Achieving Disaster-Resilient Distribution Systems via Emergency Response Resources: A Practical Approach

TL;DR

This paper introduces a practical, two-stage approach leveraging machine learning and network approximation techniques to enhance power distribution system resilience through optimized emergency response and post-disaster restoration.

Contribution

It proposes a novel, computationally efficient method combining machine learning and network relaxations for disaster resilience in power distribution systems.

Findings

The approach reduces solution time for post-disaster restoration models.

Machine learning accurately predicts outage characteristics.

Network approximations balance accuracy and computational efficiency.

Abstract

This paper presents a practical approach to utilizing emergency response resources (ERRs) and post-disaster available distributed energy resources (PDA-DERs) to improve the resilience of power distribution systems against natural disasters. The proposed approach consists of two sequential steps: first, the minimum amount of ERRs is determined in a pre-disaster planning model; second, a post-disaster restoration model is proposed to co-optimize the dispatch of pre-planned ERRs and PDA-DERs to minimize the impact of disasters on customers, i.e., unserved energy for the entire restoration window. Compared with existing restoration strategies using ERRs, the proposed approach is more tractable since 1) in the pre-disaster stage, the needed EERs are determined based on the prediction of energy shortage and disaster-induced damages using machine learning-based algorithms (i.e., cost-sensitive-RFQRF for prediction of outage customers, random forest for prediction of outage duration, and CART for prediction of disaster-induced damages); 2) in the post-disaster stage, the super-node approximation (SNA) and the convex hull relaxation (CHR) of distribution networks are introduced to achieve the best trade-off between computational burden and accuracy. Tests of the proposed approach on IEEE test feeders demonstrated that a combination of SNA and CHR remarkably reduces the solution time of the post-disaster restoration model.