扁平化科技风交通预测模型流程图插画素材

Design a publication-ready Figure 1 for a self-supervised spatio-temporal heterogeneity framework for traffic forecasting. Show original graph G and augmented graph G̃ via adaptive augmentation (traffic masking, topology perturbation). Both are encoded by a shared ST Encoder with STIDGCN-style spatio-temporal interaction (STIG), producing H and Ĥ. Include a zoom-in ST Encoder: temporal split (odd/even), TSConv, diffusion graph convolution with dynamic graph generation, bidirectional odd-even interaction, soft gated fusion, interleave, and final TSConv. Add spatial contrastive learning (region-level, soft clustering; positives: same region across views) and temporal contrastive learning (timestamp-level; positives: same time, negatives: distant time). Predict traffic from H with supervised loss L_p and jointly optimize L_p, L_s, and L_t. Use clean academic style, left-to-right layout, rounded modules, solid/dashed arrows, pastel colors, white background, SVG/PDF, 16:9, ≥300