Energy operation simulation device for retrofit objects based on rag llm

Abstract

The present invention relates to an energy operation simulation device for a retrofit object based on RAG LLM, comprising: a retrofit query receiving unit that receives a natural language query for retrofitting an object from a user; an appearance derivation unit that extracts the retrofit objective of the object based on the natural language query and inputs the retrofit objective into a RAG (Retrieval-Augmented Generation) engine to derive candidate appearance components reflecting existing retrofit cases for the object; a simulation model generation unit that generates a simulation model by selecting an optimal appearance component according to a standard condition among the candidate appearance components to generate a new retrofit design proposal and receiving at least one of the object shell structure, object light transmission structure, and object mobility relationship structure of the new retrofit design proposal as a simulation input value; a simulation execution unit that performs a simulation by outputting the energy consumption of the object before and after retrofit as a simulation output value through the simulation model; and a retrofit response providing unit that presents the energy consumption of the object before and after retrofit for the existing retrofit case and the new retrofit design proposal according to the execution of the simulation.

Description

Technology Field

[0001] The present invention relates to an energy operation simulation technology for a retrofit object, and more specifically, to a RAG LLM-based energy operation simulation device for a retrofit object capable of interpreting the energy operation purpose of the retrofit object based on natural language queries and calculating energy consumption before and after retrofit through a Large Language Model (LLM) linked with a RAG (Retrieval Augmented Generation) engine to support energy operation decision-making. Background Technology

[0003] Energy management technologies at the building and city levels are continuously being researched and adopted to reduce energy consumption, lower carbon emissions, and improve the efficiency of aging buildings. To this end, Building Energy Management Systems (BEMS), digital twin-based urban energy monitoring technologies, and simulation-based building energy forecasting technologies are widely utilized. These conventional technologies analyze building energy usage by utilizing sensor data, equipment operation history, and weather data, and provide information that can improve operational efficiency.

[0004] Meanwhile, retrofit technologies aimed at improving the energy performance of aging buildings are becoming widespread, and various methods such as envelope performance improvement, window and door replacement, equipment replacement, and insulation structural modifications are being applied for this purpose. Additionally, while building energy simulation tools like EnergyPlus and DesignBuilder are utilized to predict retrofit effects in advance, these simulations require a manual, expert-led process, and the identification of applicable retrofit elements, comparison of candidates, and prediction of performance before and after are not automated.

[0005] Furthermore, although AI-based building energy diagnostic technologies and natural language-based question-and-answer systems have recently emerged, they remain at the level of simple Q&A. Consequently, there is a lack of technology capable of interpreting retrofit objectives based on user requirements, automatically deriving design elements based on existing cases, and immediately presenting comparative changes in energy operations before and after retrofit through integration with simulations.

[0006] Ultimately, despite the partial utilization of energy analysis tools and simulations, conventional building energy diagnostic and retrofit technologies have structural limitations in that they cannot interpret retrofit objectives at the level of user requirements, cannot actively explore design elements based on existing cases, and cannot immediately compare application results. Consequently, users are required to use multiple tools in parallel to determine the direction of retrofit, and the high dependence on experts leads to a complex and time-consuming decision-making process. Prior art literature

[0008] Korean Registered Patent No. 10-2822140 (June 13, 2025) The problem to be solved

[0009] One embodiment of the present invention aims to provide a RAG LLM-based energy operation simulation device for a retrofit object that can support energy operation decision-making by interpreting the energy operation purpose of the retrofit object based on natural language queries and calculating energy consumption before and after retrofit through a Large Language Model (LLM) linked with a RAG (Retrieval Augmented Generation) engine.

[0010] One embodiment of the present invention aims to provide a RAG LLM-based energy operation simulation device for a retrofit object that enables rapid and efficient retrofit decision-making at the user level by automatically interpreting a retrofit request entered in natural language, presenting candidate exterior components through RAG-based case search, and generating a simulation model to intuitively compare energy operation results before and after retrofit. means of solving the problem

[0012] Among the embodiments, the energy operation simulation device for a retrofit object based on RAG LLM comprises: a retrofit query receiving unit that receives a natural language query for retrofitting the object from a user; an appearance derivation unit that extracts the retrofit objective of the object based on the natural language query and inputs the retrofit objective into a RAG (Retrieval-Augmented Generation) engine to derive candidate appearance components reflecting existing retrofit cases for the object; a simulation model generation unit that generates a simulation model by selecting an optimal appearance component according to standard conditions among the candidate appearance components to generate a new retrofit design proposal and receiving at least one of the object shell structure, object light transmission structure, and object mobility relationship structure of the new retrofit design proposal as a simulation input value; a simulation execution unit that performs a simulation that outputs the energy consumption of the object before and after retrofit as a simulation output value through the simulation model; and a device that presents the energy consumption of the object before and after retrofit for the existing retrofit case and the new retrofit design proposal according to the execution of the simulation. Includes a retrofit response provider.

[0013] The receiver of the retrofit query above can identify the object in the urban space, interpret the natural query based on the type of the object to perform word selection, and perform a preprocessing process to modify the sentence structure according to the word selection.

[0014] The retrofit query receiver above can automatically extract the location of the object in the city geographic information system (GIS) through the context of the natural language query, and identify the object through the extraction of the location.

[0015] The above-mentioned appearance derivation unit can determine the retrofit purpose of the object in the above-mentioned natural query as at least one of energy operation, appearance improvement, appearance repair, and change of use, and generate a RAG search query based on the above-mentioned retrofit purpose and input it into the above-mentioned RAG engine.

[0016] The above appearance derivation unit can improve search accuracy by weighting the importance of keywords in the generated RAG search query according to the retrofit purpose.

[0017] The above appearance derivation unit can derive the above candidate appearance components by performing a first search based on the type of the object in the existing retrofit case database through the above RAG engine and performing a second search based on the user's retrofit priority.

[0018] The above exterior appearance derivation unit can reconstruct the candidate exterior appearance components by reflecting weights for at least one evaluation factor among constructability, cost, and performance during the process of performing the above secondary search.

[0019] The simulation model generation unit can select the optimal exterior component by reflecting the retrofit purpose in the use of each of the candidate exterior components, and generate the new retrofit design proposal through a combination of the optimal exterior components.

