Abstract: Time series analysis is essential for engineering feature extraction,model building,and threshold identification. However,conventional methods often fail to fully capture long-range dependence,quantify trend persistence,or define critical time thresholds. To overcome these shortcomings,this study developed an improved R/S analysis method based on seepage monitoring anomaly detection. The proposed approach was applied in engineering practice and validated via numerical simulation. It reveals the fractal evolution and underlying mechanism of seepage-induced deformation in earth-rock dam core walls. The results delineate the fractal evolution of seepage across three critical stages. In the incubation stage(fractal dimension D=1.31),the time series displays significant randomness,corresponding to the initial mobilization,disordered rearrangement,and Brownian motion of fine particles,with pore channels beginning to form and connect. During the formation stage(D=1.28),the series exhibits both randomness and a weak trend,marked by a non-stationary shift in fractal characteristics and the directional loss of fine particles along potential seepage paths. From development to failure(D decreasing from 1.08 to 1.03),trend dominance intensifies markedly,long-term memory vanishes,and seepage channels expand continuously until achieving full connectivity,thereby critically compromising dam safety. These findings provide a theoretical basis for defining early-warning thresholds and informing emergency decision-making against seepage failure in earth-rock dams.