Photovoltaic power generation uncertainty forecast for microgrid energy management efficiency enhancement.

Directeurs.rices de thèses : Migan-Dubois A. & Badosa J. & Bourdin V.

Date 2020-09-28
Diplôme Sorbonne Université


Composition du jury

ALONSO, Corinne, Professeur des Universités, Rapporteur
GUERRERO, Josep M., Professeur des Universités, Rapporteur
HELIER, Marc, Professeur des Universités, Examinateur
BONNASSIEUX, Yvan, Professeur des Universités, Examinateur
MIGAN-DUBOIS, Anne, Professeur des Universités, Directeur de thèse
BADOSA, Jordi, Ingénieur de Recherche, Co-encadrant
BOURDIN, Vincent, Ingénieur de Recherche, Co-encadrant


The research work hereby presented, pretends to contribute with the well-known issue of
dealing with the uncertainty of intermittent renewable sources (solar photovoltaic particularly)
in a microgrid. More specifically, the objective of this work is to evaluate the impacts of
solar photovoltaic production uncertainty in the performance of a microgrid that serves a
smart-building, and to propose, test and validate strategies to deal with it. To tackle this
problematic the work has been divided in three parts.
First, the design and construction of a -laboratory scale- nanogrid (300W peak consumption)
has been carried out, along with a control interface to interact with the system and
collect the data. This system is intended to emulate a microgrid that is being deployed in
the Drahi-X Novation Center building (campus of Ecole Polytechnique, Palaiseau, France,
48,7ºN, 2.2ºE), so that it serves as a test-bench for different energy management scenarios.
The system conceived presents a direct-current common-bus architecture, where the power is
freely exchanged among all the elements of the microgrid. The nanogrid is equipped with
a measurements system, disconnection means for every element, as well as the capacity to
control the power transacted by the battery, to some extent. The following of the real-time
-scaled- consumption of the Drahi-X building is also possible. Some remote monitoring
and control capabilities are included, as well as the gathering of some meteorological variables.
Several months of power-flows data were gathered, that served for different analysis
regarding the electrical interactions among elements of the nanogrid. This helped to improve
the understanding of these type of systems in order to propose proper solutions for the
uncertainty issue mentioned above. The nanogrid also served as a pedagogical tool that
allowed many students to get a hands-on knowledge regarding microgrids through several
practical experiences and internship projects that were performed in our laboratory using the
nanogrid. The pedagogical and demonstrative outcomes obtained from the implementation
of the nanogrid are considered very significant objectives for the laboratory as well as for the
interests and future academic career of the author of this work.
In a following step, the topic of solar irradiance forecasts is addressed. Profiting from the
expertise of the SIRTA atmospheric observatory and the Dynamic Meteorologic Laboratory
in the domain of weather forecasting, a collaborative work was performed in order to evaluate
what is the reliability of readily available day-ahead solar irradiance forecasts which in turn,
will be used to produce predictions of photovoltaic power production. For this, an analogsensembles
method is proposed to obtain probabilistic information from the above-mentioned
deterministic forecasts, to evaluate its eventual added value for an energy management system.
The adapted analogs-ensembles method proposed demonstrated superior performance
with respect to reference -probabilistic- forecasting methods, such as persistence, monthly
climatology and a well-known commercially-available probabilistic forecasting method from
the European Centre for Medium-Range Weather Forecasts (ECMWF). The methods were
evaluated based on the quantile skill score decomposed in reliability, resolution and uncertainty,
which are state-of-the-art metrics for probabilistic forecasting assessment. Besides, the
quantile forecasts obtained from the analogs-ensembles proved to be an interesting solution to
reduce forecasting uncertainty, as they provide information about the bias of the forecasting
error. If chosen properly, this feature might be beneficial for the resource scheduling tasks
performed by an energy management system, as proven in the last part of this work.
In the third section (Chapter 4), insights of the previous parts of this work were used in
order to propose energy management strategies with the aim to deal with the uncertainty issue
of PV production. These strategies are meant to favor different services that were chosen as
indicators of performance for the study-case building, namely: the energy cost, the carbon
footprint of the energy, the grid contracted power and the grid commitment. The impact of
such strategies in the performance of the study-case microgrid is evaluated, using different
forecasting approaches. The particularity of the quantile forecasts used (i.e. providing some
degree of certainty regarding the bias of their errors), proved to improve performance on
some of the services proposed, by properly choosing the quantile to be used. It helped in
providing flexibility when facing different production, consumption and pricing scenarios.
The energy management strategies proposed were compared to different reference strategies,
including the case when no microgrid was deployed, in order to have a more meaningful
idea of the real added-value of the strategies proposed and the microgrid deployment. In
general, the EMS strategies and the quantile forecasting method proposed, outperformed
the reference strategies in almost all the scenarios studied. Therefore, the presented work
brought promising answers and elements that are worth putting in practice to further validate
their added value in a real system.