Abstract: The steam produced by boiler fuel combustion has a very important role in industrial thermal power plants. Predicting the amount of steam produced according to the working conditions of boiler is conducive to the real-time monitoring of boiler combustion efficiency. Useful features of the boiler data are extracted by preprocessing and feature engineering, a variety of machine learning algorithm models are constructed for the extracted features, and the multiple models are merged. The results show that the fused model of multiple machine learning algorithms is superior to the single model in accuracy, and the root mean square error of steam predicted by the fused model is 0.106, providing important reference for the real-time industrial monitoring of boiler combustion efficiency.

Key words: data preprocessing, feature engineering, machine learning algorithm, model fusion