A High-Precision Chaotic LSTM Network Model Optimized by FrVMD for the Earth Natural Pulse Electromagnetic Signal Intensity Trend Forecast

Potential trends in the intensity and irregularity of the Earth’s natural pulse electromagnetic field (ENPEMF) signal may be related to the occurrence of seismic hazards. To investigate this potential variation trend, radial basis function (RBF) forecast models have been applied in the field of ENPEMF signal intensity forecast, but its forecast accuracy is not high enough to be used in the field of seismic hazard monitoring and other fields that require high accuracy. In addition, although the rapid development of deep learning has improved the accuracy of time series forecasts, ENPEMF signals are susceptible to noise, which may lead to erroneous identification of pregnant earthquake information. Based on this, this article proposes a chaotic long short-term memory (LSTM) forecasting model based on fractional Fourier transform (FrFT) optimized VMD (FrVMD) to forecast the underlying ENPEMF signal intensity trend. The strategy of the proposed model is the FrVMD algorithm to decompose and reconstruct the original signal for noise reduction, whereupon forecasting the phase space reconstructed ENPEMF data by chaotic parameters optimizing the LSTM network. The model of this article is applied to the ENPEMF signal before and after the Ms7.0 magnitude earthquake in Lushan on April 20, 2013. The numerical results show that the proposed model forecasts the intensity trend of ENPEMF data more accurately, for the forecast effect and accuracy are better than the traditional RBF network, chaotic RBF network, and traditional LSTM network model, which helps to identify the pregnant earthquake information and provides support for electromagnetic monitoring analysis before geological disasters and strong earthquakes.