Seismic instruments: article

ON THE POSSIBILITY OF USING ARTIFICIAL NEURAL NETWORKS IN SEISMIC MONITORING TASKS
A.E. Hannibal1,2
1 Geological Institute of the Kola Science Centre RAS
2 Kola Branch of Geophysical Survey RAS
Journal: Seismic instruments
Tome: 54
Number: 3
Year: 2018
Pages: 5-21
UDK: 004.032.26:550.34.064
DOI: 10.21455/si2018.3-1
Keywords: neural networks, seismic monitoring, seismic data, classification
Аnnotation: The present paper is dedicated to the possibility of application of the artificial neural networks (ANN) approach to the seismic monitoring, which is understood as the task of detection and recognition of seismic events with subsequent assessment of their nature and cause. This task does not always have an effective solution, using conventional methods of analytical and numerical modeling. Therefore, the author considers the possibility of using ANN to solve it. This approach takes a certain interest in the scientific literature and a number of publications are devoted to it. In this paper, the author made an attempt to generalize and systematize the basic knowledge about neural networks, their design and working principles from his position. The principles of data processing by neural networks are structured in the article. As a basic unit of ANN, a layer is taken, which simplifies the understanding of the structure and principles of the ANN and can be especially useful in solving applied problems. The application part of the work is devoted to the peculiarities of using neural networks in seismic monitoring tasks. The main types of data typical for seismic monitoring tasks are given. Their application features in neural networks are considered. In the final, third part of the work, an example of the practical use of a neural network for the task of detecting false seismic events is given. The author built a neural classifier based on the perceptron. With it, the search for false alarms of the detector of weak seismic events was performed. The final accuracy of the detector was 88 %. Thus, the work shows by practical example that despite the comparative simplicity of the ANN, they are able to solve complex problems of seismic monitoring with significant saving of time and human labor for the preparation and processing of seismic data.