.png){kind=icon}
.png){kind=icon}
Return
Task-Oriented Predict-Then-Optimize Framework for Channel Prediction and Handover Decision
.png){kind=icon}
PRE
AI
DOI：10.1109/TCE.2026.3685113.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Reliable and real-time communication is essential for train autonomous control systems, yet ensuring seamless handover remains challenging in urban rail environments with rapidly varying channels and short overlap regions. To improve handover reliability, recent studies have introduced data-driven channel prediction models to anticipate signal variations and support proactive decision-making. However, these prediction-based methods are primarily evaluated using signal-level metrics—such as the signal-to-interference-plus-noise ratio (SINR) and reference signal received power (RSRP) prediction errors—which do not necessarily align with system-level objectives, including handover success rate and interruption time. This mismatch between prediction metrics and operational objectives leads to suboptimal decision outcomes. To address this gap, this paper proposes a Task-Oriented Predict–then–Optimize (TPO) framework that couples channel prediction and handover decision within a unified, differentiable learning pipeline. An Enhanced Informer predictor captures long-term temporal and boundary-aware signal dynamics, while an optimization-based solver generates handover actions under hysteresis and outage constraints. A learnable surrogate network bridges the non-differentiable task losses with gradient-based optimization, aligning forecasting behavior with decision quality. Simulations on urban rail scenarios demonstrate that the proposed framework reduces handover latency and interruption time while improving overall communication stability.
Keywords:
Task-Oriented Predict-then-Optimize (TPO)
channel prediction
handover decision
handover optimization
urban rail transit
Journal
IF:
10.9
Papers:
5.1K
Citations:
6.8K