Earthquake Early Warning Using Multi-Channels Echo State Extreme Learning Machine

Abstract

Predicting earthquake strong motions is crucial for mitigating seismic risks and enhancing the effectiveness of Earthquake Early Warning (EEW) systems. While conventional models are capable of high precision, they often require substantial computational resources, limiting their practicality for real-time applications. This study proposes the Multi Echo-State Extreme Learning Machine (Multi ES-ELM), an efficient and effective alternative for strong motion prediction. It compares the performance of Multi ES-ELM with two well-established models-Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN)-using multi-channel time-series data. The CNN model achieved high performance with an accuracy of 94.65 ± 0.30, recall of 92.84 ± 2.36, precision of 87.34 ± 2.35, and F1-score of 89.95 ± 0.41. In contrast, the RNN model showed significant variability, with an accuracy of 84.83 ± 19.40, recall of 84.93 ± 13.34, precision of 74.18 ± 18.15, and F1-score of 77.80 ± 16.40. Notably, the Multi ES-ELM model demonstrated competitive accuracy (93.46 ± 0.22), high recall (96.50 ± 0.52), precision (81.53 ± 0.53), and F1-score (88.38 ± 0.37), while using significantly fewer resources-only 882 parameters and a model size of 0.003 MB. These results highlight Multi ES-ELM as a highly efficient and reliable model for real-time EEW, overcoming the computational challenges of traditional approaches. Its performance and resource efficiency underscore its potential for practical implementation in seismic risk mitigation and for improving community resilience against seismic hazards.