Enhancing vehicular emissions monitoring: A raw data processing and imputation model for heavy-duty diesel vehicles using remote OBD systems

作者：Tao Li, Xin Lou, Zhuoqian Yang, Jing Zhang, Guoquan Xie, Baoli Gong, Danqi Wang, Kui Wang, Yong Peng · 发表于：International Journal of Sustainable Transportation · 年份：2025 · DOI：10.1080/15568318.2025.2538685 · 被引用次数：4 · 研究领域：Vehicle emissions and performance、Air Quality Monitoring and Forecasting、Traffic control and management

Currently, the raw data exported from the remote on-board diagnostics (OBD) monitoring systems of heavy-duty diesel vehicles exhibit significant issues with missing values in key items such as NOx concentration, posing challenges to effective emission regulation. This study proposes a systematic approach to raw data processing, provides a detailed analysis of NOx data missing patterns, and develops a weighted prediction model based on the AutoRegressive Moving Average with eXogenous variables- Long Short-Term Memory (ARMAX-LSTM) for missing data imputation. The ARMAX-LSTM model combines the capability of LSTM to capture nonlinear patterns with ARMAX’s ability to describe linear data, enhanced by dynamic weighting coefficients to improve prediction accuracy. Using Spearman rank correlation analysis, nine key parameters and the NOx downstream concentration itself were selected as input features for predicting and imputing missing values. Experimental results demonstrate that the proposed model reduces the mean squared error by 26.99% compared to the ARMAX baseline model and by 14.08% compared to the LSTM baseline model for randomly missing data. For naturally missing data segments, the model produced imputed curves with good continuity and stability, meeting engineering application requirements. This study provides technical support for improving OBD data quality and identifying high-emission vehicles, while also highlighting the model’s limitations in handling scenarios with a...