Data-driven Arctic wind energy analysis by statistical and machine learning approaches

• Main Author: Chen, Hao
• Format: Doctoral or Postdoctoral Thesis
• Language: English
• Published: UiT Norges arktiske universitet 2022
• Subjects: VDP::Mathematics and natural science: 400::Information and communication science: 420::Mathematical modeling and numerical methods: 427; VDP::Matematikk og Naturvitenskap: 400::Informasjons- og kommunikasjonsvitenskap: 420::Matematisk modellering og numeriske metoder: 427; VDP::Mathematics and natural science: 400::Information and communication science: 420::Simulation; visualization; signal processing; image processing: 429; VDP::Matematikk og Naturvitenskap: 400::Informasjons- og kommunikasjonsvitenskap: 420::Simulering; visualisering; signalbehandling; bildeanalyse: 429; VDP::Technology: 500::Electrotechnical disciplines: 540::Electrical power engineering: 542; VDP::Teknologi: 500::Elektrotekniske fag: 540::Elkraft: 542; DOKTOR-004; Arctic
• Online Access: https://hdl.handle.net/10037/26938

Norway's Arctic region is rich in wind resources and developing wind energy in the region can promote a green transition and economic development. However, the region's unique topography with fjords and mountains and cold climate conditions make wind resource assessment, generation analysis, and power forecasting particularly challenging. The accumulation of wind data and the emergence of data science give new promise to this issue. “Can advanced statistical and machine learning methods deliver effective and accurate analysis for wind energy in these Arctic landscapes that are characteristics with dramatically fluctuating wind?” The thesis systemically answers the question with the chronological order of the wind power generation process. First, a statistical probabilistic modeling approach is utilized to assess wind energy resources in particular wind speed and its volatility, both from measured and numerically modeled wind data. The accurate assessment results contribute to evaluating wind resources of sites in the Arctic region. Then, we propose a wind power curve model to monitor wind power generation for the Arctic wind park. The model involves quantifying wind turbulence, clustering meteorological data, and ensemble learning and reaching a satisfactory modeling result for the park power curve. Finally, we demonstrate that traditional machine learning methods can be used to make short-term wind power forecasts for the Arctic wind parks, and these forecasts could be improved to some extent by applying appropriate meteorological wind data, as inputs, to the forecasting models. Moreover, we developed a novel approach for turbine forecasting with appropriate data processing techniques, and loading the data into large deep learning models allows for more accurate forecasting in different terrain conditions. Further, we utilized a variety of transfer learning techniques to make it possible to refine the raw data information and transfer large accurate but slow training forecasting models to smaller and faster ...