- Ground-motion model is:
where y is in m∕s2, a1 = -0.703, a2 = 0.392, a3 = -0.598, a4 = -0.100, a5 = -7.063, a6 = 0.186,
a7 = 0.125, a8 = 0.082, a9 = 0.012 and a10 = -0.038 (do not report σ but unbiased mean square error) for
horizontal PGA; and a1 = 0.495, a2 = 0.027, a3 = -2.83, a4 = 0.235, a5 = 7.181, a6 = 1.150, a7 = 1.103,
a8 = -0.074, a9 = 0.065 and a10 = -0.170 (do not report σ but unbiased mean square error).
- Use three site categories:
- SS = 1, SA = 0.
- SA = 1, SS = 0.
- SS = 0, SA = 0.
- Use four faulting mechanisms:
- FN = 1, FT = 0, FO = 0.
- FN = 0, FT = 0, FO = 0.
- FT = 1, FN = 0, FO = 0.
- FO = 1, FN = 0, FT = 0.
- Use same data and functional form as Ambraseys et al. (2005a) and Ambraseys et al. (2005b) but exclude
six records that were not available.
- Use genetic (global optimization) algorithm to find coefficients so as to find the global (rather than a local)
minimum. Use the unbiased mean square error as the error (cost or fitness) function in the algorithm.
Use 20 chromosomes as initial population, best-fitness selection for offspring generation, uniform random
selection for mutation of chromosomes and heuristic crossover algorithm for generation of new offspring.
- Find smaller (by 26% for horizontal and 16.66% for vertical) unbiased mean square error than using
standard regression techniques.