Backpropagation neural network as earthquake early warning tool using a new modified elementary Levenberg–Marquardt Algorithm to minimise backpropagation errors

Abstract

Abstract.A new modified elementary LevenbergMarquardt Algorithm (M-LMA) was used to minimise backpropagation errors in training a backpropagation neural network (BPNN) to predict the records related to the ChiChi earthquake from four seismic stations: Station-TAP003, Station-TAP005, Station-TCU084, and Station-TCU078 belonging to the Free Field Strong Earthquake Observation Network, with the learning rates of 0.3, 0.05, 0.2, and 0.28, respectively. For these four recording stations, the M-LMA has been shown to produce smaller predicted errors compared to the Levenberg-Marquardt Algorithm (LMA). A sudden predicted error could be an indicator for Early Earthquake Warning (EEW), which indicated the initiation of strong motion due to large earthquakes. A trade-Off decision-making process with BPNN (TDPB), using two alarms, adjusted the threshold of the magnitude of predicted error without a mistaken alarm. With this approach, it is unnecessary to consider the problems of characterising the wave phases and pre-processing, and does not require complex hardware; an existing seismic monitoring network-covered research area was already sufficient for these purposes.

Predicting Ionospheric Precursors Before Large Earthquakes Using Neural Network Computing and the Potential Development of an Earthquake Early Warning System

“…Referring to the captions of Figures 7, 8, and 9, the threshold of the predicted error is 11.91 TECu in Figure 8, which is the maximum value of the predicted errors (Note: predicted errors for September 17, 18, and20, 1999, andMay 13, 2003, were not evaluated). In this study, under the condition that the predicted error is greater than 11.91 TECu, the predicted error was identified as a TEC precursor (Lin et al, 2018). The predicted results of the two BPNN models were compared with the results of past related studies in Section 2 to verify the reasonability of the selected threshold, that is, 11.91 TECu.…”

“…Therefore, for the first objective, it is expected that the TEC precursors of these two earthquakes can be detected by predicting the TEC situation using the two BPNN models. A large predicted error can be an indicator for the TEC precursors, according to the results of the study of Lin et al (2018). For the second objective, any TEC precursor is expected to be detected relating to the earthquakes with an ML ≥ 5.0.…”

“…Several studies and engineering applications have revealed that a neural network consisting of two hidden layers with few neurons can replace a network with numerous neurons in a hidden layer (Thomas et al, 2016). Therefore, in Figure 5, a four-layered BPNN model with two hidden layers is employed similarly as that in the study of Lin et al (2018). This BPNN model is based on the multilayer perceptron (MLP) neural network with an error backpropagation (EBP) algorithm.…”

“…A modified elementary Levenberg-Marquardt Algorithm (M-LMA) served as the EBP algorithm in this study (Lin et al, 2018). The M-LMA is the best method for general nonlinear problems, in which rank-deficient nonlinear least squares are encountered; that is, when ill-conditioned data are encountered, such as the nonlinear TEC and seismic data, the global minimum can be easily accessed after a single completed iteration.…”

Earthquake Magnitude and Grid-Based Location Prediction using Backpropagation Neural Network

“…Numerous studies showed the performance of neural networks in predicting earthquakes. Neural networks have been applied to an early earthquake warning system (EEWS), which is trained using backpropagation with modified Levenberg-Marquardt algorithm to minimize the error rate in the EEWS [3]. The error rate that was tried to be minimized was the error on seismic data amplitude prediction based on the Chi-chi earthquake in Taiwan in 1999.…”