Parameter-Efficient Temporal Convolutional Networks for Multivariate Time-Series Forecasting

Accurate prediction of multivariate datasets increasingly relies on deep learning models, often increasing complexity and computing requirements, particularly for multi-horizon predictions. This work proposes a model designed to significantly reduce computing requirements while retaining predictive accuracy. By replacing standard convolutions with depthwise and grouped pointwise operations, the model preserves temporal processing and captures dependencies without substantially complicating the architecture. Evaluated on five multivariate datasets across four forecasting horizons, the proposed model is compared against conventional and depthwise Temporal Convolutional Networks; lightweight variants, such as TSMixer and FITS; and advanced models like Informer, Autoformer, and iTransformer. It reduces multiply–accumulate operations by 58.9–70.8% and trainable parameters by 14.15–35.8% compared to standard temporal convolution networks. Notably, forecasting accuracy varies across datasets and horizons, with small differences between TCN variants corresponding to run-to-run variability. The findings reveal that the model excels in accuracy–efficiency trade-offs, offering an alternative for multivariate time-series forecasting when computational cost is a critical consideration.

Authors

Institutions

Publication Details

Journal
AI
Published
2026-11-12
DOI
https://doi.org/10.3390/ai7110481
Primary Topic
Time Series Analysis and Forecasting
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Parameter-Efficient Temporal Convolutional Networks for Multivariate Time-Series Forecasting

Jelena D. Velimirović, Zoran P. Stajić, Lazar Zoran Velimirovic, Aleksa Bogdanović
AI
Time Series Analysis and Forecasting
article

Parameter-Efficient Temporal Convolutional Networks for Multivariate Time-Series Forecasting

Jelena D. Velimirović, Zoran P. Stajić, Lazar Zoran Velimirovic, Aleksa Bogdanović
article en

Abstract

Accurate prediction of multivariate datasets increasingly relies on deep learning models, often increasing complexity and computing requirements, particularly for multi-horizon predictions. This work proposes a model designed to significantly reduce computing requirements while retaining predictive accuracy. By replacing standard convolutions with depthwise and grouped pointwise operations, the model preserves temporal processing and captures dependencies without substantially complicating the architecture. Evaluated on five multivariate datasets across four forecasting horizons, the proposed model is compared against conventional and depthwise Temporal Convolutional Networks; lightweight variants, such as TSMixer and FITS; and advanced models like Informer, Autoformer, and iTransformer. It reduces multiply–accumulate operations by 58.9–70.8% and trainable parameters by 14.15–35.8% compared to standard temporal convolution networks. Notably, forecasting accuracy varies across datasets and horizons, with small differences between TCN variants corresponding to run-to-run variability. The findings reveal that the model excels in accuracy–efficiency trade-offs, offering an alternative for multivariate time-series forecasting when computational cost is a critical consideration.

AIVol. 7(11)
University of Nis (RS), Mathematical Institute of the Serbian Academy of Sciences and Arts (RS)
Time Series Analysis and Forecasting
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon