Abstract: In the process of steering drilling, the triaxial acceleration signals collected at downhole near-bit measurement points are often subjected to strong interference and weak signal-to-noise ratio(SNR) conditions, posing significant challenges for signal reconstruction and attitude angle estimation. To address this issue, this paper proposes an adaptive signal extraction method, namely IGWO-EEMD, based on improved grey wolf optimizer(IGWO) combined with ensemble empirical mode decomposition(EEMD). The proposed method performs multiple EEMD decompositions to construct a candidate set of intrinsic mode functions(IMFs), and introduces an improved GWO algorithm, integrating Levy flight and spiral bubble-net strategies, to automatically search for the optimal number of reconstruction layers using SNR as the objective function, thereby achieving optimal reconstruction of weak SNR signals. Simulation results demonstrate that the proposed method significantly improves the accuracy in inclination and toolface angle estimation, with root mean square error(RMSE) reductions of 46% and 20.6% for inclination, and 44.1% and 30.1% for toolface angle, compared with random-layer and ordinary least squares(OLS) methods, respectively. Validation using actual downhole data further confirms the effectiveness and practicality of the proposed method in attitude angle reconstruction. This study provides theoretical guidance and methodological reference for weak SNR signal extraction in measurement-while-drilling applications.