With the development and progress of smart cities, ride-hailing have high passenger demand because of high spatial and temporal flexibility. However, when ride-hailing platforms dispatch vehicles, it finds that lots of vehicles wandering areas with no demand and areas with passenger demand have no service vehicles, resulting in increased vehicle costs, reduced benefits, and increased passenger waiting times. To solve this problem, it is necessary to accurately predict high demand areas to guide service vehicles to converge to these areas earlier, which helps to balance the demand and supply of the ride-sharing platform. Therefore, we propose a hierarchical clustering application algorithm based on distance and density (HDDCA) to construct virtual service areas, and a hybrid network based convolutional and time-series models is used for demand prediction. To improve the prediction accuracy, the correlation between virtual service areas is explored and used as the input of the prediction model to predict the demand for different virtual service areas in the next interval. In addition, the attention mechanism is also used to capture external factors with different weights. In this paper, experiments are conducted on the travel data set of Haikou City in 2017, and the results show that the prediction accuracy of this method is better than that of some typical models and existing deep neural network models, which validates the effectiveness.