[0020] The simulation model generation unit can generate the simulation input values ​​by performing Building Information Modeling (BIM) on the new retrofit design proposal to reflect spatial constraints, and by extracting the object envelope structure, object light transmission structure, building mobility structure, and object relationship structure from the BIM.

[0021] The simulation execution unit can calculate the energy consumption of the target object before and after the retrofit in the first step by setting virtual weather operation conditions according to seasonal outdoor temperature change scenarios in the simulation model and performing the simulation, and then calculate the energy consumption of the target object before and after the retrofit in the second step by setting a sudden temperature change in the virtual weather operation conditions and re-performing the simulation, and output the simulation output values ​​respectively.

[0022] The simulation execution unit can output the simulation output value by calculating the energy consumption of the object before and after the retrofit by setting virtual mobility operation conditions based on the acceptance of electric transport means by reflecting the object mobility relationship structure.

[0023] The above retrofit response providing unit can additionally present the feasibility of construction and estimated construction costs for the above new retrofit design plan, as well as the estimated distribution status of energy consumption according to the relationship structure of the above object. Effects of the invention

[0025] The disclosed technology may have the following effects. However, this does not mean that a specific embodiment must include all of the following effects or only the following effects; therefore, the scope of the rights of the disclosed technology should not be understood as being limited by this.

[0026] An energy operation simulation device for a retrofit object based on a RAG LLM according to one embodiment of the present invention can interpret the energy operation purpose of the retrofit object based on natural language queries and support energy operation decision-making by calculating the energy consumption before and after retrofit through a Large Language Model (LLM) linked with a RAG (Retrieval Augmented Generation) engine.

[0027] An energy operation simulation device for a retrofit target based on a RAG LLM according to one embodiment of the present invention can enable rapid and efficient retrofit decision-making at the user level by automatically interpreting a retrofit request entered in natural language, presenting candidate exterior components through RAG-based case search, and generating a simulation model to intuitively compare energy operation results before and after retrofit.

[0028] Therefore, the present invention can improve the efficiency and accuracy of the retrofit decision-making process and significantly reduce reliance on manual analysis by experts. It also has the effect of reducing costs and time by shortening the retrofit procedure and enabling energy performance predictions that closely approximate the actual operating environment, rather than limited analysis based on a single factor. Brief explanation of the drawing

[0030] FIG. 1 is a diagram illustrating an energy operation simulation system for a retrofit object according to the present invention. Figure 2 is a diagram illustrating the system configuration of the energy operation simulation device of Figure 1. Figure 3 is a diagram illustrating the functional configuration of the energy operation simulation device of Figure 1. FIG. 4 is a flowchart illustrating a method for simulating the energy operation of a retrofit target based on RAG LLM according to the present invention. Figure 5 is a diagram showing an example of a user-friendly interface based on RAG LLM. Specific details for implementing the invention

[0031] The description of the present invention is merely an example for structural or functional explanation, and therefore the scope of the present invention should not be interpreted as being limited by the examples described in the text. That is, since the examples are subject to various modifications and may take various forms, the scope of the present invention should be understood to include equivalents capable of realizing the technical concept. Furthermore, the objectives or effects presented in the present invention do not imply that a specific example must include all of them or only such effects; therefore, the scope of the present invention should not be understood as being limited by them.

[0032] Meanwhile, the meaning of the terms described in this application should be understood as follows.

[0033] Terms such as "first," "second," etc., are intended to distinguish one component from another, and the scope of rights shall not be limited by these terms. For example, the first component may be named the second component, and similarly, the second component may be named the first component.

[0034] When it is stated that one component is "connected" to another component, it should be understood that it may be directly connected to that other component, or that there may be other components in between. Conversely, when it is stated that one component is "directly connected" to another component, it should be understood that there are no other components in between. Meanwhile, other expressions describing the relationships between components, such as "between" and "exactly between," or "adjacent to" and "directly adjacent to," should be interpreted in the same way.

[0035] A singular expression should be understood to include a plural expression unless the context clearly indicates otherwise, and terms such as "include" or "have" are intended to specify the existence of the implemented features, numbers, steps, actions, components, parts, or combinations thereof, and should be understood not to preclude the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0036] In each step, identifiers (e.g., a, b, c, etc.) are used for convenience of explanation and do not describe the order of the steps; the steps may occur differently from the specified order unless a specific order is clearly indicated in the context. That is, the steps may occur in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.

[0037] The present invention may be implemented as computer-readable code on a computer-readable recording medium, and the computer-readable recording medium includes all types of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc. Additionally, the computer-readable recording medium may be distributed across networked computer systems, so that computer-readable code can be stored and executed in a distributed manner.

[0038] Unless otherwise defined, all terms used herein have the same meaning as generally understood by those skilled in the art to which this invention pertains. Terms defined in commonly used dictionaries should be interpreted as having meanings consistent with the context of the relevant technology and should not be interpreted as having an ideal or overly formal meaning unless explicitly defined in this application.

[0040] FIG. 1 is a diagram illustrating an energy operation simulation system for a retrofit object according to the present invention.

[0041] Referring to FIG. 1, the energy operation simulation system (100) of a retrofit object may include a user terminal (110), an energy operation simulation device (130), and a database (150).

[0042] The user terminal (110) may be a computing device capable of utilizing the energy operation simulation system (100) of the retrofit object and receiving the results of the energy operation simulation of the retrofit object based on natural language queries. The user terminal (110) may be implemented as a smartphone, laptop, or computer that can be connected to and operated with the energy operation simulation device (130), but is not necessarily limited thereto and may be implemented as various devices such as a tablet PC.

[0043] Additionally, the user terminal (110) can install and run a dedicated program or application to link with the energy operation simulation device (130). Through this, the user terminal (110) can utilize various simulations provided by the energy operation simulation device (130).

[0044] The energy operation simulation device (130) can be implemented as a server corresponding to a computer or program capable of receiving natural language queries from a user and performing energy operation simulations of a retrofit target based on RAG LLM. The energy operation simulation device (130) can interpret retrofit requirements through natural language-based queries and combine RAG-based case information and simulation models to compare energy operation results before and after retrofit. The energy operation simulation device (130) can generate simulation models according to user type or retrofit purpose and perform simulations that comprehensively consider various design elements such as the outer shell structure, light-transmitting structure, and mobility-linked structure of the retrofit target.

