Energy Systems Modeling

Systems Operation Planning

Integrated Resource Planning

Short-Term Operation Schedule

Renewable Resources Modeling

Advanced Renewable Modeling

Hydropower and Environmental Resource Assessment

Financial Support Tools

Energy Portfolio Management

Integrated Tools and Computational Environments

High-Performance Computing Environment

PSR presents AI research at the 8th Network Meeting for Power Market Modelling in Oslo

On November 12th, PSR took part in the 8th Network Meeting of the Professional Network for Power Market Modelling, held in Oslo, Norway. Created in 2023 by NVE, Statnett and Statkraft, the network is open to everyone who works with power market models and brings together generation companies, consultancies, software and data providers, and research institutions from the Nordic region, with the aim of improving collaboration, sharing knowledge and developing good modelling practice.

This edition was dedicated to two themes: future weather data and AI for power market modelling. It was the fourth time PSR contributed to the meeting as a speaker, this time with two presentations on the company’s research and recent advances in AI.

Luiz Carlos Costa, Technical Director, presented “Can AI Agents Learn to Operate a Hydrothermal System?”, a proof of concept based on research by Julio Alberto Dias and Mario Veiga Pereira. In the experiment, an AI agent received a goal and a set of domain capabilities, with no mention of SDDP or any other solution algorithm, and built an operating policy for a simplified Brazilian system by its own reasoning. The takeaway: AI agents can work inside complex analytical models and complement optimization, not replace it.

Raphael Chabar, Executive Director, presented “AI and GPUs at Work in Power System Planning”, covering two PSR experiences. The first is PSR Cast, which turns AI-based climate scenarios into inputs for stochastic optimization (research by Julio Alberto Dias). The second is the solution of stochastic expansion planning models on multiple GPUs, in collaboration with the NVIDIA cuOpt team (research by Rafael Benchimol, Raphael Sampaio and Guilherme Bodin). By solving the complete problem as a single linear program, with full hourly chronology in every scenario, the GPU approach handled models with up to 185 million variables, a scale the CPU reference solver could not solve.

David Parrini, Luiz Raphael Lopes and Geir Holler from PSR also attended the event, reinforcing the company’s commitment to being involved in the strategic discussions about the Nordic power market.

We thank Statnett, Statkraft and NVE for the invitation and for another edition of open, high-level technical discussion. PSR remains committed to advancing state-of-the-art tools and methodologies that support the energy transition, system reliability and efficient market operation worldwide.

Leia também

Procurando um conteúdo específico? Faça sua busca abaixo.

Your sign-in has changed

The PSR client area has been updated to provide greater security for our clients. For this reason, the first time you access the new environment you will need to create a new account, even if you already have one.

If you already use PSR Cloud, simply sign in with the same account.

Create your account