基于注意力机制的短期光伏功率预测

Short-term PV power prediction based on attention mechanism

针对传统的光伏功率预测难度大、精度低等问题,提出一种基于注意力机制的短期光伏功率预测模型,将光伏电站的历史记录数据进行处理后导入到预测模型进行训练,利用CNN局部特征提取功能能力以及BiLSTM处理序列信号的能力,再结合Attention机制对不同特征进行权重系数分配。选取澳大利亚某光伏电站数据进行模拟仿真,将Attention-CNN-BiLSTM模型与LSTM等模型进行对比,验证了该模型有更好的预测精度。

Proposing a short-term PV power prediction model based on an attention mechanism addresses the challenges of traditional PV power prediction,such as difficulty and low accuracy.The model utilizes historical data from photovoltaic power stations for training,leveraging the local feature extraction capability of CNN and the sequential signal processing ability of BiLSTM.Additionally,the Attention mechanism allocates weight coefficients to different features.Simulating with data from a specific Australian photovoltaic power station,the Attention-CNN-BiLSTM model is compared with LSTM and other models,validating its superior predictive accuracy.