Short-Term Forecasting of Passenger Demand under On-Demand Ride Services: A Spatio-Temporal Deep Learning Approach (1706.06279v1)

Abstract: Short-term passenger demand forecasting is of great importance to the on-demand ride service platform, which can incentivize vacant cars moving from over-supply regions to over-demand regions. The spatial dependences, temporal dependences, and exogenous dependences need to be considered simultaneously, however, which makes short-term passenger demand forecasting challenging. We propose a novel deep learning (DL) approach, named the fusion convolutional long short-term memory network (FCL-Net), to address these three dependences within one end-to-end learning architecture. The model is stacked and fused by multiple convolutional long short-term memory (LSTM) layers, standard LSTM layers, and convolutional layers. The fusion of convolutional techniques and the LSTM network enables the proposed DL approach to better capture the spatio-temporal characteristics and correlations of explanatory variables. A tailored spatially aggregated random forest is employed to rank the importance of the explanatory variables. The ranking is then used for feature selection. The proposed DL approach is applied to the short-term forecasting of passenger demand under an on-demand ride service platform in Hangzhou, China. Experimental results, validated on real-world data provided by DiDi Chuxing, show that the FCL-Net achieves better predictive performance than traditional approaches including both classical time-series prediction models and neural network based algorithms (e.g., artificial neural network and LSTM). This paper is one of the first DL studies to forecast the short-term passenger demand of an on-demand ride service platform by examining the spatio-temporal correlations.

• The paper presents FCL-Net, a novel deep learning architecture that fuses convolutional and LSTM layers for accurate ride service demand forecasting.
• It demonstrates that integrating spatial, temporal, and external factors reduces RMSE by 50.9%, outperforming traditional methods.
• The approach provides actionable insights for optimizing resource allocation in ride services and paves the way for future spatio-temporal forecasting research.

Overview of "Short-Term Forecasting of Passenger Demand under On-Demand Ride Services: A Spatio-Temporal Deep Learning Approach"

This paper presents an advanced deep learning approach to address the complexities of short-term passenger demand forecasting in on-demand ride services. The proposed model, named the fusion convolutional long short-term memory network (FCL-Net), adeptly integrates the spatial, temporal, and exogenous dependences that characterize this problem. Leveraging data from DiDi Chuxing in Hangzhou, China, the research demonstrates the efficacy of the model in predicting passenger demand, outperforming traditional models.

Deep Learning Architecture: FCL-Net

The FCL-Net stands out due to its novel architecture that combines conv-LSTM and standard LSTM layers with convolutional operations. This allows the model to capture spatial correlations through convolutional layers and temporal dependencies via LSTM. By applying this integrated approach, the FCL-Net is capable of handling the dynamic nature of passenger demand, influenced by both spatial constraints and temporal behaviors.

Key steps in the model include:

• Feature Selection: A spatially aggregated random forest algorithm is used for ranking and selecting significant explanatory variables, which reduces dimensionality without notably impacting accuracy.
• Incorporation of Exogenous Variables: Alongside demand intensity, variables like travel time rate, weather conditions, and temporal markers (time of day, day of week) are included, leading to a substantial reduction in RMSE by 50.9%.

Experimental Validation and Results

The FCL-Net is tested and validated against benchmark models including historical average (HA), moving average (MA), ARIMA, ANN, and LSTM. Validation on real-world data from DiDi Chuxing reveals that FCL-Net achieves superior predictive accuracy. Specifically, the inclusion of exogenous variables significantly enhances the model's performance compared to when only historical demand data is used.

• Performance Metrics: FCL-Net models achieved a remarkable reduction in RMSE compared to traditional models and displayed a minimal performance trade-off despite reduced variable dimensionality through feature selection.

Implications and Future Directions

The research provides a comprehensive tool for operators of on-demand ride services, enabling optimized resource allocation and improved service efficiency through more accurate demand predictions. The methodological advances have implications for broader applications in transportation and other domains requiring spatio-temporal forecasting.

For future research, exploring more complex interactions involving endogenous factors or integrating real-time data streams could further enhance the model's applicability and robustness. Additionally, coupling this model with real-time adaptive strategies could provide even more pronounced benefits in dynamic environments.

In summary, this paper contributes significantly to the field of on-demand transportation by introducing a sophisticated, integrated method for passenger demand forecasting, showcasing the power and potential of deep learning techniques in addressing complex, real-world problems.