Abstract
Accurately predicting nitrogen oxides (NOx) concentration in cement kilns is crucial for controlling pollutant emissions. However, due to the variable and unbalanced distribution of working conditions, achieving precise prediction of NOx concentration remains challenging. To address this issue, this study proposes an echo state network with ensemble modular outputs (EMOESN) for NOx concentration prediction under unbalanced working condition distribution. First, a two-stage working condition recognition framework is proposed. In the first stage, for the problem of multiple working conditions in cement kilns, the k-means++ clustering approach combined with calcination mechanism is adopted to generate multiple datasets with different working condition characteristics. In the second stage, to address the problem of unbalanced working condition distribution, a block bootstrap sampling method with sliding window is applied to subdivide the larger datasets into more balanced subdatasets. Second, an EMOESN is established for NOx concentration prediction during the process of cement clinker calcination, where the optimal outputs under different working conditions are determined using a variance-based weighting method. Finally, the effectiveness of the proposed prediction model is validated using a dataset from an actual cement clinker calcination process. Experimental results show that the suggested method outperforms various well-known NOx prediction methods.
Funding Information
  • National Natural Science Foundation of China (62073153)

This publication has 29 references indexed in Scilit: