基于FEEMD-PACF-BP_AdaBoost模型的风电功率超短期预测

ULTRA-SHORT-TERM FORECAST OF WIND POWER BASED ON FEEMD-PACF-BPDABOOST MODEL

针对风电功率超短期预测问题,提出基于快速集合经验模态分解(Fast Ensemble Empirical Mode Decomposition, FEEMD)、样本熵(Sample Entropy, SE)和BPAdaBoost集成神经网络组合的超短期风电功率预测模型。对风电功率原始数据,采用FEEMD方法将其分解为从一系列本征模态函数分量(IMF)和余项;运用样本熵来解决分量个数过多、计算量繁杂的问题,通过PACF(偏自相关系数)筛选出与预测值关联程度高的元素确定输入维数;选用泛化能力强的集成神经网络BPAdaBoost构建单步滚动预测模型并叠加获得最终值。实验结果表明,该组合模型提高了预测精度,具有可行性和有效性。

Aiming at the problem of ultra-short term wind power prediction, a ultra-short term wind power predication model based on fast ensemble empirical mode decomposition(FEEMD), sample entropy(SE) and BPAdaBoost integrated neural network is proposed. The original wind power data were decomposed into a series of intrinsic mode function components(IMF) and residual terms by FEEMD method. The sample entropy was used to solve the problem of too many components and complicated calculation. The input dimension was determined by selecting elements with high correlation degree with the predicted value through PACF(partial autocorrelation coefficient). BPAdaBoost, an integrated neural network with strong generalization ability, was selected to construct the single-step rolling prediction model and obtain the final value by superposition. The experimental results show that the combined model improves the prediction accuracy and it is feasible and it effective.