The invention provides a TBM tunneling parameter prediction method based on multi-target learning, which can solve the problems of subjectivity of artificial decision and singleness of decision rulesin a traditional control strategy. The method comprises: acquiring TBM tunneling parameters through an intelligent data acquisition system; performing noise reduction and enhancement processing on theTBM tunneling parameters to obtain data of each cyclic in a rising section anda stable section; extracting time domain characteristics of the rising section and the stable section in each cycle, andconstructing an evaluation index reflecting the health state of the TBM; utilizing an artificial intelligence algorithm to effectively represent a nonlinear mapping relation between the rising sectionTBM characteristic index and the stable section multi-target variable, and establishing a multi-target learning-based rock machine action model; carrying out self-adaptive adjustment on parameters ofthe rock machine action model through a nonlinear optimization method; according to the multi-target parameter information estimated by the rock machine action model, enabling the TBM driver to optimize the TBM control strategy by integrating the geological conditions of the surrounding rock and the particle distribution of the slag pieces. According to the method, TBM tunneling parameters can beestimated online, and meanwhile normal operation of the TBM is not interfered.