R-RTRL based on recurrent neural network with K-fold cross-validation for multi-step-ahead prediction landslide displacement

Abstract

The reinforced real-time recurrent learning (R-RTRL) algorithm with K-fold cross-validation for recurrent neural networks (RNNs) are applied to forecast multi-step-ahead landslide displacement (K-R-RTRL). The proposed novel method is implemented to make two-and four-ahead forecasts in Liangshuijing landslide monitoring point ZJG24 in Three Gorges Reservoir area. Based on comparison purpose, two comparative neural networks are performed, one is RTRL, the other is back propagation through time neural network (BPTT). The proposed algorithm K-R-RTRL gets superior performance to comparative networks in the final numerical and experimental results.