Topology Priors Fail the Test in Distribution-Level Load Forecasting Study

Two widely used post-processing steps for distribution-level residential load forecasting—hierarchical reconciliation and topology-aware conformal calibration—have been put to a critical test, and only one of them survives scrutiny. The study, published in Energy Informatics, evaluates whether electrical topology priors genuinely improve forecast coherence and uncertainty calibration across the buses of a distribution feeder.

The benchmark maps 114 de-identified apartment smart-meter profiles onto standard IEEE distribution feeders by mean-load rank. The authors are careful to note a key limitation: this preserves temporal shapes but alters magnitudes, coincident peaks and spatial correlation, so absolute values do not transfer to real deployments.

On the first front, minimum-trace shrinkage reconciliation does what it promises. It enforces exact coherence across aggregation levels, cutting incoherence from 4.26 percent of aggregate load down to machine precision. It also improves aggregate accuracy by 0.28 percentage points over simple bottom-up aggregation—though at a cluster-level cost of 0.11 percentage points, concentrated on the most volatile buses.

The second technique fares far worse. Under a matched-null test, a conformal layer that weights conformity scores by electrical distance produced improvements indistinguishable from those obtained when the profile-to-bus assignment was randomly permuted. In other words, the apparent benefit vanished as soon as the topology information was scrambled.

The evidence against topology priors goes deeper. Electrical distance was essentially uncorrelated with score similarity, showing a Spearman correlation of just +0.02. On a measured 906-bus feeder, the topology-based kernel performed 36 to 70 percent worse than using no topology at all. The authors conclude that electrical-topology priors are not identifiable at this aggregation scale.

By contrast, per-bus adaptive conformal inference delivered striking results without assuming exchangeability. Across four architecture families, it reduced the across-bus standard deviation of coverage by 67 to 92 percent, raising the share of buses within two percentage points of nominal coverage from 52 to 67 percent up to 97 to 100 percent—while keeping interval widths almost unchanged.

The practical payoff was tested in a network-constrained battery-dispatch study, where the calibrated intervals cut dispatch regret by 14 to 29 percent. That links the statistical improvement directly to operational value for grid operators managing uncertainty at the distribution edge.

Beyond the headline results, the paper offers a methodological contribution: a matched-null protocol for testing topology priors, which other researchers can apply before claiming that network structure improves their forecasts. For a field increasingly reliant on smart-meter data, the message is that coherence can be enforced reliably, but calibration is best earned bus by bus rather than borrowed from the grid diagram.

Subject of Research: Hierarchical reconciliation and per-bus adaptive conformal calibration for distribution-level residential load forecasting

Article Title: Hierarchical reconciliation and per-bus adaptive conformal calibration for distribution-level residential load forecasting: a critical assessment of topology priors

Article References: Hong, H., Xu, S., Yang, Y., Hu, J., Chen, Y., & Wu, J. (2026). Hierarchical reconciliation and per-bus adaptive conformal calibration for distribution-level residential load forecasting: a critical assessment of topology priors. Energy Informatics. https://doi.org/10.1186/s42162-026-00690-1

Image Credits: AI Generated

DOI: 10.1186/s42162-026-00690-1

Keywords: load forecasting, hierarchical reconciliation, conformal calibration, distribution networks, smart meters, topology priors, uncertainty quantification, battery dispatch, Energy Informatics, electrical distance, Hierarchical, reconciliation