基于小波-极限学习机的短期风电功率预测

Short-Term Power Prediction of a Wind Farm Based on Wavelet and Extreme Learning Machine

风能作为可再生绿色能源已经得到世界各国的广泛应用。由于风具有很强的随机性和不稳定性,导致风电的波动剧烈,给电网的安全稳定运行提出了挑战。本文根据风电的强随机性特点和BP网络预测精度低的缺点,提出了基于小波分解和多模型极限学习机的风电功率预测模型。通过小波分解将风电功率时间序列分解为不同频段的子序列,对各个子序列进行分析,在一定程度上弱化了风电的强随机性。再用多模型极限学习机分别对子序列建立模型,对于每一个子序列再利用设定的挂起准则将建立的子序列的模型分为两部分;对于误差小的挂起模型,无需随机方式进行在线更新,用以降低模型的误差波动;对于误差大的更新模型,采用随机方式进行在线更新。可以进一步弱化风电的强随机性。最后将各个子序列的预测结果重构。对内蒙古某风电场的风电功率预测结果表明,本文提出的方法与其他方法相比,具有更高的预测精度。

Based on the strong randomness of wind and the short precision of BP network forecasting,short-term power prediction of a wind farm based on wavelet decomposition and multiple Models extreme learning machine(WD-MMELM) is proposed.Signal is decomposed into several sequences in different band by wavelet decomposition.Decomposed time series are analyzed separately,then building the model for decomposed time series with MMELM.For each sub-model,a suspending criterion is designed to separate the whole models into two parts;the suspending models and the updating models.For the suspending models with minor error,it needn' t adopt the random selection method to update online.For the updating models with major error,it must utilize the random selection method. MMELM improves the accuracy of wind power prediction.In the last the predicted results were added.Through a wind-power simulation analysis of a wind farm in Inner Mongolia,the result shows that the method in this paper has higher power prediction precision compared with other methods.