DTAE-CCP: Decoupled Truck Activity Encoder with Causal Cascade Prediction for Truck Stop Behaviors

Accurate prediction of truck stop behavior is essential for freight transportation analysis and logistics management. However, many existing methods neglect the operational decision logic that governs the transport activities. For example, they treat stop location and stop duration as separate prediction targets, or they rely on uniform sequence modeling architectures that cannot adequately represent multi-scale temporal patterns or freight-specific operational semantics. To overcome these challenges, this paper introduces DTAE-CCP, a decision-aligned framework for truck stop behavior prediction that embeds freight operational logic into the representation and sequential prediction process through domain-aware truck activity encoding and structured sequential prediction. The framework uses a decoupled truck activity encoder that integrates heterogeneous temporal features and periodic operational patterns to characterize both long-term behavioral regularities and short-term driving dynamics, alongside a causal cascade prediction architecture that explicitly models the sequential dependence from driving time to stop location and then to stop duration while ensuring spatial feasibility. Experiments on large-scale real-world freight trajectory datasets show that the proposed method achieves the best observed performance among the compared representative baselines across the reported evaluation metrics, and ablation plus sensitivity studies indicate that aligning the architecture with freight decision logic, reinforced by domain-specific representation learning, is the primary contributor to the performance gains.