Abstract: The number of low Earth orbit (LEO) space debris continues to rise, rendering orbital decay prediction a critical task for space safety and reentry risk assessment. However, Two-Line Elements (TLE) exhibit high noise in the LEO region, coupled with strong nonlinearity in the decay process, which results in considerable errors in the long-term extrapolation of traditional methods. To address this issue, this study constructs an equal-altitude interval time series based on TLE data via height-time scatter fitting and fixed-altitude step inverse sampling. Concurrently, a sequence-to-sequence (Seq2Seq) gated recurrent unit (GRU) multi-step decay time prediction model is employed, leveraging an encoder-decoder structure to extract nonlinear decay patterns and achieve remaining lifetime prediction for different altitude segments. Experiments conducted on debris from the COSMOS 1408 and COSMOS 2251 breakup events demonstrate that the model attains mean absolute errors of 0.088 h, 0.142 h, and 0.587 h at initial altitudes of 250 km, 300 km, and 350 km, respectively. The relative error at 350 km is constrained within the range of 10%–25%, significantly outperforming the SGP4 model. This proposed method provides effective support for orbital congestion assessment and reentry timing analysis.