Development of a deep-learning based gap index for addressing demand–supply interactions in ride-sourcing services

The imbalance between demand and supply is one of the fundamental operational bottlenecks in ride-sourcing markets. Resolving these spatial mismatches requires accurate evaluation of both the gap and the related demand and supply. However, existing studies lack efficient methods to quantify gaps arising from both demand and supply. To address these issues, the study proposes a Gap Index (GI) that illustrates the relative demand–supply gaps from a global perspective under various scales, rather than merely calculating the difference between them. Furthermore, in advance GI evaluation and forecasting depends on both demand and supply and can benefit the effectiveness and efficiency of ride-sourcing operation policy, the study thus proposes a multi-task deep learning framework based model (i.e., 4-in-1 MTF model) for gap evaluation and demand and supply forecasting simultaneously. To address the intrinsic spatiotemporal correlations between demand and supply and the impact of related external factors, the proposed 4-in-1 MTF model integrates ConvLSTM, 3D-CNN, 2D-CNN, and GRU techniques to effectively extract the information of spatial correlation and temporal order dependence of demand and supply, and make the related features (e.g., weather conditions, driving speed, POI, etc.) fed into the fusion network for accuracy improvement. Experiments based on Xiamen data show that the 4-in-1 MTF model outperforms the best benchmarks, reducing RMSE by 10.0% for demand and 25.5% for supply forecasts, and 17.9% for GI. The GI is then used to identify areas with demand–supply imbalances, demonstrating the adaptability of gap spatiotemporal evaluation and flexibility across various supply scales. The spatiotemporal distribution reveals that gaps are larger during peak hours compared to off-peak hours and are primarily concentrated on the island and coastal continent. As the supply radius expands, the difference in GI values between neighboring areas decreases, providing options to adjust pick-up distance policies to help balance gap spatial distribution. Furthermore, the GI can be extended to assist in vehicle deployment policies by quantifying shortages and surpluses of idle vehicles at the study units, offering valuable insights into the management of ride-sourcing operations.