Scholarlink

<p>With the increasing proportion of renewable energy in power generation, coal-fired power units are expected to play a significant role in peak shaving within future renewable energy grids. The operating parameters of coal-fired boilers frequently change, which makes traditional prediction models less effective in predicting NO<sub><i>x</i></sub> emissions. A novel hybrid model for predicting nonlinear, non-stationary and high-noise NO<sub><i>x</i></sub> emissions in power plants was introduced in this work. Complementary Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) was employed to decompose the original feature variables into smooth Intrinsic Mode Functions (IMFs) and residuals. A feature selection algorithm based on Random Forest was utilized to filter the decomposed data, reducing data dimensionality and eliminating background noise in the harsh environment of thermal power plants. Subsequently, the Attention Mechanism (AM) and Gated Recurrent Unit (GRU)-based models were trained using the screened data. Finally, a novel heuristic optimization algorithm, the Kepler Optimization Algorithm (KOA), was employed to optimize the model parameters and enhance its prediction performance. The experimental results demonstrated that the method for handling nonlinear and high-noise data yields significant performance improvements in prediction accuracy and computational efficiency when applied to various models. The hybrid model achieves a prediction accuracy of 94.1%, showing a 4.09% improvement over the GRU model, and its training time is reduced by 30% compared to the LSTM-based model. This work is enlightening for improving the prediction accuracy and modeling efficiency of NO<sub><i>x</i></sub> emissions soft measurement methods, which is expected to reduce ammonia consumption in the denitrification system and control NO<sub><i>x</i></sub> emissions within the required range.</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

A Novel Hybrid Model for Predicting Nonlinear and Non-Stationary NOx Emission of Coal-Fired Boiler

  • 期刊论文

  • 发表日期：

    2026年5月23日

查看全文
  • Zhenghui Zhou,
  • Qichao Zhang,
  • Juan Yu,
  • Zhongxiao Zhang,
  • Xiaojiang Wu,
  • Junyi Wang,
  • Xinwei Guo

摘要

With the increasing proportion of renewable energy in power generation, coal-fired power units are expected to play a significant role in peak shaving within future renewable energy grids. The operating parameters of coal-fired boilers frequently change, which makes traditional prediction models less effective in predicting NOx emissions. A novel hybrid model for predicting nonlinear, non-stationary and high-noise NOx emissions in power plants was introduced in this work. Complementary Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) was employed to decompose the original feature variables into smooth Intrinsic Mode Functions (IMFs) and residuals. A feature selection algorithm based on Random Forest was utilized to filter the decomposed data, reducing data dimensionality and eliminating background noise in the harsh environment of thermal power plants. Subsequently, the Attention Mechanism (AM) and Gated Recurrent Unit (GRU)-based models were trained using the screened data. Finally, a novel heuristic optimization algorithm, the Kepler Optimization Algorithm (KOA), was employed to optimize the model parameters and enhance its prediction performance. The experimental results demonstrated that the method for handling nonlinear and high-noise data yields significant performance improvements in prediction accuracy and computational efficiency when applied to various models. The hybrid model achieves a prediction accuracy of 94.1%, showing a 4.09% improvement over the GRU model, and its training time is reduced by 30% compared to the LSTM-based model. This work is enlightening for improving the prediction accuracy and modeling efficiency of NOx emissions soft measurement methods, which is expected to reduce ammonia consumption in the denitrification system and control NOx emissions within the required range.