High-accuracy slope stability analysis using data-driven and attention-based deep learning model

Because of the abrupt occurrence and severe consequences of slope disasters, the analysis of slope stability has been a focal point in the field of slope disaster prevention. Traditional methods require considerable investments and fail to effectively predict the development trends of slope stability. However, the emergence of data-driven approaches based on deep learning has forged a novel avenue. Recently, a transformer model is proposed, which has an attention module to learn the high dimensionality correlation between the properties and the stability of slopes. In this study, the transformer model is used to predict the slope safety factor and evaluate the slope stability, and a dataset consisting of 72,000 slope samples is created based on the computer-generated method. The superior predictive capabilities of the transformer model are demonstrated in comparison to LSTM and Attention-LSTM models. Subsequently, the transformer-based multi-classification and regression models are discussed. The regression model outperformed in predicting slope safety factor, reaching an impressive accuracy of 99.983%. The results indicate that the deep learning approach based on the transformer model has shown great potential and advantages for slope stability analysis. Its high accuracy and short computation time will contribute to rapid on-site decision-making in geotechnical engineering applications.