Deep learning analysis for energy consumption of shield tunneling machine drive system

Abstract

Inaccurate estimation of energy from the shield driving system may result in serious energy loss and low tunneling efficiency. A deep learning network is developed in this study to optimize the energy consumption of cutterhead drives in a shield tunneling system. The proposed model, KCNN-LSTM, cooperates convolutional neural network (CNN) and long short-term memory predictor (LSTM) based on clustering algorithm to realize accurate energy consumption forecasts. First, geological conditions, shield operational parameters, and geometry are characterized and identified using a feature extraction strategy. The proposed model is used to decompose the energy consumption values and extract the sequential and implicit features of nonlinear interactions. The prediction is then applied to a real tunnel field case study. The modeling results are evaluated by comparing its performance with the reproduced Naïve predictor and other proven deep learning networks including recurrent neural network (RNN), CNN, and LSTM. The comparisons reveal that the proposed model outperforms the other conventional techniques in predictive accuracy. The KCNN-LSTM model provides an accurate and feasible tool for adjusting the energy consumption of cutterhead drives in shield tunneling.