A generative intelligent planning method and system for hospital department layout

By generating hospital layout condition codes and constructing hard constraint maps, and calculating department constraint residual vectors, the problem of hard constraints being difficult to perceive in real time during the hospital department layout generation process in existing technologies is solved, realizing efficient and complete digital generation of hospital department layouts.

CN122595422BActive Publication Date: 2026-11-10THE FIRST COMPARY OF CHINA EIGHTH ENG BUREAU LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202610665793.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-05-14
Publication Date
2026-11-10
Estimated Expiration
2046-05-14

AI Technical Summary

Technical Problem

Existing methods for generating hospital department layouts struggle to detect and locate violations of hard constraints in real time during the generation process, resulting in a lack of calculable, traceable, and closed-loop data processing relationships between generative layout models and hospital spatial constraint data.

Method used

By acquiring digital data of hospital layout, generating hospital layout condition codes, constructing a hard constraint map of hospital departments, calculating the constraint residual vector of hospital departments, and inputting it into a generative layout model, iterative correction of candidate layouts can be achieved.

Benefits of technology

It enables unified digital representation of multi-source heterogeneous data on hospital department layout, improves data retrieval and computability, and enhances constraint awareness and the completeness of digital layout output during the generation process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122595422B_ABST
    Figure CN122595422B_ABST
Patent Text Reader

Abstract

This invention discloses a generative intelligent planning method and system for hospital department layout, relating to the field of hospital department layout data processing technology. The method includes acquiring digital data of hospital layout and generating hospital layout condition codes. Based on the hospital layout condition codes, a hard constraint graph of hospital departments is constructed, and hospital department constraint residual vectors are calculated. The hospital department constraint residual vectors are input into a generative layout model, and a digital layout scheme for hospital departments is output. This invention, through closed-loop processing of hospital layout condition codes, hospital department hard constraint graphs, hospital department constraint residual vectors, and the generative layout model, transforms hospital department layout generation into a data processing flow that is coded, traversable, feedback-enabled, and iterative, achieving beneficial effects such as improved consistency of layout generation constraints, data traceability, and the output capability of digital schemes.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of hospital department layout data processing technology, specifically to a generative intelligent planning method and system for hospital department layout. Background Technology

[0002] With the development of computer-aided architectural design, spatial data modeling, generative models, and multi-source constraint computation technologies, hospital department layout planning is gradually shifting from manual experience-based configuration to intelligent generation based on electronic digital data processing. Existing technologies typically convert hospital spatial objects, department attributes, workflow directions, and constraints into matrix, vector, or graph structures by reading department lists, architectural space files, medical process records, and standard parameters. Then, optimization algorithms or generative models are used to output candidate layout schemes. This type of technology already has a certain foundation in data structuring, spatial coordinate transformation, path calculation, and candidate scheme generation.

[0003] However, existing methods for generating or optimizing hospital department layouts often treat constraints such as building boundaries, departmental adjacency, medical process distances, clean / contaminated flow separation, and infection control as post-evaluation indicators or screening conditions, lacking a unified constraint feedback mechanism for the generation process. Existing methods can typically generate candidate layouts or calculate indicators such as spatial distance, circulation line intersection, and adjacency relationships separately, but they struggle to encode multiple types of hard constraints into a residual data structure that can be directly read by the generative model. When candidate layouts exhibit boundary violations, overlaps, excessive medical process distances, clean / contaminated flow line intersections, or insufficient infection control, existing methods often require re-screening or manual adjustments, failing to feed back the specific violated departmental nodes, constraint edges, constraint types, and violation amounts to the next round of generation conditions. Therefore, existing technologies struggle to achieve real-time perception, location, and iterative correction of hard constraints in hospital department layouts during the generation phase, resulting in a lack of calculable, traceable, and closed-loop data processing relationships between the generative layout model and hospital spatial constraint data. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that existing hospital department layout data processing methods have the following problems: it is difficult to uniformly encode multi-source hospital layout data; it is difficult to convert hard constraints of hospital space into a feedback residual data structure; it is difficult for generative layout models to locate and correct constraint violations during the generation process; and it is also difficult to drive generative layout models to iteratively output digital layout schemes for hospital departments based on hospital department constraint residual vectors.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a generative intelligent planning method for hospital department layout, comprising acquiring digital data of hospital layout and generating hospital layout condition codes.

[0007] Based on the hospital layout condition coding, a hard constraint map of hospital departments is constructed, and the constraint residual vector of hospital departments is calculated.

[0008] Input the constraint residual vectors of hospital departments into the generative layout model, and output the digital layout scheme of hospital departments.

[0009] As a preferred embodiment of the hospital department layout generative intelligent planning method described in this invention, the acquisition of hospital layout digital data includes: extracting fields, converting formats, and validating data from a list of hospital departments, building space files, medical process records, and hospital layout specification files to form department function data, building space data, medical process data, and hospital layout specification data, respectively, thus obtaining hospital layout digital data. The hospital department list extracts department identifiers, department area requirements, department function categories, cleanliness / contamination levels, infection risk levels, and equipment dependency levels. The building space files extract building boundaries, floor numbers, column grid locations, corridor locations, staircase locations, elevator locations, and entrance / exit locations. The medical process records extract flow direction data for patient flow, medical staff flow, material flow, and waste flow. The hospital layout specification files extract minimum department area, minimum passage width, maximum medical process distance, minimum infection isolation distance, and clean / contamination flow line constraints.

[0010] As a preferred embodiment of the hospital department layout generative intelligent planning method described in this invention, the generation of hospital layout condition codes includes: uniformly mapping, normalizing, categorizing, and aligning the hospital layout digital data to obtain the hospital layout condition codes; mapping the building space data to floor coordinates to obtain a floor space boundary matrix; vectorizing the department function data to obtain a department attribute vector; statistically analyzing the flow direction between departments in the medical process data to obtain a department flow weight matrix; parametrically processing the hospital layout specification data to obtain a constraint parameter vector; and storing the floor space boundary matrix, department attribute vector, department flow weight matrix, and constraint parameter vector according to department identifiers and floor numbers.

[0011] As a preferred embodiment of the hospital department layout generative intelligent planning method described in this invention, the construction of a hospital department hard constraint map based on hospital layout condition coding includes: generating nodes, identifying edge relationships, and labeling constraint types for the department attribute vectors, floor space boundary matrices, inter-department flow weight matrices, and constraint parameter vectors in the hospital layout condition coding to obtain the hospital department hard constraint map. Each department is generated as a department node, and building boundaries, floor boundaries, corridors, elevators, and staircases are generated as building space constraint nodes. When there are adjacency requirements, medical process distance requirements, infection isolation requirements, clean and contaminated separation requirements, or flow passage requirements between two departments, constraint edges are generated between the corresponding department nodes. When there are boundary crossing, passage, or evacuation restrictions between a department node and a building space constraint node, constraint edges are generated between the department node and the building space constraint node, and constraint types and constraint weights are written for each constraint edge.

[0012] As a preferred embodiment of the generative intelligent planning method for hospital department layout described in this invention, the calculation of the hospital department constraint residual vector includes: performing constraint traversal, violation calculation, and residual encoding on the candidate hospital department layout and the hospital department hard constraint map to obtain the hospital department constraint residual vector; performing spatial inclusion judgment on the department boundary and building boundary to obtain the building boundary residual; performing intersection judgment on different department spatial areas to obtain the department overlap residual; comparing the traversable path distance between department nodes with medical process distance requirements with the maximum medical process distance to obtain the medical process distance residual; performing path intersection judgment on clean flow lines and waste flow lines to obtain the clean and waste flow line intersection residual; and comparing the distance between infection risk departments and ordinary departments with the minimum infection isolation distance to obtain the infection isolation residual.

[0013] As a preferred embodiment of the generative intelligent planning method for hospital department layout described in this invention, the step of inputting the hospital department constraint residual vector into the generative layout model includes: inputting hospital layout condition codes, previous round candidate hospital department layouts, and hospital department constraint residual vectors as condition inputs, generating layout states, and implementing residual feedback control to obtain the next round of candidate hospital department layouts. The hospital layout condition codes are read through the condition code input layer; candidate hospital department layouts including department identifiers, planar coordinates, floor numbers, department widths, and department heights are generated through the layout state generation layer; and hospital department constraint residual vectors are read through the residual feedback control layer. When any type of residual exceeds the corresponding hospital department layout residual control limit, the corresponding department node, constraint edge, and constraint type are written into the next round of generation conditions. When the weighted cumulative value of all types of residuals is less than the hospital department layout hard constraint energy threshold, the generation iteration stops.

[0014] As a preferred embodiment of the generative intelligent planning method for hospital department layout described in this invention, the output digital layout scheme for hospital departments includes: encapsulating, associating, and recording the residuals of candidate hospital department layouts when the generative layout model stops generating iterations, to obtain the digital layout scheme for hospital departments. This involves: encapsulating the planar positions of departments in the candidate hospital department layouts to obtain department coordinate data; encapsulating the floor numbers in the candidate hospital department layouts to obtain department floor data; encapsulating the width, height, and boundary range of departments in the candidate hospital department layouts to obtain department boundary data; associating the adjacency relationships between departments to obtain department adjacency data; associating the paths of patient flow, medical staff flow, material flow, and waste flow in the medical process data to obtain medical process path data; and recording the constraint residual vectors of hospital departments at the time of output to obtain constraint residual record data.

[0015] As a preferred embodiment of the hospital department layout generative intelligent planning system described in this invention, it includes a layout data encoding module, a hard constraint residual calculation module, and a residual feedback generation module.

[0016] The layout data encoding module is used to acquire hospital layout digital data and generate hospital layout condition codes.

[0017] The hard constraint residual calculation module is used to construct a hard constraint map of hospital departments based on the hospital layout condition coding, and to calculate the constraint residual vector of hospital departments.

[0018] The residual feedback generation module is used to input the constraint residual vectors of hospital departments into the generative layout model and output the digital layout scheme of hospital departments.

[0019] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program as steps to implement a generative intelligent planning method for hospital department layout.

[0020] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a generative intelligent planning method for hospital department layout.

[0021] The beneficial effects of this invention are: By extracting fields, converting formats, validating data, unifying coordinates, normalizing values, and aligning indexes from hospital department lists, architectural space files, medical process records, and hospital layout specifications, a unified digital representation of multi-source, heterogeneous hospital layout data was achieved. Text, tables, CAD / BIM spatial files, and process records were transformed into computer-readable floor space boundary matrices, department attribute vectors, inter-department flow weight matrices, and constraint parameter vectors. This provides a consistent data entry point for subsequent map construction and residual calculation, effectively improving data retrieval, computability, and cross-data source correlation stability.

[0022] By generating department nodes, building space constraint nodes, and constraint edges based on hospital layout condition coding, and performing spatial inclusion, region intersection, path distance, circulation line intersection, and isolation distance calculations on candidate hospital department layouts, a graph-structured and residual representation of hard constraints for hospital layouts was achieved. Constraints such as building boundaries, department overlap, medical process distance, clean / dirty circulation line intersection, and infection isolation are converted into hospital department constraint residual vectors and residual source index tables. This allows different types of spatial constraints to be uniformly traversed, located, and encoded, achieving the beneficial effect of transforming hard constraints from post-judgment conditions into feedback-able digital data.

[0023] By inputting hospital department constraint residual vectors, residual source index tables, hospital layout condition codes, and previous round candidate hospital department layouts into a generative layout model, layout state generation and iterative control based on residual feedback are achieved. This enables the generative model to adjust the location sampling range of department nodes, access grid sharing relationships, and inter-department exclusion relationships according to different residual types. After satisfying residual control limits and hard constraint energy thresholds, the model encapsulates and outputs department coordinates, floors, boundaries, adjacencies, paths, and residual record data, achieving the beneficial effects of improving constraint awareness during the generation process and the completeness of the digital layout output. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 The above is an overall flowchart of a generative intelligent planning method for hospital department layout provided in Embodiment 1 of the present invention. Detailed Implementation

[0026] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0027] Example 1, referring to Figure 1 As an embodiment of the present invention, a generative intelligent planning method for hospital department layout is provided, comprising: S1: Obtain hospital layout digital data and generate hospital layout condition codes.

[0028] The hospital department list, architectural space documents, medical process records, and hospital layout specification documents undergo field extraction, format conversion, and data validation to generate departmental function data, architectural space data, medical process data, and hospital layout specification data, resulting in digital data for the hospital layout. The hospital department list extracts department identifiers, departmental area requirements, departmental function categories, cleanliness / contamination levels, infection risk levels, and equipment dependency levels. The architectural space documents extract building boundaries, floor numbers, column grid locations, corridor locations, staircase locations, elevator locations, and entrance / exit locations. The medical process records extract flow direction data for patient flow, medical staff flow, material flow, and wastewater flow. The hospital layout specification documents extract minimum departmental area, minimum corridor width, maximum medical process distance, minimum infection isolation distance, and clean / contamination flow line constraints.

[0029] Furthermore, the hospital department list can be a structured spreadsheet, database record, or a department configuration file exported from the hospital information system. The computer reads the department name, department number, department category, and department area requirement according to a preset field mapping table, and merges different names for the same department into the same department identifier. When multiple records exist for the same department, deduplication is performed using the department identifier as the primary key. If the difference in area requirement corresponding to the same department identifier exceeds 5 square meters, the record is marked as conflicting data and written to the conflicting data cache. If the difference is less than 5 square meters, the larger area requirement value is taken as the department area requirement.

[0030] Furthermore, the architectural space files can be CAD files, BIM exported files, vector graphics files, or spatial coordinate tables. The computer reads the wall boundary lines, floor boundary lines, column grid points, door opening locations, corridor center lines, elevator shaft locations, stairwell locations, and entrance / exit locations, and converts the reading results into two-dimensional plane coordinate data and floor number data. The building boundary is represented by a closed polygon. If the distance between the first and last endpoints of a closed polygon is greater than 0.01 meters, the building boundary is determined to not meet the geometric closure check, and the corresponding boundary segment is marked as data to be repaired.

[0031] Furthermore, medical process records include at least one of the following: registration records, examination records, department transfer records, medical order transfer records, logistics and distribution records, or waste transfer records. The computer extracts flow data based on the occurrence time, originating department, target department, and type of transfer object in the records. Patient flow, medical staff flow, logistics, and waste flow are distinguished by the type of transfer object, with 15 minutes as the minimum statistical time granularity. When multiple records exist between the same originating department and target department within the same time granularity, their transfer counts are accumulated to form the basis data for subsequent inter-departmental flow statistics.

[0032] Furthermore, the hospital layout specification document includes spatial constraint parameters stored in the form of tables, rule text, or database fields. The computer extracts minimum department area, minimum corridor width, maximum medical process distance, minimum infection isolation distance, and clean / dirty flow restriction data through keyword matching and field mapping. Specifically, the minimum department area is based on the lower limit of the corresponding department category in the hospital layout specification document. The minimum corridor width is set to 1.8 meters by default, used for width verification of the passable grid in the subsequent floor space boundary matrix. The maximum medical process distance is set to 80 meters by default, and the minimum infection isolation distance is set to 30 meters by default. The default value for the minimum infection isolation distance is preset based on the spatial separation calculation requirements between infection-risk departments and general departments. When the hospital layout specification document provides different values, the value in the hospital layout specification document overrides the default value. The default value for the maximum medical process distance is preset based on the upper limit of the passable path between high-frequency medical process departments such as emergency, imaging, laboratory, and pharmacy.

