CN-121980651-A - BIM parameter optimization method for child hospital based on multi-objective GAN and LLM updating

CN121980651ACN 121980651 ACN121980651 ACN 121980651ACN-121980651-A

Abstract

A BIM parameter optimization method of a child hospital based on multi-objective GAN and LLM update comprises the following steps of firstly extracting the latest standard from a weight data source by adopting LLM combining GraphRAG technology, combining self-collected component data of the child hospital to generate a structured data set containing component parameters and corresponding standards, secondly analyzing the component position relation by utilizing NLP, constructing a knowledge graph KG model, storing the knowledge graph KG model in a Neo4j database to form a component-standard-position relation network, thirdly, periodically verifying and updating the standard and the component relation by LLM+ GraphRAG based on the KG model to generate a dynamic updated optimization constraint condition, and fourthly, generating a core optimization parameter scheme by utilizing a multi-objective GAN training generator and a discriminator based on the updated optimization constraint condition of the third step, and outputting an optimized BIM design. The invention ensures that the optimization process is adapted to the latest specifications, and solves the defects of the traditional method in multi-objective balance, standard dynamic property and engineering feasibility.

Inventors

WANG RU
LU JIALIANG
WU YUHENG
PAN JIANGTAO

Assignees

西安建筑科技大学

Dates

Publication Date: 20260505
Application Date: 20260113

Claims (8)

