제주대학교 Repository

풍력발전단지 에너지 생산량 예측 고도화를 위한 입력 기상요인 영향성 분석 및 기계학습 모델 평가

Metadata Downloads
Alternative Title
Analysis of the influence of meteorological factor and evaluation of machine learning models for advanced prediction of wind farm energy production
Abstract
Various prediction models have been proposed to precisely predict wind speed or power generation of wind turbines. However, it is difficult to propose a single prediction model that can be applied to all cases since wind energy has an intermittent nature. This characteristic means uncertainty and variability, and energy yield is dependent on the environmental conditions of each site where wind turbines are operated.
There are two techniques for predicting power generation of a wind turbine: a physical model-based and a statistical model-based approaches. A statistical model constructs a model based on already accumulated data, which analyzes the quantitative correlation between the data of wind power generation facility and meteorological factors. Currently, statistical model tends to transition to more improved approaches called machine learning and artificial intelligence models.
Input features generally used in the prediction of wind turbine power generation using machine learning algorithms are wind speed, wind direction, temperature, pressure, relative humidity, while atmospheric stability and turbulence characteristic values are rarely used. However, since atmospheric stability and turbulence characteristic values can significantly affect the actual output performance of a wind turbine, it is necessary to construct a prediction model using these factors and evaluated the prediction accuracy of the constructed model. Moreover, further studies on detailed role and scope of these factors on energy production need to be conducted as there are insufficient research results suggesting the difference in annual power generation from a wind turbine or wind farm scale due to the changes in atmospheric stability and turbulence characteristic values.
There are three objectives of this study to resolve these issues. First, atmospheric stability, turbulent kinetic energy, turbulence intensity, and wind speed shear factor for an onshore wind farm, a target site of this study, are classified by regime to analyze the effect on the output and AEP of a single wind turbine (and multiple wind turbines). Second, prediction contribution of individual meteorological factors was presented during the energy production prediction process in the wind farm using four machine learning algorithms and 13 meteorological factors. Finally, the prediction accuracy of the model is evaluated after constructing energy production prediction models using atmospheric stability and turbulence characteristic values that can be applied to wind farms.
For a single wind turbine, high AEP is shown when regime of atmosphere is medium instability, high TI and TKE. Although the wind turbulence is considered to have a negative effect on wind turbine operation, conditions with weakened turbulence and a certain level of turbulence is turned out to be more advantageous than fully developed turbulent flow or the absence of turbulence in terms of wind turbine power generation. For wind farms, the difference in annual power generation for each regime according to the atmospheric stability is approximately 5-7%. More specifically, when the atmosphere is stratified into weak instability and stability regimes, the output differences between wind turbines are 25% and 45%, respectively. The output differences are confirmed to be significant in terms of statistical point of view. In the evaluation of the accuracy of prediction model for wind farm power generation using machine learning, the models constructed using atmospheric stability and turbulence characteristic values show high prediction accuracy. In particular, input features that showed the second highest prediction contribution following wind speed during the prediction process are turbulent kinetic energy, turbulence dissipation rate, and turbulence intensity. These features are analyzed to provide very useful information for improving the accuracy of the prediction model.
The results of this study can be used for the advancement of wind turbine power production forecasting technology. Improved forecasting technology enables efficient operation of highly variable wind energy source, practically helping secure the stability and economic feasibility of wind power generation facilities
Author(s)
김대영
Issued Date
2022
Awarded Date
2022. 2
Type
Dissertation
URI
https://dcoll.jejunu.ac.kr/common/orgView/000000010610
Alternative Author(s)
Kim, Dae Young
Affiliation
제주대학교 대학원
Department
대학원 풍력특성화협동과정
Advisor
김범석
Table Of Contents
Ⅰ. 서론 1
1.1 연구 배경과 동향 1
1.1.1 풍력터빈 출력에 영향을 미치는 환경 조건 1
1.1.2 대기상태와 풍력터빈의 후류 회복성 3
1.1.3 기계학습 알고리즘을 이용한 풍력터빈 에너지 생산량 예측 6
1.2 연구 목적 8
Ⅱ. 풍력 발전단지 환경 조건과 기상자료 수집 11
2.1 측정사이트와 관측장비 구성 11
2.2 기상요인과 유효자료 15
2.2.1 기상요인 종류와 레짐 분류 15
2.2.2 자료 품질 관리 22
Ⅲ. 기계학습 모델 구축과 성능 평가 방법 27
3.1 연간발전량과 정규화된 출력 계산 27
3.2 통계검정 29
3.3 모델링과 검증 30
3.3.1 기계학습 알고리즘 31
3.3.2 하이퍼파라미터 최적화와 교차 검증 32
3.4 기상요인 예측 기여도 계산 34
3.5 예측 모델의 종류와 구성: 대기안정도와 난류 특성값의 활용 35
Ⅳ. 풍력 발전단지 에너지 생산량 예측 정확도 평가 38
4.1 기상요인에 의한 풍력터빈의 출력 특성과 연간발전량 변화 38
4.1.1 기상요인 분포 특성 38
4.1.2 단일 풍력터빈의 연간발전량 차이 50
4.1.3 풍력 발전단지의 연간발전량 차이 56
4.2 에너지 생산량 예측 모델 정확도 분석 61
4.2.1 기계학습 알고리즘의 예측 성능 61
4.2.2 기상요인 예측 기여도 비교 분석 64
4.2.3 기상요인 예측 기여도 비교 분석 74
Ⅴ. 결론 77
5.1 요약 및 결론 77
5.2 추후 연구 79
참고문헌 81
부록 81
Degree
Doctor
Publisher
제주대학교 대학원
Appears in Collections:
Interdisciplinary Programs > Multidisciplinary Graduate School Program for Wind Energy
공개 및 라이선스
  • 공개 구분공개
파일 목록

Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.