Reducibility Matrix based Model Reduction via Recurrent Neural Network Tuning

O. Alsmadi and M. Abdalla (Jordan)


Model Reduction, Recurrent Neural Network, Reducibility Matrix, Singular Perturbation.


A new order model reduction technique is presented in this paper. The reduction technique is based on a matrix reducibility concept, which incorporates a recurrent neural network algorithm for parameter tuning. The eigenvalues of the reduced order model are selected as a subset of the full order model eigenvalues. A more physically meaningful representation of the reduced order model is achieved by maintaining or preserving the full order system’s slow modes eigenvalues. Finally, simulation of a numerical example and a comparison with the singular perturbation model reduction method is provided in this paper.

Important Links:

Go Back