1. The BIM parameter optimization method for the child hospital based on the update of the multi-target GAN and the LLM is characterized by comprising the following steps of: Extracting the latest standard from a weight data source by adopting LLM (logical Link management) combined GraphRAG technology, and combining self-collected component data of a child hospital to generate a structured dataset containing component parameters and corresponding standards; step two, based on the structured data set, utilizing NLP to analyze the position relation of the component, constructing a knowledge graph KG model, and storing the knowledge graph KG model in a Neo4j database to form a component-standard-position relation network; step three, based on the KG model, periodically verifying and updating the relation between the standard and the component through LLM+ GraphRAG to generate a dynamically updated optimization constraint condition; And step four, generating a core optimization parameter scheme through a multi-target GAN training generator and a discriminator based on the updated optimization constraint condition of the step three, and outputting the optimized BIM design.
2. The method for optimizing BIM parameters of a child hospital based on multi-objective GAN and LLM update according to claim 1, wherein the step one specifically comprises: the multi-dimensional standard constraint to be met in the step comprises barrier-free design specifications, building lighting design standards, civil building sound insulation design specifications and public building energy-saving design standards, and the key optimization targets comprise the following quantitative constraint thresholds: (1) Wherein slope represents the slope (unit: degree) of the ramp, which is the upper threshold value of the included angle between the surface of the wheelchair ramp and the horizontal plane, and the slope of the wheelchair ramp is not more than 1:12, namely less than or equal to 8.DEG; (2) Wherein width represents the minimum width (unit: mm) of the door, which is the minimum clear width lower limit threshold of the wheelchair passing door opening or passage, and the clear width of the barrier-free entrance/exit should not be less than 900 mm; (3) Lighting_coeffient represents a lighting coefficient (dimensionless), is a minimum lower threshold value of indoor lighting uniformity and illuminance standard, and is more than or equal to 0.7; (4) Space_ utilization represents space utilization (dimensionless), which is a minimum lower threshold value of the proportion of effective functional area to total building area, and hospital building space utilization suggestion is more than or equal to 0.8 so as to reduce the energy consumption of the ineffective space; (5) where energy_control represents ventilation and lighting energy consumption (in kilowatt-hours) and is the upper threshold for energy consumption of the relevant system.
3. The method for optimizing BIM parameters of a child hospital based on multi-objective GAN and LLM update according to claim 2, wherein the method is characterized in that a LLM combining GraphRAG technology is adopted to extract standards from a authority data source, and structural data sets are generated by combining self-collected component data of the child hospital, and the extraction process is realized by the following formula: (6) Wherein, the As a standard text of the document, In order to provide the component data, In order to constrain the set of constraints, GraphRAG generates structural constraint through sub-graph query to cover mapping relation of component type, parameter and standard threshold value, and provides unified input for KG construction and GAN optimization.
4. The method for optimizing BIM parameters of a child hospital based on multi-objective GAN and LLM update according to claim 3, wherein the step two is specifically: based on the structured dataset of the step one, analyzing the positional relationship of the components by using NLP (spaCy), constructing a KG model, wherein the expression is as follows: (7) Wherein, the For nodes, the nodes include component nodes such as ramps, doors, windows, and standard constraint nodes, location nodes, For edges, edges include component-standard constraint edges, component-location edges, component-component topology edges, Is a set of relationship attributes; And the KG model is stored in a Neo4j database, and analysis of complex spatial relations is realized by combining entity identification and rule extraction.
5. The method for optimizing BIM parameters of a child hospital based on multi-objective GAN and LLM update according to claim 4, wherein the specific analysis implementation process is as follows: firstly, performing entity recognition and dependency syntax analysis on a position description text of a component by using an NLP tool, and recognizing a position entity and a spatial relationship modifier; Secondly, extracting the relation by combining with a domain-specific rule template, and supplementing the implicit relation through a predefined rule template or multi-hop dependency path analysis; The extracted nodes and relations are imported into the Neo4j database in batches by using a Cypher statement.
6. The method for optimizing BIM parameters of a child hospital based on multi-objective GAN and LLM update according to claim 5, wherein the third step is specifically as follows: Based on the knowledge graph constructed in the second step, the system establishes standard monitoring service, automatically executes revision detection every week or immediately executes the revision detection when the emergency standard is changed through a manual trigger interface, the revision detection adopts an increment crawling strategy, only obtains a change page from the last examination from the same authority source in the first step, and extracts a change document text through HTML analysis and PDF OCR.
7. The method for optimizing BIM parameters of a child hospital based on multi-objective GAN and LLM update according to claim 6, wherein the LLM performs a difference analysis on the extracted changed document text to generate a structured difference abstract, and after receiving the structured abstract, the output format is JSON Patch, graphRAG, a temporary constraint subgraph is constructed, and is matched with the existing main knowledge graph to identify the affected node set, and then incremental merging is performed: Performing difference analysis on the extracted changed document text by LLM, generating a structured difference abstract, and outputting the structured difference abstract in a JSON Patch format; GraphRAG after receiving the structured difference abstract, constructing a temporary constraint subgraph, carrying out graph matching with the knowledge graph constructed in the second step to identify an affected node set, and then executing incremental merging, wherein the merging process directly acts on the KG model generated in the second step, and only carries out local adjustment on a change part without reconstructing the whole graph: For the newly added standard, creating a new node and creating a corresponding constraint edge; For the change of the threshold value, directly updating the attribute of the corresponding standard node; for clause deletion, a soft deletion strategy is adopted; the update operation adopts transactional write operation to ensure KG consistency; The updating operation directly inherits the layered KG structure constructed in the second step, only modifies the affected subgraph, keeps the integrity of the original topological relation and the position analysis result, and dynamically adjusts the optimization weight of the system after the synchronization is completed; The method comprises the steps of recording operation time, change type, operators, front and back version snapshots, influence range and version numbers by a change log, supporting audit and one-key rollback, broadcasting a synchronous event through an event driving mechanism, subscribing the event by a downstream GAN training module, refreshing a local constraint cache in real time, and ensuring that the latest constraint is used in an optimization process; GraphRAG utilizes the Cypher query language of Neo4j to perform efficient sub-graph queries and optimizations.
8. The method for optimizing BIM parameters of a child hospital based on multi-objective GAN and LLM update according to claim 7, wherein the fourth step is specifically: based on the optimization constraint condition after the dynamic updating in the step three, the BIM component parameters of the children hospital are optimized through a multi-target GAN training generator and a discriminator, so that multi-target cooperative balance of barrier-free performance, functionality and energy consumption is realized, wherein the GAN training basic formula is as follows: (9) Wherein, the The device for generating the electric field comprises a generator, In order for the arbiter to be a function of the arbiter, For the input of noise, The method comprises the steps of obtaining real component parameters, outputting optimization parameters by a generator, evaluating multi-objective compliance by a discriminator, and adopting a weighted multi-objective loss function for realizing multi-objective optimization: (10) Wherein, the Based mainly on the slope of the ramp and the door calculating the deviation of the minimum width; the method is mainly based on the deviation calculation of the lighting coefficient and the space utilization rate; calculating, for energy loss, based on a deviation of ventilation and illumination energy consumption; the dynamic weight is adjusted in real time according to the KG subgraph query result updated in the step three; The training process adopts WASSERSTEIN GAN variants to improve stability, namely a generator G receives a noise vector z and current component parameters and outputs a candidate optimization parameter scheme; After the optimization is completed, the system performs engineering implementation filtering on the generated scheme, the qualified scheme is used as final output, a BIM model is written in, and the optimized parameters are fed back to KG to form closed loop iteration.

