Spatiotemporal Graph Learning Model for Environmental Risk Evolution and Dynamic Carbon Footprint Quantification in Power Grid Construction Projects

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:The HST-GNN framework and disturbance-driven carbon accounting model provide novel methodological contributions for spatiotemporal environmental modeling and dynamic carbon footprinting.

🏢実務担当者:Power grid construction companies can use this approach to improve environmental risk management and accurately quantify carbon emissions from land-use changes, aiding compliance with disclosure requirements.

🏛政策担当者:Policymakers can leverage the findings to mandate dynamic carbon accounting for large infrastructure projects, ensuring more accurate reporting and better alignment with net-zero targets.

The construction of Ultra-High Voltage (UHV) power grids in ecologically sensitive regions, such as the Shennongjia Forestry District, presents a critical dichotomy between infrastructure development and environmental conservation. Traditional environmental management approaches rely heavily on static Environmental Impact Assessments (EIA) and post-event compliance checks, failing to capture the dynamic, non-linear evolution of environmental risks and the “hidden” carbon emissions associated with Land Use, Land-Use Change, and Forestry (LULUCF). To address these challenges, this study proposes a novel HST-Heterogeneous Spatiotemporal Graph Neural Network (GNN) framework. By fusing multi-source data—including UAV hyperspectral remote sensing, IoT machinery telemetry, and on-site environmental sensors—we reconstruct the construction site as a dynamic heterogeneous graph. The HST-GNN model captures the complex spatiotemporal dependencies of risk propagation (e.g., soil erosion, pollutant diffusion) in non-Euclidean space. Furthermore, we develop a Disturbance-Driven Dynamic Carbon Accounting Model, which couples real-time machinery load factors with physics-based soil and vegetation carbon loss equations, filling the gap in quantifying LULUCF implicit emissions. Experimental results from a demonstration project indicate that the HST-GNN achieves a 94.2% accuracy in predicting environmental risk events, significantly outperforming baseline models. Moreover, the dynamic carbon accounting reveals that traditional static methods underestimate total carbon emissions by approximately 13.7%, primarily by neglecting soil organic carbon (SOC) mineralization induced by construction disturbances. This research provides a theoretical foundation and a practical decision-support tool for the digital and green transformation of power grid construction.

• openalex https://doi.org/10.3233/atde260355first seen 2026-06-24 05:01:01