PSR-BiTCN Combined Model for Predicting the Fouling Status on the Heating Surface of the Reheater

This paper proposes a method that combines phase-space reconstruction and bidirectional temporal convolution networks to accurately predict the degree of ash fouling on the heating surface of the boiler reheater. First, the phase-space reconstruction method maps the original one-dimensional chaotic time series into a high-dimensional phase space to analyze its intrinsic nonlinear dynamics. Then, the bidirectional temporal convolution network uses the reconstructed sequence for time series prediction. Finally, the prediction results are evaluated by evaluation indicators such as the root mean square error. The results show that the PSR-BiTCN model not only improves prediction accuracy by 14.3668% compared to the traditional single neural network model; but also reduces prediction error by 6.18226%. While verifying the rationality of the model, it also lays a theoretical foundation for the subsequent transition from time-based soot-blowing to state-based soot-blowing models.