\label{sec:conclusion}
We have described and shown applications of the DayFilter process, a set of temporal data-mining techniques that can find characteristic performance and can help pinpoint infrequent behavior for further evaluation. The process is designed as a forensic filter to focus the effort of an analyst on particular data subsets. The process was applied to two case studies and the results confirmed the ability of the process to find various types of diurnal patterns amongst large univariate datasets. Discord filtering for case study 1 found anomalous daily profiles which were consistent with cooling system faults observed on site. The motif and clustering aggregations process for both case studies produced profiles that may be used to calibrate a simulation model from the design phase or to inform future designs using empirical data. This process benefits creation of occupancy, lighting, and plug load schedules for whole building simulation input.
The process as defined in this work is univariate. The next question is whether DayFilter can be utilized to extract patterns across multiple data streams. This challenge is approached by performing the univariate process for all available data points, including the detailed data downstream from the metrics used in this study. A multivariate analysis of the overlapping discords could be used to further automate the process of AFDD implementation. Another enhancement could be the use of higher granularity parameter settings for finding patterns in sub-hourly measured data, as opposed to the daily profiles used here.