NO_x Emission Prediction of Coal-fired Boilers via Time-delay Features and Ensemble Learning

Accurate prediction of NO_x emission concentration is of great significance for optimizing combustion processes, enhancing energy utilization efficiency, and mitigating environmental pollution. An integrated prediction method for NO_x emission concentration at the outlet of coal-fired boilers by incorporating variable time delay characteristics was proposed. Firstly, the maximum mutual information coefficient was used to identify the time-delay relationships between influencing variables and NO_x concentration, which enabled the reconstruction of the modeling dataset. Then, based on the reconstructed dataset, the LightGBM regression model was employed for feature selection, and redundant information was removed using the Pearson correlation coefficient to retain only the key variables. On this basis, an ensemble learning framework was established, in which convolutional neural networks, bidirectional long short-term memory networks, and extreme gradient boosting were adopted as base learners, while an extreme learning machine served as the meta-learner. The Kepler optimization algorithm was further introduced to optimize model hyperparameters. Experimental validation was conducted by using real operational data from a coal-fired power plant in Hebei Province. Results show that the proposed method significantly enhances prediction accuracy, achieving a mean absolute error of 0.38×10^-6, and a root mean square error of 0.75×10^-6 in terms of volume fraction, demonstrating strong robustness and practical applicability. The model effectively improves the prediction performance of NO_x concentration, and provides solid theoretical support and reliable technical assurance for combustion process optimization control and emission monitoring.