土压实参数的准确预测对岩土工程中提高土体承载力、降低可压缩性具有实际意义，现有模型在预测进度和工程适用性上存在一定局限性，且忽略了对模型预测不确定性的量化。针对具有广泛性和代表性的226组土击实试验数据，采用遗传编码算法（genetic programming, GP）对两个重要土压实参数（最优含水量和最大干密度）进行了建模预测，分别得到了最优含水量和最大干密度的最优显示模型并将预测结果与现有预测模型所得结果进行了对比分析，结合分位数回归方法和不确定度统计量量化了GP模型。结果表明，压实参数受细粒含量和塑限的影响最大，而碎石量和液限对其影响最小，因此在实际工程中，可通过优先调整土体中的细粒土含量和塑限达到最优击实效果。此外分位数回归方法（quantile regression, QR）提供了90%置信度且平均预测区间值（mean prediction interval, MPI）均小于0.3，同时绝大多数数据均落在不确定带范围内，说明GP算法具有较强的预测能力和较高的预测精度，这种可解释的显示模型更便于工程应用。

The accurate prediction of soil compaction parameters has practical significance for improving soil bearing capacity and reducing compressibility in geotechnical engineering. The existing models have certain limitations in prediction progress and engineering applicability, and ignore the quantification of model prediction uncertainty. Genetic programming (GP) was used to model and predict two important soil compaction parameters (optimal water content and maximum dry density) for 226 groups of soil compaction test data with extensive and representativeness. The optimal display models of optimal water content and maximum dry density were obtained respectively, and the prediction results were compared with the results of existing prediction models. The GP model was quantified by combining quantile regression method and uncertainty statistics. The results show that the compaction parameters are most affected by fine grain content and plastic limit, while the gravel content and liquid limit have the least influence on them. Therefore, in practical engineering, the optimal compaction effect can be achieved by preferentially adjusting the fine grain content and plastic limit, while the gravel content (CG) and the liquid limit have the least influence on them. Therefore, in practical engineering, the optimal compaction effect can be achieved by preferentially adjusting the fine grain content (CF) and the plastic limit in the soil. In addition, the quantile regression (QR) method provides 90 % confidence and the mean prediction interval (MPI) is less than 0.3. At the same time, most of the data fall within the range of uncertain bands, indicating that the GP algorithm has strong prediction ability and high prediction accuracy. This interpretable display model is more convenient for engineering applications.