Improving Solar and Solar+Storage Screening Techniques to Reduce Utility Interconnection Time and Costs (Final Technical Report)

Jeremy Keen, Wenbo Wang, Julieta Giraldez, Dan Morris, Robert Broadwater, Jason Bank2022 · DOI

TL;DR

This NREL technical report introduces a data-driven framework to automate the modeling of "grid-edge" secondary networks—the often-unmapped wires connecting homes to transformers. By replacing conservative utility "shortcuts" with synthetic secondary models and machine learning-based screening, the researchers increased solar interconnection approval rates by up to 55 percentage points without compromising grid safety.

What Problem Does This Paper Address?

As residential solar and storage adoption skyrockets, utilities are struggling with an "interconnection bottleneck." The primary culprit isn't just a lack of hardware, but a lack of data. Most utilities have detailed digital models of their primary distribution lines, but their maps often stop at the service transformer. They lack data on the "secondary" network—the final stretch of wire leading to the customer's meter.

Without these models, utilities rely on conservative heuristics (like the "15% rule," which fast-tracks applications only if total solar capacity is below 15% of peak load). These shortcuts often overestimate voltage risks, leading to unnecessary, costly engineering studies or flat-out rejections for residential customers. This paper seeks to bridge that "grid-edge" data gap using synthetic modeling and machine learning.

Methods

The researchers developed a multi-stage pipeline to create "right-sized" models for utility decision-making:

  1. Synthetic Topology Prediction: Using decision trees, the team predicted how customers are connected to transformers. The model uses commonly available GIS data (building parcels, street locations, transformer coordinates) and applies branching rules based on distance thresholds.
  2. Conductor Prediction: A logistic regression model was trained to predict the type and capacity of wires used in these secondary circuits, using features like "Real World Object" (RWO) types and transformer ratings.
  3. Model Calibration: The team used Advanced Metering Infrastructure (AMI) voltage data to "tune" the synthetic models, adjusting line lengths to minimize the error between simulated and actual voltage readings.
  4. Data-Driven Screening: Finally, they trained a Random Forest (RF) model to predict the likelihood of an interconnection application passing or failing. This tool allows utilities to input basic project and network characteristics to get an immediate, high-confidence "pass/fail" probability without running a full, manual power-flow simulation.

Datasets

The study relied on real-world distribution data provided by two major utility partners:

  • Pepco (Exelon): Used for training and validating the secondary topology, conductor prediction, and the Random Forest screening tool.
  • Xcel Energy: Used to validate the generalizability of the secondary topology prediction methods.
  • Validation Tools: The team used Kevala’s "Trace Assist" (which uses Google Street View imagery) to verify synthetic topologies against actual field conditions.

Key Insights and Ideas

  • The "Right-Sized" Approach: One of the most important takeaways is that for residential solar, "perfect" models aren't required. A "credible" synthetic model that captures the general physics of the secondary network is far superior to a conservative heuristic.
  • Primary Errors Mask Secondary Accuracy: When comparing synthetic models to actual AMI meter data, the error was 1.65V. However, when comparing synthetic models to known GIS models, the error dropped to 0.35V. This suggests that a significant portion of "grid-edge" modeling error actually stems from inaccuracies in the primary network models provided by utilities.
  • Heuristics are Too Blunt: Traditional utility shortcuts were found to cause 30–50% more screening failures than necessary. By simply adding a synthetic secondary model, utilities can significantly reduce these "false negatives."

Key Results

  • Increased Approval Rates: The combination of synthetic secondary models and data-driven screening increased the interconnection success rate by as much as 55 percentage points compared to traditional utility heuristics.
  • High Accuracy: The synthetic secondary models achieved a Mean Absolute Error (MAE) of just 0.35 volts when compared against high-fidelity GIS models.
  • Predictive Power: The Random Forest screening tool identified the ratio of PV customers to total customers and primary solar penetration as the strongest predictors of whether a new application would pass technical screens.
  • Efficiency: The data-driven RF tool provides results nearly as accurate as detailed power-flow studies but at a fraction of the computational cost and time.

Limitations and Future Work

  • Data Quality Sensitivity: The method still relies on accurate "anchor points"—specifically the exact locations of service transformers and customer meters. If these are wrong in the utility's database, the synthetic model will be fundamentally flawed.
  • Legacy Infrastructure: The current decision trees are optimized for common residential layouts. Older, heterogeneous networks built over decades without standardized design manuals remain difficult to model accurately without manual intervention.
  • Inverter Complexity: The study assumed common smart inverter settings across the board. In reality, a mix of legacy and modern inverters with varying setpoints exists on the grid. Future work is needed to train models that can account for this "inverter heterogeneity."
  • Scaling: While the tool is ready for pilot projects, moving to a high "Technology Readiness Level" (TRL) will require more diverse training data across different voltage classes and geographic regions.

Summary updated

Abstract

Residential PV installations have increased rapidly over the last decade, and the increased application volume has caused permitting delays and lower overall adoption rates. In this project, we developed and evaluated whether data-driven secondary modeling and screening techniques can help utilities assess customer applications more accurately than traditional screening shortcuts. Secondary topologies are predicted using decision trees and commonly available information, such as service transformer, customer, and street locations. Conductors were predicted using a logistic regression method based on real world object (RWO) types, service transformer ratings, conductor length, and distance to transformer. After developing the combined primary and secondary distribution network model, hosting capacity results were used to train a random forest model to predict the pass/fail likelihood of a customer application. Powerflow based models with predicted secondaries and data-driven methods both increased the screening success rate, relative to common utility heuristics, by as much as 55 percentage points. Data-driven screening techniques were described by one utility as a "right-sized" approach for residential customers given the low-risk of small errors and the high-cost of accurate modeling.
BibTeX
@misc{keen2022improving,
  author = {Jeremy Keen and Wenbo Wang and Julieta Giraldez and Dan Morris and Robert Broadwater and Jason Bank},
  title = {{Improving Solar and Solar+Storage Screening Techniques to Reduce Utility Interconnection Time and Costs (Final Technical Report)}},
  year = {2022},
  doi = {10.2172/1891205},
}

Loading related content...

Figure