Point Mapping, Cost-effective Deployment, Intelligent Building Applications
Over the past years intelligent building applications that promise to dramatically reduce energy consumption, improve occupant comfort and streamline maintenance have been proposed. However their adoption has met a step barrier in the unexpectedly high cost in mapping data from building automation systems into these applicationsâ€™ data models. In fact the industry does not have a common convention on how to name points. Generally, names are correlated with the semantic of the variables they represent, but typically engineers have the freedom to set up variable names according to their preferences. In order to meet market requirements, an â€œidealâ€ algorithm would have four properties: 1) high accuracy and especially low false positive rate (i.e. small number of erroneously mapped points); 2) ability to infer complex relationships between data points (e.g. grouping points by equipment, classifying equipment, establishing relationship between equipment); 3) easy to use: it should leverage only readily available knowledge and data about the system (i.e. not requiring the installation of additional sensors); 4) minimizing the need of having a domain expert using it. In the last few years the research community has devoted increasing attention in automating the process of mapping data points from existing BAS.Â However, none of the published work meets the market requirements.Â Most require an expert user, some are not â€œeasy to useâ€ and none can automatically infer complex relationships. This paper presents a novel algorithm and its software implementation to tackle the mapping problem. This work contributes moving the state-of-the-art a bit furtherÂ in the sense that this algorithm can be applied to various dataset with minor or no modification (i.e. no expert in the loop), it is the first to automatically infer complex relationships and it only ingests point names as input (i.e. easiness of use). The algorithm has been tested against the largest and most diversified dataset in the literature comprising 25299 points, 7 buildings, 4 vendors and 3 distributors. Preliminary results suggest that it is able to correctly map about 92% of the point required by a test application and to successfully identify about 92% of VAVs, 100% of AHU and FCU.