CA-3192332-C - SYSTEM AND METHOD FOR AUTO-TAGGING BMS POINTS

Abstract

There is described a building management system and a method for autotagging points. Data (402, 404) associated with multiple points of a site are received, and each point is associated with a point name (402, 404) and a point descriptor. A building name is identified (406) based on the point name for each point by extracting a first part detected frequently among the data associated with the points. A point equipment is determined (408) from a second part of each point name and a point function is determined (408) from a third part of each point name. A set of point tags is generated (424) based on the point equipment, the point function, and the point descriptor. Confidence scores are created (444) for the set of point tags based on matching characteristics to a common tag set.

Inventors

• Qinpeng Wang
• Gregory Conte

Assignees

• SIEMENS INDUSTRY, INC.

Dates

Publication Date

20260505

Application Date

20210825

Priority Date

20201030

Claims (17)

1. CLAIMS What is claimed is: 1. A building management system for auto-tagging points comprising: a communication component configured to receive data associated with a plurality of points of a site, comprising a building or group of buildings, each point of the plurality of points is associated with a point name and a point descriptor; a processor configured to identify a building name based on the point name for each point by extracting a first part of a particular point name detected frequently among the data associated with the plurality of points; a conditional random field model executable by the processor and configured to determine a point equipment from a second part of each point name by predicting the point equipment with the conditional random field model and determine a point function from a third part of each point name by predicting the point function with the conditional random field model; and a neural network executable by the processor and configured to generate a set of point tags based on the point equipment, the point function, and the point descriptor, wherein the processor is configured to create confidence scores for the set of point tags based on matching characteristics to a common tag set and to assign the confidence scores to the tags of the set of point tags for each point, and wherein the processor is configured to output, based on the set of tags with confidence scores, one or more of at least a list of tags, the building name, an equipment name, an equipment tag, or an equipment location for each point.
2. 2. The building management system as described in claim 1, further comprising an expert knowledge system executable by the processor and configured to determine an equipment tag based on a group of points sharing a same or similar equipment name in response to the processor failing to identify the equipment tag by regular expressions.
3. 3. The building management system as described in claim 1, wherein the conditional random field model is trained with data specific to a building management domain.
4. 4. The building management system as described in claim 1, wherein the conditional random field model is configured to determine an equipment location from a fourth part of each point name by predicting the point equipment with the conditional random field model.
5. 5. The building management system as described in claim 1, wherein the processor is configured to generate a natural language version of the point function based on a library of abbreviations for a building management domain.
6. 6. The building management system as described in claim 1, the processor is configured to resolve ambiguity among a plurality of possible point functions based on the point equipment.
7. 7. The building management system as described in claim 1, wherein the neural network is configured to generate the set of point tags with a multi-label classification model trained with data specific to a building management domain.
8. 8. The building management system as described in claim 1, further comprising an expert knowledge system executable by the processor, wherein: the data is associated with a point unit, a point type, and a virtual-point indicator; the expert knowledge system is configured to infer point tags based on the point unit, the point type, and the virtual-point indicator; and the processor is configured to create confidence scores for the point tags inferred by the expert knowledge system based on matching characteristics to the common tag set. 21
9. 9. A method for auto-tagging points of a building management system, the method comprising: receiving data associated with a plurality of points of a site, comprising a building or group of buildings, each point of the plurality of points being associated with a point name and a point descriptor; identifying a building name based on the point name for each point by extracting a first part of a particular point name detected frequently among the data associated with the plurality of points; determining a point equipment from a second part of each point name by predicting the point equipment; determining a point function from a third part of each point name by predicting the point function; generating a set of point tags based on the point equipment, the point function, and the point descriptor; creating confidence scores for the set of point tags based on matching characteristics to a common tag set and assigning the confidence scores to the set of point tags for each point; and outputting, based on the set of tags with confidence scores, one or more of at least a list of tags, the building name, an equipment name, an equipment tag, or an equipment location for each point.
10. 10. The method as described in claim 9, further comprising determining an equipment tag based on a group of points sharing a same or similar equipment name in response to failing to identify the equipment tag by regular expressions.
11. 11. The method as described in claim 9, wherein determining the point equipment from the second part of each point name includes predicting the point equipment with a conditional random field model trained with data specific to a building management domain.
12. 12. The method as described in claim 9, wherein determining the point function from the third part of each point name includes predicting the point function with a 22 conditional random field model trained with data specific to the building management domain.
13. 13. The method as described in claim 9, further comprising determining an equipment location from a fourth part of each point name by predicting the point equipment with a conditional random field model trained with data specific to a building management domain.
14. 14. The method as described in claim 9, further comprising generating a natural language version of the point function based on a library of abbreviations for a building management domain.
15. 15. The method as described in claim 9, further comprising resolving ambiguity among a plurality of possible point functions based on the point equipment.
16. 16. The method as described in claim 9, wherein: generating the set of point tags includes predicting the set of point tags with a neural network, and the neural network is a multi-label classification model trained with data specific to a building management domain.
17. 17. The method as described in claim 9, wherein each point of the plurality of points is associated with a point unit, a point type, and a virtual-point indicator, the method further comprising: inferring point tags with an expert knowledge system based on the point unit, the point type, and the virtual-point indicator, wherein creating the confidence scores includes creating the confidence scores for the point tags inferred by the expert knowledge system based on matching characteristics to the common tag set.

