基于EMD-ELM-LSTM的短期风电功率预测

Short-Term Wind Power Prediction Based on EMD-ELM-LSTM

风电出力的波动性和随机性较强,给电网功率调度带来了极大的困难.如何准确预测风电功率对电网的功率供需平衡和安全稳定运行具有重要的意义.本文提出了一种基于经验模态分解(Empirical Mode Decomposition,EMD)、极限学习机(Extreme Learning Machine,ELM)和长短期记忆网络(Long Short-Term Memory,LSTM)的风电功率预测组合模型.首先,对数据集进行预处理,识别并处理数据集中的异常数据,并对数据进行归一化处理以降低不同数据之间的差异性;其次对风电功率进行EMD分解以得到有限分量;然后将所有数据输入ELM-LSTM模型,并根据风电功率分量的特征选择ELM或LSTM对分量进行预测;最后叠加各子序列得到风电功率的最终预测结果.为验证所提模型的有效性和先进性,利用传统的BP神经网络、LSTM网络、CNN-LSTM网络、ELM以及本文所提模型,分别对我国西南某风电场的实测数据进行预测.测试结果表明,所提EMD-ELM-LSTM组合预测模型可以有效提高风电功率预测精度.

The variability and randomness of wind power output pose significant challenges to power grid management.Accurately predicting wind power generation is of great importance for maintaining a balance between power supply and demand and ensuring the safe and stable operation of the grid.This article proposes a combined wind power prediction model based on Empirical Mode Decomposition(EMD),Extreme Learning Machine(ELM),and Long Short-Term Memory(LSTM)networks.First,the dataset is preprocessed,including the identification and handling of outliers,and the normalization of data to reduce differences between various datasets.Then,the wind power output is decomposed using EMD to obtain limited components.Subsequently,all the data is input into an ELM-LSTM model.Depending on the characteristics of the wind power components,ELM or LSTM is selected to predict each component.Finally,the predictions for all sub-sequences are combined to obtain the final wind power generation forecast.To validate the effectiveness and superiority of the proposed model,real-world data from a wind farm in southwestern China is used for prediction.Comparative tests are conducted using traditional Backpropagation(BP)neural networks,LSTM networks,CNN-LSTM networks,ELM,and the model proposed in this article.The results show that the EMD-ELM-LSTM combined prediction model can effectively improve the accuracy of wind power generation forecasts.