Triangular Type-2 Fuzzy Sets Based Weighted Regularized Extreme Learning Algorithm
Guoliang Zhao, Wei Han Wu · 2019
In this paper, triangular type-2 fuzzy sets are implemented to enhance the generalization ability of the extreme learning machine (ELM). Two algorithms named the triangular weighted regularized extreme learning machine (TriWR-ELM) and triangular type-2 weighted regularized extreme learning machine (TriT2WR-ELM) are proposed. The introduction of triangular type-2 fuzzy sets and extended t-norms renders the proposed extreme learning machines more stable than weighted and regularized ELMs, and only slightly increases the computation time owing to approximate type-reduction set of the triangular type-2 fuzzy set. Experiments are conducted on the Sinc function and thermal power plant's flue gas denitrification efficiency prediction problem to test the efficiency of TriWR-ELM and TriT2WR-ELM. Results show that the proposed two algorithms are more robust to larger outliers than conventional algorithms.