A novel stability discriminant model of landslide dams based on CNN-BiGRU optimized by attention mechanism

The aim of this paper is to propose an integrated model based on convolutional neural network (CNN), bidirectional gated recurrent unit (BiGRU) and attention mechanism, namely CNN-BiGRU-Attention model, to address the problems of poor universality and accuracy of the current rapid landslide dam stability discrimination model. The model relies on the CNN technique to extract the spatial characteristics of landslide dam indicators, and learns the linkage of data information flow in the time domain through the BiGRU, which takes into account the spatial and temporal variability of landslide dam parameters. To validate the proposed hybrid model, standalone GRU, BiGRU, CNN-BiGRU, support vector regression (SVR), as well as traditional rapid discriminant models based on data statistics were considered the benchmark models. The results indicated that the CNN-BiGRU-Attention model exhibited the best classification performance among all the peer models. The comprehensive performance of the CNN-BiGRU-Attention model outperforms most of the existing classification algorithms, with average Precision, Recall, Accuracy, F1, FM index, MCC and Rr of 88.89%, 97.56%, 94.74%, 0.93, 0.93, 0.89, and 92.46%. The proposed hybrid model can be recommended as a promising model for the stability discrimination of landslide dams.