Power System Robust State Estimation As a Layer: A Novel End-to-end Learning Approach

TL;DR

This paper introduces a novel end-to-end neural network framework for power system state estimation that explicitly incorporates physical constraints as a differentiable layer, improving accuracy and physical consistency.

Contribution

It proposes the first neural network-based RSE model with an explicit differentiable layer enforcing physical constraints, enhancing robustness and accuracy.

Findings

Significantly improves state estimation performance.

Ensures physical consistency in solutions.

Outperforms classical E2E and PINN approaches.

Abstract

Serving as an essential prerequisite for modern power system operation, robust state estimation (RSE) could effectively resist noises and outliers in measurements. The emerging neural network (NN) based end-to-end (E2E) learning framework enables real-time application of RSE but cannot strictly enforce the physical constraints involved, potentially yielding solutions that are statistically accurate yet physically inconsistent. To bridge this gap, this work proposes a novel E2E learning based RSE framework, where the RSE problem is innovatively constructed as an explicit differentiable layer of NN for the first time, ensuring physics alignments with rigors. Also, the measurement weights are treated as learnable parameters of NN to enhance estimation robustness. A hybrid loss function is formulated to pursue accurate and physically consistent solutions. To realize the proposed NN structure, the original non-convex RSE problem is specially relaxed. Extensive numerical simulations have been carried out to demonstrate that the proposed framework can significantly improve the SE performance while fulfilling physical consistency on six testing systems, in comparisons to the classical E2E learning based approach and the physics-informed neural network (PINN) approach.