The pre-combustion chamber (PCC) is commonly used to ensure stable combustion in boilers. However, when a coal-fired boiler uses a PCC combustor, the cross-sectional area and volumetric heat load in the PCC are high, which is prone to slagging, affecting the safe and stable operation of the boiler. Therefore, developing a fast and accurate prediction model is very important for judging the degree of slagging on the wall of the PCC. In recent years, artificial intelligence (AI) has been widely used in the field of thermal engineering, especially in the prediction of slagging. However, currently, using neural networks to predict the degree of boiler slagging only inputs simple parameters such as silicon ratio and acid–base ratio, without considering the actual complex flow and combustion characteristics in the furnace. In order to improve the accuracy of boiler slagging prediction, a deep parallel residual convolution neural network (DPRCNN) is proposed for automatically identifying three types of boiler wall slagging degrees. First, we simulate the boiler combustion process under various operating and structural parameters and output a dataset. Second, experimental validation is used to numerically simulate typical operating conditions, verifying the accuracy of the resulting dataset, and the generated dataset is sent to the DPRCNN model for identification. Finally, a convolutional neural network incorporating parallel thinking is proposed to predict boiler slagging. The experimental results show that the accuracy, accuracy, and area under curve (AUC) of DPRCNN reach 100%, 100%, and 100%, verifying the applicability of deep learning technology to boiler slagging prediction.