[0045] Additionally, the energy operation simulation device (130) can be connected to a user terminal (110) via a wired network or a wireless network such as Bluetooth, WiFi, LTE, etc., and can transmit and receive data with the user terminal (110) through the wired or wireless network. Additionally, the energy operation simulation device (130) can be implemented to operate in connection with an independent external system (not shown in FIG. 1). For example, the energy operation simulation device (130) can automatically extract the location of an object by linking with a city geographic information system (GIS).

[0046] The database (150) may correspond to a storage device that stores various information required during the operation of the energy operation simulation device (130). For example, the database (150) may store existing retrofit cases, retrofit design plans for objects, etc., and may store simulation models for performing energy operation simulation of retrofit objects, but is not necessarily limited thereto, and may store information collected or processed in various forms during the energy operation simulation of retrofit objects based on RAG LLM by the energy operation simulation device (130).

[0047] In FIG. 1, the database (150) is shown as a device independent of the energy operation simulation device (130), but is not necessarily limited thereto and can be implemented by being included in the energy operation simulation device (130).

[0049] Figure 2 is a diagram illustrating the system configuration of the energy operation simulation device of Figure 1.

[0050] Referring to FIG. 2, the energy operation simulation device (130) may be implemented to include a processor (210), memory (230), user input / output unit (250), network input / output unit (270), and communication port unit (290).

[0051] The processor (210) can execute procedures for processing each step in the process of operation of the energy operation simulation device (130), manage memory (230) that is read or written throughout the process, and schedule the synchronization time between volatile memory and non-volatile memory in memory (230). The processor (210) can control the overall operation of the energy operation simulation device (130) and can control the data flow between memory (230), user input / output unit (250), network input / output unit (270), and communication port unit (290) by being electrically connected to them. The processor (210) can be implemented as a CPU (Central Processing Unit) or GPU (Graphics Processing Unit) of the energy operation simulation device (130).

[0052] The memory (230) may include an auxiliary storage device implemented as non-volatile memory such as an SSD (Solid State Disk) or HDD (Hard Disk Drive) and used to store all data required for the energy operation simulation device (130), and may include a main memory device implemented as volatile memory such as RAM (Random Access Memory). Additionally, the memory (230) may store a set of instructions that execute an energy operation simulation of a RAG LLM-based retrofit object according to the present invention by being executed by an electrically connected processor (210).

[0053] The user input / output unit (250) may include an environment for receiving user input and an environment for outputting specific information to the user. For example, the user input / output unit (250) may include an input device including an adapter such as a touch pad, a touch screen, a virtual keyboard, or a pointing device, and an output device including an adapter such as a monitor or a touch screen. In one embodiment, the user input / output unit (250) may correspond to a computing device connected via remote access, and in such case, the energy operation simulation device (130) may be performed as an independent server.

[0054] The network input / output unit (270) includes an environment for connecting to a user terminal (110) via a network, and may include, for example, an adapter for communication such as a LAN (Local Area Network), MAN (Metropolitan Area Network), WAN (Wide Area Network), and VAN (Value Added Network). Additionally, the network input / output unit (270) may be implemented to provide short-range communication functions such as WiFi and Bluetooth, or wireless communication functions of 4G or higher for wireless transmission of data.

[0055] The communication port section (290) is a hardware interface for connecting to external hardware, for example, the external hardware may include a printer, a mouse, and USB hardware. The communication port section (290) can detect the connection of specific USB hardware and enable it to perform the role of an energy operation simulation device (130) for a retrofit target based on RAG LLM.

[0057] Figure 3 is a diagram illustrating the functional configuration of the energy operation simulation device of Figure 1.

[0058] Referring to FIG. 3, the energy operation simulation device (130) may include a retrofit query receiving unit (310), an appearance derivation unit (320), a simulation model generating unit (330), a simulation execution unit (340), a retrofit response providing unit (350), and a control unit (360).

[0059] The energy operation simulation device (130) does not need to include all of the above functional configurations simultaneously, and depending on each embodiment, some of the above configurations may be omitted, or some or all of the above configurations may be selectively included. Additionally, the energy operation simulation device (130) may be implemented as an independent module that selectively includes some of the above configurations, and the energy operation simulation method of a retrofit target based on RAG LLM according to the present invention may be performed through the interoperability between each module. The operation of each configuration is described in detail below.

[0060] The retrofit query receiver (310) can receive a natural language query for retrofitting an object from a user. The retrofit query receiver (310) may be configured to include an input interface module for receiving the user's natural language query in the form of text, voice, or multimodal. The retrofit query receiver (310) can directly receive a retrofit request as text or voice input from a user terminal (110), and in the case of voice input, it can be converted into natural language text by linking with a STT (Speech-to-Text) engine. In the case of multimodal input, the retrofit query receiver (310) can interpret the natural language combined with an image, BIM view, or map area selection input and transmit selected spatial object information along with the natural language query. The retrofit query receiver (310) can support various forms of user input without restrictions on the retrofit request input channel.

[0061] Additionally, the retrofit query receiver (310) can perform a sentence normalization process for the input natural language query, which includes removing particles and terminal words, shortening duplicate sentences, correcting incomplete sentences, correcting typos, and converting imperative sentences, and performing structural transformations to interpret the sentence based on meaning. In one embodiment, the retrofit query receiver (310) can perform a preprocessing process that identifies an object in an urban space, interprets the natural language query based on the type of object to select words, and modifies the sentence structure according to the word selection. During the preprocessing process, the retrofit query receiver (310) can perform an enhancement of the query preprocessing function by logically decomposing negative words, conditional statements, and comparison statements within the sentence and delivering them in a format optimized for RAG and simulation. The retrofit query receiver (310) can automatically extract the location of an object in an urban geographic information system (GIS) through the context of the natural language query and identify the object through the extraction of the location. The retrofit query receiver (310) can identify location entities and spatial indicator expressions, such as place names, building names, addresses, floor or zone information, and demonstrative pronouns (e.g., this building, there, our office, etc.) within a natural language input sentence through geographic information context analysis. The retrofit query receiver (310) can perform geographic information context analysis such as explicit location expression analysis, implicit location expression analysis, spatial indicator interpretation, and context-based interpretation. In the case of implicit expressions, the reliability of geographic information interpretation can be increased by additionally utilizing user account information, recent query object history, and the GPS of the user terminal (110). To convert the identified location information candidates into a form that can be searched by city GIS, the retrofit query receiver (310) can define them as GIS object IDs by normalizing them according to an address system, an administrative district system, and a building DB reference system.The retrofit query receiving unit (310) can verify whether the object is capable of retrofit simulation by linking normalized location information with a city GIS server or a GIS data storage unit to query the spatial object of the object and obtain building metadata such as use, floor area, number of floors, and heating and cooling separation area.

