Search

The presence of "ill-posed samples" specifically in low-volume datasets leads to accuracy decrement in the learning procedure and the generalization of neural networks. Such samples can be caused by various reasons such as noise contamination, corrupted sensors, or even, the complex distribution of physical properties governing the problem. The peak ground acceleration (PGA) datasets are definitely among the last mentioned. Focusing on speed and accuracy, a method for calculating earthquake magnitude based on the PGA data recorded at a single station along with hypocentral information has been presented in this research. Here, after training a deep neural network, the regression errors of the training data samples are clustered into two groups, namely well and ill posed using the grey wolf optimization algorithm. Instead of being removed, the data samples with low learning rates are then modified using samples selected from the other cluster in a fusional form. Then, two separate models are used and trained independently for the clusters. Next, in addition to the routine procedure of network generalization, every new sample is first checked whether is more likely to belong to which group of the clustered data, and after processing, the corresponding trained model is used. The results of the experiments show that using the proposed method results in magnitude calculation with an error order of less than 0.212 units of moment magnitude with a probability of more than 99.7%, which is superior to the conventional methods some of which were reviewed in this research. [ABSTRACT FROM AUTHOR]

One of the most challenging topics in analyzing multi-dimensional geo-spatial data such as geophysical data-sets is detecting outlier data. The issue mainly originates from the difficulty in describing "normality" or "abnormality" due to the complexity of the relationships between the data elements. Considerable number of methods have been proposed and applied for detecting outliers whether they are assumed to be noise, anomalies within the data-set or simply isolated events. A new outlier detection method reached from automatic training of Local Linear Model Tree (LOLIMOT) network, and based on the data selected by K-Nearest Neighborhood (KNN) search is proposed in this research. The procedure of selecting data pairs is through decile analysis using distances calculated during KNN data grouping. Experiment on a synthetic 12 cluster 3D data-set is indicative of the method's robust performance where calculated Cumulative Error Percentage (CEP) is 13% for the method whereas the nearest value for the KNN is 19%. Also, by applying the method on a micro-gravimetric data and an earthquake catalogue related to the north Zagros- west Alborz, and based on the output of the analyses performed, the superiority of the method in outlier detection was confirmed. [ABSTRACT FROM AUTHOR]

Grouping of earthquakes into distinct clusters is applied to improve mechanism identification and pattern recognition for active seismicity in a region. One of the important issues concerning earthquake data clustering is determining the optimum number of clusters (ONC) at the early stages of algorithms. In this paper a robust method based on K-nearest neighbor search (KNNS) is presented to achieve three goals: improving output accuracy, improving output stability, and adding the ability to weight the features used in ONC determination. By introducing a new formula, the proposed method utilizes the error calculated for clustered data based on the similarity between the members in each cluster. An outlier attenuation algorithm is also used to improve the performance of the method. Both the Krzanowski–Lai Index (KLI) and the silhouette coefficient (SC), as two conventional methods, were used to compare the results and evaluate the performance. Experiments on synthetic data sets verified the effectiveness of the method, with considerable differences found. The clustering of a real earthquake catalogue related to the seismogenic province of Zagros in Persia using our proposed methodology suggests using 13-cluster analysis for clustering based on the spatiotemporal features with the same weights, and seven-cluster analysis for a case where priority is given only to the spatial parameters of the epicenters. Under the same circumstances, the KLI and SC methods suggest three and 18 clusters, respectively. The results of the experiments on synthetic data sets indicate that the proposed method is quantitatively more stable and more accurate than the other two methods. [ABSTRACT FROM AUTHOR]