Abstract: The airflow temperature at the bottom of the mine intake shaft is an important node in the thermal calculation of underground airflow. In order to accurately predict the airflow temperature of shaft with water dropping, this paper utilized Pearson correlation coefficient analysis and genetic algorithm（GA） to optimize the coupling of the method of BP neural network. Using Pearson correlation coefficient analysis, three main feature variables are selected as input variables the BP neural network. GA is used to optimize the weights and thresholds of the BP neural network, and compared with the standard BP neural network prediction model. The research results show that the average absolute percentage errors of the prediction results of the standard BP neural network prediction model with all feature variables as input variables and the standard BP neural network prediction model based on feature variable screening as input variables are 1.25% and 2.33%, respectively, while the average absolute percentage errors of the prediction results of the GA optimized BP neural network prediction model are 0.97% and 2.21%, respectively. The GA-BP neural network prediction model has higher prediction accuracy than the standard BP neural network prediction model. The prediction model based on feature variable optimization maintains high prediction accuracy, it has also improved prediction efficiency.