.png){kind=icon}
.png){kind=icon}
Return
Fuzzy inference-based thermal uncertainty modeling for underground cable systems integrated within complex tunnel infrastructure environments
.png){kind=icon}
PRE
AI
J
Y
B
X
W
DOI：10.1007/s10973-026-16028-8.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
A modern tunnel cable system will be subject to complicated and uncertain thermal conditions and an accurate prediction of the health of the tunnel cable and risk assessment is not easy. The current Deep Learning (DL), graph model and neuro-fuzzy methods either enhance the prediction capability, or boost interpretability at the expense of high computation complexity, lack of uncertainty modeling, lack of scalability and lack of physical consistency. The proposed Hybrid Neuro-Fuzzy Context-Aware Network (HNF-CAN) which integrates the power of Graph Neural Network (GNN), Adaptive Neuro-Fuzzy Inference System (ANFIS), Physics-Informed Neural Network (PINN) and Reinforcement Learning (RL) in a single network. This proposed algorithm is split into two parts: spatially dependent data pre-process and construct graph structure learn spatially dependent features using GNN with consideration of context. These are then passed to a neuro-fuzzy module for making uncertainty-aware predictions which are easily interpretable and a physics-informed module ensures the thermal consistency. An RL agent can be used to optimize the operation of the RL agent, such as load control and maintenance. On this experiment, high R2 value of 0.97 obtained which is provided the efficiency of the proposed model. Lastly, an approach to thermal uncertainty modeling that is both physically consistent and interpretable/scalable for use in infrastructure is presented, which seems to be very promising for implementation and intelligent management of infrastructure.
Keywords:
Context-aware learning
Graph neural networks
Hybrid neuro-fuzzy systems
Physics-informed modeling
Thermal uncertainty
Tunnel cable health prediction
Journal
IF:
3.1
Papers:
1.8W
Citations:
3.2W
Organization
Cited Papers