| Summary: | Abstract. Data centers are increasingly using renewables such as wind and solar energy. RISE’s ICE data center has already solar panels and is now studying impact of adding a wind turbine into their microgrid. In this thesis, a machine learning model was developed to forecast wind power production for the data center. Data center in Luleå has several applications to utilize wind power forecasting. Renewable energy sources are intermittent, so accurate forecasting of output power reduces a need for additional balancing of energy and reserve power in an electricity grid. Renewable energy can be reserved from market for next hour or next day to maximize its use. Forecasting from 30 min to 6 hours ahead allows job scheduling to optimize usage of renewables and to reduce power consumption. Data center may target to minimize electricity cost or maximize usage of renewables for lower greenhouse gas emissions. Smart microgrid based on artificial intelligence is the way to implement the applications. Two open data sets from India and Sweden have been used in the research. The data available supports choosing of a statistical model. Random forest regression was the model used in the research. Data from India enabled to develop a model for one wind turbine. Developed model forecasted output power well. Swedish data set is from EEM20 competition, it included total wind power production in Sweden and had to be applied to approximate production of one wind turbine in Luleå. To achieve the goal output power of Luleå price region was averaged, and location for the simulation was chosen to be near Luleå. As expected, the accuracy of forecasting with Swedish data was reasonable, but approximations done reduced it. The developed model was applied to RISE’s ICE data center. Validation has been done, but final testing will take place in RISE’s simulation environment. In general, data from northern Sweden is not openly available for wind power forecasting. In addition, any scientific articles covering the geographical area were not ...
|
|---|