About The Workshop
Energy drives the modern world, but current social, economic, and climate conditions are forcing rapid change in legacy energy systems. Climate commitments are driving a fundamental transformation of these systems, while energy-intensive AI and data centers are simultaneously creating new sources of demand and innovation. While AI is reshaping the energy landscape, the energy value chain may ultimately constrain AI. This creates an open challenge: how do we continue to deliver reliable energy as the modeling and operation of clean energy becomes increasingly complex, for the benefit of all?
AI and ML methods can integrate physical knowledge with data to develop more accurate models and control policies, yet their operational feasibility remains uncertain. Many approaches lack rigorous safety guarantees, their ability to generalize to rare and critical events is poorly understood, and the inference cost and latency of large models can exceed the strict requirements of operators during critical decisions.
This workshop provides a platform for research on Physics-Governed AI for Energy, in which physical laws and system structure not only inform learning but also constrain agents, i.e., determine which decisions are permissible, and shape the very design of learning systems, from information structure to inference budget. By moving beyond physics-informed learning toward physics-governed AI for energy, this workshop seeks foundations to generate algorithms that are accurate beyond simulation, computationally and operationally viable, and demonstrably safe to deploy in critical energy infrastructure, under a shared belief that physics informs, constrains, and shapes ML systems.
Topics
We seek novel AI and ML research for energy systems in which physics plays one or more of three roles: (1) informing learning through physical priors, architectures, or training objectives; (2) constraining predictions and decisions through physical and operational limits; and (3) shaping which AI architectures are viable given network topology, observability, controllability, latency, and system timescales. Topics include, but are not limited to:
- Production: renewable generation, hydrogen, gas.
- Delivery: transmission, distribution, heat and gas networks.
- Consumption: buildings, transportation, industrial loads, data centers.
- Storage: batteries, thermal, etc.
- Physics-informed and physics-constrained machine learning.
- Neural operators and surrogate models.
- Robustness to rare events.
- Uncertainty quantification.
- Safe reinforcement learning.
- Verification and certification of learned models.
- Learning for optimization with feasibility guarantees.
- Topology- and structure-aware models.
- Decentralized and multi-agent learning.
- Real-time inference.
- Learning across timescales.
- Identifying when governing equations, parameterizations, or numerical solvers fail to reproduce physically observed behavior.
- Using data-driven learning to correct these errors.
- Verifying that the resulting models are both more accurate against measurements and consistent with physical laws and operational limits.
- Benchmarks and datasets.
- Digital twins and validation platforms.
- The energy footprint of AI systems.
Call For Papers
Key Dates
- Submission Deadline: November 20, 2026
- Notification of Acceptance: December 2, 2026
- Workshop: February 22/23, 2027
All deadlines are “anywhere on earth” (UTC-12).
Submission Format
We solicit two types of papers:
Full papers (7 pages) on foundational advances.
In-progress papers (3–4 pages) on promising directions.
Submissions must be prepared using the AAAI-27 author kit. Full papers are limited to 7 pages and in-progress papers to 3–4 pages.
Contributed work will anchor the program: every accepted paper will receive a poster presentation, selected papers will receive spotlights, and open-source contributions will receive demonstration slots. Awards will recognize outstanding research and impactful open tools, benchmarks, and platforms.
Review and Publication Policy
Submissions will undergo double-blind review; all submissions must therefore be anonymized. The workshop is non-archival: papers presented at the workshop may subsequently be submitted to journals or conferences. Accepted papers will be made publicly available in full on the workshop website. Authors who prefer that the full text of their paper not be made public may request that only the abstract be posted. Authors wishing to obtain a DOI for their work are welcome to post their paper to arXiv independently.
Submission Site
Papers are submitted via OpenReview: https://openreview.net/group?id=AAAI.org/2027/Workshop/PGAI4E
Invited Speakers
Scott Moura
UC Berkeley
José Daniel Lara
QXT Energy
Schedule
Tentative program (subject to change)
| Time | Event |
|---|---|
| 08:30 – 08:40 | Welcome and Overview |
| 08:40 – 09:15 |
Keynote 1: Scott Moura (UC Berkeley) Title: TBA
Abstract: To be announced.
|
| 09:15 – 10:30 | Contributed Session I |
| 10:30 – 11:00 | Best Paper Talk(s) |
| 11:00 – 12:00 | Poster Session I |
| 13:00 – 13:35 |
Keynote 2: José Daniel Lara (QXT Energy) Title: TBA
Abstract: To be announced.
|
| 13:40 – 14:25 | Panel Discussion: From Simulation to Operation: What Must Physics Govern? A moderated discussion with the keynote speakers and another invited industry expert on the major hurdles for AI/ML adoption in energy systems. |
| 14:35 – 15:50 | Contributed Session II |
| 15:50 – 16:00 | Awards and Closing Remarks |
| 16:00 – 17:00 | Poster Session II |
Accepted Papers
TBA
Awards
Awards will recognize outstanding research and impactful open tools, benchmarks, and platforms. Awardees: TBA.
Venue
The workshop is part of AAAI-27, held in Montréal, Canada.