Description

SYSTEM AND METHOD FOR AUTO-TAGGING BMS POINTS FIELD OF THE INVENTION [0001] This application relates to the field of building management systems (BMSs) and, more particularly, to systems and methods employing machine-learning techniques for metadata tagging points of building management systems. BACKGROUND [0002] Building management systems encompass a wide variety of systems that aid in the monitoring and control of various aspects of building operation for building owners, facility managers, system integrators, and users. Building management systems include various environmental control subsystems, such as security, fire safety, lighting, and heating, ventilation, and air conditioning ("HVAC"). Systems may include on-site and remote building components and operate with third-party subsystems. Unfortunately, since subsystems have been developed separately by different manufacturers, each subsystem operates on its own proprietary protocol. Although standardization has been attempted for building management, there are still many different standards currently in-play, such as BACnet, Modbus, SNMP, and OPC. The situation has been further complicated by the addition of wireless ofioT technologies to building management. [0003] A building management system may include a frontend framework that manages differing networks ofbackend devices on a common platform. The framework may utilize semantic data models and tagging features to manage specific and overall building management for backend devices. In particular, metadata tags of the framework associate backend devices with system features such as graphics, histories, alarms, schedules, and notes. The building system may improve its ability to detect abnormal conditions and diagnose them by utilizing the tagging framework. [0004] Although the advantages of tagging of building points are well recognized, the task of tagging them is still quite challenging. Traditionally, the numerous points of a building management system are mapped manually by a technician or operator. That process is labor intensive and costly, presenting a major impediment to scaling 1 CA 03192332 2023-3-9 WO 2022/093371 PCT /US2021/04 7 464 building data analytics solutions. Another approach is to apply simple text search and/or regular expression to identify a match of a sub-string. This other approach can be cumbersome due to inconsistent point naming conventions. Some conventional systems have applied basic methods based on machine learning, which involve a breakdown of point names to n-grams and training a neural network as a multi-class classification problem. The drawbacks of these basic machine learning approaches include inconsistent point naming conventions and the necessity for massive training data (which are rarely available in practice). Also, some parts of a point name are irrelevant to machine learning-based mapping, such as the building name, so keeping them in then-gram sequence makes little sense. SUMMARY [0005] In accordance with embodiments of the disclosure, there is provided an approach for auto-tagging points in a building management system. The approach extracts and determines metadata tags from data associated with points of a site using machine learning and, for some embodiments, expert systems. For this approach, analyses conducted by machine learning at an earlier stage allow another type of machine learning, utilized at later stage, to be trained with a small training dataset. The approach also leverages multiple information about a point, such as point type and point descripter, to achieve significant mapping performance and generate substantial output including building name, equipment name, point location, and other tags. In addition to machine learning, expert systems may allow for the consolidation of abbreviations for different equipment and resolution of conflicts of point property and tags, thereby improving the quality of predicted tags. The expert systems may also allow for determination of tag information if the machine learning system(s) fails to provide acceptable results. The approach calculates confidence scores and assigns them to the predicted tags for each point. [0006] One aspect is a building management system for auto-tagging points comprising a communication component, a processor, a conditional random field model, and a neural network The communication component is configured to receive data associated with multiple points of a site, and each point is associated with a point 2 CA 03192332 2023-3-9 WO 2022/093371 PCT /US2021/04 7 464 name and a point descriptor. The processor is configured to identify a building name based on the point name for each point by extracting a first part of a particular point name detected frequently among the data associated with the points. In particular, the processor extracts a part of a point name detected frequently among all points from the same site or group. The conditional random field model is configured t