Predict Atmosphere Electric Field Value with the LSTM Neural Network
Tao Guo, Rui Liu, Heng Yang, Lei Shi, Feifan Li, Lei Zhang, Yue Chen, Zhihui Liu, Fei Luo · 2017 International Conference on Computer Systems, Electronics and Control (ICCSEC) · 2017
lightning is a common meteorological phenomenon. Warning lightning is very important for protecting the safety of human beings and industrial infrastructures. Fluctuation of the atmosphere electric field value is closely related to the lightning's approaching. So the time-series data of atmosphere electric field can be used to warn the lightning. LSTM is one kind of the recurrent neural network, which could deal with long term information dependency. Here LSTM is applied to predict the atmosphere electric field, which help to warn the lightning. The real monitoring data from Jiangsu Province is collected, and three prediction experiments are designed to compare LSTM prediction performance. Finally, the results show the direct prediction could get accurate short term prediction value. The rolling prediction losses accuracy because of the error accumulation. The interval prediction gets satisfactory prediction with regards to accuracy and warning time.