An Investigation for Application of Numerical Optimization Techniques on the Energy System Model REMod

Master Thesis Defense

Mohammad Qanadilo

University of Freiburg / Fraunhofer ISE

Thursday, December 15, 2022, 15:00

Room 01-012, Georges-Köhler-Allee 102, Freiburg 79110, Germany

In order to achieve the global climate target of keeping warming below 2°C, a reduction of carbon emissions is required. This can be achieved using Energy System Models (ESM). ESMs are mathematical models that assist researchers in determining the optimal energy expansion strategy for the desired target. These models tend to get more complicated when more elements are introduced to the system and thus demanding a more powerful optimization technique to stay efficient. In 2012, Fraunhofer-Institute developed a model called REMod that aims to reduce energy-related CO2 emissions over the years in Germany. Throughout recent years, the scope of the model has been extended to incorporate more energy sectors and technologies to become more comprehensive. Researchers at the institute have been working diligently to improve the optimization approach to reduce the computational cost of running the model. In this master’s thesis, a numerical optimization approach called "grey-boxing" was sought after to evaluate if it could outperform the currently used black-box optimization technique. The model REMod was split into two parts. The objective functions of one part of the model are numerically formulated from the code to define the white-box of the approach. Then, the sparsity as well as the Jacobian approximation of the black-box part of the model are computed using finite differences by the mathematical library TD22. This work combines these two components into a single optimization problem and performs the fundamental computations, concluding by outlining the necessary steps for discovering a workable strategy that can be simultaneously applied to these two functions that originate from various optimization fields. Furthermore, the envisioned approach is restricted by the need to abide by the requirements of both white-box and black-box optimization problems. A complete attempt at running the envisioned algorithm was not yet made in this work, but rather the core components have been laid out.


Also online via Zoom:
Meeting ID: 627 9173 7415
Passcode: syscop2021