In view of the current detection and judgment of the environmental mode of subway air conditioning system, there is still the problem of low efficiency and low intelligence degree. BP neural network optimized by Adam is designed to detect the environmental mode of subway air conditioning system. Choose three key variables: smoke concentration, carbon dioxide concentration, temperature, as a condition of environment characteristics of pattern recognition, Adam algorithm are used to optimize the gradient descent of the traditional BP neural network model, first-order moment estimation and second-order moment estimation are used to dynamically adjust the learning rate of each parameter, to speed up the model learning, improve the identification accuracy of the network, and reduce the oscillation during convergence. The experimental results show that the convergence speed of the optimized BP neural network subway environmental mode detection model is improved by 98.88%, the average number of prediction errors is reduced by 45.6%, and the oscillation is greatly reduced in the convergence process. At the same time, compared with other machine learning multi-classification models, the accuracy of the optimized BP neural network model is 99.88%, the detection running time is 12 ms, and the overall performance is better.