基于GRU-BLS的超短期光伏发电功率预测

Prediction of Ultra Short Term Photovoltaic Power Generation Based on GRU-BLS

光伏发电功率的准确预测对电网的稳定运行具有重要的意义。针对深度学习训练耗时长和宽度学习特征提取能力弱等问题,将门控循环单元(GRU)与宽度学习系统(BLS)相融合,提出了用于超短期光伏发电功率预测的GRU-BLS模型。先使用GRU训练序列样本,再将所学习到的隐特征作为新的输入特征,最后在BLS中构造特征节点和增强节点以形成最终的特征。所建立的模型在保留深度学习高预测精度的前提下,有效地缩短了模型的训练时间。在实际的光伏发电数据集上进行实验,评估所提模型在不同季节和天气类型下的性能。实验结果表明:与长短期记忆(LSTM),GRU,BLS和LSTM-BLS等模型相比,GRU-BLS的RMSE值降低了23.89%~75.68%,且TIC值和MAPE值也得到了显著改善。

The accurate prediction of photovoltaic power generation is of great significance for the stable operation of power grid.In the light of the issues of long training time in deep learning and weak feature extraction ability in broad learning,GRU-BLS model is proposed for ultra short term photovoltaic power generation prediction by combining the gated recurrent units(GRU)with broad learning systems(BLS).The GRU is firstly used to train the sequence samples,and then the learned hidden features are employed as the new input features.Finally,all feature nodes and enhancement nodes are constructed in the BLS to form the final features.The established model effectively shortens the model training time while retaining the high prediction accuracy of deep learning.Experiments are conducted on an actual photovoltaic power generation dataset to evaluate the performance of the proposed model under different seasons and weather types.The experimental results show that the RMSE value of GRU-BLS is reduced by 23.89%~75.68%,the values of TIC and MAPE are also significantly improved compared with long and short term memory(LSTM),GRU,BLS and LSTM-BLS.