Predicting thermal efficiency of natural gas boilers using a deep learning ensemble model

Abstract

The purpose of this study is to develop and validate a high-accuracy and robust prediction model for the thermal efficiency of natural-gas-fired boilers using real plant data. To this end, a novel RNN–CNN stacking ensemble model is proposed, in which Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs) serve as base learners and a Gradient Boosting Regressor (GBR) is adopted as the meta-learner. Using 1 year of historical operational data collected from three natural-gas boilers in a cogeneration plant, the ensemble model was trained and benchmarked against the individual RNN and CNN models. The outcomes of the study demonstrate that, compared with the RNN, the ensemble model achieves average reductions of 23.2% in RMSE, 48.1% in MAPE, and 76.5% in MAE. Furthermore, compared with the CNN, it yields average reductions of 21.0% in RMSE, 50.90% in MAPE, and 80.6% in MAE. The outcome of the study indicates that the proposed ensemble model can provide accurate boiler-efficiency predictions and has good potential for supporting online efficiency monitoring and operational optimization in natural-gas-fired power plants.