基于Kmeans和CEEMD-PE-LSTM的短期光伏发电功率预测

Prediction of Short-term Photovoltaic Power Generation Based on Kmeans and CEEMD-PE-LSTM

针对光伏发电功率时间序列随机性和波动性强的特点,提出一种基于Kmeans和完备总体经验模态分解(CEEMD)、排列熵(PE)、长短期记忆(LSTM)神经网络结合的短期光伏功率预测模型。先通过Kmeans算法选出预测日的相似日;然后采用CEEMD将发电功率和影响因素数据的原始序列分解为多个固有模态分量,并用排列熵算法对模态分量进行重构;最后对重构后的子序列分别进行LSTM建模预测,再将子序列预测结果叠加起来确定光伏发电功率预测值。试验结果表明,所提预测模型与单独的LSTM预测模型和EMD-PE-LSTM预测模型相比,功率预测精度明显提高,为电网调度提供了一定参考。

The time series of photovoltaic power generation has strong randomness and volatility characteristics.A short-term photovoltaic power prediction model based on the combination of Kmeans and complete ensemble empirical mode decomposition(CEEMD),permutation entropy(PE),long short-term memory(LSTM)neural network is proposed.First,the similar days of the prediction days is chosen by the Kmeans algorithm.Then,the CEEMD is used to decompose the original sequence of power generation and influencing factors into multiple intrinsic modal components,and the modal components are reconstructed with PE algorithm.Finally,the reconstructed sub-sequences are respectively predicted with LSTM model,and the sub-sequence prediction results are superimposed to determine the predicted value of photovoltaic power generation.Compared with LSTM and EMD-PE-LSTM prediction models,the example results show that the proposed prediction model has obvious improvement in power prediction accuracy,which provides a certain reference for power grid dispatching.