[0033] Furthermore, data validation includes field integrity validation, numerical range validation, primary key uniqueness validation, and spatial topology validation. Among these, department identifier, floor number, building boundary, department area requirement, starting department, and target department are mandatory fields, requiring a 100% integrity rate. The integrity rate of non-mandatory fields must be no less than 95%; if it falls below 95%, a list of missing fields is generated. Numerical range validation determines whether the department area requirement, corridor width, and distance parameters are positive numbers. Primary key uniqueness validation checks for duplicate department identifiers. Spatial topology validation determines whether building boundaries are closed and whether building boundaries under the same floor number self-intersect.

[0034] The hospital layout digital data is uniformly coordinated, numerically normalized, categorized, and indexed for alignment to obtain hospital layout condition codes. Floor coordinate mapping is performed on the building space data to obtain floor space boundary matrices. Departmental function data is vectorized to obtain departmental attribute vectors. Inter-departmental flow direction statistics are performed on the medical process data to obtain inter-departmental flow weight matrices. Based on the cumulative number of patient, medical staff, material, and waste flow flows originating from or targeting each department, the corresponding department's utilization intensity is generated. Hospital layout specification data is parameterized to obtain constraint parameter vectors. The floor space boundary matrix, departmental attribute vectors, inter-departmental flow weight matrices, and constraint parameter vectors are stored according to department identifiers and floor numbers.

[0035] Furthermore, unified coordinate system refers to converting architectural spatial data from different sources to the same hospital building planar coordinate system. The computer uses the lower left corner of the building boundary of each floor as the local coordinate origin for that floor, with meters as the unified unit of length, and retains three decimal places. For millimeters, centimeters, or feet in CAD files or BIM exported files, the units are first converted according to the unit identifier in the file metadata, and then the floor number is written into the coordinate data, so that the same planar coordinates can be distinguished, stored, and retrieved on different floors.

[0036] Furthermore, numerical normalization refers to converting departmental area requirements, equipment dependency levels, infection risk levels, cleanliness / contamination levels, flow frequency, and distance parameters into computer-comparable numerical ranges. Specifically, departmental area requirements and distance parameters are normalized to between 0 and 1 based on the minimum and maximum values ​​of the corresponding fields within the same item. Infection risk levels, cleanliness / contamination levels, and equipment dependency levels are converted into integer codes using a level mapping table generated from the level fields in the hospital layout specification document, and then normalized. When the maximum and minimum values ​​of a field are equal, the normalized value of that field is set to 0.5 to avoid generating invalid codes.

[0037] Furthermore, category embedding refers to converting departmental function categories, cleanliness / contamination levels, infection risk levels, and equipment dependency levels into fixed-length numerical vectors. The computer assigns a category number to each departmental function and maps this number to a 16-dimensional category vector. The cleanliness / contamination level, infection risk level, and equipment dependency level are each mapped to a 4-dimensional level vector. These are then concatenated with the normalized departmental area requirements, departmental usage intensity, and departmental identifier index to obtain the departmental attribute vector. This attribute vector is stored using the departmental identifier as an index and is used as part of the hospital layout condition coding for subsequent steps. Departmental usage intensity is determined by the normalized cumulative number of flows of patients, medical staff, materials, and wastewater originating from or targeting that department.

[0038] Furthermore, index alignment refers to the computer using department identifiers and floor numbers as a combined index to write the floor space boundary matrix, department attribute vector, inter-department flow weight matrix, and constraint parameter vector into the same coded data table. Specifically, the floor space boundary matrix records the grid occupancy status of the layoutable areas on each floor; the inter-department flow weight matrix records the intensity of patient flow, medical staff flow, material flow, and waste flow between any two departments; and the constraint parameter vector records the standard parameters of the corresponding department or corresponding department pair. When a department lacks a floor number, its floor number is marked as null, and the department is added to the set of floors to be assigned, temporarily excluding it from floor boundary matrix matching.

