Thunderstorm Prediction Method Based on CNN-BiLSTM Using BEADS*

Abstract

Atmospheric electric field signal (AEFS) usually superimposes low-frequency noise, which has a negative effect on thunderstorm monitoring. A thunderstorm prediction method based on Convolutional Neural Network (CNN) and Bidirectional Long Short Term Memory (BiLSTM) is proposed. Firstly, AEFS is divided into useful, baseline and noise components by BEADS. After getting the estimated value of the useful signal, the denoised signal is obtained. Then, the AEFS prediction model is built based on BiLSTM. After inputting the AEFS spatial features extracted by CNN into the model, a CNN-BiLSTM hybrid model for thunderstorm prediction is formed. After analyzing the performance of the method, we carried out the experiment in thunderstorm weather. Results show that the SNR of AEFS processed by BEADS is improved effectively. It's worth noting that the determining coefficients before and after BEADS are all above 94.12%, showing a good effect. Finally, the effectiveness of the method is proved again by the coincidence between the predicted results and the radar chart.