The invention provides an ozone concentration distribution prediction method and system based on space-time deep learning. The method includes: obtaining an ozone concentration distribution map of a current moment, and obtaining meteorological data of a to-be-predicted moment; and processing the ozone concentration distribution map of the current moment and the meteorological data of the to-be-predicted moment through a trained ozone concentration prediction model based on the meteorological data, and obtaining an ozone concentration distribution map of the to-be-predicted moment. According tothe method and system, ozone concentration distribution map sequences and meteorology-time sequences are obtained through processing by methods including interpolation; historical data of a period isprocessed by employing a recurrent neural network, and trend characteristics of ozone concentration changes are extracted; historical data of a day ago or a week ago is processed by employing a convolutional neural network, and periodic characteristics of ozone are employed to the maximum; and the meteorological data and time data of the predicted moment are regarded as extra input, and the prediction accuracy is further improved by employing the influences of meteorology and time on the ozone.