Deep Statistical Solver for Distribution System State Estimation

Benjamin Habib,Elvin Isufi,Ward van Breda,Arjen Jongepier,Jochen L. Cremer

2023-06-26

Abstract:Implementing accurate Distribution System State Estimation (DSSE) faces several challenges, among which the lack of observability and the high density of the distribution system. While data-driven alternatives based on Machine Learning models could be a choice, they suffer in DSSE because of the lack of labeled data. In fact, measurements in the distribution system are often noisy, corrupted, and unavailable. To address these issues, we propose the Deep Statistical Solver for Distribution System State Estimation (DSS$^2$), a deep learning model based on graph neural networks (GNNs) that accounts for the network structure of the distribution system and for the physical governing power flow equations. DSS$^2$ leverages hypergraphs to represent the heterogeneous components of the distribution systems and updates their latent representations via a node-centric message-passing scheme. A weakly supervised learning approach is put forth to train the DSS$^2$ in a learning-to-optimize fashion w.r.t. the Weighted Least Squares loss with noisy measurements and pseudomeasurements. By enforcing the GNN output into the power flow equations and the latter into the loss function, we force the DSS$^2$ to respect the physics of the distribution system. This strategy enables learning from noisy measurements, acting as an implicit denoiser, and alleviating the need for ideal labeled data. Extensive experiments with case studies on the IEEE 14-bus, 70-bus, and 179-bus networks showed the DSS$^2$ outperforms by a margin the conventional Weighted Least Squares algorithm in accuracy, convergence, and computational time, while being more robust to noisy, erroneous, and missing measurements. The DSS$^2$ achieves a competing, yet lower, performance compared with the supervised models that rely on the unrealistic assumption of having all the true labels.

Machine Learning,Systems and Control

What problem does this paper attempt to address?

The problems that this paper attempts to solve are several key challenges in distribution system state estimation (DSSE), especially the lack of observability and the high - density distribution system structure. Specifically:

1. **Lack of observability**: There is a shortage of real - time measurement data in the distribution system, making it difficult for traditional state estimation algorithms to be directly applied.
2. **Data quality problems**: Measurement data are usually noisy, corrupted or missing, which further affects the accuracy of state estimation.
3. **Scarcity of label data**: Due to the lack of real - label data, data - driven methods based on machine learning perform poorly in distribution system state estimation.

To solve these problems, the paper proposes a deep statistical solver (DSS2) based on graph neural networks (GNNs). This model improves distribution system state estimation in the following ways:

- **Weakly - supervised learning**: By using physical information and power flow equations as constraints, the need for ideal label data is reduced, and it can learn from noisy data.
- **Graph neural network**: A hypergraph is used to model the network structure of the distribution system and its heterogeneous components, so as to better capture the complexity of the system.
- **Physical constraints**: Stability constraints of voltage levels, phase angles and line loads are added to the loss function to ensure that the model output conforms to physical laws.

Through these improvements, DSS2 not only improves the accuracy and robustness of state estimation, but also significantly shortens the computational time. The experimental results show that the performance of DSS2 on the IEEE 14 - bus, 70 - bus and 179 - bus systems is better than that of the traditional weighted least - squares (WLS) algorithm, and it shows stronger robustness in dealing with noisy, incorrect and missing measurement data.