Interval Prediction Method for Cable Tension Based on Adaptive VMD and Improved BOBiLSTM
Author:Zicong Cao1 Author Unit: 1 .School of Information and Electrical Engineering, Hunan University of Science and Technology, Xiangtan 411201 Hunan, China 2.Hunan Sine Electronic Technology Company Limited, Xiangtan 411100 Hunan,China
Abstract：Monitoring cable tension is paramount in structural health monitoring (SHM) of cable-stayed bridges. However, the traditional method of using vibration frequencies has a problem of inaccurately identifying the fundamental frequency in cable tension monitoring, resulting in errors in cable tension calculation. To address this issue, we propose an interval prediction method for cable tension. This method is based on adaptive variational mode decomposition (VMD) and an improved Bayesian optimization algorithm that optimizes the Bidirectional Long
Short-Term Memory network (BO-BiLSTM). By utilizing this approach, we can effectively monitor the structural health of cable-stayed bridges without the need to identify the fundamental frequency. Firstly, we establish an adaptive VMD model by introducing circular mapping to enhance the traditional Grey Wolf optimizer (GWO). This enhanced model was then combined with the lasso algorithm to achieve sparse representation of the modal components of VMD decomposition, effectively accomplishing the goals of feature extraction and denoising. Secondly, a fftness function composed of interval coverage and average interval bandwidth was constructed to enhance the Bayesian optimization algorithm. This improved algorithm was used to optimize the hyperparameters of the BiLSTM model, resulting in improved prediction accuracy of cable tension. Finally, non-parametric kernel density estimation with optimized kernel bandwidth is used to estimate the cable tension errors. Experimental results show that the proposed model has a high cable tension interval coverage rate of 97.56％. Even in situations where the fundamental frequency cannot be accurately identiffed, the interval coverage rate of this method can reach 94.73％. This method provides an accurate method for monitoring cable tensions in cable-stayed bridges.
