Operational modal analysis based on neural network with singular value decomposition

Abstract

Neural network can mine data features, and has strong anti-noise ability and applicability. The operational modal analysis (OMA) method based on back propagation neural network (BPNN) is proposed in this paper. Firstly, the dataset is preprocessed based on the input and output functions, which increases the anti-noise ability of the proposed method and simplifies the training by reducing the model parameters. Secondly, a three-layer BP neural network is established to identify parameters as accurately as possible with minimal network complexity and training data. In addition, an improved resilient back propagation (RPROP) algorithm is a fast and accurate batch learning methods for neural networks, which is used in the BPNN. Finally, simulation and experimental results show that the superior learning capabilities of BPNN even with few neurons and hidden layers. The proposed method has the advantages of high accuracy, strong generalization ability and fast convergence speed.