Modeling and prediction of key parameters of circulating fluidized bed boiler based on Transformer

Abstract

BackgroundA data-driven approach that integrates Transformer and convolutional neural network is proposed to address the challenge of traditional data-driven models, which often exhibit significant deviations in predicting continuous processes. MethodsThe proposed method uses the operating data of a circulating fluidized bed boiler over a specific time period as the input for the model. Firstly, the Transformer is used to extract dynamic features from the first half of the input data. Then, the real-time feature expression is enhanced in the second half of the input data by combining it with the convolutional neural network. Finally, the fused feature is obtained using the deep neural network and utilized to predict key parameters. Significant findingsThe experimental results show that fusing the features extracted by the Transformer and the convolutional neural network can reduce the impact of boiler operation time delay and ensure better performance.