[0062] Additionally, the retrofit query receiver (310) can determine the type of object using GIS metadata or a building attribute DB once the object is identified. Here, the type of object may include buildings such as commercial buildings, houses, and multi-unit dwellings, but is not necessarily limited to these and can be expanded to include mobility such as electric vehicles, eVTOLs, and drones. The retrofit query receiver (310) can select domain-corresponding words by combining the type of the identified object with the query context. For example, if the user's natural language query is "too cold in winter," the retrofit query receiver (310) can select words such as "heat loss," "insulation performance," and "window airtightness" by using the building's exterior wall or window as the type of object. Once the word selection is complete, the retrofit query receiver (310) can analyze the sentence structure and reconstruct it into a structured sentence with separated intent, purpose, and conditions. For example, the original sentence “What should be done because our building is cold in winter” can be transformed into a structured sentence that reveals the type of object and the purpose of the retrofit, such as “Analyze the retrofit requirements for improving the insulation performance of exterior walls and windows.”

[0063] The exterior extraction unit (320) can extract the retrofit purpose of an object based on a natural language query and input the retrofit purpose into a RAG (Retrieval-Augmented Generation) engine (325) to derive candidate exterior components that reflect existing retrofit cases for the object. The exterior extraction unit (320) can analyze domain keywords within the natural language query and classify the retrofit purpose. Here, a domain keyword is a term having a technical meaning related to the shell structure, facility structure, or energy operation characteristics of the retrofit object, and refers to a key word extracted from a natural language query to identify the retrofit purpose. The domain keyword includes technical concepts for defining the retrofit purpose or specifying simulation conditions, and can be used as reference information in the process of extracting the retrofit purpose, searching for RAG-based cases, and deriving candidate exterior components. In one embodiment, the exterior derivation unit (320) analyzes domain keywords and contextual information related to envelope performance, load influence, and energy operation, such as heat loss, insulation, light transmission, shading, airtightness, and cooling load, included in a natural language query, and classifies them into at least one of the target categories, such as insulation improvement, light transmission control, envelope reinforcement, and window performance improvement. The exterior derivation unit (320) performs morphological analysis, dependency parsing, and Named Entity Recognition on the natural language query and identifies the object location, building type, envelope elements, and retrofit requirements to convert the unstructured input sentence into a structured semantic unit.

[0064] The exterior extraction unit (320) can convert the retrofit purpose into a searchable query form for the RGA engine (325) once the retrofit purpose is determined. In one embodiment, the exterior extraction unit (320) can determine the retrofit purpose of the object from a natural language query as at least one of energy operation, exterior improvement, exterior repair, and change of use, and generate a RAG search query based on the retrofit purpose and input it into the RAG engine (325). The exterior extraction unit (320) can determine the retrofit purpose according to a pre-set classification rule by extracting domain keywords, object type information, and user constraints included in the natural language query. Domain keywords may include heat loss, insulation, airtightness, light transmission, shading, cracks, corrosion, leakage, extension, change of use, etc. User constraints may include budget, construction period, construction difficulty, etc. Here, although the retrofit purpose is classified into energy operation, exterior improvement, exterior repair, and change of use, it is not necessarily limited to these and can be determined in various ways depending on the type of object. In one embodiment, the appearance derivation unit (320) may calculate a priority by considering the weight of domain keywords, the suitability of object types, and user constraints when multiple objectives are simultaneously satisfied, determine the objective with the highest priority as the primary objective, and determine the objectives with the remaining ranks as auxiliary objectives. The appearance derivation unit (320) may generate a structured search query suitable for case search of the RAG engine (325) in response to the determined retrofit objectives. The appearance derivation unit (320) may generate a purpose-oriented RAG search query by selecting a template prompt corresponding to the retrofit objectives. The template prompt may be designed to derive search results with high suitability for objectives by reflecting elements such as energy loss factors, appearance conditions, light transmission structures, mobility loads, and thermal insulation performance in a combination form for each objective.A RAG search query may include at least some of the purpose field, target type field, application area field, environment and climate field, constraint field, and evaluation metric field. For example, a RAG search query may be composed of {purpose; target_type; apply_zone; environment; constraints; metrics} = {energy operation; building use; south exterior wall; climate zone; budget limit; expected savings rate}. The exterior derivation unit (320) can automatically map the fields from the preprocessing results of the natural language query and, if there are missing items, use the target metadata or basic profile to supplement the missing fields.

[0065] Additionally, the appearance derivation unit (320) can improve search accuracy by weighting the importance of keywords in the generated RAG search query according to the retrofit purpose. In one embodiment, the appearance derivation unit (320) can calculate the importance by analyzing the semantic relevance that the keywords have with the retrofit purpose for the keywords in the RAG search query. The appearance derivation unit (320) can calculate the importance based on at least one of the relevance between the retrofit purpose and the keywords, the suitability of the object type, the environmental and climate impact, and the case suitability history value. For example, if the retrofit purpose of the object is energy operation, keywords in the series of energy-loss, insulation, infiltration, SHGC, and conduction can be weighted. The appearance derivation unit (320) can generate a weight vector by reflecting the importance of the keywords in the search query and adjust the search score of each keyword using at least one of a weighted similarity calculation method, a priority filtering method, and a score correction method. The weighted similarity calculation method is a method that incorporates keyword weights into the cosine similarity or dot-product operation during the search. The priority filtering method is a method that first compares important keywords to derive a primary candidate group and then applies the remaining keywords. The score correction method is a method that recalculates the final ranking score by adding keyword weights to the basic score. The appearance derivation unit (320) can reduce the influence of keywords unrelated to the core purpose during the RAG search process through weight application operations and ensure that keywords directly corresponding to the purpose are reflected first.

