A machine learning approach for ground response analysis

Ground response is an integral part of the dynamic response of the overlying infrastructure. When a large number of ground response analyses is needed, for instance in the case of seismic hazard analysis on a large geographical scale or, when assessing uncertainty in seismic structural response due to site effects, computational efficiency is critical. Machine-learning (ML) models can predict response at a very low cost and require fewer parameters. An ML procedure is proposed here to model the ground response using artificial neural networks (ANNs) and deep-learning methods. Ground response is affected by various parameters, including the mechanical properties of the underlying soil strata and their geometry. In the current study, soil response is assumed elastic while seismic excitation is modelled through harmonic waves with unit amplitude and various frequencies in a given range of interest. The physical soil model is approximated with a recurrent neural network (RNN), which is combined with long short-term memory (LSTM) units to handle time-series data. The RNN model receives as input the har-monic waves and gives as output the estimated ground acceleration responses. The latter are compared against the target output signals derived from conventional ground response analyses to evaluate the accuracy of the model. To train the RNN, the ADAptive Moment estimation (ADAM) optimisation algorithm is applied, given its suitability for treating large datasets. To increase the computation efficiency of the RNN algorithm, the full-length input signal is split in several batches of fixed length. To tune the ML parameters and assess their influence, analyses are conducted for two loss metrics, three activation functions, and three batch lengths. It is shown that the computational demands of traditional analytical methods could be overcome by ML techniques that can efficiently process complex and large datasets of ground responses.

KOURIS Leonidas;  GKOKTSI Kyriaki;  GKATZOGIAS Konstantinos;  POHORYLES Daniel;  TSIONIS Georgios;  KRAUSMANN Elisabeth; 

2026-01-07

National Technical University of Athens (NTUA), Greece

JRC142008

978-618-5827-06-9 (online),   

2623-3347 (online),   

https://doi.org/10.7712/120125,    https://www.eccomasproceedia.org/conferences/thematic-conferences/compdyn-2025/12404,    https://publications.jrc.ec.europa.eu/repository/handle/JRC142008,   

10.7712/120125.12404.26203 (online),