Affinity-Driven Transfer Learning for Load Forecasting.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Energy Systems

Background:

• Accurate load forecasting is crucial for efficient energy management.
• Traditional transfer learning methods face challenges in selecting appropriate source tasks.
• Measuring task similarity is key to effective knowledge transfer in forecasting.

Purpose of the Study:

• To introduce a novel task affinity score for quantifying task similarity in transfer learning.
• To develop the Affinity-Driven Transfer Learning (ADTL) algorithm for enhanced load forecasting.
• To demonstrate the superiority of the task affinity score over existing metrics.

Main Methods:

• Developed a task affinity score to measure similarity between diverse tasks.
• Proposed the Affinity-Driven Transfer Learning (ADTL) algorithm integrating pre-trained models and datasets.
• Validated the approach using synthetic, AEMO, and Smart Australian energy datasets.

Main Results:

• The task affinity score outperformed traditional metrics in task selection.
• The ADTL algorithm significantly improved load forecasting accuracy on unseen datasets.
• Empirical validation confirmed the robustness and effectiveness of the proposed method.

Conclusions:

• The task affinity score is a powerful tool for refining transfer learning in load forecasting.
• The ADTL algorithm offers a robust framework for accurate energy load predictions.
• This research advances the application of transfer learning in the energy sector.