中文摘要：

考虑到直接对经验模式分解（EMD）所得多个分量分别建模预测会引入多重随机误差和产生较大预测工作量，提出一种基于游程检测法重构原则的EMD—Elman神经网络组合的风电短时功率预测算法，运用游程检测法对风电出力时间序列EMD得到系列本征模态函数IMF和趋势项Res进行波动性程度检测，将波动程度相似、变化规律相近的分量依照finetocoarse顺序重构成高频分量、低频分量和趋势项。然后针对性地对3个分量分别建立较准确的Elman神经网络短时多步预测模型，可减少预测分量建模数，提高预测精度和预测速度，最后将三分量预测结果自适应叠加。还分别给出两种预测模型的算例，对比分析发现EMD．Elman组合预测模型的精度优于Elman神经网络单一预测模型。

英文摘要：

Because multiple random errors directly forecasting components composed based on EMD and Elman neural network would be brought in forecasting results and extra work may be added through by empirical mode decomposition （EMD）, a wind power prediction method relying on runs-test reconstruction was proposed which reduced prediction components and improved forecast accuracy. Firstly, wind power output time series were composed by EMD, then components would be reconstructed into three parts by runs-test method in line with fine-to-coarse order. Then more accurate short-term step predictive models of Elman neural network aiming at the three components were established to reduce the number of predicted component and improve the prediction accuracy and prediction speed. At last, the three- component predictions were adaptive superimposed. Two examples of prediction models were also given, respectively. Through comparative analysis of calculated predictive indices, it was found that the prediction precision of EMD-Elman combined forecast model was higher than that of Elman neural network.