Prediction model of TBM response parameters based on a hybrid drive of knowledge and data

Abstract

Accurate prediction of tunnel boring machine (TBM) performance parameters and rock condition perception can effectively guide equipment construction. Relying on data from the Yinsong project in Jilin Province, China, this paper proposed two predictive models for TBM response parameters (cutterhead torque and total thrust): a data-driven model using only raw data and a hybrid drive of knowledge and data model (hybrid driven-model) incorporating derived parameters. This paper explored model optimization from input feature (X₁), dataset size, and machine learning algorithms to further compare the two models. Results demonstrate that the hybrid-driven model exhibits better learning efficiency, and its derived parameters in the input feature better reflect the surrounding rock conditions, thereby achieving high-precision prediction of response parameters. Additionally, in terms of surrounding rock feature extraction, selecting key rock fragmentation parameters randomly during the loading phase for 30 s as X₁ proves to be optimal. Regarding algorithms, deep learning algorithms further enhance predictive performance. The response parameter prediction model constructed in this paper can better extract surrounding rock conditions, laying a solid foundation for optimizing control parameters.