基于VMD和双重注意力机制LSTM的短期光伏功率预测

提出了一种基于变分模态分解(VMD)和双重注意力机制长短期记忆(LSTM)的短期光伏功率预测方法。针对光伏功率信号的波动性和非平稳性,利用VMD将光伏功率输出分解为不同频率的分量,使用LSTM对各分量进行预测,并在LSTM基础上引入特征和时序双重注意力机制。为自主挖掘光伏功率输出与各气象特征之间的关联关系,避免传统方法依赖于专家经验关联规则阈值的限制,引入特征注意力机制实时计算各气象特征量的贡献率,并对特征权重进行修正;同时,为挖掘当前时刻光伏功率输出与历史时序信息之间的关联关系,引入时序注意力机制自主提取历史关键时刻点信息,提高长时间序列预测效果的稳定性。基于中国西南某实际光伏发电站数据进行预测实验,并与其他方法进行对比,验证了该方法的有效性。

Short-term Photovoltaic Power Prediction Based on Variational Mode Decomposition and Long Short-term Memory with Dual-stage Attention Mechanism

This paper proposes a short-term photovoltaic (PV) power prediction method based on variational mode decomposition (VMD) and long short-term memory (LSTM) with dual-stage attention mechanism. Aiming at the fluctuation and instability of PV power signals, VMD is used to decompose the PV power output into different components with different frequencies. LSTM is used to predict the individual components. A dual-stage attention mechanism including the feature attention mechanism and temporal attention mechanism is adopted based on LSTM. For expressing the association between PV power output and various meteorological characteristics and avoiding the limitation of the threshold relying on the expert experience in the traditional method, the feature attention mechanism is introduced to calculate the contribution rate of different meteorological characteristic variables and modify feature weights in real time. Meanwhile, in order to obtain the association between the PV power outputs at the current time and the historical time series, the temporal attention mechanism is introduced to automatically extract historical key time point information, which improves the stability of prediction performance of long-term sequences. The prediction experiment is carried out based on the data of an actual PV power station in Southwest China. The effectiveness of the proposed algorithm is verified by the comparison with other methods.