Abstract
Artificial Neural Network (ANN) is used in this article to develop attenuation relationships for three peak ground motion parameters, namely, peak ground acceleration (PGA), peak ground velocity (PGV), and peak ground displacement (PGD). This article demonstrates the capability of ANN to capture the key physical aspects of seismic wave attenuation and region specific earthquake characteristics. Limited strong ground motion data and no particular functional form except for few constraints are used in the development of ANN based attenuation relationships. The database consists of 358 records (2 horizontal components of ground acceleration at each station) from 42 European shallow earthquakes. The surface magnitude (Ms), distance of site from surface projection of the rupture (R), and broad categories of soil type (soft soil, stiff soil, and rock formation) are the three input parameters. The Ms ranges from 5.5–7.9 and R ranges from 3 – 260 Km. The model is trained using 75% (134 data points) of the total data, while the remaining 25% (45 data points) of the total data is used to test the performance of the trained neural network models. The ANN is able to derive attenuation relationships which are consistent with the theory of ground motion attenuation phenomena. ANN can, therefore, be used as an alternative method to conventional regression techniques for developing attenuation relations, particularly for regions where limited earthquake data is available.
Acknowledgment
The authors would like to thank Dr. Jullian Bommer for providing the processed strong motion data. We also thank Drs. Gail Atkinson and Jullian Bommer for their useful comments on the results of the Artificial Neural Network and their suggestions for improvement.
Notes
Kramer, S. L. [1996] Geotechnical Earthquake Engineering, Prentice-Hall, MATLAB [2000] The Language of Technical Computing, The Math Works Inc. (Copyright 1984–2000)