Automated classification of datapoint types in building automation systems using time series

To apply energy efficiency monitoring and optimizing systems for heating, ventilation and air conditioning (HVAC) systems at scale metadata of the system components must be extracted automatically and stored in a machine-readable way. The present work deals with the automated classification of datapoints using only historical time series data and methods from the field of artificial intelligence. The dataset used to conduct the research contains multiple time series from a total of 76 buildings and covers the months of January to November of 2018. Based on these data, relevant features of the time series are defined. Using these features, different classification models as well as the influence of seasonal effects are evaluated. Overall, our approach provides very good results and assigns on average about 92 % of all datapoints to the correct class. The datapoints of the worst recognized class are still correctly classified to about 88 % (validation data) and 90 % (test data), respectively. Possible reasons for incorrect classified datapoints are discussed and a promising solution is proposed. The obtained results show that in practice these methods can reduce the effort of creating and validating digital twins and building information modeling (BIM) for retrofits.