A Robust Preprocessing and Topology-Aware Feature Engineering Framework for Power System State Estimation under Sparse and Outlier-Affected Measurements

Cosic, Said and Vokony, Istvan (2025) A Robust Preprocessing and Topology-Aware Feature Engineering Framework for Power System State Estimation under Sparse and Outlier-Affected Measurements. IEEE ACCESS. ISSN 2169-3536 (In Press)

Preview

Text Robust_framework_state_estimation.pdf - Accepted Version Download (983kB) | Preview

Abstract

T This paper presents a comprehensive, AI-based state estimation framework designed to address the complexities introduced by reduced inertia, fast-changing topologies and erratic measurements. The proposed framework introduces a layered preprocessing and feature-engineering workflow that consists of data preprocessing techniques and advanced feature engineering that embeds spatial and temporal context to the integration of topological awareness and machine learning-based estimation. A custom-designed artificial neural network is trained with tailored optimization strategies and regularization mechanisms to enhance convergence, deliver high estimation accuracy and strong resilience to bad or corrupted data. Additionally, robust solver techniques are integrated to further improve estimation reliability under challenging conditions. The proposed framework achieves consistently high accuracy across networks of varying size, redundancy, and renewable penetration, outperforming the conventional SE, especially in low-observability grids. Robustness testing confirms superior performance under bad data and topology errors, with success rates up to 25% higher than the conventional method, though gross measurement errors remain challenging. Sensitivity analysis shows that the algorithmic solver, temporal and spatial features, and robust preprocessing are the strongest contributors to both accuracy and robustness, with their influence increasing as measurement sparsity intensifies.

Actions (login required)

Edit Item