Ambient Effect Filtering Using NLPCA-SVR in High-Rise Buildings

Abstract

The modal frequencies of a structure are affected by continuous changes in ambient factors, such as temperature, wind speed etc. This study incorporates nonlinear principal component analysis (NLPCA) with support vector regression (SVR) to build a mathematical model to reflect the correlation between ambient factors and modal frequencies. NLPCA is first used to eliminate the high correlation among different ambient factors and extract the nonlinear principal components. The extracted nonlinear principal components are input into the SVR model for training and predicting. The proposed method is verified by the measured data provided in the Guangzhou New TV Tower (GNTVT) Benchmark. The grid search method (GSM), genetic algorithm (GA) and fruit fly optimization algorithm (FOA) are applied to determine the optimal hyperparameters for the SVR model. The optimized result of FOA is most suitable for the NLPCA-SVR model. As evaluated by the hypothesis test and goodness-of-fit test, the results show that the proposed method has a high generalization performance and the correlation between the ambient factor and modal frequency can be strongly reflected. The proposed method can effectively eliminate the effects of ambient factors on modal frequencies.

Eliminating environmental and operational effects on structural modal frequency: A comprehensive review

“…To investigate and remove the effects of temperature and traffic volume, the Couple‐ARIMA model was developed, which is better than LR and MPR 57 ; Laory et al 66 built and evaluated five methods, MLR, ANN, SVR, regression tree (R‐Tree), and RF, and found that SVR and RF have better performance in frequency predictions. To characterize the dependency of modal frequency on temperature and wind speed, the NLPCA‐ANN and NLPCA‐SVR combined models were conducted on the Yonghe Bridge and the Guangzhou New TV Tower 63 ; Ma et al 49 successfully established PCA‐GPR model with consideration of the temperature, humidity, wind speed, and direction. To consider multiple sources of environmental and loading factors, Zhang et al 42 used GPR to effectively quantify the adverse influences of temperature, wind speed and peak acceleration from modal frequencies; Soria et al 67 established a dynamic MLR model and concluded that the corrected frequency has a closer range of variability than the original; Bayesian network model (BNM) was developed to quantify the uncertainty of different environmental variables (temperature, humidity, wind speed, and traffic volume) of Xinguang Bridge at both the parameter and model levels 68 .…”

“…Related methods may be linear or nonlinear. Among them, the linear method includes PCA, factor analysis (FA), and independent component analysis (ICA); the nonlinear method includes nonlinear PCA (NLPCA) and the kernel mapping of linear methods 63 . Ma et al 49 utilized PCA to integrate and compress the size of multiple environmental features as the input of the environment‐driven GPR model; Spiridonakos et al 64 employed ICA to capture the higher‐order statistical information of environmental variables; Li et al 24 utilized NLPCA to capture nonlinear variations in temperature and wind speed on modal frequencies of a cable‐stayed bridge.…”