关键词: 液化概率等级, 场地液化评估, 机器学习, 标贯

Abstract: Currently, earthquake liquefaction probability evaluation methods generally suffer from complex formula forms, inconsistent probability calculation values among multiple formulas, and insufficient significance of liquefaction probability at the site level. Based on these issues, after evaluating the consistency and discriminatory power of multiple probability formula methods, reasonable probability calculation values for existing measured historical samples were selected to calibrate the liquefaction probability levels of the samples. A preliminary liquefaction prediction model was established using the classification tree method and then optimized based on China's seismic design recommendations to develop a new method for liquefaction probability level evaluation. The new method is divided into two parts: single borehole and site liquefaction probability evaluation, which can provide a quantitative basis for liquefaction risk assessment in practical engineering. Compared with existing methods, the new method omits the traditional formula calculation process, making the evaluation process simpler and the results more intuitive. New Zealand earthquake liquefaction data and an actual engineering site were selected as test samples to validate the reliability and rationality of the new method at both the single borehole and site levels. The verification results show that the new method achieves a 93% accuracy rate for single borehole predictions and demonstrates better rationality in site evaluations compared to existing norms. The new method can provide methodological support for evaluating liquefaction probability at engineering sites and has practical engineering significance.

Key words: liquefaction probability levels, site liquefaction evaluation, machine learning, standard penetration test