Functie

Future Topics in Energy Informatics

Bedrijfsgegevens

As part of our mission towards sustainability, VITO sponsors and co-promotes researchers that persue a doctorate degree or postdoctorate research in topics relevant for society. The research group algorithms, models, and optimization (AMO) within VITO’s unit for energy technology at EnergyVille focusses on discovering, leveraging, and valorizing flexibility in energy supply and demand as a key to increasing the stability of energy systems as well as enabling an increase of renewable energy sources. We design and implement pragmatic models for various kinds of flexibility and build high-performance, data driven, control- and scheduling algorithms that optimize energy provision and consumption in operational and commercial contexts. We push the state of the art in data science, forecasting, optimal control, power flow, and energy management.

Beschrijving

Currently, AMO offers topics in flexibility aggregation, thermal modeling and optimal control, uncertainty quantification, deep learning, and optimization.

Doctoral research topics

AMO doctoral research topics aim at solving academically challenging fundamental issues with immediate practical relevance and impact in today’s energy systems

Risk-aware quantification and aggregation techniques for optimal valorization of energy flexibility

This Ph.D. proposal concerns the extension and deepening of the state-of-the-art in optimal valorization and aggregation of energy flexibility and focusses on the following novelties:

  1. A focus on uncertainty and unpredictability in the behavior and the use of energy consuming/producing devices.
  2. The quantification of the uncertainty using appropriate data science techniques, enabling us to calculate the risk of unexpected unavailability of offered flexibility to take into account when valorizing the energy flexibility in risk-critical business models.
  3. A two phase approach that first addresses the uncertainty related to the user of the devices that offer flexibility and then considers other types of stochastic uncertainty, including intrinsic uncertainty about the current state and about the effect of actions.
  4. A Risk analysis aiming at the effective availability of the flexibility to the customer/aggregator/user of the flexibility.
  5. New and novel Demand-Response optimization algorithms that take into account  the quantified risks in cooperation with energy market expertise at VITO.

 

We aim to testing and validating the approach in various multi sector scenarios.

Level

Doctorate research (Ph.D)

A picture containing person, wall, person, indoor

Description automatically generatedVITO promoter

Dr. Carlo Manna, Ph.D.

University partner

Prof. Chris Develder, UGent, BE

Improving building thermal comfort levels under uncertainty by automatic tuning of optimal control parameters

The performance of any model predictive control (MPC) strategyis in residential thermal energy management depends on the accuracy of the mathematical model describing the thermal loads and capacities, and on the quality of the forecasts of disturbances, such as weather and occupant behaviour. Deviating model parameters and inaccurate disturbance predictions are shown to give rise to increased energy costs and comfort violations. A distinction can be made between (1) parametric uncertainties, which means uncertainties that can be decreased (though not fully eliminated) by adaptation improvement of model parameter (e.g., through learning), and (2) additive uncertainties (such as forecast uncertainty) that cannot be substantially decreased but that can be dealt with in the decision-making process.

In this Ph.D. proposal we aim to combining the strengths of MPC and reinforecement learing (RL) to continuously and in real time improve controller model parameters, objective function weights for conflicting objectives, controller hyperparameters, and so on. We will focus on forward uncertainty quantification. RL-MPC and stochasticMPCap models will be used. The sensitivity of these models will be analysed with respect to parameter choices which model the building. Similarly, a refinement/adjustment of how the control enters the mathematical model will be performed. The research will also investigate how to model outside uncertain disturbances like forecasts of weather and user behaviour.

The proposed solutions will be applied in optimal control with aim of achieving a preset thermal comfort level inbuildings. In particular we are interested in the improvementthat can be achieved by temporal adjustment of the model parameters using aproperly tuned method.

Level

Doctorate research (Ph.D)

A picture containing person, smiling, posing, dressed

Description automatically generatedVITO promoter

Brida Mbuwir

University partner

Prof. Lieve Helsen, KUL, BE

Uncertainty quantification in consumption patterns and flexibility of thermal loads

Advanced control strategies, and model predictive control (MPC) in particular, are gaining widespread interest for building climate control, since they can systematically save energy and/or costs with simultaneous thermal comfort improvement, as well as adapt the energy demand according to the available renewable/residual supply. As outlined in the previous topic, MPC however suffers from parametric uncertainties (p) and additive uncertainties (a) that substantially limit the performance of these approaches.

