基于RGA-BiLSTM模型的太阳辐照度预测

Solar irradiance prediction based on the RGA-BiLSTM model

目的:精确预测太阳辐照度对于电力系统设计者和电网运营商有效管理太阳能系统至关重要。为了进一步提高太阳辐照度的预测精度,我们提出了一种新的基于RGA-BiLSTM模型的方法来预测太阳辐照度。方法:该模型集成了双向长短时记忆(bi-directional long and short-term memory,BiLSTM)网络、遗传注意力(genetic attention,GA)以及残差网络(residual network,ResNet),用于多变量太阳辐照度预测。首先利用BiLSTM模型对复杂的太阳辐照度数据进行特征提取;其次,为了提高神经网络的效率和性能,使用遗传注意力机制自适应关注更重要的特征信息;最后,将残差网络融合到模型中,避免深度网络退化问题,从而加快了网络的收敛速度。结果:实验数据表明,我们的模型相较于改进前的模型(EA-LSTM),平均绝对百分比误差在四季的研究中平均降低了9.12%。结论:使用基于RGA-BiLSTM的多模型混合方法能够有效提升太阳辐照度的预测精度。

Aims:Accurate prediction of solar irradiance is essential for power system designers and grid operators to effectively manage solar energy systems.In order to further improve the prediction accuracy of solar irradiance,we proposed a new model based on RGA-BiLSTM to predict solar irradiance.Methods:The model integrated bidirectional long and short-term memory(BiLSTM)networks,genetic attention(GA)and residual networks(ResNet)for multivariable solar irradiance prediction.Firstly,the BiLSTM model was used to extract features from complex solar irradiance data.Secondly,the genetic attention mechanism was used to adaptively focus on more important feature information to improve the efficiency and performance of neural networks.Finally,the residual network was integrated into the model to avoid the deep network degradation and accelerate the convergence speed of the network.Results:The experimental data showed that the mean absolute percentage error was 9.12%,which was lower than that of the original model(EA-LSTM)in the four seasons.Conclusions:Using the RGA-BiLSTM-based multi model hybrid method can effectively improve the prediction accuracy of solar irradiance.