[0066] Additionally, the exterior derivation unit (320) can derive candidate exterior components by performing a first search based on the type of object in the existing retrofit case database through the RAG engine (325) and a second search based on the user's retrofit priority. During the second search, the exterior derivation unit (320) can reconstruct the candidate exterior components by reflecting weights for at least one evaluation factor among constructability, cost, and performance.

[0067] RAG (Retrieval-Augmented Generation) is a hybrid natural language processing (NLP) technique that retrieves relevant information from an external knowledge base (e.g., documents, databases) and generates a natural language response based on that information. RAG consists of a module that retrieves documents related to a query from an external knowledge base and a language generation module (e.g., GPT, BERT-based LM, etc.) that generates response sentences based on the retrieved documents. Here, the RAG engine (325) refers to a search-combination type artificial intelligence processing engine that receives a search query including retrofit purpose, object type, and conditions, retrieves documents, history, design cases, and construction method information semantically related to the query from an external knowledge repository or retrofit case database, and outputs a goal-oriented generation result based on the retrieved information. The RAG engine (325) may be configured to include a similarity embedding-based retriever and a generative language model generator, and operates to improve the reliability of the response of the generative model based on the search results. The RAG engine (325) can operate to first search for documents containing retrofit cases, shell design information, energy improvement methods, and passive design techniques that are semantically similar to the query for an input search query, and then combine some or all of the searched documents into the context of a generative model to generate retrofit candidate elements, techniques, and design proposals suitable for the object. By combining knowledge search and evidence-based generation, the RAG engine (325) can reduce hallucination and provide reliable retrofit information based on actual cases.

[0068] In one embodiment, the appearance derivation unit (320) can generate a primary RAG search query containing type information of the object and input it into the RAG engine (325) to search for retrofit cases that match the type as a primary candidate group. At this time, the appearance derivation unit (320) can sort the search results based on the Type Match Score to primarily exclude elements that are unsuitable for the structure of the object. To reflect the user's retrofit purpose and priority in the cases searched in the primary search, the appearance derivation unit (320) can perform a re-search by generating a secondary RAG search query that includes the user's priority and using the primary candidate group as input. In the secondary search, the appearance derivation unit (320) can apply priority weights to adjust the final similarity score of the RAG search results, thereby enabling the selective derivation of only candidates that match the user's strategy, even if they are of the same type. This can be defined by the following mathematical formula 1.

[0069] [Mathematical Formula 1]

[0070] Final Similarity Score = (Type Fit × α) + (Previous Fit Score × β)

[0071] However, the priority reflection can be strengthened by setting β>α.

[0072] The exterior extraction unit (320) can generate candidate exterior components by selecting the top N cases (where N is a natural number) from the secondary search results and extracting exterior components included in each case, such as exterior wall insulation, Low-E glass, curtain wall shading system, shading film, and leakage prevention method. Meta information such as the application envelope area, purpose suitability score, simulation feasibility, and expected performance indicators may be stored together with the candidate exterior components.

[0073] The simulation model generation unit (330) can generate a new retrofit design plan by selecting an optimal exterior component according to standard conditions among candidate exterior components and can generate a simulation model that receives at least one of the object shell structure, object light transmission structure, and object mobility relationship structure of the new retrofit design plan as a simulation input value. Here, the standard conditions may include energy performance, cost, or constructability. In one embodiment, the simulation model generation unit (330) can select an optimal exterior component by reflecting the retrofit purpose in the use of each candidate exterior component and generate a new retrofit design plan through a combination of the optimal exterior components. Specifically, the simulation model generation unit (330) can calculate a suitability score by reflecting at least one retrofit purpose among energy operation, exterior improvement, exterior repair, and change of use based on the use of each candidate exterior component, such as exterior wall insulation reinforcement, window replacement, sunshade installation, roof heat insulation layer reinforcement, etc. Suitability can be calculated according to the following mathematical formula 2.

[0074] [Mathematical Formula 2]

[0075] Suitability = ∑(w_p(k)·f_k(candidate_use, target_type))

[0076] Here, w_p(k) is the weight of the k-th evaluation item (e.g., potential for U-value improvement, SHGC improvement, leakage reduction, visual change, etc.) for objective p (e.g., energy operation), and f_k(·) represents the contribution based on the candidate's use and object type.

[0077] The simulation model generation unit (330) can calculate a comprehensive score for each candidate by integrating the purpose suitability score and the standard conditions, and can determine the final optimal candidate by applying an auxiliary judgment rule when the score difference between candidates is less than a threshold value. For example, the auxiliary judgment rule may include minimizing the risk of structural interference, compliance with regulations (e.g., lighting, evacuation, energy-saving design standards, etc.), and whether construction is permitted during operation. The simulation model generation unit (330) can select one or more optimal candidates for each use and combine the optimal candidates to generate a single-element or complex-element design proposal candidate combination, and can verify the validity of the generated design proposal candidate combination according to a compatibility matrix and interaction rules. The simulation model generation unit (330) can generate candidate combinations through rule-based search, heuristic search, or limited combination optimization, and can exclude unreasonable combinations such as violations of regulations, structural non-compatibility, or excessive construction periods early on. The simulation model generation unit (330) can determine the top N design proposals among the combinations that passed verification as new retrofit design proposals. In one embodiment, the simulation model generation unit (330) can generate simulation input values ​​by performing Building Information Modeling (BIM) on a new retrofit design to reflect spatial constraints and extracting the object envelope structure, object light transmission structure, building mobility structure, and object relationship structure from the BIM. The simulation model generation unit (330) can reflect the elements of the new retrofit design (e.g., exterior wall insulation reinforcement, window replacement, sunshade, roof heat shielding layer, EV charging infrastructure, etc.) into a 3D information model through a BIM authoring tool or an Industry Foundation Classes (IFC) compatible engine, manage version and revision information for each element, and assign metadata such as geometry, material, property set, orientation, and level to the BIM object.In addition, the simulation model generation unit (330) can analyze the retrofit application area by querying the spatial topology (Room / Zone / Level / Surface / Opening relationship) of the BIM model, structural members (columns, beams, slabs, shear walls), equipment shafts and ducts, and legal minimum separation distance, evacuation, and lighting conditions, and can perform crash detection for each design element to detect structural interference, equipment interference, and violations of window opening dimensions, and can propose automatic correction plans such as replacement use, position adjustment, and module quantity change if necessary. The spatial constraint results can be output as correction parameters (e.g., reduction of window module width, limitation of sunshade projection length, adjustment of EV charger spacing) along with a status such as applicable, conditional, or impossible.

