Application of Deep Learning Algorithms for Predicting Consolidation Settlement

Significant amount of consolidation settlement can occur in construction sites with soft clayey soil deposits. Accurate prediction is important to prevent serious issues, such as tilting and overturning of structures, as demonstrated in Busan Newport, South Korea. Observational methods, which perform regression analysis to predict settlement, are generally applied. However, the methods tend to produce inaccurate predictions when measurement records are limited. Therefore, this study applied deep learning algorithms to enhance the prediction accuracy of settlement. Three distinct models are developed based on artificial neural network, long short–term memory, and gated recurrent unit (GRU) algorithms. The models’ performance was evaluated across 277 scenarios, including 216 from the Busan Newport and 61 from an additional case study. The scenarios were classified based on the average degree of consolidation, mirroring real–world conditions. The performance of the deep learning models was compared against observational methods including the hyperbolic and Asaoka methods. According to analysis, the deep learning models demonstrated a 58 % reduction in root mean square error compared with the observational methods. Statistical analysis showed that deep learning models effectively reduced standard deviation and 90th percentile values, even with limited data. The GRU model, in particular, showed superior accuracy with the lowest statistical variation. This research highlights the potential of deep learning models for practical applications in predicting consolidation settlement.