On February 25th at 12:00 PM, the DEEC TALK dedicated to the theme “Meet the New Faculty” will take place in EA1 (North Tower).

This session will feature Zita Marinho, who will present a perspective on Model-Based Reinforcement Learning, exploring the alignment between internal models and value functions, self-consistency signals, and guided exploration mechanisms.

It will also include Jorge Alberto Mendes de Sousa, who will provide an introduction to the organization of electricity markets in the Iberian context, price formation mechanisms, and the main current research challenges, including flexibility markets and multi-market trading strategies.

TALK 1: From internal consistency to exploration: a perspective on self-correcting agents

Abstract

This talk focuses on Model-Based Reinforcement Learning (MBRL) for acting and planning.  I will present work that examines the importance of aligning an agent's internal world model with its value function. Self-consistency shows how losses can affect an RL agent's representation, planning, and action abilities. This consistency signal can also be used for guiding exploration.  Filos et al. demonstrated that the discrepancy between these value components is a critical signal for detecting uncertainty and guiding an agent to explore to improve a model. Finally, I will discuss how external guidance can provide a sufficient exploratory signal to enhance model self-repair capabilities, shifting the paradigm from internal state monitoring to external guidance.

 

About the speaker

Zita Marinho is an Assistant Professor at the University of Lisbon, Instituto Superior Técnico. She previously worked as a Senior Research Scientist at Google DeepMind, on reinforcement learning and meta-learning. She holds a dual PhD/MSc in Robotics from the Robotics Institute at Carnegie Mellon University and IST, University of Lisbon, as part of the CMU/Portugal program, where she worked on semi-supervised learning algorithms for robotics and planning. 

Her current research interests focus on improving AI algorithms to support scientific discovery and adaptation. She has worked in the intersection of reinforcement learning and natural language processing. She is particularly interested in studying how agentic AIs can learn more effectively from human/machine interactions, and how to explore effectively in continuous settings. 

TALK 2: Electricity Markets Structure, Price Formation Mechanisms and Research Opportunities

Abstract

The energy transition, driven by the increasing penetration of renewable energy sources, the electrification of the economy, and the decentralization of resources, is transforming the operation of power systems. In this context, electricity markets play a central role in the efficient coordination of resources and in providing appropriate economic signals for investment and operation.

This presentation introduces the organization of electricity markets in the Iberian context (MIBEL), addressing the different time horizons and contracting mechanisms. The principles of price formation and the most commonly used settlement models will be discussed. Finally, current research challenges will be presented, with a focus on flexibility markets and multi-market trading strategies supported by forecasting and optimization methods.

1. Organization of electricity markets (Iberian context)

Forward markets (OMIP): Financial and physical contracts for risk hedging and revenue/cost stabilization over horizons ranging from weeks to years.

Day-ahead and intraday markets (OMIE): Centralized optimization based on buy and sell bids for each period, allowing market participants to adjust positions as uncertainty decreases.

System services markets (REN, in Portugal): Procurement of reserves and ancillary services (e.g., FCR, aFRR, mFRR) to ensure real-time system security and stability.

Key message: The different markets operate in a complementary manner, reflecting the evolution of uncertainty and the operational needs of the system.

2. Price formation and settlement mechanisms

Supply and demand curves: Aggregation of agents’ bids and determination of the market clearing point.

Marginal pricing (PAC – pay-as-cleared): All matched agents receive/pay the market clearing price vs Pay-as-bid (PAB): Each agent is remunerated at its bid price.

Implications: Bidding strategies and strategic behavior; Economic efficiency and transparency; Impact on incentives for flexibility and investment.

Key message: The design of the pricing mechanism directly influences agents’ behavior and the overall efficiency of the system.

3. Current research challenges

a) Flexibility markets: Development of local and regional mechanisms for procuring flexibility services (e.g., congestion management, voltage support, DER integration), involving new resources such as BESS, demand response, and electric mobility.

b) Multi-market trading with algorithmic strategies: Coordinated participation across multiple markets (forward, day-ahead, intraday, system services), combining:

Price, renewable generation, and demand forecasting;

Optimization models under uncertainty;

Automated bidding strategies and dynamic updating of market positions.

Key message: The growing complexity and interdependence of markets require integrated approaches based on data, forecasting, and automated decision-making.

About the speaker

Jorge Alberto Mendes de Sousa is an Associate Professor (with Habilitation) at Instituto Superior Técnico in the Department of Electrical and Computer Engineering / Energy Scientific Area.

He holds a five-year degree and a Master’s in Electrical and Computer Engineering from Instituto Superior Técnico, specializing in Energy Systems, and a PhD in Economics from Nova SBE, specializing in Industrial Organization/Energy Markets.

In 2025, he was elected by the European Council to the Administrative Board of the European Union Agency for the Cooperation of Energy Regulators (ACER).

He is the founder and current Vice-President of APEEN – the Portuguese Association for Energy Economics, having served as its President from 2021 to 2025.

He was Coordinating Professor at the Instituto Superior de Engenharia de Lisboa (ISEL), where he served as President (2016–2020), Head of the Department of Electrical Energy and Automation Engineering (2014–2016), and Coordinator of the BSc in Electrical Engineering. At ISEL, he also founded the postgraduate program in Renewable Energy Engineering and Management (EGER) and the ISEL Energy Week program.

He has also been a Visiting Professor at the Faculty of Science and Technology/UNL and at the State Engineering University of Armenia (SEUA).

Research and development activities

Senior Researcher at INESC-ID and member of its Scientific Council, conducting research in electric power generation and transmission systems, energy markets, risk management, sustainable mobility, hydrogen systems, and renewable energy.

Author of more than 100 publications in peer-reviewed international journals and conferences. He has participated in and led more than 20 scientific and industry collaboration projects funded by public and private entities. He regularly supervises PhD and MSc theses. From 2008 to 2011, he was an Associate Researcher at MIT / Center for Energy and Environmental Policy Research.

Member of the scientific committees of several international conferences, having served as Chairman of the European Energy Markets Conference (EEM08, EEM15 and EEM25), as well as of the Iberian Conference on Energy Economics (EEIC2016) and the International Conference on Energy Economics (EEiC2023).

Professional experience

Before entering academia, he collaborated with several companies and institutions in the electricity sector, including EDP – Energias de Portugal, EDF – Électricité de France, INTERG – Energy Institute, and the Army Engineering Services Directorate.