Abstract for: A novel way to measure (dis)similarity between model behaviors based on dynamic pattern features

This paper presents a novel way of quantifying pattern-wise (dis)-similarity between two time-series data. The approach filters out all numerical information from a given time-series data, and generates a qualitative description of it in terms of atomic behavior modes. The comparison of two data-series, hence the similarity calculation is based on these qualitative descriptions. Different from early examples in the field, the proposed approach focuses purely on pattern features, and does not require to be trained for a fixed set of patterns beforehand. During preliminary tests, it is observed that the algorithm performs very well, and the computational cost in terms of time is quite low. Using the proposed (dis)-similarity calculation, it is possible to present model results in a more objective and quantified manner. Apart from that, such a quantification enables the use of advanced computational techniques in various phases of the modeling cycle