[0078] In one embodiment, the simulation model generation unit (330) can extract the outer shell structure of the object from BIM. The outer shell structure may include an exterior wall, windows, a roof, a floor, an insulation layer, a waterproofing layer, etc. The simulation model generation unit (330) can extract the thickness, thermal conductivity, density, ratio, surface coefficient, and thermal bridge correction coefficient of the exterior wall, roof, and floor surfaces and layers (including the insulation layer and waterproofing layer), and calculate the normal direction, orientation, area, and shadow effect (relationship with the shading object) for each surface to extract the outer shell structure. In one embodiment, the simulation model generation unit (330) can extract the light transmission structure of the object from BIM. The light transmission structure may correspond to the transmittance of the windows. The simulation model generation unit (330) can extract glass specifications (VLT, SHGC, U-value) of windows and curtain walls, frame thermal characteristics, spacer and gas filling information, opening dimensions and opening / closing characteristics, and analyze the mutual positional relationship with shading (louvers, overhangs, screens) to extract solar blocking coefficients and control rules (based on solar radiation and illuminance). Additionally, the simulation model generation unit (330) can extract building mobility structures and object relationship structures as object mobility relationship structures in BIM. The object mobility relationship structure may correspond to Hub-to-Hub or Building-to-Hub configurations, such as electric vehicle charger configurations. The simulation model generation unit (330) can extract the number of EV chargers, rated capacity, placement coordinates, circuit connections, load groups, and operation schedules, and estimate wiring paths and peak power effects with the power room and distribution board, enabling conversion into power load inputs for the simulation. The simulation model generation unit (330) can generate a relationship graph between objects and encode rules related to surface proximity, opening-shade dependency, space-equipment linkage (ventilation, lighting, charger), evacuation, lighting, solar radiation, and thermal bridge into the relationship graph so that it can be used for connecting boundary conditions of simulation input and automatically synthesizing control logic.

[0079] The simulation model generation unit (330) can eliminate the risk of design feasibility and legal conflicts in advance by reflecting BIM-based spatial constraints, and can consistently insert operational logic such as solar radiation, illuminance, and peak power into the model through automatic control synthesis based on relationship graphs that automatically generate high-reliability simulation inputs reflecting the actual operating environment by simultaneously extracting and normalizing the envelope, light transmission, mobility, and relationship structures.

[0080] The simulation execution unit (340) can perform a simulation of the energy consumption of the object before and after retrofit as a simulation output value through a simulation model. The simulation model refers to an analysis model for energy analysis configured with input variables such as building physical elements including the object's envelope structure, light transmission structure, mobility structure, and object relationship structure, weather conditions, occupancy and operation schedules, control rules, and equipment performance parameters, in order to evaluate the energy consumption characteristics and operating environment of the object before and after retrofit in a virtual space. The simulation model can reflect the thermal characteristics of the envelope, such as the exterior wall, roof, floor, insulation layer, and waterproofing layer; the thermal characteristics, such as the transmittance and shading conditions of windows and doors; load characteristics according to the configuration of the electric vehicle charging infrastructure; heat and air flow characteristics between zones; and external boundary conditions. It can also make it possible to quantitatively compare performance changes before and after by generating a model before retrofit application (Baseline model) and a model after retrofit application (Retrofit model) under the same conditions.

[0081] In one embodiment, the simulation execution unit (340) can calculate the energy consumption of the object before and after retrofit in a first step by setting virtual weather operation conditions according to seasonal outdoor temperature change scenarios in the simulation model and performing the simulation, and then calculate the energy consumption of the object before and after retrofit in a second step by setting rapid temperature changes in the virtual weather operation conditions and re-performing the simulation, thereby outputting the simulation output values. The simulation execution unit (340) can generate a four-season virtual outdoor scenario reflecting heating and cooling loads based on a seasonal outdoor temperature change curve based on regional weather data, and can set a weather scenario in which solar radiation, humidity, wind speed, occupancy schedule, indoor temperature set values, etc., are fixed and only the outdoor temperature is changed according to seasonal characteristics for performance comparison under the same conditions. At this time, a separate outdoor scenario reflecting rapid temperature changes (e.g., widening of daily temperature range, cold wave / heat wave pattern, non-linear outdoor fluctuation) can be prepared in preparation for the second calculation conditions. The simulation execution unit (340) can apply virtual weather operating conditions based on seasonal outdoor temperatures to the simulation model to execute the Baseline model (before retrofit) and the Retrofit model (after retrofit), respectively, and calculate the scenario-based annual or seasonal primary energy consumption including cooling and heating loads, total energy consumption (EUI), peak power, HVAC operating amount, and EV load interaction. Then, the simulation execution unit (340) can calculate the secondary energy consumption by re-executing the simulation while maintaining the same operating conditions as the primary scenario and changing the outdoor temperature variable to a rapid climate change scenario, thereby reflecting the thermal responsiveness of the retrofit element, the stability of cooling and heating load control, and the advantages and disadvantages of shading and airtightness performance under conditions of rapid outdoor temperature change.

