近期，我院本科生王梓维（第一作者），教师范伟（通讯作者）等的研究成果“Dynamic prediction of NOx generation concentration based on Kolmogorov–Arnold Network integrated deep learning method for a 660 MW coal-fired boiler”在《Energys》（IF=9.4）上发表。
论文简介如下：
本研究面向 660 MW 燃煤机组，提出将 Kolmogorov–Arnold 网络（KAN）与深度学习方法融合的 NOx 生成浓度动态预测模型。通过结合 KAN 的可解释性与深度学习的特征提取能力，构建多模型协同框架以提升预测精度与稳健性。研究采用 LSTM、Bi-LSTM、GRU 与 TCN 等方法，并利用 Optuna 优化超参数。结果表明，该方法在实际工况下优于传统模型，实现更准确的 NOx 排放预测，为锅炉优化控制与减排提供重要技术支撑。
Nitrogen oxides (NOx) are among the most significant pollutants produced by coal-fired power plants. Accurate prediction of NOx concentrations at the boiler outlet is crucial for optimizing unit control and effectively reducing emissions. This paper investigates the integration of deep learning techniques with Kolmogorov–Arnold Networks (KANs) to model the relationship between operational parameters and NOx generation concentration in a 660 MW coal-fired boiler. By leveraging the strengths of both methodologies, the study aims to enhance the accuracy and robustness of NOx generation concentration predictions. Four deep learning methods including Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), Gated Recurrent Units (GRU), and Temporal Convolutional Network (TCN) are employed to extract dynamic features from the auxiliary variables. An automated framework, Optuna, is utilized to optimize the hyperparameters of these deep learning algorithms. The extracted dynamic features are then used as inputs for KANs to predict NOx generation concentration. The results demonstrate that the proposed method outperforms conventional deep learning approaches in real-world NOx generation concentration prediction tasks, providing more accurate results. This approach opens new avenues for temporal forecasting models and underscores the potential of KANs as a powerful tool in predictive analytics.
