基于经验模态分解和深度学习的短期风电功率预测

Short-term wind power forecasting using empirical mode decomposition and deep learning

精准的风电功率预测有利于全网电力平衡、系统安全稳定运行和节能减耗。提出一种基于经验模态分解(empirical mode decomposition, EMD)、核主成分分析(kernel principal component analysis, KPCA)和长短期记忆(long short-term memory, LSTM)神经网络的短期风功率预测模型。采用EMD技术将多维气象序列分解为多个固有模态分量,以挖掘原始数据的主要特征并消除噪声;引入KPCA进行降维处理,提取数据的非线性特征;使用LSTM神经网络对特征提取的序列进行学习并完成预测,获得风电功率预测的最终结果。使用所提出的模型对新疆某一风电场风电功率进行预测,将预测结果与其他模型对比。结果表明,该预测模型能改善预测性能,降低风电功率预测误差。

Accurate wind power prediction plays a crucial role in maintaining the overall power grid balance and ensuring the safe and stable operation of the system.A short-term wind power prediction model based on empirical mode decomposition(EMD),kernel principal component analysis(KPCA),and long short-term memory(LSTM)was proposed.The EMD technique was employed to decompose multidimensional meteorological sequences into intrinsic mode components,aiming to extract the essential features of the raw data and eliminate noise.KPCA was introduced for dimensionality reduction,extracting nonlinear features from the data and reducing redundancy generated by external conditions.Finally,LSTM neural networks were utilized to learn from the extracted feature sequences and make predictions,ultimately providing wind power forecasts.The proposed model was applied to predict wind power at a specific wind farm in Xinjiang,China.Simulation results demonstrate that the approach improves prediction performance and reduces wind power forecasting errors.