Bayesian Distribution System State Estimation with Flow Matching
Organizations: Intelligent Electrical Power Grids Group, Delft University of Technology, 2628 CD Delft, The Netherlands
Abstract
Medium-voltage feeders have few real-time measurements, so distribution system state estimation (DSSE) rests on pseudo-measurements of the secondary substations, a prior in Bayesian terms. Weighted least squares keeps the power flow exact but reduces this prior to a mean and variance, while supervised deep learning estimators also learn the power flow and are tied to one measurement set. We propose PoSE (posterior sampling for state estimation), which keeps the Bayesian structure of DSSE and learns only what the distribution system operator does not know, namely the distribution of substation injections, modelled by a neural network trained with flow matching on historical data. The power flow equations, grid topology and measurements enter only at sampling, through a noise-aware Kalman-type correction. The neural network therefore never approximates the power flow, but every posterior sample is a physically consistent operating point, and one trained prior serves any measurement set, noise level and topology. On the IEEE 33-bus feeder, PoSE lowers the probabilistic injection error (CRPS) by 6.7-7.5% against a full-covariance Gaussian prior in all 12 tested settings. It provides reliable uncertainty: the 90% intervals contain the true injection almost exactly as often as exact posterior samples would, whereas those of supervised estimators contain it in only 37-78% of cases. On the 240-bus unbalanced three-phase Iowa feeder, it gives the lowest injection, voltage and branch-current errors of all estimators. A fast variant matches this accuracy at 0.2-0.6s per snapshot and, without retraining, beats Gaussian priors and retrained supervised estimators on seven reconfigured topologies.
Figures & tables
| Approach | Model family | Modelled quantity | Deep learning | Output | No retraining | Noise | AC- consistent |
|---|---|---|---|---|---|---|---|
| WLS, Gaussian pseudo-meas. [ 1 ] | Diagonal Gaussian | Injections, | Point, covariance | Explicit | |||
| GMM pseudo-meas. [ 6 ] | Gaussian mixture | Injections, | Point, covariance | Explicit | |||
| Learned pseudo-meas. [ 7 ] | Linear regression | Injections, | Point, covariance | Explicit | |||
| End-to-end pseudo-meas. [ 8 ] | MLP | Injections, | Point, covariance | Explicit | |||
| Bayesian deep estimator [ 9 ] | MLP | Voltages, | Point | Training only | |||
| GAN-based estimator [ 10 ] | GAN; MLP | Inj., ; volt., | Point | Training only |
| DSSE | Bayesian inverse problem |
|---|---|
| Pseudo-measurements | Prior |
| Real-time measurements and their accuracy | Likelihood |
| Unobservable feeder | Ill-posed inverse problem |
| WLS with pseudo-measurements | MAP under a Gaussian prior |
| Estimate with uncertainty | Posterior |
| Topology | Open branches | Loss (kW) | (p.u.) | (p.u.) |
|---|---|---|---|---|
| Original | 33, 34, 35, 36, 37 | 202.68 | 0.9131 | – |
