BiLSTM-based thunderstorm prediction for IoT applications

Abstract

Although the market demand for smart devices (SDs) in the Internet of Things (IoT) era is surging, the corresponding thunderstorm protection measures have rarely attracted attention. This paper presents a thunderstorm prediction method with elevation correction, to reduce the thunderstorm damage to SDs by visually tracking thunderstorm activities. First, a self-made three-dimensional atmospheric electric field apparatus (3DAEFA) deployed in IoT is developed to collect real-time AEF data. A 3DAEFA-based localization model is established, and the localization formula after correction is derived. AEF data predicted by the bi-directional long short-term memory (BiLSTM) model are input to this formula to obtain thunderstorm point charge localization results. Then, the localization skill is evaluated. Finally, the proposed method is assessed in experiments, under single and multiple point charge conditions. There are significant reductions of at least 33.1% and 8.8% in ranging and elevation angle errors, respectively. Particularly, this post-prediction correction reduces the deviation of fitted point charge moving paths by at most 0.189 km, demonstrating excellent application effects. Comparisons with radar charts and existing methods testify that this method can effectively predict thunderstorms.