Scholarlink

<p>Boilers contribute substantially to carbon emissions and pollutants in industrial sectors. Precise furnace temperature modeling is vital for combustion optimization and efficiency improvement. Yet, modeling is difficult due to the interaction between fast-changing dynamic and slow-varying static data. This paper presents a hybrid framework for boiler temperature modeling (HFBTM). HFBTM has several advantages. Its modular structure, using a multi-layer dense network ace model (S3M) for dynamic features, allows for efficient handling of different data types. The hybrid feature fusion module, with weighted integration, comprehensively uses features, enhancing multi-step temperature prediction accuracy. Compared to traditional single-dynamic models, HFBTM reduces information redundancy, curbs error propagation, and integrates static and dynamic features end-to-end. Experimental results show that in a 30-step prediction, HFBTM reduces RMSE by 4.6% and MAE by 2.4%, validating its effectiveness.</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

HFBTM: Hybrid Framework for Boiler Temperature Modeling via Static-dynamic Feature Decoupling and Weighted Fusion

  • 期刊论文

  • 发表日期：

    2025年9月9日

查看全文
  • Yusen Gang,
  • Chen Peng,
  • Chuanliang Cheng

摘要

Boilers contribute substantially to carbon emissions and pollutants in industrial sectors. Precise furnace temperature modeling is vital for combustion optimization and efficiency improvement. Yet, modeling is difficult due to the interaction between fast-changing dynamic and slow-varying static data. This paper presents a hybrid framework for boiler temperature modeling (HFBTM). HFBTM has several advantages. Its modular structure, using a multi-layer dense network ace model (S3M) for dynamic features, allows for efficient handling of different data types. The hybrid feature fusion module, with weighted integration, comprehensively uses features, enhancing multi-step temperature prediction accuracy. Compared to traditional single-dynamic models, HFBTM reduces information redundancy, curbs error propagation, and integrates static and dynamic features end-to-end. Experimental results show that in a 30-step prediction, HFBTM reduces RMSE by 4.6% and MAE by 2.4%, validating its effectiveness.