In recent years, the rapid expansion of ride-hailing platforms has reshaped the landscape of modern urban transportation systems. The growth of these services has underscored the importance of vehicle dispatching and resource allocation, making demand forecasting a crucial operational focus. Existing methods often overlook key contextual information in the urban environment when exploring the spatiotemporal dependencies of demand sequences. In this study, we propose a novel approach: Weather-Conditioned Multi-Graph Network (WGNN). This method captures the similarity of supply-demand relationships in different spatial regions and integrates weather information into the spatiotemporal dependency model for predicting ride-hailing demand. Specifically, two methods are proposed for constructing region-related graphs, utilizing spatiotemporal attention and multi-graph convolution to model spatial correlations. Weather information and supply-demand relationships are modeled through cross-attention, and multi-period information is captured through a time block. Additionally, cross-region contrastive learning is employed to dynamically capture effective information components, reducing the impact of redundant information and achieving precise demand prediction. Experimental evaluations on real-world ride-hailing order datasets demonstrate that the proposed model outperforms state-of-the-art baselines in predictive performance.