Problem
Research questions and friction points this paper is trying to address.
short-term turning-movement forecasting
negative forecasts
neural networks
traffic flow prediction
LTV-CTDNet: Compositional Turning Decomposition for Short-Term Turning-Movement Forecasting
🤖 AI Summary
This study addresses the issues of negative outputs and the absence of total volume constraints in neural network-based traffic prediction by proposing a compositional turning decomposition framework. The method constructs a lightweight encoder that integrates historical data, temporal embeddings, and positional information. Combined with linear time-varying networks and LiDAR data, it decouples the prediction task into non-negative total flow estimation and proportion estimation, thereby ensuring physical plausibility by design. Experimental results demonstrate that the proposed model achieves a Mean Absolute Error (MAE) of 1.819 and a Root Mean Square Error (RMSE) of 3.807. It completely eliminates negative predictions without requiring any post-processing, enabling high-accuracy and interpretable short-term turning movement forecasting.
📝 Abstract
Short-term turning-movement forecasts can support signal control and corridor operations, but unconstrained neural networks may produce physically impossible negative counts or outputs that are not explicitly tied to an approach-demand total. This study introduces the Linear Temporal-Variable Compositional Turning Decomposition Network (LTV-CTDNet), a forecasting framework designed to combine competitive accuracy with structurally admissible outputs. LTV-CTDNet was evaluated using seven months of 15-minute LiDAR observations from eight monitored corridor locations in Nashville, Tennessee. Its lightweight encoder combines recent turning-movement history, weekly time-slot embeddings, and location embeddings. The Compositional Turning Decomposition framework separately predicts nonnegative approach totals and within-approach turning proportions, then reconstructs movement forecasts from these components. Among the evaluated predefined configurations, LTV-CTDNet achieved a movement-level MAE of 1.8189 and RMSE of 3.8072. Its accuracy gains over the strongest sequence models were modest, but it produced no negative forecasts, while unconstrained learned models generated negative values in approximately 10.6% to 29.2% of raw forecast cells. The framework enforces nonnegative outputs and exact agreement between each model-predicted approach total and the sum of its component movements by construction, providing directly interpretable forecasts without clipping or coherence correction.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Compositional Turning Decomposition
Short-Term Forecasting
Nonnegative Constraints
Physically Admissible Outputs
Lightweight Encoder
🔎 Similar Papers
No similar papers found.
