Scholarlink

The growing concern regarding the detrimental impact of ammonia (NH3) emissions from diesel vehicles on air quality underscores the necessity for precise real-world monitoring to ensure effective environmental protection. Traditional testing methods are inadequate in capturing the complete lifecycle of NH3 emissions from in-use diesel vehicles. This research introduces a novel prediction model for NH3 emissions, which combines Convolutional Neural Network (CNN) and Transformer architectures. By integrating CNN's local feature extraction with Transformer's global dependency modeling, the model achieves precise predictions of NH3 emissions under realistic driving conditions. The training data used for the model was obtained from on-road emissions measurements of a heavy-duty diesel vehicle belonging to the N3 class. Feature selection was performed using the Pearson correlation coefficient, while Bayesian optimization was applied to refine key hyperparameters and enhance model performance. The SHAP algorithm identified crucial factors influencing NH3 emissions. Results demonstrate that the proposed model outperforms traditional Long Short-Term Memory (LSTM), Random Forest (RF), and standalone Transformer models, with R2 values of 0.986, MAE value of 0.663, and MSE value of 2.285 respectively. This research presents an efficient and reliable method for monitoring NH3 emissions while providing new insights into the underlying factors driving NH3 emissions from diesel vehicles under real-world conditions.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Novel CNN-Transformer Fusion Framework for Accurate Prediction of NH3 Emissions from Diesel Vehicles

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xiaoxin Bai,
  • Xiangyang Guo,
  • Chunling Wu

摘要

The growing concern regarding the detrimental impact of ammonia (NH3) emissions from diesel vehicles on air quality underscores the necessity for precise real-world monitoring to ensure effective environmental protection. Traditional testing methods are inadequate in capturing the complete lifecycle of NH3 emissions from in-use diesel vehicles. This research introduces a novel prediction model for NH3 emissions, which combines Convolutional Neural Network (CNN) and Transformer architectures. By integrating CNN's local feature extraction with Transformer's global dependency modeling, the model achieves precise predictions of NH3 emissions under realistic driving conditions. The training data used for the model was obtained from on-road emissions measurements of a heavy-duty diesel vehicle belonging to the N3 class. Feature selection was performed using the Pearson correlation coefficient, while Bayesian optimization was applied to refine key hyperparameters and enhance model performance. The SHAP algorithm identified crucial factors influencing NH3 emissions. Results demonstrate that the proposed model outperforms traditional Long Short-Term Memory (LSTM), Random Forest (RF), and standalone Transformer models, with R2 values of 0.986, MAE value of 0.663, and MSE value of 2.285 respectively. This research presents an efficient and reliable method for monitoring NH3 emissions while providing new insights into the underlying factors driving NH3 emissions from diesel vehicles under real-world conditions.