The effective observation of burst events in solar radio research has been impeded by various interference signals, especially interference signals with a wide frequency range and high intensity, as they can partially or completely obscure the observation of burst events. Image processing methods that directly remove the interference signal channels and subtract the average of the interference signal channel are not suitable for processing all types of interference signals. This paper proposes the use of a specific kind of recurrent neural networks, called long short-term memory networks, to predict the value of the radio frequency interference signals with high intensity of the burst event in the solar radio spectrum. The predicted interference can then be removed in accordance with the principle that signals can be linearly added. Therefore, predicted value is subtracted from the data containing the burst event signals and the RFI signals (The radio frequency interference signals to be processed in this article refer to the signal of the broadcast signal that can be received in the frequency range, the signal transmitted by the mobile phone, and the signal transmitted by the sea vessel, and the like) to remove the interference. Then, in order to reduce the error caused by the stepwise prediction in the network and further improve the prediction accuracy, this paper analyzes the characteristics of the value of the radio interference and applies the digital mapping method to convert the prediction problem into the classification problem in the time series. The experimental results show that the proposed method can effectively remove the radio interference in the solar spectrum and clearly show the burst events.

太阳射电研究中对爆发事件的有效观测一直受到各种干扰信号的阻碍，特别是频率范围宽、强度高的干扰信号会部分或完全掩盖爆发事件的观测。直接去除干扰信号通道并减去干扰信号通道平均值的图像处理方法并不适合处理所有类型的干扰信号。本文提出使用一种特定类型的循环神经网络，称为长短期记忆网络，来预测太阳无线电频谱中具有高强度突发事件的射频干扰信号的值。然后可以根据信号可以线性相加的原理去除预测的干扰。所以，从包含突发事件信号和RFI信号的数据中减去预测值（本文要处理的射频干扰信号是指在该频率范围内可以接收到的广播信号的信号，手机，以及海轮发射的信号等）来消除干扰。然后，为了减少网络中逐步预测带来的误差，进一步提高预测精度，本文分析了无线电干扰值的特点，并应用数字映射方法将预测问题转化为分类问题在时间序列中。