.png){kind=icon}
.png){kind=icon}
Return
Flexible fusion models for improving the early warning of river floodings, based on a generalization of Takagi-Sugeno-Kang inference system
.png){kind=icon}
PRE
AI
I
X
L
P
J
G
H
DOI：10.1016/j.fss.2026.110025.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Among natural disasters, floods are the most common and impactful type of natural disaster in the world, causing significant structural damage and loss of life. Anticipating their occurrence plays a crucial role in reducing their impact. However, this predictability is often challenging due to the combination of various meteorological conditions and flooding patterns that can deviate from the norm. The Adaptive Neuro-Fuzzy Inference System (ANFIS) appears to be the solution to perfectly incorporate fuzzy inference systems and Artificial Neural Networks (ANNs), thus overcoming the drawbacks of the lack of interpretability of ANNs and the shortage of self-learning in fuzzy theory. In this paper, we propose a novel Adaptive Neuro-Fuzzy Inference System (ANFIS) with the implementation of a new generalization of the Takagi-Sugeno-Kang inference system for making short-term river flow predictions aimed at anticipating floods caused by overflow up to 12 hours in advance. This generalization is based on a special type of aggregation function, namely the so-called non-disjunctive fusion functions, which enhance the capabilities of the traditional system, leading to performance improvements. Our model, when combined with predictions from a Long Short Term Memory neural network in an ensemble, outperforms other state-of-the-art methods. Furthermore, the fusion of recurrent and nonrecurrent models offers increased reliability in real-world scenarios, particularly regarding time-consistency, responsiveness to sudden changes, and the ability to make reasonable predictions when recurrence is not applicable after periods of missing data. Finally, we show that the proposed models are flexible enough to be adapted to datasets from different rivers.
Journal
IF:
2.7
Papers:
7.6K
Citations:
1.5W
Organization
Cited Papers