A unified new-information-based accumulating generation operator based on feature decoupling for multi-characteristic time series forecasting

Diverse internal system laws and complex external environments often generate multi-characteristic time series, presenting challenges for forecasting methods in terms of model adaptability and practical feasibility. To address this issue, this paper introduces a novel approach by enhancing the Accum...

Full description

Saved in:

Bibliographic Details
Published in:	Applied soft computing Vol. 154; p. 111310
Main Authors:	Ding, Song, Cai, Zhijian, Ye, Juntao, Ma, Bianjing
Format:	Journal Article
Language:	English
Published:	Elsevier B.V 01.03.2024
Subjects:	Accumulating generation operators Grey models Heuristic intelligent algorithms New information priority Time series forecasting Time series forecasting Grey models Accumulating generation operators Heuristic intelligent algorithms New information priority
ISSN:	1568-4946, 1872-9681
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	Diverse internal system laws and complex external environments often generate multi-characteristic time series, presenting challenges for forecasting methods in terms of model adaptability and practical feasibility. To address this issue, this paper introduces a novel approach by enhancing the Accumulating Generation Operator (AGO), commonly employed in grey prediction models, as a sequence pre-process technique for forecasting models. Our innovation lies in the development of a unified new-information-based AGO (UNAGO), which distinguishes itself by multiple weight adjustment effects encompassing unconstrained scaling, exponential, and equal-accumulation terms. This technique can significantly enhance the adaptive variability of the accumulating weight structure and mitigate possible incompatibility. To validate our approach, we conduct comprehensive comparisons using UNAGO against seven existing AGOs within grey models. We select five datasets that span different industries and domains, and involve forecasts with various data trend characteristics, data lengths, and prediction horizons. Despite precision comparisons, we conduct further experiments regarding sample-size and trend-reversal analyses and robustness tests on heuristic intelligent algorithms. Empirical results show that the new model with UNAGO exhibits optimal predictive accuracies with improvement rates of over 51% in terms of MAPE values compared to all its rivals, demonstrating its strongest adaptability and robustness with multiple data trend characteristics and short-to-long terms forecasts. Additionally, the robustness tests, including algorithm comparisons, Monte Carlo simulations, and hyper-parameter sensitivity analysis, validate UNAGO's optimal heuristic intelligent algorithm and confirm its computational stability. •Accumulating generation operators (AGOs) for data preprocessing are systematically introduced and analyzed.•A novel unified accumulating generation operator based on the feature decoupling is initially designed.•The proposed AGO outperforms the other existing AGO methods significantly in terms of forecasting accuracy.•The proposed AGO has strong adaptability and robustness over other competitors.
ISSN:	1568-4946 1872-9681
DOI:	10.1016/j.asoc.2024.111310