Browsing by Author "Koski, Teemu Petteri"

Now showing 1 - 1 of 1

Tietokonekeskuksen mikroverkko-järjestelmän energianhallinta
(2010) Koski, Teemu Petteri
School of Electrical Engineering | Master's thesis
This thesis focuses on energy management in micro grid. Micro grid is limited around a data centre and the components that are directly connected with the data centre. These components are: energy sources and energy consumers such as servers and reserve power. Energy management in micro grid is done with a controller with an optimization algorithm. Optimization is done with Mesh Adaptive Direct search (MADS) which is used to control generator's power output. The MADS algorithm is compared with Genetic Algorithm (GA). The MADS algorithm was chosen to be used in this thesis because there wasn't significant difference in total costs but MADS worked much faster than GA. Energy sources that were chosen in the thesis were: a diesel generator, a micro turbine, a wind turbine, a fuel cell and a photovoltaic cell. Each energy source is introduced individually with its positive and negative characteristic each chapter introduces one computer model that can be found in literature and is being used in this thesis. In order to MADS function to work, cost functions were introduced for each energy sources. The cost functions were formulated so that costs were functions of produced energy. The cost functions were based on manufacturers' datasheets or research results in literature. Prediction methods were introduced for renewable energy sources that didn't have any cost functions. The prediction methods didn't work very well compared to the straightforward prediction where the next predicted value is the same as the latest measurement. In the last part a complete micro grid model is built and simulated in different working conditions: normal condition, during an islanding process and how fuel and electricity prices affect optimization and total costs. Micro grid was combined by using the energy source models, an electricity consumption model of the data Günter and the optimization controller. Lastly the MADS controller was compared with a self-implemented supervisor controller. Comparing results clearly points out some problems with MADS algorithm like slowness to find a new optimum point when prices change. Also MADS controller output contains unnecessary variation for energy sources. Finally, it is discussed how the optimization controller should be improved if the controller would be implemented in the data centre. Also some future work is pointed out.