Bidirectional coupled-gate long short-term memory: a simplified and efficient time series prediction model

A bidirectional coupled-gate long short-term memory model is proposed in order to improve the prediction accuracy of discrete time series data. The model is applied to the prediction of tunnel settlement development trend. In order to simplify the long short-term memory topology and improve the learning efficiency, the forget gate and the input gate of the long short-term memory are combined into coupled-gate to update the unit state of the network. The bidirectional coupled-gate long short-term memory model is proposed to capture more time series information, and the root mean square prop algorithm is used as the optimizer in the model training process to eliminate the phenomenon that the parameters are prone to large oscillations when they are update in a certain direction. The prediction of the settlement trend of a subway tunnel is used to verify the validity of the model. The results show that the mean absolute error, mean relative error and root mean square error of the bidirectional coupled-gate long short-term memory model for tunnel settlement prediction can be reduced to 0.1967mm, 0.68% and 0.2624mm respectively. It shows that the bidirectional coupled-gate long short-term memory model proposed in this paper has better functional approximation ability and robustness.