Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fault Detection for agents on power grid topology optimization: A Comprehensive analysis

Malte Lehna, Mohamed Hassouna, Dmitry Degtyar, Sven Tomforde, Christoph Scholz

TL;DR

This work collects the failed scenarios of three different agents on the WCCI 2022 L2RPN environment, and proposes a multi-class prediction approach to detect failures beforehand and evaluates five different prediction models.

Abstract

Optimizing the topology of transmission networks using Deep Reinforcement Learning (DRL) has increasingly come into focus. Various DRL agents have been proposed, which are mostly benchmarked on the Grid2Op environment from the Learning to Run a Power Network (L2RPN) challenges. The environments have many advantages with their realistic grid scenarios and underlying power flow backends. However, the interpretation of agent survival or failure is not always clear, as there are a variety of potential causes. In this work, we focus on the failures of the power grid simulation to identify patterns and detect them in advance. We collect the failed scenarios of three different agents on the WCCI 2022 L2RPN environment, totaling about 40k data points. By clustering, we are able to detect five distinct clusters, identifying common failure types. Further, we propose a multi-class prediction approach to detect failures beforehand and evaluate five different prediction models. Here, the Light Gradient-Boosting Machine (LightGBM) shows the best failure prediction performance, with an accuracy of 82%. It also accurately classifies whether a the grid survives or fails in 87% of cases. Finally, we provide a detailed feature importance analysis that identifies critical features and regions in the grid.

Fault Detection for agents on power grid topology optimization: A Comprehensive analysis

TL;DR

This work collects the failed scenarios of three different agents on the WCCI 2022 L2RPN environment, and proposes a multi-class prediction approach to detect failures beforehand and evaluates five different prediction models.

Abstract

Optimizing the topology of transmission networks using Deep Reinforcement Learning (DRL) has increasingly come into focus. Various DRL agents have been proposed, which are mostly benchmarked on the Grid2Op environment from the Learning to Run a Power Network (L2RPN) challenges. The environments have many advantages with their realistic grid scenarios and underlying power flow backends. However, the interpretation of agent survival or failure is not always clear, as there are a variety of potential causes. In this work, we focus on the failures of the power grid simulation to identify patterns and detect them in advance. We collect the failed scenarios of three different agents on the WCCI 2022 L2RPN environment, totaling about 40k data points. By clustering, we are able to detect five distinct clusters, identifying common failure types. Further, we propose a multi-class prediction approach to detect failures beforehand and evaluate five different prediction models. Here, the Light Gradient-Boosting Machine (LightGBM) shows the best failure prediction performance, with an accuracy of 82%. It also accurately classifies whether a the grid survives or fails in 87% of cases. Finally, we provide a detailed feature importance analysis that identifies critical features and regions in the grid.
Paper Structure (30 sections, 1 equation, 8 figures, 6 tables)

This paper contains 30 sections, 1 equation, 8 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Contribution
  3. Related Work
  4. Methodology
  5. Descriptive Analysis with Clustering
  6. Forecasting Models
  7. Experimental Setting
  8. Agents
  9. General Data Collection
  10. Cluster Data:
  11. Forecasting Data:
  12. Metrics
  13. Results
  14. Clustering Results
  15. Cluster Identification:
  16. ...and 15 more sections

Figures (8)

  • Figure 1: 3D visualization of the five clusters, where each point represents a failure of the agents. The axis in the plot are the $gen^p_{mean}$ on the y-axis, $\#lines_{dis}$ on the x-axis and $\#sub_{changed}$ as z-axis. The points are colored according to their respective clusters.
  • Figure 2: Box plot of the survival time $t_{survived}$ of each cluster. We further report the median survival time. The ticks on the x-axis correspond to a full day, e.g., 288 steps. Note that an agent must survive a total of 2016 time steps to complete the chronics successfully.
  • Figure 3: Average probability distribution of for the ground truth of $obs_{t=survived},obs_{t=n-5},obs_{t=n-3}$ and $obs_{t=n-1}$. The probability output is averaged for all observations. Black lines visualizes the kernel density estimation across the classes.
  • Figure 4: Feature importance of the 30 most important features according to the model. The color indicates the type of variable.
  • Figure 5: Top 10 important lines (red), generators (green), and loads (yellow) for failure prediction. The numbers on the marked lines correspond to the line ids of Figure \ref{['fig:feature_importance']}. Sub-grids are separated with dotted lines. For better clarity, we grayed out the less important elements of the grid and highlighted 3 significant regions (A,B,C) of important grid features.
  • ...and 3 more figures