Construction of a New Energy Settlement Index System and Market Operation Status Evaluation Model Based on Graphormer

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

🔬研究者:Provides a novel Graphormer-TCN architecture for energy market operation assessment, useful for AI×ESG research.

🏢実務担当者:Offers a tool to optimize new energy settlement and improve market revenue for industrial energy users.

🏛政策担当者:Demonstrates how AI can support market design and operational efficiency in renewable energy integration.

With the gradual introduction of renewable energy into industrial energy systems, assessing the market operation of new energy settlement mechanisms in industrial parks has become a research task of practical significance. To address the complexity and uncertainty in this process, this paper proposes an innovative Graphormer–Temporal Convolutional Network (Graphormer-TCN) fusion model, aiming to construct a robust, end-to-end tool for assessing the market operation of new energy settlement mechanisms in industrial energy systems. The study establishes a unified input format using an indicator system covering actual generation, forecast deviation, curtailment, settlement, ancillary revenue, economic benefits, capacity factor, and equipment health. The Graphormer spatial layer models the transmission topology as a weighted graph, explicitly encoding connectivity and conduction strength via self-attention biases (edge weight position encoding and node centrality) to capture non-local network effects. The TCN time series layer then utilizes optimized causal convolutions to perform multi-scale analysis on the spatial encodings, characterizing short-term mutations and daily patterns to score market operation status. The model demonstrates stable, high accuracy with low average mean absolute errors (MAE) across wind, photovoltaic (PV), and storage (e.g., 0.017, 0.014, 0.015 over 24 hours). Crucially, this enhanced model reduces the average market transaction response time to approximately 120.5 milliseconds, significantly improving the operational efficiency of clustered industrial loads in high-frequency energy management, enabling rapid response to changes in production load, and driving a 9.6% increase in market operating revenue. This effectively promotes the efficient participation of industrial energy users in renewable energy assets and low-carbon transformation. Since the model directly uses power-network topology and conduction relationships, the work has a natural connection with electromagnetic energy-network operation in industrial parks.

• semanticscholar https://doi.org/10.7716/aem.v15i3.3522first seen 2026-08-17 05:27:45 · last seen 2026-08-19 05:19:48