Scholarlink

<p>This study presents the automated large reservoir high-order fuzzy cognitive maps (AL-RHFCM) forecasting framework, an enhanced extension of the LR-HFCM method designed for univariate time series prediction. The LR-HFCM approach, rooted in the principles of R-HFCM, integrates fuzzy time series (FTS), fuzzy cognitive maps (FCMs), and reservoir computing (RC), forming a hybrid model. Functionally, it operates as a variant of the echo state network (ESN), consisting of an input layer, a large intermediate reservoir, and an output layer trained using LASSO regression. The unique architecture of LR-HFCM employs a large reservoir with multiple sub-reservoirs, each capturing distinct dynamics of input time series through varying combinations of concepts and orders. The proposed AL-RHFCM builds upon this structure by fully automating the selection of sub-reservoirs while maintaining randomly assigned and fixed weights within each sub-reservoir throughout training. The performance of AL-RHFCM is validated using five distinct time series datasets, demonstrating significant improvements over a range of baseline models. These results establish AL-RHFCM as a robust and efficient technique for time series forecasting, capable of capturing complex patterns and dynamics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated forecasting model leveraging large reservoirs of randomized fuzzy cognitive maps

  • 期刊论文

  • 发表日期：

    2025年11月11日

查看全文
  • Omid Orang,
  • Fabricio J. Erazo-Costa,
  • Petrônio C. L. Silva,
  • Guilherme de Alencar Barreto,
  • Frederico Gadelha Guimarães

摘要

This study presents the automated large reservoir high-order fuzzy cognitive maps (AL-RHFCM) forecasting framework, an enhanced extension of the LR-HFCM method designed for univariate time series prediction. The LR-HFCM approach, rooted in the principles of R-HFCM, integrates fuzzy time series (FTS), fuzzy cognitive maps (FCMs), and reservoir computing (RC), forming a hybrid model. Functionally, it operates as a variant of the echo state network (ESN), consisting of an input layer, a large intermediate reservoir, and an output layer trained using LASSO regression. The unique architecture of LR-HFCM employs a large reservoir with multiple sub-reservoirs, each capturing distinct dynamics of input time series through varying combinations of concepts and orders. The proposed AL-RHFCM builds upon this structure by fully automating the selection of sub-reservoirs while maintaining randomly assigned and fixed weights within each sub-reservoir throughout training. The performance of AL-RHFCM is validated using five distinct time series datasets, demonstrating significant improvements over a range of baseline models. These results establish AL-RHFCM as a robust and efficient technique for time series forecasting, capable of capturing complex patterns and dynamics.