Abstract

We developed a method for seismic sensor data recovery and simulated seismic sensor operation using artificial neural networks. We selected two parameters to train the artificial neural network on: the time between seismograph recordings of longitudinal (primary) and transverse (secondary) seismic waves, as well as the time between primary seismic wave recordings by two seismographs located at a certain distance from each other. We used data on 2636 earthquakes that occurred in 2020 in the Republic of Dagestan for our seismograph data recovery. The existing 19 seismic stations recorded less than 60 % of the total number of earthquakes. To recover seismic data, we trained the neural network twice for each seismic sensor, the first time involving zero time differences regarding seismic wave arrival at seismographs that did not record the time, and the second time involving time differences recovered from the results of training the neural network for the first time. Times between seismic wave recordings with known data were used as inputs to train the artificial neural network, while time differences to be determined were designated as outputs. The trained neural network displays a correlation coefficient related to real time intervals between seismograph recordings of seismic waves that exceeds 0.99919. The paper provides root-mean-square error plots of the neural network operation by epochs of its training, plots demonstrating how training results calculated by the neural network matchthe initial data, and histograms of neural network operation errors

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call