DTTrans: PV Power Forecasting Using Delaunay Triangulation and TransGRU.

Area of Science:

• Renewable Energy Systems
• Power Systems Engineering
• Meteorological Forecasting

Background:

• High penetration of renewable energy necessitates accurate photovoltaic (PV) power forecasting for grid stability.
• PV power output is inherently uncertain due to stochastic weather conditions, posing challenges for grid management.
• Existing forecasting methods struggle with weather forecast errors, impacting reliability.

Purpose of the Study:

• To develop a novel short-term PV forecasting technique robust against weather forecast errors.
• To improve the accuracy of PV power predictions for grid integration and scheduling.
• To investigate the benefits of PV aggregation for virtual power plants using the proposed forecasting framework.

Main Methods:

• Proposed a novel forecasting technique using Delaunay triangulation with three enclosing weather stations.
• Developed a TransGRU model, combining Transformer encoder and Gated Recurrent Unit (GRU), for robust feature representation from weather data.
• Constructed a framework integrating Delaunay triangulation and the TransGRU model for short-term PV power forecasting.

Main Results:

• The proposed TransGRU framework demonstrated a 7-15% improvement in normalized mean absolute error (NMAE) compared to state-of-the-art methods.
• PV aggregation for virtual power plants showed significant error compensation, resulting in 41-60% improvement.
• The aggregated framework achieved an average forecasting error as low as 3-4%.

Conclusions:

• The Delaunay triangulation and TransGRU-based framework offers a significant advancement in short-term PV power forecasting accuracy.
• The proposed method effectively mitigates the impact of weather forecast errors on PV power predictions.
• PV aggregation within the proposed framework substantially enhances forecasting performance and grid management capabilities.