Project Overview
This project focuses on developing reliability evaluation techniques and management frameworks for load prediction models in distribution lines integrated with renewable energy. By combining machine learning, explainable AI (XAI), and MLOps methodologies, we aim to ensure stable and sustainable operation of prediction models in real-world power grids.
β’
Research Subject: Research on the reliability evaluation technique and management methodology for load prediction models of distribution lines integrated with renewable energy resources
β’
Program: Basic Research
β’
Funding Agency: Korea Electric Power Corporation (KEPCO)
β’
Research Period: Nov 2022 β Oct 2024 (24 months)
β’
Research Background
With the rapid growth of renewable energy such as solar and wind, power grid operation has become increasingly complex and uncertain. Traditional load prediction models often fail to maintain stable performance under volatile conditions. While machine learning and deep learning have improved prediction accuracy, the lack of reliability evaluation and lifecycle management has limited their sustainable deployment.
This project addresses these issues by introducing explainable AI for interpretability and MLOps frameworks for continuous monitoring and management, ensuring that prediction models remain trustworthy and adaptable in dynamic grid environments.
Research Objectives
1.
Load Pattern Analysis: Extract and profile distribution line load characteristics and perform clustering to identify representative patterns.
2.
Explainable AI (XAI) for Model Interpretation: Apply XAI techniques to analyze and visualize the relationship between load features and model predictions.
3.
Reliability Metrics Development: Design new reliability indicators for evaluating stability, accuracy, and sustainability of load prediction models.
4.
MLOps-based Lifecycle Management: Build a framework for continuous monitoring, evaluation, retraining, and improvement of deployed models.
5.
Expected Impact: Provide reliable, explainable, and cost-efficient load prediction solutions that support stable renewable energy integration and smart grid optimization.
