基于DBSCAN-BIN算法的风功率曲线构建方法及应用研究

Research on Construction Method and Application of Wind Power Curves Based on the DBSCAN-BIN Algorithm

摘要: 针对风功率曲线建模中异常数据干扰与传统方法平滑性不足的问题，提出了一种基于DBSCAN-BIN的风功率曲线建模方法。首先，为有效清洗原始数据中的异常数据，提出基于分位数回归与3σ准则的异常数据处理方法，进而提升数据分布的合理性与表征性。其次，为精准剔除高风速段离散异常点，基于DBSCAN密度聚类算法提出一种高风速段离散异常数据处理方法。再次，利用BIN方法分箱统计构建风功率曲线，实现降噪与拟合的协同优化。最后，基于MAE、MAPE、RMSE及R²模型评估指标，与BIN、KNN、RF、DTR等方法进行实验对比。实验结果表明：所提方法与传统方法相比评估指标更佳。其中，MAE为28.6177、MAPE为0.096、RMSE为39.4059、R²为0.9934。所提方法构建曲线全局过渡无突变、局部无冗余波动，对异常数据敏感度低，为风功率预测、机组性能评估、故障诊断及运维调度提供可靠支撑。

Abstract: To address the issues of abnormal data interference in wind power curve modelling and the insufficient smoothness of traditional methods, this paper proposes a DBSCAN-BIN based wind power curve modelling method. Firstly, to effectively remove abnormal points from raw data, an outlier processing method based on quantile regression and the 3σ criterion is developed, which improves the rationality and characterisation capability of data distribution. Secondly, to accurately eliminate discrete outliers in the high wind speed interval, a processing strategy for discrete abnormal data is presented based on the DBSCAN density clustering algorithm. Thirdly, the BIN method is employed to construct the wind power curve via binning statistics, realizing the coordinated optimisation of noise reduction and curve fitting. Finally, taking MAE, MAPE, RMSE and R2 as model evaluation metrics, comparative experiments are carried out with BIN, KNN, RF, DTR and other traditional algorithms. The experimental results demonstrate that the proposed method achieves superior overall performance over conventional methods. The corresponding evaluation indexes are MAE of 28.6177, MAPE of 0.096, RMSE of 39.4059 and R2 of 0.9934. The wind power curve established by the proposed method presents smooth global variation without abrupt changes and free of redundant local fluctuations, and shows low sensitivity to abnormal data. This method can provide a reliable reference for wind power prediction, wind turbine performance evaluation, fault diagnosis, as well as operation and maintenance scheduling.