The second case study is an office building in a temperate, continental climate in Switzerland. It is a facility with gas-heated hot water and electric chilled water cooling systems. The building is used as a case study facility for innovative building systems. Measurement system data has been collected for over three years from various building sub-metering systems. This analysis is abbreviated as compared to the first as less time was spent on the investigation of the motifs and discords created in this example; it is included to contrast the basic capabilities of DayFilter from Case Study 1 for a dataset in a continental climate with a different type of univariate data stream. For this application, we investigate the electricity profiles for the whole building. Whole building electricity should be more diverse in this example since more subsystems are contributing to the profiles, as compared to only cooling energy in the previous example. A dataset of 9,769 hourly data points spanning from July 1, 2012 to August, 12, 2013 are implemented using the DayFilter technique on 407 daily profiles. Figure \ref{fig:sankeyheatmap2} illustrates the process as applied to this case study with parameters \(A\)=3 and \(W\)=4..
Ten motif candidates were created in the SAX process due to the larger diversity of operating profiles because of the distinct heating and cooling seasons that exist in a temperate climate as well as the different operating schedules and system types. Additionally, this analysis is using whole building electricity consumption as opposed to only cooling electricity, thus increasing the potential diversity of load patterns. The most notable difference in the cooling season patterns compared to cast study 1 is in the presence of many more daily SAX words starting with b or c. It was discovered that this facility has an ice storage system which is utilizing electricity in the early morning hours to shift the demand during the cooling season. Motif candidate patterns fitting in this category include \(cccb\), \(caaa\), \(bccb\), \(bbcb\), \(bbbb\), and \(baaa\). The remaining motif candidates are more common during the winter with \(accb\) and \(abcb\) exemplifying weekday profiles and \(abbb\) and \(aaaa\) representing weekends and holidays.
There are 22 discord candidates in this case study with anywhere between 1 and 8 days in each discord. The most obvious anomaly in the discord dataset are the top consuming \(cccc\), \(ccbb\), \(cbcb\), \(cbcc\) patterns. These discords represent days in the dataset in which the whole building consumption remained high for most hours of the day. Unlike case study 1, these discord patterns were not investigated in more detail for this paper. However, the wider range of discords is not surprising due to the increased complexity of the whole building electricity data stream as compared to a subsystem like cooling.