ABSTRACT
Application of data mining methods is discussed for detection of temporal patterns and association rules between those patterns in strong ground motion acceleration time series. The patterns are interpreted in the context of seismic wave fields, whose identification, classification and quantification are of interest for non-stationary seismic load modeling. Two approaches are discussed. The first aims at classification according to patterns which are already known or supposed based on empirical knowledge and a-priori modeling conclusions. The second one aims at identification of significant unknown patterns. Both are based on AI methods for supervised and unsupervised learning. Ongoing work is discussed, especially, an approach for learning of association rules between several time series, based on clustering of local trends, interval state sequences and pattern support.
