Abstract: Aiming at the problem that it is difficult to identify the operating conditions under the conditions of strong noise, non-stationary and lack of labels in the industrial production process of vertical mill, this paper proposes a hierarchical unsupervised working condition identification method. The method takes 'data pre-cleaning-abnormal condition identification-weighted clustering' as the core, and constructs a three-level robust condition discovery architecture. Firstly, combined with the vertical milling process and field operation experience, the pressure difference of the mill and the vibration value of the vertical mill are selected as the testing parameters under operating conditions. Then, the progressive hierarchical unsupervised learning algorithm is adopted, and the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm is used to clean the shutdown and discrete noise data. The iForest algorithm is used to identify the abnormal operating conditions, and the weighted K-means + + is used to realize the clustering of typical operating conditions, so as to realize the automatic identification of all operating conditions and the generation of operating condition labels. Finally, the BO-LightGBM (Bayesian Optimized Light Gradient Boosting Machine) algorithm is used to train the generated condition labels and establish the corresponding condition classification model. Based on the industrial field historical data of a slag plant, the simulation experiment is carried out. The results show that the proposed method can accurately identify the operating conditions of the vertical mill. The accuracy of the final classification model reaches 98.83%, and the single running time is only 0.61 s. It solves the problem of difficult identification of vertical mill operating conditions under complex industrial data conditions.