Correctly assessing seepage amount of the regulated reservoir is vital to water resources efficlent utilization and engineering safety. A regulated reservoir in mudstone area without seepage control measures was used as the engineering prototype. Based on the mutual verification of field monitoring and numerical simulation, the estimation formula of the seepage amount of the regulated reservoir was determined. In addition, the sensitivity analysis of the factors affecting the seepage amount of the regulated reservoir was carried out, and the seepage prediction models were established based on the different optimization algorithms. The research results indicate that the regulated reservoir without lying geomembranes exists in a certain amount of seepage every day, accounting for about 7 to 10 % of the total reservoir capacity. The sensitivity ranking of influencing factors in descending order is as follows: reservoir bottom width, water depth, hydraulic conductivity, and saturated volumetric water content. The prediction accuracy of Bayesian regression is significantly better than that of traditional regression models under small training samples. This research approach provides a highly accurate and strongly robust solution to the small sample engineering prediction problem.
Niu et al. (Sat,) studied this question.