Metro is an essential component of urban transport and has received attention for its intelligibility and sustainability. In this paper

a self-feedback wind energy harvester (SWEH) based on a piston wind pavilion is designed consisting of four modules

namely: a wind energy harvesting module (WEHM)

an energy conversion module (ECM)

an energy storage module (ESM)

a machine learning module (MLM). The SWEH collects the wind energy passing through the piston wind kiosk and gathers information about the voltage generated by the inlet rotor. The MLM module uses a deep learning model based on Convolutional Neural Networks (CNN) to enable SWEH to recognize real-time ventilation status. Experiments show that the maximum output power of the prototype is 0.7 W and the maximum average power is 781 mW. Feedback to get the predicted airflow information and feedback system device of the electrical signal on the deep learning network can reach a 99.92% recognition rate. The experimental results show that the SEWH can collect redundant piston wind energy passing through the metro piston kiosk for application to the appropriate sensors

and the feedback from the SWEH possesses the potential to provide a form of identifying information for energy saving in metro air conditioning.