Search

A novel channel-selection convolutional neural network (CS-CNN) is proposed to predict NOx emission from coal-fired boilers under steady-state and transient load conditions. First, a new channel-selection convolutional layer (CS-CL) is presented to replace regular convolutional layer (RCL). The CS-CL evaluates the channel importance of the input variables, selects the Top-C important channels and releases the hyperparameters of the remaining low-importance channels, thus contributing to maximize the utilization of the parameter resources of the model. The advantages of using CS-CLs are the preservation of the great majority of manipulated variables involved in combustion control among the input variables and the prevention of the model overfitting problem due to the redundancy of input variables. Second, a sliding window-based preprocessing method is applied to the historical data of the boiler which is divided into four-dimensional (4D) tensors. Then, comparative tests are performed on long short-term memory (LSTM) model, baseline CNN and CS-CNN using the historical data of a 670 MW boiler. The results of tests showed that CS-CNN has higher prediction performance. Finally, in order to increase the interpretability of the deep learning black box model, this study analyzes the working mechanism of the CS-CNN through ablation analysis and visualization of model parameters. • Channel importance assessment of input variables by channel selection convolutional layer. • Data-driven model for complete burner manipulation variables. • Novel convolutional neural network for predicting NOx emission from coal-fired boilers. • Model visualization analysis of convolutional neural network for industrial time series data. [ABSTRACT FROM AUTHOR]

The selective catalytic reduction (SCR) denitrification efficiency of coal-fired boilers is used to assess the clean production of power plants. Accurate predictions of NOx emissions at the SCR inlet can help improve the denitrification efficiency. However, the generation and distribution of NOx in a furnace involves complex combustion, thermodynamics, and fluid mechanics processes. The physical-based method for NOx emission models is complex and requires expensive computational resources. To address these issues, deep learning was employed to develop a prediction approach based on the random forest (RF) algorithm and lightweight convolutional neural network (CNN). To avoid curse of dimensionality, the random forest algorithm was utilized to select the importance of the candidate variables. The selected variables were in line with the boiler NOx generation and combustion control trends. The lightweight CNN was developed to build the model that predicts NOx emissions. The key was that a Cross-Channel Communication (C3)-Block was introduced into the lightweight CNN to resolve the issue in which there was no communication between feature maps in the same convolutional layer. This model was named C3–CNN. The predictive effectiveness of the C3–CNN-based model was assessed on the real historical dataset of a 600 MW down-fired boiler. The experimental results indicated that the proposed method reduced the model depth of the CNN (saved 3 convolutional layers, 1.5 times parameters, and 1 times FLOPs than 6-layer CNN) while also ensuring the model prediction accuracy (RMSE = 13.52704 ± 0.21036 mg/m3, MAE = 9.89313 ± 0.18288 mg/m3, R 2 = 0.93337 ± 0.00222), which was suitable for the online optimization of industrial pollutant control and may assist cleaner production. [Display omitted] • A novel regression approach based on convolutional neural networks (C3–CNN) is proposed. • C3–CNN can make NOx emission predicts for coal-fired boilers. • C3–CNN improves efficiency and accuracy of NOx emission predicts under steady-state and transient conditions. • C3–CNN's lightweight framework encourages cross-channel communication between feature maps of the same convolutional layer. • Random forest algorithm is utilized to prioritize the model's manipulated variables for high-dimensional industrial variables. [ABSTRACT FROM AUTHOR]