Description

BIM parameter optimization method for child hospital based on multi-objective GAN and LLM updating Technical Field The invention relates to the technical field of artificial intelligence and Building Information Models (BIM), in particular to a BIM parameter optimization method for a child hospital based on Multi-objective generation countermeasure GENERATIVE ADVERSARIAL Network (GAN) and large-scale language model (Large Language Model, LLM) updating. Background As a special medical building for infants and disabled children, the BIM design of the hospital meets strict barrier-free specifications and simultaneously meets the requirements of child friendly functions and the aim of green energy conservation. Traditional parameter optimization mainly relies on designer to carry out manual iteration adjustment through BIM software, for example guarantees first that ramp slope AND gate width reaches standard, then manual regulation window wall compares in order to promote lighting coefficient, verifies space utilization through energy consumption simulation software at last. The serial and manual leading flow is long in time consumption, and dynamic balance among multiple targets is difficult to realize, for example, increasing the window area can promote lighting but increase energy consumption, reducing gradient to meet wheelchair traffic, and possibly compressing corridor width to influence fire evacuation and streamline efficiency. In the prior art, a multi-objective generation countermeasure network (GAN) has been applied in the field of BIM parameter optimization, for example, generating a facade scheme meeting lighting standards by using a conditional GAN, or optimizing building volume by using WASSERSTEIN GAN to reduce energy consumption. However, the method has the following general defects that firstly, an optimization target is single or weight is fixed, the multi-dimensional and dynamic change constraint requirements of the child hospital are difficult to adapt, secondly, the standard input is a static threshold value, regular revisions of the standard cannot be met, thirdly, the generated result lacks closed loop feedback with real component data, and the optimization scheme is difficult to directly implement engineering. In recent years, a Large Language Model (LLM) is applied in the aspects of standard analysis and knowledge dynamic update, for example, threshold values are extracted from unstructured canonical documents in real time through GraphRAG technology, and a knowledge graph is constructed, but the existing research is limited to a standard management level, and a collaborative optimization closed loop is not formed with GAN yet, so that the method is applied to intelligent generation of BIM parameters of a child hospital. The traditional optimization method also faces special challenges in the scene of the children hospital that sufficient lighting needs to be ensured in a game area, continuous and barrier-free wheelchair channels need to be ensured, the complex constraints are mutually coupled, and the comprehensive consideration of manpower is difficult. The existing GAN tool can generate a local optimal solution, but cannot sense the relation between standard change and space topology, so that an optimization result is disjointed with an actual engineering. In summary, how to construct a BIM parameter optimization method which can achieve multi-objective collaborative optimization, standard real-time updating, intelligent reasoning of component relation and deep adaptation to the scene of the child hospital becomes a key problem in the development of the current technology. Disclosure of Invention In order to overcome the technical problems, the invention aims to provide a BIM parameter optimization method for a child hospital based on multi-objective GAN and LLM updating, which realizes collaborative optimization of parameters such as slope gradient, door width, lighting coefficient, space utilization rate and the like through a multi-objective generation countermeasure network (GAN), and utilizes LLM combined GraphRAG technology to update standard constraint and component relation in a Knowledge Graph (KG) in real time, so as to ensure that an optimization process is suitable for the latest standard, and solve the defects of the traditional method in multi-objective balance, standard dynamic and engineering implementation. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: A BIM parameter optimization method for a child hospital based on multi-objective GAN and LLM updating comprises the following steps: extracting the latest standard from a weight data source by adopting LLM (logical web management) combined GraphRAG technology, and combining self-collected component data (including parameters of components such as a ramp, a door and the like, such as gradient, width, lighting coefficient and position information) of the components