Control Methods for Microgrids (Möhrke et al. 2014)
2. September 2014Linking financial engineering to energy system planning
29. September 2014Multi-Objective Optimization of Micro Grids using Evolutionary Algorithms (Wanitschke 2014)
Alexander Wanitschke
Master thesis
(September 2014)
In this thesis a multi-objective evolutionary algorithm (MOEA) is designed that is particularly suited to solve optimization problems formulated with the micro grid simulation tool SMOOTH(Simulation Model for Optimized Operation and Topology of Hybrid energy systems), developed at RLI.
The theoretical background of multi-objective optimization approaches and the history and theory of evolutionary optimization is surveyed in order to identifiy the most appropriate algorithm design to handle the characteristics of the SMOOTH optimization problem. After a number of potentially suited algorithm operators is identified, 16 algorithm variants are tested and compared on a test problem using Monte-Carlo-Simulations. The most appropriate algorithm variant, called SMOOTH-MOEA, exhibited robust convergence to optimality as well as feasibility. It showed to be highly parallelizable, which can reduce optimization runtime by a factor at the order of the number of parallel workers. SMOOTH-MOEA incorporates two newly proposed algorithm operators. Firstly, evolutionary polynomial mutation exhibited superior performance on the test problem compared to classic uniform mutation. Secondly, Leyland’s method for pareto front tail preservation, called tail box, was advanced to become the idea of tail bands, which proofed to enhance SMOOTH-MOEA’s convergence on the test problem.
The theoretical background of multi-objective optimization approaches and the history and theory of evolutionary optimization is surveyed in order to identifiy the most appropriate algorithm design to handle the characteristics of the SMOOTH optimization problem. After a number of potentially suited algorithm operators is identified, 16 algorithm variants are tested and compared on a test problem using Monte-Carlo-Simulations. The most appropriate algorithm variant, called SMOOTH-MOEA, exhibited robust convergence to optimality as well as feasibility. It showed to be highly parallelizable, which can reduce optimization runtime by a factor at the order of the number of parallel workers. SMOOTH-MOEA incorporates two newly proposed algorithm operators. Firstly, evolutionary polynomial mutation exhibited superior performance on the test problem compared to classic uniform mutation. Secondly, Leyland’s method for pareto front tail preservation, called tail box, was advanced to become the idea of tail bands, which proofed to enhance SMOOTH-MOEA’s convergence on the test problem.