Abstract: Based on sparse borehole data in geotechnical engineering and the coupled Markov chain model, this paper proposes a stratum inversion model that accounts for spatial variability of strata. In the vertical direction, a vertical probability transition matrix is constructed using borehole data. Considering the non-uniform dip angle variations of strata in both horizontal and vertical directions, an intelligent calculation method for determining Walther parameter combinations is introduced by integrating the particle swarm optimization algorithm. This approach reasonably extends one-dimensional vertical stratigraphic conditions to two-dimensional stratigraphic conditions. The resulting stratigraphic conditions determined through the above methods can serve as input parameters for the intelligent temporal prediction of shield tunneling parameters. By jointly considering the influence of historical data from adjacent excavation rings and current data from the ongoing excavation ring, a feature fusion method is proposed to eliminate dimensional differences between historical and current data. Based on a Bi-LSTM sequential deep learning model, an intelligent prediction model for shield tunneling parameters that integrates current information is established. Using a shield tunneling section in Qingdao as a case study, the inversion performance of the proposed optimized coupled Markov chain model is analyzed. Compared with the traditional coupled Markov chain model, the accuracy and efficiency of the proposed model are improved by 93% and 79%, respectively. Based on the developed optimized coupled Markov chain model and the intelligent prediction model for shield tunneling parameters, the R² value of the cutterhead torque reaches 0.82 on the test set, indicating high predictive performance. The proposed framework, which integrates stratum inversion and intelligent sequential prediction of tunneling parameters, provides high-quality and efficient support for both construction efficiency and safety throughout the shield tunneling process.