Abstract:This work proposes a novel approach for multiple time series forecasting. At first, multi-way delay embedding transform (MDT) is employed to represent time series as low-rank block Hankel tensors (BHT). Then, the higher-order tensors are projected to compressed core tensors by applying Tucker decomposition. At the same time, the generalized tensor Autoregressive Integrated Moving Average (ARIMA) is explicitly used on consecutive core tensors to predict future samples. In this manner, the proposed approach tactically incorporates the unique advantages of MDT tensorization (to exploit mutual correlations) and tensor ARIMA coupled with low-rank Tucker decomposition into a unified framework. This framework exploits the low-rank structure of block Hankel tensors in the embedded space and captures the intrinsic correlations among multiple TS, which thus can improve the forecasting results, especially for multiple short time series. Experiments conducted on three public datasets and two industrial datasets verify that the proposed BHT-ARIMA effectively improves forecasting accuracy and reduces computational cost compared with the state-of-the-art methods.

What problem does this paper attempt to address?

This paper attempts to solve the problem of Multiple Short Time Series Forecasting, especially the challenges faced when dealing with multiple short - time series. Specifically, the paper proposes a new method - Block Hankel Tensor ARIMA (BHT - ARIMA), aiming to improve the prediction accuracy and reduce the computational cost. ### Main problems addressed in the paper: 1. **High computational cost**: The existing ARIMA models need to predict multiple time series one by one, resulting in high computational costs, especially when the number of time series is large. 2. **Failure to consider internal relationships**: Traditional ARIMA models do not consider the internal relationships between related time series, which may limit their performance. 3. **Difficulties in predicting short - time series**: The time series in many practical applications are relatively short, and the existing linear models, tensor methods and deep - learning methods are not effective in dealing with these short - time series. ### Solutions: The paper proposes solutions through the following steps: 1. **Multidimensional Delay - embedding Transformation (MDT)**: - Use MDT to transform multiple time series into high - order Block Hankel Tensors (BHT). This step takes advantage of the low - rank structure of time series in the embedding space. 2. **Low - rank Tucker Decomposition**: - Apply low - rank Tucker decomposition to project high - order tensors onto compressed core tensors. Maximize the retention of the temporal continuity between core tensors through orthogonal factor matrices (projection matrices) to better capture the internal temporal correlations of time series. 3. **Generalized Tensor ARIMA**: - Generalize the classical ARIMA model to tensor form and directly apply it to the core tensor. Predict future time points by training the core tensor, and then obtain the prediction results of all time series through inverse Tucker decomposition and inverse MDT. ### Main contributions: 1. **Introduction of MDT and tensor decomposition**: - For the first time, combine MDT and tensor decomposition and apply them to time series prediction. Experiments prove that this strategy is also effective for existing tensor methods. 2. **Exploration of internal correlations**: - Explore the internal correlations of multiple time series in the low - rank embedding space through low - rank Tucker decomposition. In addition, study the effect of relaxing the orthogonality constraint (only impose the orthogonality constraint on non - temporal patterns), and find that it can further improve the prediction accuracy and reduce the parameter sensitivity. 3. **Generalized tensor ARIMA**: - Generalize the classical ARIMA model to tensor form and combine it with Tucker decomposition in a unified framework. The experimental results show that BHT - ARIMA is significantly superior to nine state - of - the - art methods in terms of efficiency and effectiveness, especially when dealing with multiple short - time series. ### Experimental verification: The paper conducts experiments on three public datasets and two industrial datasets to verify the effectiveness and superiority of BHT - ARIMA. The experimental results show that BHT - ARIMA not only improves the prediction accuracy but also reduces the computational cost. In conclusion, by introducing Block Hankel Tensors and Generalized Tensor ARIMA, this paper provides an efficient and accurate method for multiple short - time series prediction.

Block Hankel Tensor ARIMA for Multiple Short Time Series Forecasting

TCP-ARMA: A Tensor-Variate Time Series Forecasting Method

Distributed ARIMA Models for Ultra-long Time Series

Low-rank Tensor Autoregressive Predictor for Third-Order Time-Series Forecasting

Parallel Extraction of Long-term Trends and Short-term Fluctuation Framework for Multivariate Time Series Forecasting

Ultra-short-term wind speed and wind power forecast via selective Hankelization and low-rank tensor learning-based predictor

HiMTM: Hierarchical Multi-Scale Masked Time Series Modeling with Self-Distillation for Long-Term Forecasting

Robust Multi-Dimensional Time Series Forecasting

Generalizable Memory-driven Transformer for Multivariate Long Sequence Time-series Forecasting

A Short-Term Tropical Cyclone Intensity Forecasting Method Based on High-Order Tensor (student Abstract)

Low-Rank Autoregressive Tensor Completion for Multivariate Time Series Forecasting

N-HiTS: Neural Hierarchical Interpolation for Time Series Forecasting

MPM: Multi Patterns Memory Model for Short-Term Time Series Forecasting

Forecasting Urban Traffic States with Sparse Data Using Hankel Temporal Matrix Factorization

Masked Multi-Step Multivariate Time Series Forecasting with Future Information

TCLN: A Transformer-based Conv-LSTM Network for Multivariate Time Series Forecasting

Multi-step Forecasting of Short-Term Traffic Flow Based on Intrinsic Pattern Transform

Deep Video Prediction for Time Series Forecasting

Adaptive Multi-Scale Decomposition Framework for Time Series Forecasting

Low-Rank Temporal Attention-Augmented Bilinear Network for financial time-series forecasting

Forecasting with Multiple Seasonality