Sign in Subscribe
Machine learning

Machine Learning Model Predicts NOx Emission Concentration with High Accuracy

Strategic Insights

— 2 min read

Researchers at Taiyuan University of Technology have developed a fusion model that combines mechanism-based and machine learning methods to predict NOx emission concentration in CFB units with high accuracy. The model uses a one-dimensional semi-empirical model to simulate hydrodynamics, combustion, and NOx generation in the furnace, and a gated recurrent unit (GRU) neural network to fine-tune the initial NOx emission concentration. The results show that the proposed model is superior to single mechanism models and other neural networks, and the combination of mechanism and machine learning methods enables the fusion model to have both high prediction accuracy and physical significance.

Key Takeaways:

  • The deep peak shaving of thermal power units has raised higher requirements for ultra-low emission of the units, making accurate monitoring of NOx emission concentration in CFB units a prerequisite for NOx emission control.
  • Mechanism-based NOx emission concentration prediction methods are explanatory but may have large prediction errors due to their inability to simulate the thermal inertia of boilers well sometimes.
  • Machine learning-based methods have high prediction accuracy but poor interpretability due to lack of physical significance.
  • The proposed fusion model uses a one-dimensional semi-empirical model to simulate hydrodynamics, combustion, and NOx generation in the furnace and predict initial values of NOx emission concentration.
  • A parameter optimization method is introduced to make the mechanism model more realistic, considering that CFB units often operate in off-design conditions to meet peak shaving requirements.
  • The gated recurrent unit (GRU) neural network is introduced as an error correction model to fine-tune the initial NOx emission concentration.
  • The fusion model is demonstrated to be effective in both steady state and dynamic process, and the results show that the proposed model is superior to the single mechanism model, GRU model, and other neural networks.
  • The proposed model has both high prediction accuracy and physical significance, enabling the combination of mechanism and machine learning methods in NOx emission concentration prediction.
  • The research has been peer-reviewed and the results are published in the Energy journal.

Statistics:

  • 2025: The year the research was published in the Energy journal.
  • 330: The issue number of the Energy journal where the research was published.
  • 1335: The page number of the Energy journal where the research was published.
  • CFB units: 2 units were used as research objects to demonstrate the effectiveness of the proposed model.
  • 1-dimensional semi-empirical model: Used to simulate hydrodynamics, combustion, and NOx generation in the furnace.
  • GRU neural network: Introduced as an error correction model to fine-tune the initial NOx emission concentration.

Sources:

  • NewsRx. Investigators from Taiyuan University of Technology Zero in on Machine Learning (A Nox Emission Concentration Prediction Method for Cfb Unit Based On One-dimensional Semi-empirical Model Corrected By Gru Network). Journal of Engineering. September 1, 2025; p 1335.
  • Energy. A Nox Emission Concentration Prediction Method for Cfb Unit Based On One-dimensional Semi-empirical Model Corrected By Gru Network. Volume 330.
  • Pergamon-elsevier Science Ltd, The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, England. (Elsevier - www.elsevier.com; Energy - www.journals.elsevier.com/energy/)

Read more