Research on Earthquake Prediction Methods Based on Pre-trained Models and Danger Theory

In response to the problem of uneven data samples in earthquake prediction, this paper proposes a Transformer-DCA prediction model based on the fusion of Transformer architecture and dendritic cell algorithm (DCA). By integrating the sequence feature extraction advantages of Transformer and the small sample processing capability of DCA, seismic prediction experiments were carried out using multi-component monitoring data of AETA seismic monitoring system. Experimental results show that compared with comparative models such as GBDT, SVM, LSTM and LSTM-DCA, Transformer-DCA has achieved significant improvements in accuracy and recall indicators, with the recall rate optimized from 0.9065 to 0.9375, confirming that the DCA algorithm has a key promoting effect on feature classification. Through 50 random resampling verification, the model demonstrates excellent generalization performance. This study provides a new method to improve the reliability of earthquake prediction and has certain practical application value.