Agentic Optimization for Resilience in Hydropower Reservoir Systems
The growing electricity demand in the U.S. requires a resilient energy strategy. Hydropower is a major component of that strategy but is subject to environmental and human stressors that impact its dependability. AI-driven reinforcement learning, called "Agentic" learning, presents new opportunities to address two key challenges: 1) Optimizing hydropower reservoir management to increase electricity generation without additional infrastructure, while still meeting the needs of water stakeholders; and 2) Designing more resilient operational policies that can adapt to natural disruptions and changing human objectives. A third challenge, and perhaps the most difficult, is building trust in these Agentic tools to ease their adoption in operational settings. Overcoming these technical challenges could play a crucial role in developing a more resilient and responsive energy strategy for the U.S.
Near-Term Opportunity: Evidence from a preliminary study of the Navajo Reservoir and hydropower facility in New Mexico demonstrates an opportunity to use Agentic AI to "squeeze" more electricity out of existing hydropower facilities while simultaneously providing resilience against destabilizing natural and human elements. Reservoir management is fundamentally an optimization problem where many objectives must be met. Management and resource planners thus design policies to allocate water in a way that maximizes benefit across all users (including hydropower generation) while considering uncertain forecasts and changing objectives. Agentic learning centered around 'Agents' that mimic hydropower operators has significant potential as a tool that designs realistic policies that are flexible and can demonstrate resilience to shocks of different kinds.
Success Measures: The research aims to quantify historical hydropower potential lost due to suboptimal management, validate component models against standard metrics, and demonstrate the resilience of learned policies to environmental and anthropogenic drivers. Collaborating with reservoir operators, the team plans to develop standardized resilience tests to assess policy robustness. Given the unique nature of each system, partnerships with National Labs will be leveraged to develop tailored Agentic technologies and accelerate R&D through shared insights. Implementation success is defined by the scale of deployment, with the team aiming to have Agentic policies adopted by a wide range of management entities. Partnerships with operating entities are crucial to ensure the approach adheres to system realities, builds trust, and provides data for R&D.
Citation Formats
TY - DATA
AB - The growing electricity demand in the U.S. requires a resilient energy strategy. Hydropower is a major component of that strategy but is subject to environmental and human stressors that impact its dependability. AI-driven reinforcement learning, called "Agentic" learning, presents new opportunities to address two key challenges: 1) Optimizing hydropower reservoir management to increase electricity generation without additional infrastructure, while still meeting the needs of water stakeholders; and 2) Designing more resilient operational policies that can adapt to natural disruptions and changing human objectives. A third challenge, and perhaps the most difficult, is building trust in these Agentic tools to ease their adoption in operational settings. Overcoming these technical challenges could play a crucial role in developing a more resilient and responsive energy strategy for the U.S.
Near-Term Opportunity: Evidence from a preliminary study of the Navajo Reservoir and hydropower facility in New Mexico demonstrates an opportunity to use Agentic AI to "squeeze" more electricity out of existing hydropower facilities while simultaneously providing resilience against destabilizing natural and human elements. Reservoir management is fundamentally an optimization problem where many objectives must be met. Management and resource planners thus design policies to allocate water in a way that maximizes benefit across all users (including hydropower generation) while considering uncertain forecasts and changing objectives. Agentic learning centered around 'Agents' that mimic hydropower operators has significant potential as a tool that designs realistic policies that are flexible and can demonstrate resilience to shocks of different kinds.
Success Measures: The research aims to quantify historical hydropower potential lost due to suboptimal management, validate component models against standard metrics, and demonstrate the resilience of learned policies to environmental and anthropogenic drivers. Collaborating with reservoir operators, the team plans to develop standardized resilience tests to assess policy robustness. Given the unique nature of each system, partnerships with National Labs will be leveraged to develop tailored Agentic technologies and accelerate R&D through shared insights. Implementation success is defined by the scale of deployment, with the team aiming to have Agentic policies adopted by a wide range of management entities. Partnerships with operating entities are crucial to ensure the approach adheres to system realities, builds trust, and provides data for R&D.
