Unsupervised quantitative judgment of furnace combustion state with CBAM-SCAE-based flame feature extraction.

The furnace combustion state of coal-fired power plants is difficult to accurately monitor during low-load and dynamic operation conditions, thus hindering the secure and economic operation of such power plants. In this study, a novel combustion state judgment system based on flame images of coal-fired boilers is proposed. First, an unsupervised deep learning model CBAM-SCAE, which is integrated with a convolutional block attention module (CBAM) and stacked convolutional autoencoder (SCAE), is constructed to extract combustion features from flame images. Next, the feature dimension is reduced by principal component analysis to calculate the combustion state index and fluctuation degree, which are then used to evaluate combustion stability. Finally, the results of the proposed judgment system are compared with the combustion-related variables and the flame detection system equipped in the power plant. The correlation coefficient between the combustion state index and the combustion-related variables exceeds 0.7; in addition, the unstable state of flame combustion is detected 22 s in advance by the proposed judgment system compared with the flame detection signal. Furthermore, the judgment system does not rely on manual data labeling during the evaluation process, which facilitates its practical application. • A quantitative system is defined to evaluate combustion stability of the furnace. • A CBAM-SCAE model is constructed to extract combustion features from flame images. • The attention mechanism is introduced to focus on flame global characteristics. • The judgment system does not rely on manual data labeling during the evaluation. • Unstable combustion is detected 22s in advance compared with plant equipped signal. [ABSTRACT FROM AUTHOR]