Spatio-Temporal Forecasting of Retaining Wall Deformation: Mitigating Error Accumulation via Multi-Resolution ConvLSTM Stacking Ensemble
Researchers have developed a novel ensemble framework using Convolutional Long Short-Term Memory (ConvLSTM) networks to improve long-term forecasting of retaining wall deformation. This multi-resolution approach integrates models trained on different temporal input scales, effectively mitigating error accumulation. Validation against both simulated and real-world data confirmed that the ensemble consistently outperformed individual ConvLSTM models, especially for multi-step predictions, highlighting its enhanced stability and accuracy in geotechnical AI forecasting. AI
IMPACT Enhances predictive stability and accuracy in AI-driven geotechnical forecasting, potentially leading to safer infrastructure development.