The hydrogeological structure of Xietan landslide is very complex; its slidebody

slipband and sliding-disturbed zone can be regarded as porous continuous medium

but its infiltrative bedrock is fractured. An equivalent continuum model is used to model fractured rock; and a finite element model of saturated-unsaturated seepage analysis of Xietan landslide is established. Then back-propagation neural network is used to substitute finite element seepage analysis; and saturated permeability coefficient of every stratum is optimized by genetic algorithm based on monitoring data during reservoir filling. The saturated permeability coefficient of slidebody by back analysis is 4.89×10-2 cm/s

which is consistent with field testing result(1.78?10?2 cm/s to 3.2×10?2 cm/s); the saturated permeability coefficient of slipband by back analysis is 4.66×10?5 cm/s

which is very larger than indoor testing result (2.74×10?7 cm/s to 5.73×10?7 cm/s). The scale effect of soil sample is reflected by this difference.