Multi-potential subspace information fusion earthquake short-term and temporary prediction method based on LSTM (Long Short Term Memory)

The invention discloses a multi-potential subspace information fusion earthquake short-term and temporary prediction method based on LSTM (Long Short Term Memory), which is mainly oriented to an electromagnetic data set, comprises time domain features and frequency domain features, and comprises the following steps: firstly, preprocessing the data set, specifically comprising data normalization and data resampling; taking a time domain feature and a frequency domain feature of electromagnetic data as two different observation views, fusing and clustering data of the two views by using kernel canonical correlation analysis, solving a plurality of potential subspaces of the data of the two views and canonical correlation variables corresponding to the potential subspaces, and taking the potential subspaces and the canonical correlation variables as original data of model training; manufacturing a training sample and a test sample by using a sliding window method according to the characteristics of a seismic tim