1. Physics-informed machine learning for battery state estimation and degradation-aware energy management.

The BET unit is seeking a motivated MSc student interested in the development of data-driven methods for battery monitoring and management. Accurate knowledge of the internal condition of a battery is a prerequisite for its safe and economically efficient operation, yet quantities such as state of charge, state of health and degradation rate cannot be measured directly and must be inferred from voltage, current and temperature signals. Purely data-driven estimators achieve good accuracy but extrapolate poorly and offer little physical interpretability, while purely physics-based observers are limited by parameter uncertainty and ageing-induced model mismatch.
The thesis will focus on the development of a physics-informed machine-learning framework in which electrochemical or equivalent-circuit model structures are embedded into the learning process, so that the estimator remains consistent with the underlying battery physics while adapting to operational data. Battery models are already available in-house, in OpenModelica and PyBaMM, and will serve as the starting point. The student will parametrize and validate the physical model against experimental or field measurements, train and benchmark the hybrid estimator across different operating conditions and subsequently integrate the estimated states into an energy-management system that schedules charging and discharging with an explicit degradation cost, quantifying the resulting trade-off between operating cost and battery lifetime.

Ideal candidate requisites:

• MSc student in Energy Engineering, Electrical Engineering, Mechatronics Engineering, Computer Science or Data Science;
• Programming skills for modelling (e.g., PyBaMM, Modelica, MATLAB/Simulink) and data analysis (e.g., Python, MATLAB);
• Basic knowledge of battery modelling and machine-learning methods; is an added value;
• Ability to summarize scientific results with presentations;
• Good communication and relational skills;
• Skills in problem solving;
• Ability to work in a collaborative environment with good autonomy.

Reference people: Saman Korjani (skorjani@fbk.eu), Amir Naebi Toutounchi (anaebitoutounchi@fbk.eu)


2. Real-time MPC of storage for distribution-grid congestion and reinforcement learning for resilient grid operation.

The BET unit is seeking a motivated MSc student interested in the development of advanced control strategies for battery storage in distribution networks. Distribution System Operators (DSO) are increasingly confronted with congestion and voltage problems caused by distributed generation, electric vehicles and flexible loads, while simultaneously being required to maintain service continuity under disturbances and outages. Battery energy storage can serve both needs, but the two objectives call for different control philosophies: real-time Model Predictive Control (MPC) provides explicit network-constraint handling and forecast-based decisions over a receding horizon, whereas Reinforcement Learning learns an offline-trained control policy that penalizes constraint violations during training and can then be executed online in real time, making it suitable as a future automatic control layer for the DSO.
An MPC framework for storage dispatch is already available in-house and will be adapted by the student to a representative DSO network, providing both a working congestion-management baseline and the simulation environment for the main part of the work, which is the design and training of a Reinforcement Learning (RL) controller for resilient operation including islanding and critical-load prioritization. The student will adapt and validate the MPC on the DSO test network, design and train the reinforcement learning agent on an equivalent environment, and benchmark both approaches under normal and disturbed operating conditions in terms of constraint satisfaction, operating cost, battery degradation and restoration performance.

Ideal candidate requisites:

• MSc student in Energy Engineering, Electrical Engineering, Control/Automation Engineering or Computer Science;
• Programming skills for modelling and optimization (e.g., Python, MATLAB/Simulink);
• Basic knowledge of power system analysis, optimal management/control and/or reinforcement learning is an added value;
• Ability to summarize scientific results with presentations;
• Good communication and relational skills;
• Skills in problem solving;
• Ability to work in a collaborative environment with good autonomy.

Reference people: Saman Korjani (skorjani@fbk.eu)

3. AI-driven forecasting and multi-market optimization of battery energy storage.

The BET unit is seeking a motivated MSc student interested in the development of forecasting and optimization methods for battery participation in electricity markets. The revenue that a battery energy storage system can extract from these markets depends less on its physical capabilities than on the quality of the information available at bidding time, since the various energy market products must be committed hours in advance under uncertain renewable generation, net load and price trajectories. Deterministic point forecasts are insufficient for this purpose: what the scheduler requires is a representation of the plausible range of future conditions, so that bids remain profitable across outcomes rather than optimal for a single expected path.
The thesis will focus on the development of an integrated scenario-based forecasting and scheduling framework, in which statistical and machine-learning models produce short-term predictions of market prices together with the underlying energy indicators that drive them, such as renewable generation and consumption, and generative or scenario-based methods translate the associated uncertainty into a set of coherent stochastic realizations rather than a single trajectory. Particular attention will be given to the stability and robustness of the forecasting layer, assessing how the quality of the scenario set degrades over time and under changing market conditions, and how such degradation of the forecasting core propagates into the bidding decisions. The student will build and benchmark the forecasting layer against standard baselines, generate and reduce the scenario set, and formulate a stochastic or robust optimization model for battery participation across multiple market segments, quantifying achievable revenues, the risk profile of the resulting bidding strategy and the sensitivity of both to forecast accuracy.

Ideal candidate requisites

• MSc student in Computer Science, Data Science, ICT, Energy Engineering or Electrical Engineering, with an interest in data-driven methods applied to energy markets and, more generally, to energy systems;
• Programming skills for data analysis, machine learning and optimization (e.g., Python), including the use of standard scientific, ML and solver libraries, version control with Git, and familiarity with notebook-based workflows;
• Basic knowledge of time-series forecasting, generative models for scenario generation, stochastic or robust optimization, electricity markets and/or large language models is an added value; 
• Ability to summarize results with presentations;
• Good communication and relational skills;
• Skills in problem solving;
• Ability to work in a collaborative environment with good autonomy.

