Missing data in monitoring systems represent a major data quality issue that is frequently seen in buildings. Induced by malfunctioning sensors, poor network stability and other environmental or human causes, missing data create gaps that can last for a few seconds up to multiple days. Relatively long gaps generate inaccuracies that can significantly affect building operations. There is thus a need for a methodology for imputation in building data analysis. Various imputation techniques are available in literature, so it is difficult to select the best methodology to use for a specific gap in building operation data. For instance, the selection of the best methodology could change according to the duration of the gap and the physical quantity that is measured by a sensor. Multiple parameters, which are usually related to the indoor environment (temperature, humidity, CO2…) or to energy consumption (HVAC energy use, electricity consumption, hot water…) are measured in buildings. All these parameters behave differently and should consequently be treated differently in data imputation. In this work, we compare the imputation performance of multiple methodologies in order to identify the best to use based on the duration of a data gap and on the nature of the measured physical quantity. Interpolation (linear, cubic, spline…), regression and more advanced techniques such as Fourier series are used in the analysis. Our comparison is made from a large dataset coming from the monitoring of 10 dwellings. The parameters listed above are all measured in the monitored dwelling with a 1-minute frequency. Two years of data are available. We create artificial gaps in this dataset and fill them with different imputation techniques. The mean errors computed by each technique are then compared to identify the most accurate approach. A chart providing the best technique to use according to the gap duration and to the nature of the data was thus developed.
| File Size : | 1
file
, 5.2 MB |
| Note : | This product is unavailable in Russia, Belarus |
| Number of Pages : | 9 |
| Product Code(s) : | D-CH-24-C105 |
| Published : | 2024 |
