To improve the accuracy of taxi customer pick-up prediction and the operation efficiency of taxi service

a CorrelationNet with dropconnect method is proposed. The method includes two stages

spatiotemporal features identification and regularization with dropconnect. First

the prediction model adds the spatiotemporal correlation analysis mechanism to the deep neural network

and determines the spatiotemporal features for taxi demand prediction

forming a new deep learning network called CorrelationNet. Then

the prediction model uses dropconnect method to train the new deep learning network Correlationnet to reduce over fitting. Finally

a case study is carried out in Guangzhou to verify the model; support vector machine regression (SVR)

artificial neural network (ANN) and CorrelationNet methods are adpoted to evaluate the performance of the proposed method by using the same taxi GPS data. The experiments results show that this method is better than other methods and is more suitable for passenger taxi demand prediction.