Abstract: Long-term big data monitoring of atmospheric corrosion behavior of Q345 carbon steel was carried out in 4 typical atmospheric environments in China by electric couple atmospheric corrosion monitoring sensor. Environmental data such as temperature, relative humidity, and air pollution index were recorded. The relationship between carbon steel corrosion behavior and environmental factors in outdoor dynamic atmosphere were investigated, and the key factors affecting the atmospheric corrosion behavior of carbon steel in the environmental factors were obtained by machine learning. A test method for predicting the corrosion behavior of carbon steel through environmental factors was established. The results show that relative humidity and temperature were the most important factors affecting the corrosion of carbon steel. Machine learning could predict the corrosion trend of carbon steel through environmental factors, but there was still a certain error between the predicted value and the actual value. So it was necessary to repeat training the model and constantly revise the algorithm.