|Title||Fitness landscape analysis for the operational planning of distributed energy resources|
The analysis of so-called fitness landscapes is often used to better understand the structure and characteristics of optimization problems and
thus to select suitable algorithms. A fitness landscape is defined by the search space, which contains all possible solutions of the problem, a
distance measure, which indicates how far apart these solutions are, and an objective function, which assigns a certain value to each solution
and for which the optimal value must be found. For continuous problems, the fitness landscape can therefore be described as a landscape where
the search space is the lower ground and the landscape surface is raised according to the values of the objective function. Analogous to a
geographic landscape, fitness landscapes can consist of peaks, valleys, plains, ravines, cliffs, plateaus, and basins. Examining these landscape
features provides insight into how difficult it is to find an optimum, i.e., the highest mountain peak (maximization) or the lowest valley
(minimization) (link: https://www.researchgate.net/publication/225336568_A_Comprehensive_Survey_on_Fitness_Landscape_Analysis ).
The scheduling of decentralized energy resources is an optimization problem in which the schedules of all plants have to be coordinated to
achieve a common target schedule (link: https://www.annals-csis.org/Volume_18/drp/pdf/160.pdf ). Here, the analysis of the problem properties
should also provide valuable information for the selection and parameterization of optimization algorithms. In this case, the search
space is composed of the flexibilities, i.e., the operating possibilities of all plants.
|Home institution||Department of Computing Science|
|Type of work||conceptual / theoretical|
|Type of thesis||Master's degree|
|Author||Paul Hendrik Tiemann, M. Sc.|
There are two challenges in fitness landscape analysis for distributed operational planning. On the one hand, many landscape features rely on
so-called random walks. This involves changing candidate solutions gradually, resulting in a trajectory through the search space. To apply this
to asset flexibility, a suitable sampling strategy must be developed to create a set of valid schedules.
On the other hand, the search spaces of the individual plants are coupled, since the schedule selection of other plants influences how well the
target schedule is jointly achieved. Existing concepts for fitness calculation in coupled search spaces must therefore be adapted accordingly.
To evaluate the fitness landscape features, small optimization scenarios will be created and solved by agent-based distributed optimization
heuristics. Subsequently, correlation analysis will be used to investigate the strength of the fitness landscape features.
- Goal-oriented and independent way of working
- Programming skills in Python