Abstract: Coal mine safety risk pre-warning is a key part of the safety guarantee in coal mining process. When recording hidden safety hazard data in underground mines,collecting relevant data is susceptible to external environmental factors(such as light,temperature,humidity,etc.) and human factors.This leads to errors in the classification of safety risk levels,which makes hidden safety hazards unable to be effectively dealt with. In order to solve problem of inaccurate classification of safety risk in coal mines and other related companies,a coal mine safety risk intelligent hierarchical control and information pre-warning system based on improved particle swarm optimization(PSO) and convolutional Neural Networks(CNN) was designed. The system adopts an intelligent data screening model based on improved PSO,and uses the feature of PSO algorithm to find the global optimal solution to filter out unreasonable data,and reduces problems of inaccurate manual calculation of data or inaccuracies in the process of collecting mining area information. The improved CNN’s intelligent risk classification model realizes high-precision classification of security risk levels through collection and fusion of data features.The experiment and application results show that system has a recall rate of 85.6% anda precision rate of 91.7%. Compared with other systems,the recall rate has increased by 4.2%,and the precision rate has increased by 2.8%,which greatly improved accuracy of security risk classification,and the timely warning effect on occurrence of security risks is significant. This method effectively reduces frequency of potential safety hazards.