Prediction of cutterhead torque change trend of shield machine based on partial state visible HMM and LSTM

Abstract

Cutterhead torque is a key parameter to determine the normal operation of shield machine. Accurately predicting the change trend of cutterhead torque can provide decision support for shield operators to control operating parameters. Therefore, a prediction method of cutterhead torque change trend based on partial state visible Hidden Markov Model (HMM) and Long Short-Term Memory (LSTM) network is proposed in this paper. First, the Spearman correlation coefficient is introduced to analyze the correlation of the tunneling parameters, and the historical tunneling data of the cutter torque, pitch angle and related parameters are obtained to establish a database. Secondly, the pitch angle prediction model is established based on LSTM network, and the predicted pitch angle is used as the observation sequence, and the implicit relationship between cutterhead torque and pitch angle is mined through HMM to obtain the probability matrix. Finally, the cutterhead torque prediction model is established, and the pitch angle sequence is used as input to output the corresponding cutterhead torque change trend. The results show that the method can accurately predict the dynamic change trend of the cutterhead torque in the case of less known parameter states, and the effectiveness of the method are verified in this paper.