Abstract: In response to the low accuracy of power prediction and poor fitting of prediction curves of wind turbines under the background of big data of wind power,a wind power prediction algorithm based on eXtreme Gradient Boosting (XGBoost)—General Regression Neural Network (GRNN) was proposed,which established a wind turbine power prediction model considering segmented wind power data.A Supervisory Control And Data Acquisition (SCADA) data segmentation method based on wind turbine operating state characteristics and wind speed distribution model was proposed,and a power correlation index framework was constructed based on multidimensional data analysis.A segmented power prediction algorithm combined GRNN based on improved XGBoost variables for wind turbines was proposed to get more accurate and less error-prone power prediction values.Furthermore,the predictive performance of the proposed model was evaluated based on prediction deviation and curve fitting.Finally,the 20 wind turbines in the Saihanba wind farm of Inner Mongolia were analyzed as an example.The experimental results showed that compared to traditional prediction methods,the R2 mean value of the proposed method was enhanced by at least 0.0101.Compared with the full segment data prediction,R2 with the segmented prediction was enhanced by 0.0084.The curve fitting rate of the proposed model was 0.9184 that enhanced by at least 0.036 compared with the other four models.

Key words: big data of wind power, wind turbines, extreme gradient boosting, general regression neural network, piecewise power prediction algorithm

摘要： 针对风电大数据背景下风电机组功率预测准确性、预测功率曲线契合率低等问题,提出一种基于XGBoost-GRNN的风功率预测算法,建立考虑分段式风电数据的风电机组功率预测模型。首先,提出基于风电机组运行状态特征、风速分布模型的SCADA数据分段划分方法,并基于数据多维度分析构建功率关联指标架构。其次,提出一种基于改进极端梯度提升(XGBoost)变量的广义神经网络(GRNN)联合风电机组分段式功率预测算法,以获取准确性较高、误差较小的功率预测值。进一步,基于预测偏差、曲线契合率等指标评估所提预测模型的预测性能。最后,以内蒙古塞罕坝风电场20台风电机组为例进行实验分析,结果表明:与传统预测方法相比,所提方法R2均值至少提高了0.010 1;与全段数据预测相比,分段式预测R2提高了0.008 4。所提模型预测曲线契合率为0.918 4,相比其余4种模型预测曲线契合率至少提高了0.036。针对风电大数据背景下风电机组功率预测准确性、预测功率曲线契合率低等问题,提出一种基于XGBoost-GRNN的风功率预测算法,建立考虑分段式风电数据的风电机组功率预测模型。首先,提出基于风电机组运行状态特征、风速分布模型的SCADA数据分段划分方法,并基于数据多维度分析构建功率关联指标架构。其次,提出一种基于改进极端梯度提升(XGBoost)变量的广义神经网络(GRNN)联合风电机组分段式功率预测算法,以获取准确性较高、误差较小的功率预测值。进一步,基于预测偏差、曲线契合率等指标评估所提预测模型的预测性能。最后,以内蒙古塞罕坝风电场20台风电机组为例进行实验分析,结果表明:与传统预测方法相比,所提方法R2均值至少提高了0.010 1;与全段数据预测相比,分段式预测R2提高了0.008 4。所提模型预测曲线契合率为0.918 4,相比其余4种模型预测曲线契合率至少提高了0.036。

关键词: 风电大数据, 风电机组, 极端梯度提升, 广义神经网络, 分段式功率预测算法