arXiv Introduces PoSE: A Novel Bayesian Distribution-Based Approach for Distribution System State Estimation

Summary:arXiv has released a new research paper introducing PoSE (Posterior Sampling for State Estimation), a novel approach to address the data sparsity challenge in medium-voltage distribution system state estimation (DSSE). PoSE maintains the Bayesian structure of DSSE and learns only the distribution of substation injections through a neural network trained with flow matching on historical data. By incorporating power flow equations, grid topology, and measurements through a noise-aware Kalman-type

---

Technical Mechanism Analysis

The core idea of PoSE (Posterior Sampling for State Estimation) is to combine Bayesian inference with deep learning to address the data sparsity challenge in distribution system state estimation (DSSE). Traditional methods like weighted least squares maintain the exactness of power flow but reduce the prior to a mean and variance, losing uncertainty information. Supervised deep learning models, on the other hand, are limited to specific measurement sets and lack generalization capabilities. PoSE addresses these issues through the following approaches:

1. Bayesian Structure Preservation: PoSE retains the Bayesian structure of DSSE and learns only the neural network model of the substation injection distribution.
2. Flow Matching Training: Utilizes flow matching techniques to train on historical data, capturing the complex patterns of the injection distribution.
3. Noise-Aware Kalman Correction: Incorporates power flow equations, grid topology, and measurements through a noise-aware Kalman-type correction during sampling, ensuring that each posterior sample is a physically consistent operating point.

Engineering Trade-offs and Performance

PoSE demonstrates strong performance in multiple benchmarks:

• IEEE 33-bus Feeder: In all 12 tested settings, PoSE's CRPS (probabilistic injection error) is reduced by 6.7-7.5% compared to a full-covariance Gaussian prior.
• 240-bus Unbalanced Three-Phase Feeder: PoSE outperforms other estimators in terms of injection, voltage, and branch-current errors.
• Fast Variant: PoSE's fast variant achieves a per-snapshot inference time of 0.2-0.6 seconds and, without retraining, beats Gaussian priors and retrained supervised estimators on seven reconfigured topologies.

Developer Implementation and Deployment Recommendations

PoSE offers a new technical path for distribution system state estimation, particularly in scenarios with high data sparsity and uncertainty. Developers can consider the following recommendations for implementation:

1. Data Preparation: Collect historical substation injection data and perform flow matching training to generate the prior model.
2. Model Integration: Integrate the PoSE model with existing DSSE systems to leverage its Bayesian inference capabilities for improved accuracy and reliability.
3. Real-Time Optimization: Further optimize the fast variant of PoSE to meet real-time requirements for real-time applications.
4. Cross-Platform Adaptation: The architecture of PoSE makes it easy to deploy on different hardware platforms, and developers can optimize it according to actual needs.

Conclusion

PoSE demonstrates the great potential of combining Bayesian inference with deep learning in distribution system state estimation. Its excellent performance in multiple benchmarks indicates that the method has significant advantages in improving the accuracy and reliability of state estimation.

---

Source: ArXiv Machine Learning (cs.LG) (2026-11-02)

— END —

Tags: #Distribution Grid Follow #Bayesian Inference Follow #Deep Learning Follow #State Estimation Follow #arXiv Follow