Reference people: Saman Korjani (skorjani@fbk.eu)

4. Grid-forming control in hybrid AC/DC microgrids - Develop and optimize converter control strategies for hybrid AC/DC microgrid operation.

The BET unit is seeking a motivated MSc student interested in the design and experimental validation of converter control strategies for hybrid AC/DC microgrids. The growing share of DC-connected resources such as PV, batteries and EV chargers alongside conventional AC loads is driving interest in these systems. In islanded operation, where no external grid is available to set the voltage and frequency reference, grid-forming converters must coordinate control across both the AC and DC sides of the system. In practice, the converters available for such systems are often commercial units with fixed internal control, so the coordination problem must be addressed at the level of their externally accessible settings, such as droop coefficients and voltage or power setpoints.
The thesis will study grid-forming coordination for the interlinking converter connecting an AC subgrid and a DC subgrid, using MATLAB/Simulink to model the system and determine suitable setpoint parameters. The core of the work is experimental: the identified parameters will be applied to the hybrid AC/DC microgrid available in the laboratory, where the student will define and run a test campaign under disturbance scenarios such as load steps, acquire and process the resulting voltage, frequency and power measurements, and compare the measured response with the simulated one in order to refine both the model and the parameter selection. Particular attention will be given to the practical constraints imposed by the commercial converters and to the tuning of the accessible settings on the real hardware.

Ideal candidate requisites

• MSc student in Electrical Engineering, Power Electronics, Energy Engineering or Mechatronics Engineering;
• Programming and modelling skills (e.g., MATLAB/Simulink, Python);
• Basic knowledge of power electronics, converter control and/or microgrid operation is an added value;
• Interest in experimental work and laboratory testing;
• Ability to summarize scientific results with presentations;
• Good communication and relational skills;
• Skills in problem solving;
• Ability to work in a collaborative environment with good autonomy.

Reference people: Amir Naebi Toutounchi (anaebitoutounchi@fbk.eu), Edoardo Macchi (emacchi@fbk.eu)


5. Physics-informed graph neural networks for grid monitoring.

The BET unit is seeking a motivated MSc student interested in the application of machine learning to power system monitoring. Distribution grid monitoring and state estimation increasingly rely on data-driven tools to cope with sparse measurement infrastructure. Graph Neural Networks (GNNs) are a natural fit given the graph-structured nature of electrical networks, but purely data-driven GNNs can produce estimates that violate basic physical laws such as the power flow equations. Embedding physical constraints into the learning process can improve accuracy and consistency compared to conventional methods.
The thesis will implement a physics-informed GNN for state estimation on a standard IEEE test feeder, embedding power flow constraints into the loss function. The student will build the network model and training dataset, design and train the physics-informed architecture, and evaluate its performance under varying measurement availability and noise conditions, benchmarking the results against a standard estimation method such as Weighted Least Squares in terms of estimation accuracy, physical consistency and computational cost.

Ideal candidate requisites

• MSc student in Electrical Engineering, Energy Engineering, Computer Science or Data Science with an interest in power systems;
• Programming skills for machine learning and data analysis (e.g., Python, PyTorch);
• Basic knowledge of power system analysis, state estimation and/or graph neural networks is an added value;
• Ability to summarize scientific results with presentations;
• Good communication and relational skills;
• Skills in problem solving;
• Ability to work in a collaborative environment with good autonomy.

Reference people: Amir Naebi Toutounchi (anaebitoutounchi@fbk.eu)


6. Flexible data centres with energy storage for grid congestion management.

The BET unit is seeking a motivated MSc student interested in the modelling and optimization of flexible loads for grid support. Data centres represent one of the fastest-growing loads on distribution and transmission networks, and their concentration at single connection points makes them a direct contributor to congestion; at the same time, a substantial part of their consumption is technically flexible, since non-critical workloads can be shifted in time, cooling systems possess thermal inertia, and on-site energy storage and backup systems are already installed for reliability reasons. Exploiting these degrees of freedom turns the data centre from a passive constraint into a controllable resource for the network operator, provided that the flexibility is quantified under realistic technical limits rather than assumed.

The thesis will focus on the development of an optimization model in which non-critical workload scheduling, cooling flexibility, batteries and backup systems are co-optimized to relieve distribution or transmission congestion. The student will characterize the available flexibility of each of these resources, formulate and solve the scheduling problem under network and service-level constraints, and quantify the technical and economic potential of the resulting strategy together with the operational risks it entails for the data centre.

Ideal candidate requisites

• MSc student in Energy Engineering, Electrical Engineering, Computer Science or Data Science, with an interest in the interaction between digital infrastructure and power systems;
• Programming skills for modelling and optimization (e.g., Python, MATLAB);
• Basic knowledge of optimization methods, power system operation and/or data centre energy systems, such as cooling and backup infrastructure, is an added value;
• Ability to summarize scientific results with presentations;
• Good communication and relational skills;
• Skills in problem solving;
• Ability to work in a collaborative environment with good autonomy.

Reference people: Saman Korjani (skorjani@fbk.eu), Edoardo Macchi (emacchi@fbk.eu)