System identification of dynamic structure by the multi-branch BPNN

Abstract

System identification is one of the critical factors to control structural vibration with high quality and evaluate whether control method can be applied or not. In this paper, a kind of multi-branch back propagation neural network (BPNN) model is proposed to identify a structural dynamic system. In this model, the primary factors that affect structural dynamic response, namely structure state variables and seismic inputs, are separately treated as the branches of the model, that is expected to enhance prediction precision. The aim of identification is to make the trained model be able to accurately predict structural future dynamic response. When the model is established, it can be trained with collected dynamic response and seismic wave data. In this paper, a numerical example is given. The analytic result turns out that the proposed identification model can accurately predict structural future dynamic response after being trained.