Preserving Normal Power Curve Data With Sparse Density via Wind Speed-Power Correlation Trend Cleaning Method

Abstract

Stochastic wind conditions and curtailment lead to a sparse distribution of normal data compared to outliers on the Wind Power Curve (WPC). This results in the removal of sparse normal data during the data cleaning process, hampering short-term wind power assessment and forecasting. To address this issue, this paper proposes a decision boundary construction method that utilizes the wind speed-power correlation trend to retain normal WPC data. First, leveraging the positive correlation between wind speed and power, an incremental trend search strategy is used to obtain the trend curve. Building on this curve, a scatter motion trend algorithm is introduced to eliminate densely clustered curtailed power data. Finally, a kernel function-based 3-sigma boundary construction method is suggested to further reduce the influence of remaining clustered outliers on decision boundaries. The proposed method is compared to eight advanced algorithms using data from 17 wind turbines across three wind farms, demonstrating superior performance, especially in scenarios with sparse normal data.