EUVIMEDEUVIMED entdecken
Crossref

Nonlinear Residual Prediction Method for Site Based on Lasso Regression and Monte Carlo Sampling

Journal of Civil Engineering and Urban Planning · 2026 · 8 (1)

DOI: 10.23977/jceup.2026.080118 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Civil Engineering and Urban Planning, 2026.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

: Local ground effects exert significant modulation on seismic motion characteristics. Traditional one-dimensional equivalent linearization methods exhibit severe high-frequency "over-damping" phenomena under strong earthquake strains, while quantifying nonlinear residual responses remains challenging due to multidimensional coupling effects from seismic sources, paths, and spectral characteristics. Leveraging massive strong ground motion records from Japan's KiK-net network, this study proposes a machine learning-based approach for nonlinear system deviation prediction and uncertainty quantification through feature dimensionality reduction and probabilistic sampling. First, 15 multidimensional physical parameters encompassing seismic sources, paths, and ground motion responses were extracted. The Lasso regression algorithm (with L1 regularization) was employed for dynamic feature ±1 optimization and dimensionality reduction to construct a multivariable nonlinear residual prediction model. Residual prediction was further implemented using Monte Carlo (MC) sampling combined with Jaccard similarity index. Results demonstrate that dominant nonlinear residual parameters exhibit distinct frequency-dependent behaviors: high-frequency components are controlled by input intensity (e.g., peak response spectra under underground conditions), while mid-to-long periods are governed by spectral evolution characteristics (e.g., average period). Compared to traditional single-parameter empirical models, the Lasso multivariable model significantly improves full-cycle prediction accuracy. MC-sampled multiple standard deviations objectively reconstruct ground response uncertainties while maintaining high Jaccard similarity with real observation datasets. This study effectively compensates for high-frequency energy misconsumption in one-dimensional theories, providing a reliable data-driven approach for accurate residual estimation.

Zitieren

APA

(2026). Nonlinear Residual Prediction Method for Site Based on Lasso Regression and Monte Carlo Sampling. Journal of Civil Engineering and Urban Planning, 8(1). https://doi.org/10.23977/jceup.2026.080118

Vancouver

Nonlinear Residual Prediction Method for Site Based on Lasso Regression and Monte Carlo Sampling. Journal of Civil Engineering and Urban Planning. 2026;8(1). doi:10.23977/jceup.2026.080118

BibTeX
@article{euvimed2026,
  title = {Nonlinear Residual Prediction Method for Site Based on Lasso Regression and Monte Carlo Sampling},
  journal = {Journal of Civil Engineering and Urban Planning},
  year = {2026},
  volume = {8},
  number = {1},
  doi = {10.23977/jceup.2026.080118}
}
Kostenlos starten →