanism, which ensures prediction accuracy and consistency across different hierarchies while sharing signals between them. DeepHGNN leverages the predictability differences across hierarchical time series by aggregating and utilizing knowledge from all hierarchies to improve overall prediction accuracy. Through extensive evaluations against various state-of-the-art models, DeepHGNN demonstrates its superior performance in terms of prediction accuracy, providing new understanding and benchmarks for graph-based methods in hierarchical time series. Additionally, the paper discusses the limitations of existing methods, such as assumption limitations, inapplicability to a large number of time series, and poor performance on large datasets. DeepHGNN addresses these issues by adapting to evolving hierarchical structures, enhancing model adaptability and prediction accuracy.

DeepHGNN: Study of Graph Neural Network based Forecasting Methods for Hierarchically Related Multivariate Time Series

Hierarchical Joint Graph Learning and Multivariate Time Series Forecasting

Connecting the Dots: Multivariate Time Series Forecasting with Graph Neural Networks

HiSTGNN: Hierarchical Spatio-temporal Graph Neural Networks for Weather Forecasting

MTHetGNN: A Heterogeneous Graph Embedding Framework for Multivariate Time Series Forecasting

Adaptive Dependency Learning Graph Neural Networks

Graph Deep Learning for Time Series Forecasting

Dynamic Personalized Graph Neural Network with Linear Complexity for Multivariate Time Series Forecasting

GEANN: Scalable Graph Augmentations for Multi-Horizon Time Series Forecasting

ForecastGrapher: Redefining Multivariate Time Series Forecasting with Graph Neural Networks

Graph-based Time Series Clustering for End-to-End Hierarchical Forecasting

A Study of Joint Graph Inference and Forecasting

A Multi-Granularity Hierarchical Network for Long- and Short-Term Forecasting on Multivariate Time Series Data

FourierGNN: Rethinking Multivariate Time Series Forecasting from a Pure Graph Perspective

Parallel Multi-Scale Dynamic Graph Neural Network for Multivariate Time Series Forecasting

Messages Are Never Propagated Alone: Collaborative Hypergraph Neural Network for Time-Series Forecasting.

TimeGNN: Temporal Dynamic Graph Learning for Time Series Forecasting

Learning the Evolutionary and Multi-scale Graph Structure for Multivariate Time Series Forecasting

Discrete Graph Structure Learning for Forecasting Multiple Time Series

Learning and Integration of Adaptive Hybrid Graph Structures for Multivariate Time Series Forecasting.

Scalable Spatiotemporal Graph Neural Networks