In order to diagnose the overheating situation of boiler metal wall pipes under the premise of ensuring real-time and accuracy, 10 measurement points prone to overheating on the high-temperature reheaters of boilers were selected as the research objects. At the same time, the influencing factors of the outlet temperature of the high-temperature reheaters in an ultra-supercritical unit were analyzed, and the correlation between each influencing factor and the metal temperature of the wall was obtained using grey correlation analysis. 22 variables with correlation coefficients greater than 0.8 were selected as research variables related to overheating. An ALSTM-ResFCN model was proposed, which was based on LSTM-FCN and added residual structures and attention mechanisms, and used focal-loss for multi-label classification model construction. Taking the high-temperature reheaters pipe wall of a 660 MW ultra-supercritical boiler as the diagnosis object, the precision rate for overheating events was 72.7%, the recall rate was 80.2%, and the F1 score was 0.76.