An Improved Load Forecasting Method Based on the Transfer Learning Structure under Cyber-Threat Condition.

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Smart grids enhance power system efficiency and reliability using AI and communication tech.
• Cyber threats pose a significant risk, potentially causing data loss and disrupting grid operations.
• Accurate load forecasting is crucial for effective power planning and scheduling.

Purpose of the Study:

• To propose a transfer learning-based framework for smart grid scheduling.
• To reduce reliance on local data for grid operations, enhancing resilience to cyber attacks.
• To achieve low operating costs for smart grid scheduling.

Main Methods:

• A transfer learning approach for power forecasting to ensure high-quality load prediction with minimal data.
• An adaptive time series prediction method addressing covariate shift for improved model generalization.
• An optimal economic power scheduling model for day-ahead operations, incorporating a shared energy storage station.

Main Results:

• The proposed framework demonstrates high-quality load prediction capabilities even with limited local training data.
• The adaptive time series method enhances the generalization ability of the forecasting models.
• The integrated approach enables cost-effective day-ahead scheduling for smart grids.

Conclusions:

• Transfer learning offers a robust solution for smart grid scheduling challenges, particularly in data-scarce or cyber-attack-prone environments.
• The developed framework improves load forecasting accuracy and operational efficiency.
• This approach contributes to more reliable and economically viable smart grid operations.