The research goal is to identify the ore grindability online in this paper. Through a detailed analysis of the interaction between the grinding process parameters,it is concluded that the change of the grinding process operation index can reflect the change of ore grindability to a certain extent. In order to solve the problems of high dimensionality and strong coupling of industrial data,on the basis of focusing on the mechanism of KPLSR and SAE two algorithms that can extract key features,the actual collected grinding process data is used as a sample to compare multiple algorithms horizontally and vertically,and finally the advantages of deep learning algorithm SAE in industrial complex data analysis are verified. The algorithm structure with the best performance in the test is used as the online recognition model of the ore grindability intelligent identification system,and the average absolute error of the model can reach 0.61 kW·h/t in industrial applications,and the average error rate is about 4.79%,which can meet the needs of on-site production.