Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Ανάπτυξη μοντέλων τεχνητών νευρωνικών δικτύων για τον προσδιορισμό του χρόνου διαδρομής σεισμικών κυμάτων. Εφαρμογή σε δεδομένα σεισμικής τομογραφίας από το σταθμό Ανθούπολης του Αττικού Μετρό

Kakaroglou Evangelos

Full record

---

Summary

The time of the first-arrival in seismic data, is essential for performing refraction statics computations or diving-wave tomography calculations. Picking first-arrival times, is a pattern recognition problem and, as such, involves a substantial amount of human effort.Furthermore, with the number of seismic sensors growing, it is becoming increasingly difficult for analysts to pick seismic phases both, manually and comprehensively, yet such efforts are fundamental. Despite the improvements made in recent years with the automatic phase picking methods, it is still very difficult to match the performance of experienced analysts, with a subtler issue being the fact that, different seismic analysts may pick phases in different ways, which most of the times can introduce bias in the data. From the above it becomes obvious that a more reliable and accurate way of automated phase picking needs to be introduced.With the use of neural networks becoming increasingly popular in geophysical applications, because of their abilities as universal approximators, they are tools that can approximate any continuous function with an arbitrary precision, and as such they make for a very promising way to automatically pick first break events and edit seismic trace data.In this thesis, a number of seismic data is imported, edited and then a number of neural networks is trained. The best performing one is chosen and used to automatically pick first break events on seismic data. Using the neural network’s output against the output from an experienced user, useful conclusions are derived.Having a larger input data set improves the output of the network and the number of nodes positively contributes to the output, but after a certain point there is no improvement.

Available Files

Services