| description abstract | The advancement of computer technology has led to the increased utilization of new algorithms, such as machine learning, in various fields including underground engineering. The estimation of jacking force plays a critical role in the construction of rectangular jacked tunnels. Conventional prediction techniques often rely on empirical models and statistical analysis, posing challenges in accurately forecasting the jacking force for intricate tunnel structures. To overcome this obstacle, a method for predicting tunnel jacking force is proposed, which integrates a convolutional neural network (CNN) and long short-term memory network (LSTM). By utilizing geometric and operational parameters as inputs, the CNN extracts data features, which are subsequently inputted into the LSTM network for time-series modeling. This model effectively processes continuous jacking force data by comprehending the complex correlations within the data set, resulting in more precise predictions of future jacking force values. Comparative analysis with traditional methods such as the artificial neural network, single CNN model, and LSTM network demonstrates that the CNN-LSTM model significantly reduces prediction errors in tunnel jacking force estimation, thereby enhancing model accuracy. Consequently, the efficacy of the CNN-LSTM model has been validated, showcasing the benefits of employing deep-learning techniques for predicting jacking force in pipe jacking tunnel construction. An understanding of the dynamic variations of jacking force is crucial for engineers to enhance safety assessments in tunnel construction and to optimize the management and reduction of material and energy consumption throughout the construction process. This study introduces a deep-learning model that combines convolutional neural networks (CNNs) and long short-term memory (LSTM) networks to predict jacking force in rectangular pipe jacking tunnels. We gathered feature data pertinent to thrust from a practical shallow-buried tunnel project, which included variables such as jacking distance, frictional resistance, horizontal deviation of the cutter head, elevation deviation of the cutter head, and the rotation angle of the cutter head. Following the preprocessing of this data to eliminate outliers and noise, we developed a data set specifically for jacking force prediction. This data set was subsequently utilized to train the CNN-LSTM model. The findings demonstrate that the CNN-LSTM model significantly reduces prediction errors and exhibits a superior model fit compared to conventional approaches, achieving a fit score of 86% on the test data set. Furthermore, the model’s accuracy and generalizability could be enhanced by integrating additional thrust-related data from diverse geological strata. | |
