Adaptive Modeling of Uncertainties for Traffic Forecasting

Ying WuYongchao YeAdnan ZebJames J.Q. YuZheng Wang

Mar 2023

摘要

Deep neural networks (DNNs) have emerged as a dominant approach fordeveloping traffic forecasting models. These models are typically trained tominimize error on averaged test cases and produce a single-point prediction,such as a scalar value for traffic speed or travel time. However, single-pointpredictions fail to account for prediction uncertainty that is critical formany transportation management scenarios, such as determining the best- orworst-case arrival time. We present QuanTraffic, a generic framework to enhancethe capability of an arbitrary DNN model for uncertainty modeling. QuanTrafficrequires little human involvement and does not change the base DNN architectureduring deployment. Instead, it automatically learns a standard quantilefunction during the DNN model training to produce a prediction interval for thesingle-point prediction. The prediction interval defines a range where the truevalue of the traffic prediction is likely to fall. Furthermore, QuanTrafficdevelops an adaptive scheme that dynamically adjusts the prediction intervalbased on the location and prediction window of the test input. We evaluatedQuanTraffic by applying it to five representative DNN models for trafficforecasting across seven public datasets. We then compared QuanTraffic againstfive uncertainty quantification methods. Compared to the baseline uncertaintymodeling techniques, QuanTraffic with base DNN architectures deliversconsistently better and more robust performance than the existing ones on thereported datasets.

展开全部

图表提取

论文十问

参考文献

被引用

提问与回答

暂无人提供速读十问回答

被引用

笔记

问答

领域