Semantic information alignment of BIMs to computer-interpretable regulations using ontologies and deep learning

A semantic information alignment method is proposed to align the representations used in building information models (BIMs) to the representations used in energy regulations. Compared to existing alignment efforts, which are either manual or semi-automated, the proposed method aims to automate the alignment process for supporting fully automated energy compliance checking. A first-level simple alignment method is proposed to align single design information instances to single regulatory concepts, in which (1) domain knowledge is used for interpreting the meaning of concepts to recognize candidate instances, and (2) deep learning is used for capturing the semantics behind the words to measure semantic similarity and select the matches. A final complex alignment method is proposed to recognize the instance groups belonging to a regulatory requirement, in which (1) supervised and unsupervised searching algorithms are used to identify the instance pairs, and (2) network modeling is used to group and link the instance pairs to the requirement. The proposed method showed 93.4% recall and 94.7% precision on the testing data.