Abstract: Limited by construction conditions, some concrete face rockfill dams have not built down- stream weirs, and only a certain number of osmometers are arranged, which can not reflect the dam seep- age situation through the leakage volume. Due to the complexity of seepage control of concrete face rock - fill dams, it is difficult to obtain the relationship between seepage pressure and seepage volume behind the concrete face by analytical method. In this study, the artificial neural network method is used to es- tablish the relationship model between seepage pressure and seepage volume of the CFRD based on its nonlinear mapping ability, and the measured values are used to verify it. The results by the leakage esti- mation method based on the training samples obtained from the three-dimensional finite element seep- age model are close to the measured results, which can provide a new way for leakage estimation of earth rock dams limited by geological conditions.

Key words: concrete face rockfill dam, seepage volume, three-dimensional finite element seepage model

摘要： 部分混凝土面板堆石坝受建设条件限制，未建设坝后量水堰，仅布置一定数量渗压计，无法用渗流量反映坝体渗流情况。由于混凝土面板堆石坝防渗复杂，难以采用解析方法获得面板后渗压和渗流量关系。本研究采用人工神经网络方法，依靠其非线性映射能力，建立面板坝渗压和渗流量关系模型，并用测量值进行验证。结果表明，基于三维渗流有限元模型获取的训练样本所建立的估算渗流量方法，其结果与实测结果接近，可以为受地质条件限制的土石坝工程提供渗流量估算新途径。

关键词: 混凝土面板堆石坝, 渗流量, 三维有限元渗流模型