Abstract: To solve the problem of less attention to spatial factor and multidimensional influence factor in parking occupancy rate prediction, the Granger causality model was expanded and a parking analysis model that involves the spatial-temporal correlation and multi-dimensional influencing factors was proposed. Based on the model, a prediction algorithm for parking occupancy rate was designed by using neural network. The simulation experiments were conducted on the CityPulse dataset provided by the European Union FP7 project. Results show that neural network learning prediction methods based on multidimensional spatial-temporal causality have improved the prediction accuracy compared to other prediction methods based on spatial-temporal correlation. In a sample of parking lot with a capacity of 56 cars, the mean absolute error of parking occupancy rate prediction after 30 minutes and after 1 hour are 2.488 and 3.418, respectively, and prediction results with absolute error less than 20% and 10% account for a larger proportion of all predictions.