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

Area of Science:

• Structural Engineering
• Data Science
• Environmental Monitoring

Background:

• Modal frequencies of structures are sensitive to environmental changes (temperature, wind).
• Accurate modeling is crucial for structural health monitoring and safety.
• Existing methods may not fully capture complex, nonlinear relationships.

Purpose of the Study:

• To develop a robust mathematical model correlating ambient factors with modal frequencies.
• To effectively eliminate the influence of environmental variables on structural responses.
• To enhance the accuracy and generalization performance of modal frequency prediction.

Main Methods:

• Nonlinear Principal Component Analysis (NLPCA) for feature extraction and dimensionality reduction of ambient factors.
• Support Vector Regression (SVR) for modeling the relationship between extracted components and modal frequencies.
• Optimization of SVR hyperparameters using Grid Search Method (GSM), Genetic Algorithm (GA), and Fruit Fly Optimization Algorithm (FOA), with FOA showing superior performance.

Main Results:

• The proposed NLPCA-SVR model demonstrated high generalization performance in predicting modal frequencies.
• The model effectively captured and reflected the strong correlation between ambient factors and modal frequencies.
• Validation using Guangzhou New TV Tower (GNTVT) Benchmark data confirmed the method's efficacy.

Conclusions:

• The integrated NLPCA-SVR approach provides an effective means to model and mitigate the impact of ambient factors on structural modal frequencies.
• This method offers a significant advancement for accurate structural health monitoring systems.
• The study highlights the potential of advanced machine learning techniques in civil engineering applications.