Virtual Smart Metering in District Heating Networks via Heterogeneous Spatial-Temporal Graph Neural Networks

The project addresses a highly specialized niche: District Heating Networks (DHN). While the code is currently in a 'reference implementation' state (0 stars, 6 days old, tied to an arXiv paper), it represents a sophisticated application of GNNs to physical infrastructure. The defensibility is low (3) because the primary value is the research methodology rather than a hardened software product; however, the niche is defensible against general-purpose AI labs (frontier_risk: low) who lack the domain-specific hydraulic and thermal data to compete. The real threat comes from industrial giants like Siemens, Danfoss, or Schneider Electric, who already manage these networks and could integrate similar HSGNN architectures into their existing SCADA or Digital Twin platforms (platform_domination_risk: medium). The 5 forks relative to 0 stars suggest initial interest from other researchers or industry specialists. The moat here is not the code itself, but the domain-specific knowledge required to model supply/return pipes and heat exchangers as heterogeneous graph nodes.