Abstract: Aiming at the problem of delayed CO concentration measurement in the grate furnace system of domestic waste incineration, the correlation between the continuous images of high-temperature flue gas and the CO concentration level at the first flue of the domestic waste incineration grate furnace was studied, and a real-time monitoring method for the CO concentration level at the first flue based on a three-dimensional convolutional neural network model was proposed. Firstly, a large number of high-temperature flue gas images in the furnace and CO concentration data in the flue were obtained by high-temperature industrial cameras and high-temperature laser flue gas analyzers TDLAS to produce a “high-temperature flue gas image sequence-CO concentration level” dataset. Secondly, the data set was used to train a CO concentration level classification model based on the Slow-Fast three-dimensional convolutional neural network model. The classification accuracy of this classification model on the validation set can reach 95.40%, which is about 7.8 percentage points higher than that of the traditional single-frame image classification algorithm, and the classification result is highly stable, which better meets the actual needs of the project. Finally, the fine-tuned CO concentration level classification model was deployed on a domestic waste incineration grate furnace system, and the online effect evaluation was carried out. The model performed inference once per second. Compared with the results of high-temperature laser flue gas analyzer TDLAS and continuous flue gas emission monitoring system CEMS, it was found that the recall rate and alarm accuracy of the model reached 90.7% and 68.5% respectively, and the model achieved an early warning of about 214 seconds for all CO exceeding the standard conditions in CEMS. The comparison results proved that the proposed method had high application value and prospects in improving the environmental protection performance of municipal solid waste incinerator operation.

Key words: waste incineration grate furnace, CO concentration level, three-dimensional convolutional neural network, online monitoring