To be in accordance with the thinking habits of human and improve the efficiency of reasoning, thisstudy argues to stratify weighted bases. The study shows that the existing compilation approach to non-stratifiedweighted bases can also be applied to COMPILE stratified weighted bases; however, its time and space costs arerelatively high because of redundant information in the compilation results. The paper proposes a novel compilationapproach, which can remove the redundant information in the process of compilation, and presents two optimizationtechniques to further improve the time efficiency. As with the existing approach, re-compiling a stratified weightedbase is not required whenever the weights associated with soft constraints change with time. The approach is testedby compiling random instances into ROBDD-normal bases, and the preliminary experimental results show that thetime and space costs of this approach are lower than the existing approach for most instances.