基于UNet和注意力机制的太阳光球层磁场预测模型

Prediction Model of Solar Photosphere Magnetic Field Based on UNet and Attention Mechanism

在UNet神经网络结构的基础上,引入多头自注意力(Multi-head Self Attention,MHSA)神经网络层,构建了一套太阳光球层磁场预测模型。通过在SDO/HMI观测数据上进行训练,此模型具备在一定精度上预测太阳磁场的能力。模型的测试结果还表明,MHSA的引入显著提升了模型的预测能力。

We present a prediction model of solar photosphere magnetic field in this paper.It is based on the UNet neural network,combining with the Multi-head self attention layer.After being trained on the SDO/HMI observation data,this model has the ability to predict the photosphere magnetic field at a certain accuracy.Furthermore,the results also shows that the MHSA layer makes significant improvement to the model’s performance.