Modal decomposition integrated model for ultra-supercritical coal-fired power plant reheater tube temperature multi-step prediction

Abstract

Overheating of reheater tubes in ultra-supercritical coal-fired power plants can affect the efficiency and safety of power generation (PG). To avoid reheater tubes overheating and bursting, this study proposes a modal decomposition integrated model (MDIM) for multi-step prediction of the reheater tube temperature to help managers adopt appropriate measures based on the predicted temperature changes. Considering the non-smooth and non-linear nature of the original temperature data, this study applies complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) to decompose the temperature data to remove noise and extract non-linear features effectively. In this study, the residual network18-convolutional block attention module (ResNet18-CBAM), transformer, gate recurrent unit (GRU), and temporal convolutional network (TCN) are applied to predict each component with different degrees of complexity after decomposition. The results of each component are integrated by the multilayer perceptron (MLP). The proposed MDIM is evaluated with various metrics. For single-step prediction, mean absolute error (MAE), mean absolute percentage error (MAPE), and root mean square error (RMSE) are 0.037, 0.0062, and 0.066, respectively. In the case of 48-step prediction, the corresponding values are 1.97, 0.33, and 2.4. Therefore, the proposed MDIM achieves outstanding results in both single-step and multi-step prediction.