Several computational methods to handle uncertainties in mathematical models have recently been developed at KU Leuven; for example, quasi-Monte Carlo methods in forward uncertainty quantification for PDEs with random diffusion coefficients. Quasi-Monte Carlomethods are multivariate quadrature methods which are especially suitable forhigher dimensional integrals/expectations, having the possibility of vanquishing the curse of dimensionality under certain conditions. These methods can also be used for inverse problems, such as determining parameters of a mathematical model given data, and for function approximation. In particular, Bayesian inversion allows the formulation of a point estimator for the parameters of a mathematical model in terms of (high-dimensional) integrals. If the mathematical model is relatively expensive then it can be approximated by a Gaussian process surrogate. The Gaussian process emulator can beinterpreted as a kernel interpolation method for function approximation with added uncertainty. Recently new construction methods for quasi-Monte Carlopoint sets for function approximation were developed which can also be used forkernel interpolation methods. Last but not least, in the dissertation of Van Barel at KUL, multilevel Monte Carlo and quasi-Monte Carlo methods were developed for robust optimization under uncertainty.

In this Ph.D. research, we aim to improve this methodology in direct, practical, application to optimal control problems in residential and commercial thermal energy management.

Level

Doctorate research (Ph.D.)

A picture containing person, smiling, posing, dressed

Description automatically generatedVITO pomoter

Dr. Brida Mbuwir, Ph.D.

University partner

Prof. Dirk Nuyens, KUL, BE

Coming soon: Behavioral operations research and multi objective optimization in user centric residential energy management (in planning)

This Ph.D. takes a user centric view to energy management and combines user comfort with energy price and market operation and the aim of increasing the share of renewable energy in multi-objective, multilateral, optimization.

Level

Doctorate research (Ph.D.)

A person in a suit

Description automatically generated with low confidenceVITO promoter

Dr. Sarnavi Mahesh, Ph.D.

University partner

Prof. Carlos Henggeler, UCoimbra, PT

 

Postdoctorate research topics

AMO postdoctoral research topics aim to push the state of the art in data science and optimal control for renewable energy while at the same time both building a strong academic profile and as well as proven industrial relevance for the researcher.

Supporting the participation of aggregator of a large number of prosumers through deep learning based optimization

Recent years have seen an increasing interest in Demand Response (DR) as a means to provide flexibility and increasing the penetration of renewable resource sin the power system. At the aggregator level,  where there is a multitude and variety of devices and appliances across the aggregator’s portfolio (building HVAC, EVs, water tanks, etc.), the process of control and scheduling is rendered infeasible without automating a substantial part or even the whole process.

Real-life DR systems are increasingly an aggregation of a large number of heterogeneous systems, users and devices. Modelling such heterogeneous system is challenging for the following reasons:

  • The lack of complete information and data from the different components;
  • The unavailability or the high complexity of the mathematical models for the processes to be controlled/optimized;
  • The traditional approaches (mainly from the mathematical optimization and control theory) result in complex and intractable problems as the number of devices and therefore the variables involved increase;
  • There are many sources of uncertainty (e.g., dynamic prices, user behavior, renewable resources etc.) which introduce additional complexities to the ones described to the previous points.

Moreover, the increasing integration of renewable resources in the electric power grid has increased the need for producers and virtual producers (i.e., aggregators) to offer/trade or correct their offer close to real-time (e.g., intraday/continuous intraday market) in order to better deal with the high uncertainty due to the availability of renewables.

For these reasons, fast stochastic optimization and control methods are needed. However, the current approaches (MILP, MPC etc.) cannot guarantee a good anytime solution for real-time control and scheduling due to the aforementioned computational issues.

Recently, there has been a large body of literature in deep learning that has sought to approximate or speed up optimization and control models. As described in reviews on topics such as combinatorial optimization, the underlying idea is to combine Machine Learning methods (to cope with complex stochastic models) and model-based approaches (integrating the knowledge about the process).

The aim of this project is to investigate and implement innovative machine learning based methodologies for fast optimization and control approaches, particularly focused to support aggregators to trade/offer energy flexibility close to real-time.

Level

Postdoctorate research (Postdoc)

A picture containing person, wall, person, indoor

Description automatically generatedVITO promoter

Dr. Carlo Manna, Ph.D.

Context

This postdoc position will apply for a Marie Curie Postdoctural Fund (MSCA-PF) scholarship

Please indicate in your motivation letter which position you are interested in.