[0082] In one embodiment, the simulation execution unit (340) can calculate the energy consumption of the object before and after retrofit and output a simulation output value by performing a simulation by setting virtual mobility operation conditions according to the acceptance of an electric transport means by reflecting the object mobility relationship structure. Here, the electric transport means may include transport means that move using electricity, such as EVs and UAMs. The object mobility relationship structure may include an electric vehicle charger, a UAM vertical take-off and landing pad, a power supply module, a distribution board, an ESS, a load control device, etc. The simulation execution unit (340) generates virtual mobility operation conditions including the operation pattern of electric transport means, charging and movement cycles, and load patterns by time period based on the mobility relationship structure of the object, and executes the Baseline model (before retrofit) and the Retrofit model (after retrofit) respectively under the same mobility conditions to calculate the total energy consumption, mobility load contribution, peak power change, and HVAC impact, and by comparing the results, can derive indicators such as the change in total energy consumption before and after retrofit, the energy reduction effect in response to HVAC due to light transmission and exterior improvement, analysis of the contribution to peak power mitigation, and analysis of greenhouse gas reduction amount. In particular, the simulation execution unit (340) can quantitatively analyze whether retrofit elements such as window insulation improvement, shading improvement, and leakage reduction maintain their effectiveness or are offset even in situations where the mobility load increases.

[0083] The retrofit response providing unit (350) can present the energy consumption of the object before and after retrofit for existing retrofit cases and new retrofit design proposals according to the performance of the simulation. In one embodiment, the retrofit response providing unit (350) can collect data on the energy consumption of the object before and after retrofit, peak change amount, and load-specific items (heating and cooling, lighting, plug, mobility load, etc.), and also collect the same indicators for existing retrofit case candidates based on RAG, and automatically calculate before and after comparison indicators as shown in Table 1 below.

[0084] [Table 1]

[0085]

[0086] Additionally, the retrofit response providing unit (350) can rank existing cases and new design proposals according to reduction rate, cost efficiency, construction difficulty, mobility load response efficiency, etc., and can provide the compared data visualized as graphs, charts, text summaries, and contribution analysis results by item. The retrofit response providing unit (350) can automatically generate descriptive sentences of the effects of the retrofit design along with the visualization results.

[0087] In one embodiment, the retrofit response providing unit (350) may additionally present the feasibility of construction and estimated construction costs for the new retrofit design plan, as well as the estimated distribution status of energy consumption according to the relationship structure of the target object. The retrofit response providing unit (350) may calculate the feasibility of construction and estimated construction costs for the new retrofit design plan based on energy consumption information before and after retrofit derived from the simulation results. The retrofit response providing unit (350) may evaluate the feasibility of construction for candidate exterior components included in the new retrofit design plan by referring to spatial and structural constraint information from the IBM model and the simulation model. Here, the evaluation indicators may include at least one of the risk of structural interference, accessibility to construction, operational maintenance feasibility during construction, and compliance with regulations. If constraints are detected during the feasibility evaluation, the retrofit response providing unit (350) may output alternative usage suggestions, alternative methods for each construction phase, or construction zone division plans together. The retrofit response providing unit (350) can calculate the estimated construction cost based on the material specifications, construction scope, construction type classification, quantity information (BOQ), and estimated construction period of the retrofit design plan, and can also provide the energy saving effect relative to the cost. The retrofit response providing unit (350) can analyze the distribution status of which element within the building consumes the energy after retrofit based on the object relationship structure, and can provide indicators including the consumption rate by zone, the consumption rate by facility, the distribution by time of day, and the interaction analysis. The retrofit response providing unit (350) can combine the energy saving effect, construction difficulty, construction cost, and energy distribution status to present a single response to the user, and can provide it in the form of a UI, API, PDF report, dashboard, etc.

[0088] The control unit (360) controls the overall operation of the energy operation simulation device (130) and can manage the control flow or data flow between the retrofit query receiving unit (310), the appearance derivation unit (320), the simulation model generating unit (330), the simulation execution unit (340), and the retrofit response providing unit (350).

[0090] FIG. 4 is a flowchart illustrating a method for simulating the energy operation of a retrofit target based on RAG LLM according to the present invention.

[0091] Referring to FIG. 4, the energy operation simulation device (130) can receive a natural language query for retrofitting an object from a user through a retrofit query receiver (310) (step S410). The retrofit query receiver (310) supports multi-channel input such as text, voice query, image query, and drawing query, and can adjust the query interpretation method and prompt according to the profile of the queryer (e.g., architect, general public, public official, etc.).

[0092] Additionally, the energy operation simulation device (130) can derive the retrofit purpose of the object based on a natural language query through the appearance derivation unit (320) and input the retrofit purpose into the RAG (Retrieval-Augmented Generation) engine (325) to derive candidate appearance components that reflect existing retrofit cases for the object (step S430). The energy operation simulation device (130) can generate a new retrofit design proposal by selecting the optimal appearance component according to the standard conditions among the candidate appearance components through the simulation model generation unit (330), and can generate a simulation model that receives at least one of the object shell structure, object light transmission structure, and object mobility relationship structure of the new retrofit design proposal as a simulation input value (step S450). The standard conditions may include energy performance, cost, or constructability. The shell structure includes an exterior wall, windows, a roof, a floor, an insulation layer, a waterproof layer, etc., the light transmission structure includes the transmittance of the windows, and the object mobility relationship structure may include the configuration of an electric vehicle charger.

[0093] Additionally, the energy operation simulation device (130) can perform a simulation in which the energy consumption of the object before and after retrofit is output as a simulation output value through a simulation model in the simulation execution unit (340) (step S470). The energy operation simulation device (130) can present the energy consumption of the object before and after retrofit for existing retrofit cases and new retrofit design proposals according to the execution of the simulation through the retrofit response providing unit (350) (step S490).

[0095] Figure 5 is a diagram showing an example of a user-friendly interface based on RAG LLM.

[0096] Referring to FIG. 5, the energy operation simulation device (130) can be implemented to provide a user-friendly interface (UI) based on RAG LLM to a user terminal (110) to visualize urban spatial information and energy consumption data, and to allow the user to search for and select retrofit targets and view simulation results through natural language queries. The energy operation simulation device (130) can be linked with an urban spatial information system (GIS) to visualize multiple building information in the form of a 3D map on the UI of FIG. 5 and automatically interpret the user's query into retrofit purpose, target location, building type, energy operation conditions, etc., to provide search and recommendation results. By selecting an area of ​​interest (530) in the map-based UI (510) or directly clicking on a specific building, the user can intuitively check the energy consumption before and after retrofit of the building, candidate exterior components for application, priority retrofit recommendation elements, estimated costs, carbon reduction effects, etc. The user-friendly interface exemplified in Fig. 5 is linked with RAG LLM, and when a natural language command is received from a user, it can combine RAG search results and simulation results to automatically generate a retrofit analysis screen optimized for user requirements.

