Accuracy of Service Area Estimation Methods Used for Critical Infrastructure Recovery

Okan Pala, David C. Wilson, Russell Bent, Steve Linger, James ArnoldCritical Infrastructure Protection 2014 · DOI

TL;DR

This paper evaluates four methods for estimating the service areas of critical infrastructure (specifically electric power substations) when direct data is unavailable. It finds that weighted approaches generally outperform their unweighted counterparts, with weighted cellular automata showing the best aggregate statistical accuracy and weighted Thiessen polygons excelling in point-specific accuracy.

What Problem Does This Paper Address?

Accurately understanding the service areas of critical infrastructure, such as power substations, is crucial for effective disaster response and recovery. When infrastructure damage occurs, decision-makers need to know which consumers are affected to prioritize restoration efforts. However, the precise source-sink relationships (which substation serves which consumer) are often proprietary or unavailable, especially during emergencies. This forces reliance on estimation methods. The challenge lies in quantifying the accuracy of these estimation methods to ensure reliable decision-making.

Methods

The paper compares four methods for estimating electric power substation service areas:

  • Thiessen Polygons: A standard geometric method that divides a plane into regions based on the nearest point.
  • Weighted Thiessen Polygons: An enhancement of Thiessen polygons where the size of each region is influenced by a weight (in this case, the power output of the substation).
  • Cellular Automata (CA): A computational model that simulates systems by updating the states of cells in a grid based on local rules.
  • Weighted Cellular Automata: A CA approach where cell growth or influence is weighted, again using substation power output.

The accuracy of these methods is evaluated using a reference dataset of a real-world power distribution network in a mid-sized U.S. city. Two types of accuracy are assessed: aggregate impacts (like total population or economic activity within a service area) and point data impacts (how well specific assets are correctly assigned to a service area).

Datasets

The study uses a reference dataset for a mid-sized U.S. city, which includes:

  • Transmission network data
  • Substation locations and their power outputs (in MW)
  • Reference substation service areas (defined by an expert)
  • Economic and population data from the 2010 LandScan dataset and daytime/nighttime population information.

No new benchmark datasets were introduced; the study relies on existing real-world data.

Key Insights and Ideas

  • Weighting Matters: Incorporating real-world data like substation power output significantly improves the accuracy of both Thiessen polygon and cellular automata methods. Unweighted methods are less realistic.
  • Different Strengths for Different Needs: The "best" method depends on the analysis. For broad impact assessments (e.g., total economic loss), weighted CA is superior. For pinpointing specific asset impacts, weighted Thiessen polygons are more accurate.
  • CA Edge Cases: Cellular automata can produce unrealistic results at the edges of the study area due to their growth mechanisms, which can be mitigated with boundary constraints.
  • Data Scarcity is a Major Hurdle: The paper highlights that the lack of detailed, high-quality data is the primary driver for needing these estimation methods, and obtaining such data remains a significant challenge.

Key Results

  • Aggregate Accuracy: Weighted cellular automata showed the best performance in terms of aggregate statistical accuracy, yielding the smallest mean and sum of differences for population and economic impact metrics.
  • Point Accuracy: Weighted Thiessen polygons achieved the highest overall point accuracy (68.9%), followed closely by weighted cellular automata (59.5%).
  • Proximity Confidence: For points closest to the service source (within the nearest 25% of the service area), weighted cellular automata had the highest accuracy (91%), with weighted Thiessen polygons also performing well (86%).
  • Neighborhood Relaxation: When considering neighboring polygons (allowing a point to be associated with a set of sources rather than a single one), all methods showed dramatic accuracy improvements, exceeding 95%.

Limitations and Future Work

Limitations: * The study used "out of the box" parameters for cellular automata, which may not represent their full potential. Parameter tuning could significantly improve CA performance but is highly application-specific. * The paper notes that certain situations, like substations with very large outputs serving small areas, can lead to lower accuracy in the reference data itself. * The accuracy of CA at the edges of the study area was identified as an issue, suggesting the need for boundary constraints.

Future Work: * Investigate other critical infrastructures to identify commonalities and unique aspects. * Develop new comparison metrics, such as comparing substation loads to expected consumption. * Explore formal probability-based methods to handle the inherent error and uncertainty in service area algorithms.

Summary updated

BibTeX
@inproceedings{pala2014accuracy,
  author = {Okan Pala and David C. Wilson and Russell Bent and Steve Linger and James Arnold},
  title = {{Accuracy of Service Area Estimation Methods Used for Critical Infrastructure Recovery}},
  booktitle = {Critical Infrastructure Protection},
  year = {2014},
  doi = {10.1007/978-3-662-45355-1\_12},
}
Figure