In this paper, the spatio-temporal data mining is studied by adopting the ontology theory. The method of spatio-temporal data mining based on ontology is presented and a new spatio-temporal forecasting model based on ontology is designed. The model uses two measures of spatio-temporal autocorrelation: spatio-temporal adjacency matrix and spatio-temporal contiguity measure. We use this model in the forecast about medium and long term coarse granularity in the railway passenger flow, and the results show that using the method of spatio-temporal forecasting based on ontology can obtain a good forecasting precision.