EML Energy Conference 2026

Inaugural EML Energy Conference 2026: Electricity Demand Forecasting in the AI Era

February 20, 2026

The Economic Machine Learning (EML) Lab hosted its inaugural Energy Conference, bringing together industry practitioners, policymakers, and researchers for focused discussions on AI-powered load forecasting. The event highlighted EML Lab's development of transparent, self-diagnostic forecasting systems tailored to economic, financial, and energy data.

Distinguished External Speakers

The conference featured three distinguished external speakers who brought diverse perspectives on energy forecasting:

• Honorable Greg Ballard, Former Mayor of Indianapolis, provided leadership perspectives on energy policy and innovation at a critical moment for Indiana. His remarks emphasized the intersection of rising energy costs and the ongoing AI transformation across the economy, underscoring the importance of developing forecasting capabilities that both reduce consumer costs and support workforce preparation for the AI era.
• Kenneth B. Medlock III, James A. Baker III and Susan G. Baker Fellow in Energy and Resource Economics; Senior Director, Center for Energy Studies, Rice University, presented virtually, offering insights into electricity market dynamics. Drawing on experience spanning trading operations, industry consultations, and energy research leadership, he explained how renewable integration and battery adoption are increasing market volatility and creating new arbitrage opportunities. His analysis highlighted why accurate real-time forecasting has become central to operational and strategic decision-making.
• Kristian Doty, Principal Forecasting Analyst, Midcontinent Independent System Operator (MISO), participated throughout the day, presenting detailed insights into MISO's short- and medium-term load forecasting processes and operational workflows. His engagement with EML's live dashboard demonstration reinforced the operational relevance of the Lab's approach.

AI-Powered Load Forecasting: The ART Framework

EML's AI-powered load forecasting system is fully automated and continuously self-updating. What distinguishes this system from conventional black-box forecasters is its ART framework:

• Adaptability: Producing reliable forecasts in evolving environments
• Robustness: Reducing exposure to model instability and failure risks
• Transparency: Supporting automated self-diagnosis and self-revision

Unlike opaque forecasting systems that provide predictions without explanation, EML's approach integrates counterfactual decomposition to analyze the structural drivers of its forecasts.

Live Demonstration: Self-Diagnosis and Self-Revision in Real Time

A continuous dashboard display demonstrated the system's ability to diagnose and refine itself in real time. For 272 days of day-ahead load forecasts (January 1 – September 20, 2025), the dashboard displayed:

• Forecasts alongside realized loads
• Counterfactual decompositions capturing day-type and temperature effects unique to each day

This decomposition allows operators to understand not only the forecast outcome, but the components contributing to it. When deviations occur, the system identifies which elements require adjustment. The result is a forecasting framework that supports trust, operational decision-making, and systematic improvement.

The Intellectual Contribution

The central contribution demonstrated at the conference was the development of forecasting systems capable of self-diagnosis and self-revision through counterfactual decomposition. Rather than focusing solely on forecast accuracy, the approach emphasizes interpretability and structured self-revision. This reflects a broader insight: economic, financial, and energy data are characterized by weak signals embedded in high noise and shaped by institutional and behavioral dynamics. Standard machine learning tools are not designed for this environment. By customizing ML methods guided by economic theory and advanced econometric methodologies, EML preserves interpretability while retaining predictive strength.

Distinctive Features of the Conference

The conference centered on a live demonstration of a fully operational forecasting system. Participants observed:

• Consistently small forecast errors using publicly available data
• Fully automated operation without discretionary human adjustments
• Clear identification of improvement pathways through structured diagnostics

The discussions that followed focused on operational integration, data collaboration, and future partnerships.

Community and Forward Path

Though intentionally small in scale, the conference facilitated substantive exchange among practitioners, policymakers, researchers, and students. The focused format enabled detailed discussion of grid-level operational challenges and potential collaboration.

The event established a foundation for continued engagement with industry and policy partners and for pursuing applied research initiatives in energy forecasting and AI-driven economic modeling.

The conference marks an important step in EML Lab's broader effort to rethink how econometric structure and machine learning should be jointly deployed in operational forecasting, particularly in settings where transparency, interpretability, and structured self-diagnosis are essential for responsible decision-making.

For more information about EML Lab's research and upcoming events, visit eml-lab.github.io.

Dear Conference Participants,

Thank you for making our inaugural EML Energy Conference on "Electricity Demand Forecasting in the AI Era" such a meaningful event.

I am especially grateful to our external speakers:

Honorable Greg Ballard for providing leadership perspectives on how AI-driven forecasting connects to Indiana's energy costs and broader economic transformation;

Kenneth Medlock for clarifying how renewables, battery adoption, and market volatility make accurate real-time forecasting increasingly critical;

Kristian Doty for offering invaluable insights into MISO's operational forecasting and engaging deeply with our live demonstration.

What we were able to showcase was more than a forecasting model. We demonstrated a fully operational AI system built around what we call the ART framework — Adaptability, Robustness, and Transparency.

The live dashboard displaying 272 days of day-ahead forecasts, together with counterfactual decompositions of day-type and temperature effects, illustrated our central idea: a forecasting system should not only predict, but also explain, diagnose, and improve itself in real time.

This combination of econometric foundations and customized machine learning — specifically tailored to economic, financial, and energy data — distinguishes our approach from black-box forecasters. Transparency enables trust, operational relevance, and continuous improvement.

Equally important were the discussions. The focused format allowed for substantive exchange among practitioners, policymakers, researchers, and students. The engagement from all of you — including our student fellows and visiting scholars — made the day especially productive. We are encouraged by the interest expressed in further collaboration and partnership.

A fuller summary of the conference is available above, including details on the ART framework and our next steps. Presentation files and a link to our dashboard will also be posted on the EML website.

Thank you again for being part of this milestone event. Your participation helped establish an important foundation for EML's work in energy forecasting and future engagement with industry and policy partners.