Research on Short-Term Wind Power Forecasting by Data Mining on Historical Wind Resource
Research on Short-Term Wind Power Forecasting by Data Mining on Historical Wind Resource
复制标题
基于历史风资源数据挖掘的短期风电预测研究
作者：
B. Tang;Yan Chen;Qin Chen;M. Su
为了提高短期风电功率预测的准确性，设计了一种基于数据挖掘的历史风资源短期风电功率预测方法。该方法首先剔除风电涡轮机和气象监测设备造成的数据污染，然后利用气象数据的时空相关性，利用Lomnaofski优化模型对数据进行补充。其次，利用连续时间相似性聚类（CTSC）方法对风场特征进行了分析，并对相似样本进行了筛选。为了提高确定性预测的精度和预测误差，建立了能够逼近非线性解的径向基函数神经网络（RBF）确定性预测模型。此外，提出了模糊信息粒化与Elman神经网络相结合的风电功率区间预测方法（FIG-Elman），以获取预测区间。RBF-CTSC模型的确定性预测精度高，能够准确描述风速的随机性、波动性和非线性特性。此外，平均绝对误差（MAE）和均方根误差（RMSE）降低了新的模型。FIG-Elman结果的区间预测表明，区间宽度减少了18.85%，区间覆盖概率增加了10.94%。
In order to enhance the accuracy of short-term wind power forecasting (WPF), a short-term wind power forecasting method based on historical wind resources by data mining has been designed. Firstly, the spoiled data resulting from wind turbine and meteorological monitoring equipment is eliminated, and the missing data is added by the Lomnaofski optimization model, which is based on the temporal-spatial correlation of meteorological data. Secondly, the wind characteristics are analyzed by the continuous time similarity clustering (CTSC) method, which is used to select similar samples. To improve the accuracy of deterministic prediction and prediction error, the radial basis function neural network (RBF) deterministic forecasting model was built, which can approximate nonlinear solutions. In addition, the wind power interval prediction method, combining fuzzy information granulation and an Elman neural network (FIG-Elman), is proposed to acquire forecasting intervals. The deterministic prediction of the RBF-CTSC model has high accuracy, which can accurately describe the randomness, fluctuation and nonlinear characteristics of wind speed. Additionally, the mean absolute error (MAE) and root mean square error (RMSE) are reduced by the new model. The interval prediction of FIG-Elman results show that the interval width decreased by 18.85%, and the coverage probability of interval increased by 10.94%.