Transactions of China Electrotechnical Society  2023, Vol. 38 Issue (7): 1793-1807    DOI: 10.19595/j.cnki.1000-6753.tces.211893
Current Issue| Next Issue| Archive| Adv Search |
Cleaning Method of Wind Speed Outliers for Wind Turbines Based on Velocity and Correlation Constraints
Li Yang1, Shen Xiaojun1, Zhang Yangfan2,3, Wang Yu2,3
1. College of Electronic and Information Engineering Tongji University Shanghai 200092 China;
2. State Grid Jibei Electric Power Research Institute Beijing 100045 China;
3. Grid-connected Operation Technology for Wind-Solar-Storage Hybrid System State Grid Corporation Key Laboratory Beijing 100045 China

Download: PDF (2166 KB)   HTML (1 KB) 
Export: BibTeX | EndNote (RIS)      
Abstract  The wind speed sensors of wind turbines, due to long-term exposure to the harsh operating environment, are prone to instability or misalignment, resulting in a large amount of abnormal data in the collected raw data. The existing data cleaning methods for wind turbines, which primarily focus on the mapping relations between wind speed and wind power in power curves, are designed for wind power generation performance evaluation or wind power curve fitting. Few anomaly detection methods of wind speed data are proposed for the performance evaluation and health management (PEHM) for wind speed sensors. Consider the fact that different applications have different boundaries and requirements for data cleansing, this paper proposes a wind speed data cleaning method based on temporal and spatial correlation characteristics to deeply mine various abnormal data, especially the weak fault data located in or around the wind power curve. The anomaly detection results are helpful to find more information of wind speed sensers behaviors, which can be furtherly employed for PEHM.
Firstly, the typical abnormal wind speed data in wind farms and their temporal features are summarized, which are classified into the sudden-changes and gradual-changes wind speed outliers. Then, a combined method using velocity and correlation constraints is proposed to detect various outliers. Since the local features of wind speed data are prone to partial coverage and annihilation when using long time scale data to model their temporal relations and correlations, this paper proposes a change-points detection algorithm based on shape binary segmentation to perform temporal interval partitioning for the global wind speed series. Thus, the velocity and correlation constraints are constructed for each sub-series in temporal and spatial dimensions to conduct data cleaning tasks.
The experiment results on two datasets, including a simulated dataset and an actual wind farm dataset, show that, the proposed method is capable of detecting various abnormal wind speed data for wind turbines. The precision rates of the proposed anomaly detection method for sudden-changes and gradual-changes abnormal wind speed data reach 95.57% and 100%, respectively. Statistics on the abnormal data indicate that the gradual-changes abnormal wind speed data holds the dominate proportion among the detected outliers and the rate of abnormal data reaches 21.87%. An explanation for that is the wind capture performance of the selected sensor decrease severely, instead, the anti-electromagnetic performance is perfect. Interestingly, there are many scatters inside the wind speed and power curve that are detected as abnormal wind speed data by the proposed method. However, such scatters are detected as normal data for power curves fitting and wind power generation performance evaluation, which furtherly proves that the boundaries and definitions for outliers vary for different data category and digital applications.
The following conclusions can be drawn from the case analysis: (1) The shape binary segmentation-based change-points detection method is capable of effectively identifying the moments when the temporal features of wind speed series change significantly. The temporally segmented local wind speed sequences are mor suitable to correlations modelling with higher accuracy and reliability. (2) The proposed data cleaning method based on velocity and correlation constraints provides accurate detection for various abnormal wind speed data in wind farms, especially for the weak fault data. (3) This paper presents a general data cleaning framework for time series based on spatial and temporal correlation characteristics, which takes the differences of the sensitivity to abnormal data in practical applications into account and can accurately identify various abnormal wind speed data with precision rate of 95.57% and 100% for sudden-changes and gradual-changes outliers.
Key wordsWind turbines      wind speed outliers      data cleaning      shape-based binary segmentation      velocity-correlation constraints     
Received: 18 November 2021     
PACS: TK83  
Service
E-mail this article
Add to my bookshelf
Add to citation manager
E-mail Alert
RSS
Articles by authors
Li Yang
Shen Xiaojun
Zhang Yangfan
Wang Yu
Cite this article:   
Li Yang,Shen Xiaojun,Zhang Yangfan等. Cleaning Method of Wind Speed Outliers for Wind Turbines Based on Velocity and Correlation Constraints[J]. Transactions of China Electrotechnical Society, 2023, 38(7): 1793-1807.
URL:  
https://dgjsxb.ces-transaction.com/EN/10.19595/j.cnki.1000-6753.tces.211893     OR     https://dgjsxb.ces-transaction.com/EN/Y2023/V38/I7/1793
Copyright © Transactions of China Electrotechnical Society
Supported by: Beijing Magtech