基于稀疏表示方法的短期风电功率预测

Short-Term Wind Power Forecasting Based on Sparse Representation Method

针对短期风电功率预测,提出一种基于稀疏表示特征提取的建模方法。为了构建预测模型,将历史风电功率数据构成具有时延的输入-输出数据对,将时延输入数据向量作为初始字典,由K-均值奇异值分解(K-SVD)算法将其进行稀疏分解与变换至稀疏域以得到学习后的字典,由正交匹配追踪(OMP)算法获取相应的稀疏编码向量,再将该向量作为极限学习机(ELM)或支持向量机(SVM)的输入来构建全局回归模型。为了验证所提出的方法的有效性,将所提出的方法用于短期风电功率预测中,在同等条件下与单一SVM、ELM方法和非字典学习的其他稀疏表示建模方法进行了比较。实验结果表明,不同的稀疏表示建模方法均能取得很好的预测效果,其中所提出的方法具有更好的预测效果,显示出其有效性。

For short-term wind power forecasting,a feature extraction modeling based on sparse representation method is proposed.The historical wind power data was composed to the time-lagged input-output data pairs to build the forecasting model,the time-lagged input data vectors was taken as the initial dictionary.K-means singular value decomposition(K-SVD)algorithm was used sparsely decompose initial dictionary and transformed it into the sparse domain to obtain the dictionary after learning.The corresponding sparse coding vectors were obtained by orthogonal matching pursuit(OMP)algorithm,and then the sparse vectors were taken as the input vectors of the extreme learning machine(ELM)or support vector machine(SVM)to build a global regression model.In order to verify the effectiveness of the proposed method,it was applied to short-term wind power forecasting,compared with single SVM,ELM method as well as other forecasting methods using sparse representation with non-dictionary learning under the same conditions.Experimental results show that the forecasting methods based on sparse representation technology can achieve good forecasting results,and the proposed methods may significantly improve the accuracy of wind power forecasting and show their effectiveness.