A Neural Network approach to the identification of Electric Earthquake Precursors |
| |
| Institution: | 1. Université Grenoble Alpes, Univ. Savoie Mont-Blanc, CNRS, UMR CNRS 5204, EDYTEM, 73370 Le Bourget du Lac, France;2. Université de La Réunion, Laboratoire GéoSciences Réunion, F-97744 Saint Denis, France;3. Université de Paris, Institut de Physique du Globe de Paris, CNRS, F-75005 Paris, France;4. Research Institute of Earthquake and Volcano Geology, Geological Survey of Japan, AIST, Tsukuba, Japan |
| |
| Abstract: | Reliable recognition of Electric Earthquake Precursors (EEP) is mainly prevented by disturbances of magnetotelluric (MT) origin, making EEP's interpretation difficult. However, assuming that EEP's are not accompanied by a significant magnetic anomaly, the MT disturbances can be reduced by applying the method of residual electric field calculations. In the present work, we present a Dynamic Neural Network (DNN) based prediction procedure. A DNN is employed to predict the behaviour of an ellectrotelluric time series with ultimate objective its application to earthquake prediction using electric precursors. To achieve such a goal, a learning phase is necessary, during which the synaptic weights of the DNN are adjusted according to an appropriate update law so as to minimize an error signal. Then the method proceeds to the prediction phase, where using the weight values resulted at the end of the learning phase the DNN predicts, the actual signal accurately. The proposed approach is applied to measured electrotelluric data. |
| |
| Keywords: | |
| 本文献已被 ScienceDirect 等数据库收录！ |
|
