Data-Driven Uncertainty Representation for Scheduling Flexibility in Energy Systems

Name: Mahtab Kaffash

Prof. dr. ir. Y. Willems, chair
Prof. dr. ir. G. Deconinck, supervisor

The operation of energy systems are facing serious challenges by increasing the penetration of renewable energy resources (RES). This thesis focuses on the optimal operation of photovoltaic (PV) systems installed on the roof of buildings. Due to inherent uncertainty in the generation of PV systems and the fact that the output of these systems is uncontrollable, unlike conventional generators, integrating these systems to the grid causes some challenges in the operation of distribution grid. To mitigate these
challenges, PV systems can be combined with energy storage devices to provide flexibility and compensate the mismatch between local load and PV power generation. In order to optimally operate these storage devices combined with PV systems, an estimation of the PV power generation is required.

The main goal of this thesis is to develop algorithms to represent and manage the uncertainty of PV systems combined with different energy storage devices, which provide flexibility to the system. Several approaches that estimate and represent PV power generation have been proposed in literature. The first approach includes deterministic methods, where PV generation is shown as a single value at each time step ahead. Among these deterministic approaches, data-driven methods have received a great attention in the most recent research works due to their great performance and recent development in machine learning methods and sensors to provide real data. Although using this deterministic estimation to optimally operate storage devices has a low computation time, this approach ignores the possible error in the estimation of PV generation, which may result in inaccurate and cost-ineffective solution.

To solve the challenges of deterministic approaches, stochastic approaches have been used to estimate PV power generation. In this approach, the PV generation at each time step ahead is represented by a set of values assigned to a probability. The main advantage of this approach is to provide information to the decision maker about the stochastic behavior of a random variable. In recent years, more researchers were interested in estimating PV power generation using stochastic approaches and different models were proposed, which rely on weather forecast data. While these models could be economically feasible for large scale of PV installation, like utility level, it would not be cost-effective for smaller scale size, which is installed on the roof of buildings.

Therefore, this thesis attempts to build on recent literature in the forecast of PV power generation with focus on industrial and commercial level and contribute to the development of control algorithm for energy storage devices under uncertainty of PV generation. To this end, first, two different data-driven models to represent the uncertainty in the PV generation are proposed in this thesis. Then, a two-stage stochastic optimization is developed to manage a PV-battery system to deal with uncertainty of PV system. While the results show a stable and unbiased solution, the high computation time of stochastic optimization makes it difficult to apply this method on more complex systems, such as multi-energy systems (MES) where different energy carriers are coupled together. In the final phase of this thesis, an interval optimization is proposed to deal with uncertainty in a MES to reduce the computation time of the problem. The proposed algorithms are beneficial for end-users to reduce their energy cost, for grid operator to avoid voltage issue and congestion management, and for technology provider companies to include smarter and easy to implement controller in their energy management system devices.

Prof. dr. ir. J. Driesen
Prof. dr. ir. J. Suykens
Prof. dr. ir. K. Bruninx
Prof. dr. ir. Z. Lukszo
(Delft University of Technology, the Netherlands)
Prof. dr. ir. S. Shariat Torbaghan
(Wageningen University & Research, the Netherlands)