AU - Schwenk, Jon
A2 - Garcia-Cardona, Cristina
A3 - Bennett, Katrina
A4 - Singh, Shubhendu
A5 - Ranasinghe, Rajiv
A6 - Brelsford, Christa
DB - Energy-Water Resilience
DP - Open EI | National Laboratory of the Rockies
DO -
KW - hydropower
KW - AI
KW - reservoir management
KW - energy resilience
KW - agentic optimization
KW - technology innovation
KW - modeling
KW - resource planning
KW - droughts
KW - flooding
KW - stakeholder collaboration
KW - forecasting
KW - energy production
LA - English
DA - 2026/01/16
PY - 2026
PB - LANL
T1 - Agentic Optimization for Resilience in Hydropower Reservoir Systems
UR - https://ewr.openei.org/submissions/64
ER -
Schwenk, Jon, et al. Agentic Optimization for Resilience in Hydropower Reservoir Systems. LANL, 16 January, 2026, Energy-Water Resilience. https://ewr.openei.org/submissions/64.
Schwenk, J., Garcia-Cardona, C., Bennett, K., Singh, S., Ranasinghe, R., & Brelsford, C. (2026). Agentic Optimization for Resilience in Hydropower Reservoir Systems. [Data set]. Energy-Water Resilience. LANL. https://ewr.openei.org/submissions/64
Schwenk, Jon, Cristina Garcia-Cardona, Katrina Bennett, Shubhendu Singh, Rajiv Ranasinghe, and Christa Brelsford. Agentic Optimization for Resilience in Hydropower Reservoir Systems. LANL, January, 16, 2026. Distributed by Energy-Water Resilience. https://ewr.openei.org/submissions/64
@misc{EWR_Dataset_64,
title = {Agentic Optimization for Resilience in Hydropower Reservoir Systems},
author = {Schwenk, Jon and Garcia-Cardona, Cristina and Bennett, Katrina and Singh, Shubhendu and Ranasinghe, Rajiv and Brelsford, Christa},
abstractNote = {The growing electricity demand in the U.S. requires a resilient energy strategy. Hydropower is a major component of that strategy but is subject to environmental and human stressors that impact its dependability. AI-driven reinforcement learning, called "Agentic" learning, presents new opportunities to address two key challenges: 1) Optimizing hydropower reservoir management to increase electricity generation without additional infrastructure, while still meeting the needs of water stakeholders; and 2) Designing more resilient operational policies that can adapt to natural disruptions and changing human objectives. A third challenge, and perhaps the most difficult, is building trust in these Agentic tools to ease their adoption in operational settings. Overcoming these technical challenges could play a crucial role in developing a more resilient and responsive energy strategy for the U.S.
Near-Term Opportunity: Evidence from a preliminary study of the Navajo Reservoir and hydropower facility in New Mexico demonstrates an opportunity to use Agentic AI to "squeeze" more electricity out of existing hydropower facilities while simultaneously providing resilience against destabilizing natural and human elements. Reservoir management is fundamentally an optimization problem where many objectives must be met. Management and resource planners thus design policies to allocate water in a way that maximizes benefit across all users (including hydropower generation) while considering uncertain forecasts and changing objectives. Agentic learning centered around 'Agents' that mimic hydropower operators has significant potential as a tool that designs realistic policies that are flexible and can demonstrate resilience to shocks of different kinds.
Success Measures: The research aims to quantify historical hydropower potential lost due to suboptimal management, validate component models against standard metrics, and demonstrate the resilience of learned policies to environmental and anthropogenic drivers. Collaborating with reservoir operators, the team plans to develop standardized resilience tests to assess policy robustness. Given the unique nature of each system, partnerships with National Labs will be leveraged to develop tailored Agentic technologies and accelerate R\&D through shared insights. Implementation success is defined by the scale of deployment, with the team aiming to have Agentic policies adopted by a wide range of management entities. Partnerships with operating entities are crucial to ensure the approach adheres to system realities, builds trust, and provides data for R\&D. },
url = {https://ewr.openei.org/submissions/64},
year = {2026},
howpublished = {Energy-Water Resilience, LANL, https://ewr.openei.org/submissions/64},
note = {Accessed: 2026-11-15}
}
Details
Data from Jan 16, 2026
Last updated Jan 16, 2026
Submitted Jan 16, 2026
Contact
Jon Schwenk
Authors
Keywords
hydropower, AI, reservoir management, energy resilience, agentic optimization, technology innovation, modeling, resource planning, droughts, flooding, stakeholder collaboration, forecasting, energy productionDOE Project Details
Project Name White Papers on Ideas to Advance Energy-Water Resilience
Project Lead
Project Number WP-064
