The invention provides an ozone concentration prediction method and system based on spatio-temporal data and statistical learning. The method comprises the steps that a reference feature and a reference result corresponding to each historical moment of each environment monitoring station are acquired; a training validation set and a test set are acquired; a statistical learning algorithm is used to perform model parameter selection and training on the training validation set through 10-fold cross-validation, a prediction model is obtained, and testing is performed on the test set; and according to a target feature corresponding to a current moment of a to-be-tested environment monitoring station, the prediction model and the ozone concentration corresponding to the current moment of the to-be-tested environment monitoring station, the ozone concentration of the to-be-tested environment monitoring station at the current prediction moment is obtained. Through the ozone concentration prediction method and system, the ozone concentration, meteorological conditions and point source discharge are introduced into the feature extraction process through the spatio-temporal data and statistical learning to precisely depict the reason for ozone concentration rise; and meanwhile, by the adoption of incremental analysis, compared with previous statistical models, accuracy is greatly improved, universality is high, and the ozone concentration prediction method and system are suitable for different environment monitoring stations.