Evolutionary Multi-Objective Optimization of Micro Grids (Wanitschke 2015)
Alexander Wanitschke
A new simulation environment called SMOOTH (Simulation Model for Optimized Operation and Topology of Hybrid energy systems) is being developed at RLI as an optimization tool for cost minimization or maximization of sustainability. Evolutionary algorithms (EA) were found to be along the most useful and promising methods in hybrid energy system design so they are chosen as the principle optimization approach for SMOOTH. While existing EAs for hybrid energy systems optimization are capable of dealing with some of SMOOTH’s characteristics, none of them can treat all characteristics simultaneously. Thus, it was necessary to compose a new multi-objective evolutionary algorithm (MOEA) to meet the requirements of SMOOTH.
Discussion
Past and current research in the field of evolutionary optimization was reviewed with regard to the requirements of the optimization problem formulated in SMOOTH, so that for each of the three main steps in the EA heuristic (recombination, mutation and selection) two candidate subroutines could be identified, along with a newly developed approach for speeding up convergence of the optimization process, called tail band. For comparison, a bi-objective test problem with SMOOTH’s characteristics was formulated. To compensate for the semi-stochastic nature of EAs a sufficient number of optimization runs was conducted, so that the algorithm variants could be compared with statistical significance regarding ultimate optimization success, speed of convergence, scope of constraint violations and diversity within the MOEA’s solution population.
Conclusions and Outlook
The algorithm variant identified as having superior performance, called SMOOTH-MOEA, demonstrated effective and reliable optimization behavior on the test problem. It converged the solution to a sensible tradeoff curve between the objectives of minimized LCOE and maximized SSR while satisfying the constraints 98% of the time. Due to its selection subroutine, SMOOTH-MOEA was found to be highly parallelizable, distributing the optimization function evaluations among separate workers with a parallel efficiency of over 80%. It can be expected that SMOOTH-MOEA is a suitable optimization approach for any multi-objective optimization problem with the same or similar characteristics as SMOOTH. Its parallelizability allows reducing optimization time by a factor at the order of the number of available workers.