A machine learning model for parameter correlation analysis and structural deformation prediction

Tunnelling through steel reinforced concrete piles using shield machine involves high risk and creates an engineering challenge. Protecting safety of the superstructure is beneficial for the success of the project. The deformation of the superstructure is a complicated process and is affected by a variety of parameters. This paper proposes a novel model, MK-LSTM, combining maximal information coefficient (MIC), K-median and long short-term memory (LSTM) network to analyze the correlation between the various parameters and to predict of the superstructure deformation during shield driving. The MIC-K-median algorithm is developed to analyze the correlation between the different input parameters and structural deformation, then to preprocess the input parameters based on their correlation coefficients. The prediction accuracy and efficiency using different dimensions of input parameters are analyzed through the LSTM model and the optimal input parameter dimensions are selected. The conclusions show that: 1) the MK-LSTM algorithm can reduce the computational complexity and the impact of noise in raw data; 2) the prediction accuracy of the proposed MK-LSTM model is better than baseline models and some parameters with low correlation (e.g., chamber earth pressure) do not significantly improve the prediction effect. The sorting index can provide reference for construction.