| Loss-reducing | 7, 9, 14, 32, 37 | 139.55 | 0.9378 | 0.0412 |
| All ties closed | 7, 9, 14, 28, 32 | 139.98 | 0.9413 | 0.0458 |
| Tie 8–21 | 7, 34, 35, 36, 37 | 158.39 | 0.9299 | 0.0192 |
| Tie 9–15 | 14, 33, 35, 36, 37 | 196.42 | 0.9167 | 0.0134 |
| Tie 12–22 | 11, 33, 34, 36, 37 | 156.79 | 0.9267 | 0.0416 |
| Base PV | High PV | |||||||||
| Method | CRPS | CRPS vs WLS [95% CI] | V RMSE | Branch | Cov. | CRPS | CRPS vs WLS [95% CI] | V RMSE | Branch | Cov. |
| kW | kW | p.u. | % | kW | kW | p.u. | % | |||
| WLS | 8.27 | Reference | 1.04 | 4.27 | 0.941 | 11.56 | Reference | 1.20 | 5.59 | 0.952 |
| Correlated Gaussian | 6.94 | 1.33 [ 1.47, 1.20] | 0.87 | 3.53 | 0.906 | 7.87 | 3.68 [ 4.11, 3.27] | 0.94 | 3.92 | 0.904 |
| Pooled Gaussian | 6.76 | 1.51 [ 1.66, 1.38] | 0.86 | 3.45 | 0.902 | 7.68 | 3.88 [ 4.30, 3.47] | 0.92 | 3.84 | 0.900 |
| Gaussian mixture | 6.91 | 1.36 [ 1.49, 1.22] | 0.93 | 3.54 | 0.879 | 7.89 | 3.67 [ 4.09, 3.27] | 1.14 | 3.98 | 0.881 |
| Base PV | High PV | |||||||||
| Method | V hit | V FA | V Brier | I Brier | I err p95 | V hit | V FA | V Brier | I Brier | I err p95 |
| % | % | % | % | % | % | |||||
| WLS | 85.0 | 0.10 | 2.9 | 1.9 | 32.4 | 91.5 | 0.23 | 5.0 | 12.9 | 40.5 |
| Correlated Gaussian | 91.5 | 0.15 | 2.1 | 1.6 | 30.7 | 95.5 | 0.22 | 3.4 | 6.4 | 31.9 |
| Pooled Gaussian | 91.5 | 0.13 | 2.0 | 1.6 | 30.6 | 95.7 | 0.20 | 3.3 | 6.3 | 31.3 |
| Gaussian mixture | 96.9 | 0.31 | 2.7 | 1.6 | 30.9 | 91.7 | 0.21 | 5.1 | 7.0 | 31.8 |
| Base PV | High PV | |||||
| Guidance | S1 | S2 | S3 | S1 | S2 | S3 |
| Correlated Gaussian (reference) | 7.07 (0.908) | 6.94 (0.906) | 6.65 (0.900) | 8.06 (0.906) | 7.87 (0.904) | 7.39 (0.901) |
| Projection | 6.82 (0.874) | 14.76 (0.689) | 15.11 (0.632) | 7.90 (0.878) | 17.65 (0.708) | 17.82 (0.657) |
| Isotropic | 6.80 (0.876) | 6.69 (0.874) | 6.36 (0.875) | 7.87 (0.881) | 7.70 (0.878) | 7.24 (0.879) |
| DPS | 6.54 (0.868) | 6.48 (0.866) | 6.33 (0.841) | 7.50 (0.871) | 7.42 (0.863) | 7.15 (0.843) |
| PoSE | 6.54 (0.872) | 6.42 (0.874) | 6.15 (0.872) | 7.52 (0.875) | 7.33 (0.876) | 6.87 (0.877) |
| Base PV | High PV | |||||||
|---|---|---|---|---|---|---|---|---|
| S1 | S2 | S1 | S2 | |||||
| Method | Orig. | Reconf. | Orig. | Reconf. | Orig. | Reconf. | Orig. | Reconf. |
| WLS | 8.45 | 8.45–8.46 | 8.27 | 8.23–8.31 | 12.09 | 12.05–12.09 | 11.56 | 11.48–11.66 |
| Correlated Gaussian | 7.07 | 7.06–7.07 | 6.94 | 6.92–6.99 | 8.06 | 8.06–8.06 | 7.87 | 7.86–7.96 |
| Pooled Gaussian | 6.88 | 6.88–6.88 | 6.76 | 6.73–6.80 | 7.87 | 7.87–7.87 | 7.68 | 7.66–7.77 |
| Gaussian Mixture | 7.04 | 7.03–7.04 | 6.91 | 6.91–6.97 | 8.08 | 8.06–8.08 | 7.89 | 7.88–7.97 |
| Method | CRPS | CRPS | RMSE | Branch | Cov. | Time |
|---|---|---|---|---|---|---|
| kW | kvar | p.u. | % | 90% | s | |
| WLS | 1.30 | 0.69 | 0.21 | 5.70 | 0.881 | 0.11 |
| Correlated Gaussian | 1.22 | 0.74 | 0.16 | 5.41 | 0.884 | 0.20 |
| Pooled Gaussian | 1.25 | 0.66 | 0.18 | 5.43 | 0.934 | 0.12 |
| Gaussian mixture | 1.36 | 0.77 | 0.18 | 5.65 | 0.735 | 0.43 |
| PoSE-fast (ours) | 1.13 | 0.66 | 0.15 | 4.79 | 0.810 | 0.58 |