Integrating Artificial Intelligence in Earthquake Arrival Process for Enhanced Seismic Probabilistic Risk Assessment

Kee, Ernie; Bui, Ha; Farshadmanesh, Pegah; Mohaghegh, Zahra

Permalink

https://hdl.handle.net/2142/121854 Copy

Description

Title

Integrating Artificial Intelligence in Earthquake Arrival Process for Enhanced Seismic Probabilistic Risk Assessment

Author(s)

• Kee, Ernie
• Bui, Ha
• Farshadmanesh, Pegah
• Mohaghegh, Zahra

Issue Date

2023

Keyword(s)

• Artificial intelligence (AI)
• Earthquake arrival
• Poisson process
• Marked point process
• Seismic probabilistic risk assessment
• Probalistic Risk Assessment (PRA)
• Nuclear power plants (NPPs)

Abstract

This paper examines the widespread assumption commonly employed in Probabilistic Risk Assessment (PRA) practice for commercial Nuclear Power Plants (NPPs) that earthquake arrivals adhere to a Poisson process. This assumption considerably influences risk management strategies in NPP design, operation, and emergency planning. Although the Poisson process assumption appears reasonable due to extended inter- arrival times and the presumed independence of successive arrivals, it fails to account for the empirically observed "clustering" behavior of earthquakes. This oversight in the current PRA methodology could result in erroneous damage probability estimates. Given the limited data available for calibrating seismic PRA models for PRA use, probability theory becomes the primary tool to decipher likely event sequences. The derived conditional probabilities represent long-run average availabilities of equipment, human actions, and other factors involved in event sequences. However, conditioning these probabilities on the predicted frequency of the triggering event may introduce potential errors in the analysis. This paper questions the validity of the Poisson arrival process assumption and suggests the consideration of self-exciting process models. Advancements in Artificial Intelligence (AI) have proved instrumental in providing a more accurate representation of localized earthquake arrival processes. However, to the best of our knowledge, no application of AI has been employed to relax the Poisson process arrival assumption in seismic PRA. The goal of this paper is to lay the groundwork for a novel approach that utilizes AI to better inform the earthquake arrival process, thereby relaxing this prevailing assumption in seismic PRA. Persistence in current assumptions, such as long-run average availabilities and the Poisson arrival process, could inevitably lead to skewed event probabilities and frequencies in seismic PRA analyses. The theoretical basis for this error is mainly associated with unrealistic assumptions regarding the average availability at the time of the seismic event arrival and the clustering nature of seismic event arrivals. Both factors could bias equipment performance during successive shocks. In response, a new approach is proposed in this paper that relaxes the need for the Poisson arrival process assumption and the Lack of Anticipation Assumption (LAA). This methodology models the NPP subject to earthquake damages using a marked point process, accommodating any arbitrary arrival process. Uncertainty can be estimated by conducting repeated independent simulations of the NPP, like lifetime testing on equipment. Given the high success rate of protective systems in a commercial NPP, either accelerated testing or the utilization of high-capacity computing facilities may be necessary for efficient simulations.

Type of Resource

text

Language

eng

Handle URL

https://hdl.handle.net/2142/121854

Owning Collections