New framework quantifies uncertainty in distributed energy adoption across diverse grid structures

Distributed energy resources (DERs), including rooftop solar panels and other customer-owned technologies, are rapidly transforming the way electricity moves through the grid. A new forecasting framework developed by researchers at Carnegie Mellon University could help utilities predict where adoption will accelerate, quantify the uncertainty surrounding those predictions, and prepare infrastructure before local networks become overwhelmed.

The study, published in the Annals of Applied Statistics, addresses a problem that is becoming increasingly urgent as households and businesses generate more of their own electricity. Utilities must anticipate changing electricity demand, determine where circuits and substations may require upgrades, and maintain reliable service despite adoption patterns that can vary dramatically from one neighborhood to the next.

Traditional forecasting methods often produce a single estimate of future solar or DER adoption. While such projections can be useful, they may conceal the uncertainty that matters most to planners. A forecast that predicts 20% adoption in a service area, for example, does not reveal whether the realistic range is 15% to 25% or 5% to 40%. Those differences can determine whether a utility needs to reinforce a local circuit immediately or can safely delay investment.

The Carnegie Mellon researchers developed a method called Hierarchical Probabilistic Conformal Prediction for Distributed Energy Resources Adoption. The approach combines data-driven forecasting with conformal prediction, a statistical technique designed to create prediction intervals with measurable reliability. Instead of offering one definitive number, the method generates a range of plausible outcomes, allowing decision makers to see both the expected level of adoption and the uncertainty around it.

The framework is specifically designed for the hierarchical organization of electric distribution systems. Individual customers are connected to local circuits, circuits feed into substations, and substations form part of larger service territories. Forecasts made independently at each level can contradict one another—for example, the projected adoption across several circuits might exceed the forecast for the substation that contains them. The researchers’ method incorporates these relationships so that predictions remain statistically valid and logically consistent across the grid hierarchy.

This feature is technically important because grid infrastructure is planned at multiple scales. A cluster of rooftop solar installations may have little effect on a utility’s overall territory but create voltage or reverse-power-flow challenges on a specific circuit. Reverse power flow occurs when customer generation sends electricity back toward the distribution system, potentially changing operating conditions for equipment designed primarily to deliver power in one direction. Identifying such concentrated growth early can help utilities target upgrades where they will have the greatest impact.

To test the framework, the researchers used customer-level solar installation data from Indianapolis, Indiana. The detailed data allowed them to examine adoption patterns at a fine spatial scale rather than treating the entire service area as uniform. Their results indicated that the approach could produce more reliable and actionable forecasts than existing techniques, particularly when planners need to understand which circuits or substations may experience unusually rapid growth.

The model’s uncertainty estimates could also improve long-term decisions about capacity, resilience, and investment timing. Utilities could use the forecast ranges to evaluate multiple scenarios, such as moderate, high, or unexpectedly concentrated DER adoption. Regulators could then assess whether proposed infrastructure investments are robust under different futures instead of relying on a single central projection. This kind of scenario-based planning may reduce the risk of both underbuilding, which can threaten reliability, and overbuilding, which can increase costs for customers.

The researchers say the work has already moved beyond academic testing. Wenbin Zhou, a PhD student in machine learning and public policy at Carnegie Mellon’s Heinz College, said the approach was adopted for an Indiana utility’s 2025 integrated resource plan, where it helped inform long-term planning for distributed energy adoption and grid infrastructure. The project also received second place in the 2026 Innovative Applications in Analytics Award at the INFORMS Analytics+ Conference, highlighting the growing role of advanced statistical methods in energy planning. As DER adoption continues to expand, tools that combine detailed local data, hierarchical modeling, and transparent uncertainty estimates could become essential to building a grid capable of accommodating millions of individual energy decisions.

News Publication Date: 10-Jun-2026

Web References: https://doi.org/10.48550/arXiv.2411.12193

References: Carnegie Mellon University researchers’ study; U.S. National Science Foundation

Keywords

Distributed energy resources, rooftop solar, conformal prediction, probabilistic forecasting, electric grids, power systems, substations, circuit planning, infrastructure investment, renewable energy, machine learning, uncertainty quantification