# US 12,181,844 B2 · Building management system with natural language model-based data structure generation > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,181,844 B2 - Document ID: US12181844B2 - Status: Granted - Kind: B2 - Application number: 18/419,464 - Filed: 2024-01-22 - Published: 2024-10-17 - Granted: 2024-12-31 - Earliest claimed: 2023-04-12 - Assignees: Tyco Fire & Security GmbH - Applicants: Tyco Fire & Security GmbH - Inventors: Rajiv Ramanasankaran; Krishnamurthy Selvaraj - Web page: https://patentgrep.com/patent/US12181844B2 ## CPC - G05B 13/027: {using neural networks only} - G06F 16/248: Presentation of query results - G06N 3/0895: Weakly supervised learning, e.g. semi-supervised or self-supervised learning - G05B: CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS (systems for controlling or regulating non-electric variables G05D; systems for regulating electric or magnetic variables G05F; control devices or systems insofar as characterised by mechanical features only G05G) - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract Systems and methods are disclosed relating to building management systems with language model-based data structure generation. For example, a method can include receiving a query to select, from a plurality of data sources of a building management system, a selected one or more data sources according to a characteristic indicated by the query in at least one of a natural language representation or a semantic representation. The method can further include applying the query as input to a machine learning model to cause the machine learning model to generate an output indicating the selected one or more data sources, the machine learning model configured using training data comprising sample data and metadata from the plurality of data sources. The method can further include presenting, using at least one of a display device or an audio output device, the output. ## Claims ### Claim 1 A method, comprising: receiving, by one or more processors associated with a building management system, an unstructured query in the form of at least one of a natural language representation or a semantic representation, the unstructured query including a request for one or more of lineage information, availability information, or quality information regarding a category of data associated with the building management system, the category of data relating to one or more of equipment failure, sustainability, comfort, operations, energy, air quality, or net-zero emissions; applying, by the one or more processors, the unstructured query as input to a machine learning model, the machine learning model configured using training data comprising sample data and metadata from a plurality of data sources of the building management system; generating, by the one or more processors using the machine learning model, an output indicating the one or more of the lineage information, the availability information, or the quality information regarding the category of data; and presenting, by the one or more processors using at least one of a display device or an audio output device, the output. ### Claim 2 (depends on 1) The method of claim 1, wherein the machine learning model comprises at least one of a large language model (LLM), a generative artificial intelligence (AI) model, or a neural network comprising a transformer. ### Claim 3 (depends on 2) The method of claim 2, wherein the machine learning model is configured by fine tuning the machine learning model using the training data. ### Claim 4 (depends on 1) The method of claim 1, wherein at least one of the query is received or the output is presented via a conversational interface. ### Claim 5 (depends on 1) The method of claim 1, wherein the training data includes at least one of master data, timeseries data, transaction data, or a data definition. ### Claim 6 (depends on 1) The method of claim 1, further comprising retrieving, by the one or more processors, context from a data manager pertaining to the unstructured query, and wherein applying the unstructured query as input to the machine learning model includes applying the unstructured query and the context retrieved from the data manager to the machine learning model. ### Claim 7 (depends on 6) The method of claim 6, wherein the context includes information relating to at least one of an access controls list, data usage history, or extract, transform, and load (ETL) job processing logs. ### Claim 8 (depends on 1) The method of claim 1, further comprising: receiving, by the one or more processors, feedback regarding the output; and applying, by the one or more processors, the feedback to the machine learning model. ### Claim 9 (depends on 8) The method of claim 8, wherein the training data used to configure the machine learning model is stored within a data manager, and the method further comprises updating, by the one or more processors, the data manager using one or more of the unstructured query, the output, or the feedback. ### Claim 10 A building management system, comprising: one or more processors to: receive an unstructured query in the form of at least one of a natural language representation or a semantic representation, the unstructured query including a request for one or more of lineage information, availability information, or quality information regarding a category of data associated with the building management system, the category of data relating to one or more of equipment failure, sustainability, comfort, operations, energy, air quality, or net-zero emissions; apply the unstructured query as input to a machine learning model, the machine learning model fine-tuned using training data comprising sample data and metadata from a plurality of data sources of the building management system, wherein the machine learning model comprises at least one of a large language model (LLM), a generative artificial intelligence (AI) model, or a neural network comprising a transformer; generate, using the machine learning model, an output indicating the one or more of the lineage information, the availability information, or the quality information regarding the category of data; and present, using at least one of a display device or an audio output device, the output. ### Claim 11 (depends on 10) The building management system of claim 10, wherein the one or more processors are to retrieve a context pertaining to the unstructured query from a data manager, and wherein applying the unstructured query as input to the machine learning model includes applying the unstructured query and the context retrieved from the data manager to the machine learning model. ### Claim 12 A method, comprising: retrieving, by one or more processors, a plurality of data elements from at least one data source coupled with a building management system, the plurality of data elements comprising at least one of sample data regarding one or more items of equipment associated with the building management system or metadata indicative of a definition of the sample data; structuring, by the one or more processors, the plurality of data elements into training data for a machine learning model by mapping the plurality of data elements to a building knowledge graph data structure corresponding to the one or more items of equipment; and applying, by the one or more processors as input to the machine learning model, the training data to configure the machine learning model according to the training data. ### Claim 13 (depends on 12) The method of claim 12, wherein mapping the plurality of data elements to the building knowledge graph data structure comprises mapping the plurality of data elements to corresponding nodes of the building knowledge graph data structure. ### Claim 14 (depends on 12) The method of claim 12, further comprising: applying, by the one or more processors as input to the machine learning model, context data regarding at least one of access controls for the at least one data source, usage of the at least one data source, or data processing log information regarding the at least one data source. ### Claim 15 (depends on 14) The method of claim 14, wherein applying the context data includes combining the context data with corresponding data elements of the plurality of data elements prior to structuring the plurality of data elements into the training data for the machine learning model. ### Claim 16 (depends on 14) The method of claim 14, further comprising combining the context data with corresponding data elements based on mapping of the plurality of data elements to a plurality of nodes of the building knowledge graph data structure that are respectively associated with the one or more items of equipment. ### Claim 17 (depends on 12) The method of claim 12, wherein the sample data includes at least one of analytics data, sensor data, or telemetry data and the metadata includes one or more of the definition of the sample data, master data, timeseries data, or transaction data. ### Claim 18 (depends on 12) The method of claim 12, wherein the training data is initial training data and the initial training data is applied to configure the machine learning model in at least one first instance, and the method further comprises: subsequent to applying the training data to configure the machine learning model in the at least one first instance, retrieving, by the one or more processors, from the at least one data source, a plurality of differential data elements comprising at least one of differential sample data regarding the one or more items of equipment or differential metadata indicative of a definition of the differential sample data; structuring, by the one or more processors according to the at least one of the differential sample data or the differential metadata, the plurality of differential data elements into differential training data; and applying, by the one or more processors as supplemental input to the machine learning model, the differential training data to update or reconfigure the machine learning model according to the differential training data. ### Claim 19 (depends on 12) The method of claim 12, wherein the machine learning model comprises at least one of a large language model (LLM), a generative artificial intelligence (AI) model, or a neural network comprising a transformer. ### Claim 20 (depends on 19) The method of claim 19, wherein the machine learning model is configured by fine tuning the machine learning model using the training data.