Bus passenger volume forecasting is the basis of traffic planning and dispatching, but weather and temperature often have a great impact on the short-term fluctuation of passenger volume, which bring great difficulty to bus planning. With the popularity of bus smart card, it is possible to use historical bus smart card data to forecast bus passenger volume. This paper presents a prediction model of integrated learning algorithm based on XGBoost, which uses the data combine the weather, temperature and other data affecting traffic trip with Shaoxing bus smart card data, and applied to forecast Shaoxing's total passenger volume and commuter line passenger volume. Next, seven hyperparameters of the model are adjusted by GridSearchCV module. In the end, the average absolute error obtained in the verification set is 11.1146 and the accuracy of passenger volume prediction every 3 minutes reaches 95.6%, which indicates the model can predict the passenger volume more accurately.

CICTP