Predicting Ride-Hailing Service Demand via RPA-LSTM

Abstract

Ride hailing is a representative practice of rapidly developing sharing economy, which not only brings people convenience on mobility, but also benefits the utilization of vehicular resources. For ride-hailing service providers, it is necessary to supervise the demands of customers so that they could carry out reasonable dispatches. To achieve it, a number of data mining methods have been applied. But most of these predicting methods only consider this problem from the temporal aspect, neglecting the connection inter and intra blocks of the city. Blocks are the outputs of region partition, which divides the whole city into a series of divisions. At the same time, the partitioned map, specifying borders of blocks, is useful for platforms to improve the process of supervising, dispatching, and recommending. However, the majority of current partition methods is static and cannot effectively utilize features extracted from historical and geographical operational data. In this work, we propose to conduct partition task with taking distinct situations in different blocks into consideration. In particular, we propose a novel region partition assisted long short term memory neural network for ride-hailing service demand prediction. This is a two-phase model, consisting of the grid-merging region partition and the application of artificial neural network. In experiment part, we show how the region partition method assists LSTM to better forecast the demands, and compare it with other partition techniques. Finally, for evaluating the performance of RPA-LSTM, we conduct extensive experiments on a real-world dataset provided by Didi Chuxing. The experimental results clearly demonstrate that our method can effectively improve the accuracy of prediction.