Abstract: At present, the humidification dynamic compaction method is widely used for foundation treatment in loess high-fill sites. However, existing specifications lack guidelines for humidification dynamic compaction parameter designs, and the humidification effect is not considered in the formula for calculating the effective reinforcement depth, resulting in insufficient theoretical guidance for on-site treatment and potential safety hazards. Therefore, based on the loess high-fill site, dynamic compaction tests are carried out under different tamping energy levels, tamping point spacing, and humidification conditions. The reinforcement effectiveness is evaluated based on indicators such as tamping settlement, tamping pit diameter, number of tamping passes, and the physical and mechanical properties of compacted soil. The parameters of dynamic compaction are optimized using a BP neural network-based proxy model and a genetic algorithm. This paper summarizes the effective reinforcement depth of soil layers under different dynamic compaction parameter designs; improves the dynamic compaction parameter design for loess high-fill sites treated by the humidification dynamic compaction method in the specification; and optimizes dynamic compaction parameters for high-fill sites. Additionally, this study incorporates the humidification effect into the Menard formula and establishes a design method with an evaluation system for dynamic compaction parameters. These findings provide an important reference for the treatment of similar loess high-fill sites.