Scholarlink

<p>This study proposes an XGBoost target (XGB-T)-based reinforcement learning framework for optimizing reconstruction cycles in TDLAS-based exhaust gas temperature measurement. Residual features derived from deviations between thermocouple and TDLAS data were utilized, and predictive performance was enhanced through SMOTE-based data augmentation and hyperparameter optimization. Four experimental models were evaluated using accuracy, ROC-AUC, execution time, RMSE, MAE, and MaxAE. Final validation was conducted using deterministic and robust modes to assess reproducibility and generalizability. The results indicated that the framework achieved relatively stable predictive performance and computational efficiency. These findings suggest that the proposed framework offers a practical foundation for developing adaptive cycle selection strategies in real-time combustion diagnostics.</p>

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

Optimizing TDLAS analysis cycles for exhaust gas temperature measurement using XGBoost reinforcement learning

  • 期刊论文

  • 发表日期：

    2026年1月13日

查看全文
  • Doowon Choi,
  • Hyeonae Jang

摘要

This study proposes an XGBoost target (XGB-T)-based reinforcement learning framework for optimizing reconstruction cycles in TDLAS-based exhaust gas temperature measurement. Residual features derived from deviations between thermocouple and TDLAS data were utilized, and predictive performance was enhanced through SMOTE-based data augmentation and hyperparameter optimization. Four experimental models were evaluated using accuracy, ROC-AUC, execution time, RMSE, MAE, and MaxAE. Final validation was conducted using deterministic and robust modes to assess reproducibility and generalizability. The results indicated that the framework achieved relatively stable predictive performance and computational efficiency. These findings suggest that the proposed framework offers a practical foundation for developing adaptive cycle selection strategies in real-time combustion diagnostics.