Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ProDER: A Continual Learning Approach for Fault Prediction in Evolving Smart Grids

Emad Efatinasab, Nahal Azadi, Davide Dalle Pezze, Gian Antonio Susto, Chuadhry Mujeeb Ahmed, Mirco Rampazzo

TL;DR

This paper tackles fault prediction in evolving smart grids by introducing continual learning to adapt models as fault patterns and grid zones evolve. It proposes ProDER, a Prototype-based Dark Experience Replay method that extends DER++ with Semantic Prototype Alignment and a prototype-aware memory strategy to stabilize representations and reduce forgetting. Across four class- and domain-incremental scenarios on an IEEE-13 node feeder dataset, ProDER consistently achieves the best performance among CL methods, with minimal accuracy degradation such as $0.045$ for fault-type prediction and $0.015$ for fault-zone prediction, approaching the joint-training upper bound. The work demonstrates the practicality of replay-based CL for real-world, deployment-ready fault prediction in smart grids, and highlights memory-efficient strategies and prototype-driven regularization as key drivers of robustness.

Abstract

As smart grids evolve to meet growing energy demands and modern operational challenges, the ability to accurately predict faults becomes increasingly critical. However, existing AI-based fault prediction models struggle to ensure reliability in evolving environments where they are required to adapt to new fault types and operational zones. In this paper, we propose a continual learning (CL) framework in the smart grid context to evolve the model together with the environment. We design four realistic evaluation scenarios grounded in class-incremental and domain-incremental learning to emulate evolving grid conditions. We further introduce Prototype-based Dark Experience Replay (ProDER), a unified replay-based approach that integrates prototype-based feature regularization, logit distillation, and a prototype-guided replay memory. ProDER achieves the best performance among tested CL techniques, with only a 0.045 accuracy drop for fault type prediction and 0.015 for fault zone prediction. These results demonstrate the practicality of CL for scalable, real-world fault prediction in smart grids.

ProDER: A Continual Learning Approach for Fault Prediction in Evolving Smart Grids

TL;DR

This paper tackles fault prediction in evolving smart grids by introducing continual learning to adapt models as fault patterns and grid zones evolve. It proposes ProDER, a Prototype-based Dark Experience Replay method that extends DER++ with Semantic Prototype Alignment and a prototype-aware memory strategy to stabilize representations and reduce forgetting. Across four class- and domain-incremental scenarios on an IEEE-13 node feeder dataset, ProDER consistently achieves the best performance among CL methods, with minimal accuracy degradation such as for fault-type prediction and for fault-zone prediction, approaching the joint-training upper bound. The work demonstrates the practicality of replay-based CL for real-world, deployment-ready fault prediction in smart grids, and highlights memory-efficient strategies and prototype-driven regularization as key drivers of robustness.

Abstract

As smart grids evolve to meet growing energy demands and modern operational challenges, the ability to accurately predict faults becomes increasingly critical. However, existing AI-based fault prediction models struggle to ensure reliability in evolving environments where they are required to adapt to new fault types and operational zones. In this paper, we propose a continual learning (CL) framework in the smart grid context to evolve the model together with the environment. We design four realistic evaluation scenarios grounded in class-incremental and domain-incremental learning to emulate evolving grid conditions. We further introduce Prototype-based Dark Experience Replay (ProDER), a unified replay-based approach that integrates prototype-based feature regularization, logit distillation, and a prototype-guided replay memory. ProDER achieves the best performance among tested CL techniques, with only a 0.045 accuracy drop for fault type prediction and 0.015 for fault zone prediction. These results demonstrate the practicality of CL for scalable, real-world fault prediction in smart grids.

Paper Structure

This paper contains 33 sections, 13 equations, 12 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Introduction
  3. Related works
  4. Fault Prediction
  5. Continual Learning
  6. Methodology
  7. Fault Type and Fault Zone Prediction Models
  8. Problem formulation
  9. Scenarios
  10. Scenario 1 – Fault Type Prediction with Two New Fault Classes (Class-Incremental)
  11. Scenario 2 – Fault Type Prediction with One New Fault Class (Class-Incremental)
  12. Scenario 3 – Fault Type Prediction with Known Faults in a New Grid Zone (Domain-Incremental)
  13. Scenario 4 – Fault Zone Prediction with New Zone Labels (Class-Incremental)
  14. Continual Learning Methods
  15. ProDER: Our Proposed Approach
  16. ...and 18 more sections

Figures (12)

  • Figure 1: Smart Grid Representation. Example of a smart grid that connects several key components: transmission networks, smart meters, power plants, sensors, and consumers (houses and factories). Moreover, different zones (colored circles) can be interconnected, enabling seamless energy flow and coordinated operation across the network.
  • Figure 2: Task sequence for Scenario 1: The first task includes 3 fault types, and each of the remaining 4 tasks introduces 2 new fault types, covering all 11 faults.
  • Figure 3: Task sequence for Scenario 2: The first task includes 3 fault types, and each of the remaining 8 tasks introduces 1 new fault type, covering all 11 faults.
  • Figure 4: Task sequence for Scenario 3: All tasks contain the same set of 11 fault types, but each task presents data from a different zone or domain. As a result, the next task expand the knowledge of previously encountered classes.
  • Figure 5: Task sequence for Scenario 4: The first task includes 2 fault zone classes, and each of the following 2 tasks introduces 1 new fault zone class, resulting in 3 tasks covering all fault zones.
  • ...and 7 more figures