Permanent URI for this collection
1 - 5 of 5
ItemComparing Machine Learning and Optimization Approaches for the N − k Interdiction Problem Considering Load Variability( 2023-01-03)Power grids must be operated, protected, and maintained such that a small number of line failures will not result in significant load shedding. To identify problematic combinations of failures, we consider an N-k interdiction problem that seeks the set of k failed lines (out of N total lines) that result in the largest load shed. This is naturally formulated as a bilevel optimization problem with an upper level representing the attacker that selects line failures and a lower level modeling the defender's generator redispatch to minimize the load shedding. Compounding the difficulties inherent to the bilevel nature of interdiction problems, we consider a nonlinear AC power flow model that makes this problem intractable with traditional solution approaches. Furthermore, since the solutions found at a particular load condition may not generalize to other loading conditions, operators may need to quickly recompute these worst-case failures online to protect against them during operations. To address these challenges, we formulate and compare the performance of three simplified methods for solving the N-k interdiction problem: a state-of-the-art optimization approach based on a network-flow relaxation of the power flow equations and two newly developed machine learning algorithms that predict load sheds given the state of the network.
ItemPPGN: Physics-Preserved Graph Networks for Real-Time Fault Location in Distribution Systems with Limited Observation and Labels( 2023-01-03)Electric faults may trigger blackouts or wildfires without timely monitoring and control strategy. Traditional solutions for locating faults in distribution systems are not real-time when network observability is low, while novel black-box machine learning methods are vulnerable to stochastic environments. We propose a novel Physics-Preserved Graph Network (PPGN) architecture to accurately locate faults at the node level with limited observability and labeled training data. PPGN has a unique two-stage graph neural network architecture. The first stage learns the graph embedding to represent the entire network using a few measured nodes. The second stage finds relations between the labeled and unlabeled data samples to further improve the location accuracy. We explain the benefits of the two-stage graph configuration through a random walk equivalence. We numerically validate the proposed method in the IEEE 123-node and 37-node test feeders, demonstrating the superior performance over three baseline classifiers when labeled training data is limited, and loads and topology are allowed to vary.
ItemA Generalized Approach to Contingency Screening with System Islanding( 2023-01-03)This paper introduces a generalized contingency analysis approach that can identify critical pairs of contingencies (N-2 contingencies) that result in severe reliability violations or large numbers of islanded system components, bridging the gap between academic research and practical applications. We formulate the practical problem of N-2 contingency analysis in a clear mathematical format. We propose a generalized contingency screening approach compatible with all types of contingencies, including generator failure, load failure, open branches, and their mixtures that can lead to islanding. The proposed approach is demonstrated in a real Texas system, showing its effectiveness in critical contingency identification and its ability to enable possible practical contingency analysis applications.
ItemDistributed Power System State Estimation Using Graph Convolutional Neural Networks( 2023-01-03)State estimation plays a key role in guaranteeing the safe and reliable operation of power systems. This is a complex problem due to the noisy and unreliable nature of the measurements that are obtained from the power grid. Furthermore, the laws of physics introduce nonconvexity, which makes the use of efficient optimization-based techniques more challenging. In this paper, we propose to use graph convolutional neural networks (GCNNs) to learn state estimators from data. The resulting estimators are distributed and computationally efficient, making them robust to cyber-attacks on the grid and capable of scaling to large networks. We showcase the promise of GCNNs in distributed state estimation of power systems in numerical experiments on IEEE test cases.
ItemIntroduction to the Minitrack on Resilient Networks( 2023-01-03)