The traditional pattern recognition method of abnormal coal mill system has low accuracy and is difficult to meet the industrial production demand. In this paper, we propose a coal mill outlet pressure prediction and anomaly pattern recognition method based on particle swarm optimization three-dimensional residual neural network. The method firstly adopts the 3σ criterion to eliminate abnormal data and performs mean filtering preprocessing, constructs a 3D residual neural network integrating soft thresholding sub-network and distraction mechanism, and optimizes the network hyper-parameters using particle swarm algorithm. Experiments show that, compared with the traditional method, this model is significantly better than the genetic algorithm in convergence speed, shortening the training time by more than 30%; it performs excellently in prediction accuracy, with the MAE value reduced to 21.473, the MAPE reduced to 0.0333, and the RMSE reduced to 20.069, which is reduced by 10.4%, 38.4%, and 15.3% compared with the traditional ResNet, respectively. In terms of anomaly pattern recognition, the model’s recognition accuracy for current anomaly, temperature anomaly, pressure anomaly, flow anomaly, and rotational speed anomaly reaches 94.7%, 97.5%, 97.7%, 96.1%, and 98.5%, respectively, with an accuracy far exceeding that of traditional classification algorithms. The results confirm that the particle swarm optimized 3D residual neural network has significant advantages in the prediction of coal mill outlet pressure and identification of abnormal states.
