Approximation of interval models by neural networks
Xifan Yao, Shengda Wang, Shaoqiang Dong · 2005
An approach to approximate interval models by neural networks is proposed. The networks are structured according to the corresponding interval models, which makes them different from the existing interval backpropagation networks. The approach can incorporate analytical knowledge as well as expert's knowledge in the network and can provide transparency to the network. Furthermore, since the networks are linear, they are guaranteed to converge to the minimum. The proposed approach is applied to static interval systems as well as dynamic interval systems. Simulation results indicate that these relative simple interval networks achieve good approximation.