[0097] In addition, as illustrated in FIG. 5, retrofit classification (energy operation, exterior improvement, change of use, mobility response), data filters (gross floor area, use, building age, energy consumption range), and simulation conditions (weather conditions, time of day, season) can be intuitively adjusted through the left menu area (550) of the UI screen, and the entire screen can be updated immediately after post-processing by RAG LLM. Therefore, users can explore retrofit targets in an interactive manner and make simulation-based decisions without professional energy engineering knowledge.

[0099] The energy operation simulation device for a retrofit target based on RAG LLM according to the present invention can improve the efficiency and accuracy of the retrofit decision-making process by rapidly interpreting retrofit requirements through natural language-based queries and intuitively comparing energy operation results before and after retrofit by combining RAG-based case information and simulation models.

[0100] Furthermore, the processes of searching for candidate exterior components, generating design proposals, and performing simulations are automated, which can significantly reduce reliance on manual analysis by experts. Additionally, by adaptively controlling the method of generating simulation models and presenting results according to user type or retrofit purpose, information can be delivered at a level appropriate to the user, thereby improving the user experience during the decision-making process.

[0102] Although the present invention has been described above with reference to preferred embodiments, those skilled in the art will understand that various modifications and changes can be made to the invention without departing from the spirit and scope of the invention as described in the following claims. Explanation of the symbols

[0104] 100: Energy Operation Simulation System for Retrofit Objects 110: User terminal 130: Energy operation simulation device 150: Database 210: Processor 230: Memory 250: User I / O Section 270: Network I / O Section 290: Communication port section 310: Retrofit query receiver 320: Appearance extraction unit 325: Retrieval-Augmented Generation (RAG) Engine 330: Simulation Model Generation Unit 340: Simulation Execution Unit 350: Retrofit response provider 360: Control unit

Claims

Claim 1 A retrofit query receiving unit that receives a natural language query from a user for the retrofit of an object; an exterior derivation unit that extracts the retrofit purpose of the object based on the natural language query, generates a RAG search query based on the retrofit purpose and inputs it into a RAG (Retrieval-Augmented Generation) engine, performs a first search based on the type of the object in an existing retrofit case database through the RAG engine, performs a second search based on the user's retrofit priority to derive candidate exterior components reflecting existing retrofit cases for the object, sorts the search results based on the Type Match Score during the execution of the first search to exclude elements unsuitable for the structure of the object in the first search, and applies priority weights in the second search to adjust the final similarity score of the RAG search results to selectively derive only candidates suitable for the retrofit purpose even if they are of the same type; and calculates a purpose suitability score by reflecting the retrofit purpose in the use of each candidate exterior component, and uses criteria conditions including energy performance, cost, or constructability as the purpose suitability A simulation model generation unit that calculates a comprehensive score for each candidate by integrating with scores to select an optimal exterior component, generates a new retrofit design proposal through a combination of the optimal exterior components, and generates a simulation model that receives at least one of the object shell structure, object light transmission structure, and object mobility relationship structure of the new retrofit design proposal as a simulation input value; a simulation execution unit that performs a simulation that outputs the energy consumption of the object before and after retrofit as a simulation output value through the simulation model; and a unit that presents the energy consumption of the object before and after retrofit for the existing retrofit case and the new retrofit design proposal according to the execution of the simulation. Energy operation simulation device for a retrofit target based on RAG LLM including a retrofit response providing unit. Claim 2 A RAG LLM-based energy operation simulation device for a retrofit object according to claim 1, wherein the retrofit query receiving unit identifies the object in an urban space, interprets the natural query based on the type of the object to perform word selection, and performs a preprocessing process to modify the sentence structure according to the word selection. Claim 3 In paragraph 2, the retrofit query receiving unit automatically extracts the location of the object in a city geographic information system (GIS) through the context of the natural language query, and identifies the object through the extraction of the location, characterized in that it is a RAG LLM-based energy operation simulation device for a retrofit object. Claim 4 A RAG LLM-based energy operation simulation device for a retrofit object according to claim 1, wherein the appearance derivation unit determines the retrofit purpose of the object in the natural language query as at least one of energy operation, appearance improvement, appearance repair, and change of use, and improves search accuracy by weighting the importance of keywords in the generated RAG search query according to the retrofit purpose. Claim 5 delete Claim 6 delete Claim 7 In claim 4, the appearance derivation unit is characterized by reconstructing the candidate appearance components by reflecting weights for at least one evaluation factor among constructability, cost, and performance during the execution of the second search, in a RAG LLM-based energy operation simulation device for a retrofit target. Claim 8 delete Claim 9 An energy operation simulation device for a retrofit object based on RAG LLM according to claim 1, wherein the simulation model generation unit performs Building Information Modeling (BIM) on the new retrofit design proposal to reflect spatial constraints, and extracts the object shell structure, object light transmission structure, and object mobility relationship structure from the BIM to generate the simulation input values. Claim 10 A RAG LLM-based energy operation simulation device for a retrofit object, characterized in that, in claim 1, the simulation execution unit performs the simulation by setting virtual weather operation conditions according to a seasonal ambient temperature change scenario in the simulation model to first calculate the energy consumption of the object before and after the retrofit, and performs the simulation again by setting a rapid temperature change in the virtual weather operation conditions to secondarily calculate the energy consumption of the object before and after the retrofit, and outputs the simulation output values ​​respectively. Claim 11 A RAG LLM-based energy operation simulation device for a retrofit object, characterized in that, in claim 1, the simulation execution unit sets virtual mobility operation conditions based on the acceptance of an electric transport means by reflecting the mobility relationship structure of the object and performs the simulation to calculate the energy consumption of the object before and after the retrofit and outputs the simulation output value. Claim 12 An energy operation simulation device for a retrofit object based on RAG LLM, wherein, in claim 1, the retrofit response providing unit additionally presents the feasibility of construction and the estimated construction cost of the new retrofit design proposal, and the estimated distribution status of energy consumption according to the mobility relationship structure of the object.

Images