Proper orthogonal decomposition(POD) dimensionality reduction combined with machine learning to predict the vibration characteristics of stay cables at different lengths

For the study of vibration characteristics of stay cables, the response amplitude and frequency are typically obtained through programming in MATLAB software, which can be time-consuming. Based on this, this paper proposes the use of Proper Orthogonal Decomposition (POD) combined with four types of machine learning models to form corresponding hybrid models for rapid prediction of response amplitude and frequency. Initially, the displacement responses of the stay cable under various spans are collected using MATLAB programming to form a snapshot matrix; subsequently, the POD dimensionality reduction is applied to extract the POD mode shapes from the snapshot matrix. Four typical machine learning models are then employed: Support Vector Machines (SVM), Back Propagation Neural Network (BPNN), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM). These models are integrated with the POD dimensionality reduction to create hybrid models, and the four hybrid models (POD-SVM, POD-BPNN, POD-LSTM, POD-CNN) are compared for their predictions of response amplitude and frequency under different spans. Finally, the prediction efficacy and error rates of these hybrid models are analyzed. The study demonstrates that the hybrid models provide faster and more accurate predictions of target response values and frequencies compared to MATLAB programming. Among these models, the POD-BPNN hybrid model shows the best performance in predicting response amplitude and frequency with the smallest error. © 2024