Exploring the Feature Space of Activity Space. A Time Series Feature Approach to Quantifying and Comparing Points of Interest (POI) Temporal Signatures

Keywords: temporal signatures, place-based activity, points of interest, occupancy patterns
Abstract Type: Paper Abstract

Abstract

Current research quantifying POI temporal signatures – also commonly referred to as building occupancy patterns – is generally focused on making comparisons. Comparisons of occupancy patterns are made across different use categories, locations, or to that of industry Standard Occupancy Schedules such as ASHRAE Standard 90.1, all with the goal of noting and interpreting the differences between them. The most widely used process of comparison is some version of 1) constructing a single representative pattern from all of the building occupancy patterns within each cluster of interest (e.g. use category, location, etc.) then 2) quantifying the pair-wise similarity using a single distance measure between each cluster’s representative pattern such as Earth Movers Distance or cosine similarity. Although widely applied, this approach to comparing occupancy patterns leaves much to be desired. Collapsing a set of occupancy patterns into a single representative pattern loses valuable information about the range of observed behaviors within a set of occupancy patterns, forces pair-wise univariate comparison, and impedes the ability to compare sets of occupancy patterns. Additionally, these approaches are all based on relative comparisons. This results in only being able to numerically summarize occupancy patterns in relation to some other pattern. Making these measures difficult to compare across studies and adding the requirement to establish a meaningful comparison pattern which may not be appropriate or practical for all research applications. In this talk we will discuss an alternative approach for comparing POI occupancy patterns based on times series feature analysis that helps overcome these limitations.