Automated monitoring applications for existing buildings through natural language processing based semantic mapping of operational data and creation of digital twins

Abstract

Buildings' operation constitutes 36% of the German energy consumption. Currently, operators lack the knowledge on energy-saving techniques. There is a shortage of cost-effective and easily-implementable solutions to evaluate building performance. The cause of this problem lies with the semantically heterogeneous operational data used in technical applications. Integrating the data into monitoring applications demands substantial and costly manual efforts. This paper presents a method that enables automated generation of technical monitoring applications for existing buildings. The method outlined represents existing automation stations as digital twins and employs artificial intelligence to map the heterogeneous data to a standard and create semantic digital twins of buildings. The paper introduces a method using natural language processing for the semantic processing of data. The developed method involves a four-stage process for classification of data points, which are subsequently mapped to a uniform vocabulary. To classify the data points, language models were trained on a created dataset of 54,125 data points. Following successful training, the models can process semantically heterogeneous data points. The results, demonstrating an average F1-Score of over 95%, indicate that the developed method is suitable for automating data point mapping. The models were implemented as an Industry 4.0 service and integrated into an application.