Continuous Model Adaptation Using Online Meta-Learning for Smart Grid Application

Area of Science:

• Engineering
• Artificial Intelligence
• Computer Science

Background:

• Deep learning advances offer insights into complex systems like smart grids.
• Existing predictive models struggle with real-time data adaptation due to fixed training.

Purpose of the Study:

• To propose a novel online meta-learning (OML) algorithm.
• To enable continuous adaptation of predictive models using real-time data.

Main Methods:

• Developed an online meta-learning (OML) algorithm.
• Utilized a meta-optimizer to adapt base-learner parameters in real-time.
• Compared OML against traditional machine learning (ML) and online base learning.

Main Results:

• Both ML and OML significantly outperformed online base learning.
• OML demonstrated superior performance over ML and online base learning under limited data conditions.
• OML showed enhanced adaptability when training and real-time data exhibited divergent time-variant patterns.

Conclusions:

• The proposed OML algorithm offers superior adaptability for engineering systems.
• OML effectively addresses the limitations of fixed predictive models in dynamic environments.
• OML is particularly beneficial for smart grid applications with evolving data characteristics.