[0039] It should be noted that this step transforms hospital departments, building spaces, medical processes, and regulatory constraints into a unified hospital layout condition code through field extraction, format conversion, coordinate unification, numerical normalization, and index alignment. This design enables heterogeneous data to form a computable, searchable, and associative digital structure, providing stable input for subsequent hard constraint map construction and residual calculation.

[0040] S2: Construct a hard constraint map of hospital departments based on hospital layout condition coding, and calculate the constraint residual vector of hospital departments.

[0041] The hospital layout condition coding process involves node generation, edge relationship identification, and constraint type labeling for the department attribute vectors, floor space boundary matrices, inter-department circulation weight matrices, and constraint parameter vectors, resulting in a hard constraint map of hospital departments. Each department is generated as a department node, and building boundaries, floor boundaries, corridors, elevators, and staircases are generated as building space constraint nodes. When there are adjacency requirements, medical process distance requirements, infection isolation requirements, clean / contaminated separation requirements, or circulation requirements between two departments, constraint edges are generated between the corresponding department nodes. When there are boundary crossing, passage, or evacuation restrictions between a department node and a building space constraint node, constraint edges are generated between the department node and the building space constraint node, and constraint types and weights are written for each constraint edge.

[0042] Furthermore, node generation involves using the department identifier from the department attribute vector as the primary key of the department node, and writing the department area requirement, department function category, cleanliness / contamination level, infection risk level, and equipment dependency level into the department node attributes. The building boundary, floor boundary, corridor, elevator, and staircase in the floor space boundary matrix are used as building space constraint nodes, and the corresponding floor number, planar coordinate range, access status, and space type are written into the building space constraint node attributes. The hospital department hard constraint graph is stored using both an adjacency list and a constraint matrix. The adjacency list records the constraint edges between nodes, and the constraint matrix records the constraint type and constraint weight.

[0043] Furthermore, during edge relationship identification, the computer uses the inter-departmental streamline weight matrix and constraint parameter vector to determine whether the nodes between departments meet the edge formation conditions. When there is a clear adjacency requirement, medical process distance requirement, infection isolation requirement, or clean / contaminated separation requirement between two department nodes, a constraint edge is directly generated. When only streamline passage requirements exist, if the normalized streamline intensity between two department nodes is not less than 0.2, a constraint edge is generated. 0.2 serves as the hospital department streamline edge formation threshold, derived from normalizing the streamline intensity within a 15-minute statistical granularity and identifying department pairs with streamline intensity exceeding 20% ​​of the maximum streamline intensity for the entire project as the streamline department pairs required to participate in layout constraint calculations.

[0044] Furthermore, the constraint weights are calculated jointly based on the constraint type indicator value and streamline intensity, and are used to distinguish the calculation priority of different constraint edges in the hard constraint map of hospital departments. The calculation of the constraint weights is expressed as follows:

[0045] in, Indicates the first Each department node and the first Constraint weights between departmental nodes. This indicates the adjacency requirement indicator value; it is 1 if an adjacency requirement exists, and 0 otherwise. This indicates the required distance for medical processes; it is set to 1 if a medical process distance requirement exists, and 0 otherwise. The normalized streamline intensity is obtained by normalizing the cumulative intensity of patient flow, medical staff flow, material flow, and waste flow in the interdepartmental streamline weight matrix, and its value ranges from 0 to 1. This indicates the infection isolation requirement; it is set to 1 if an infection isolation requirement exists, and 0 otherwise. This represents the indicator value for clean and contaminated area separation requirements. It is set to 1 if such requirements exist, and 0 otherwise. 0.2, 0.25, 0.2, 0.2, and 0.15 are the default weighting coefficients for adjacency requirements, medical process distance requirements, flow path requirements, infection isolation requirements, and clean and contaminated area separation requirements, respectively. The sum of the default weighting coefficients is 1. When weight configuration is provided in the hospital layout specification data, the weight configuration in the hospital layout specification data replaces the default weighting coefficients.

[0046] Furthermore, the formula for calculating constraint weights is applied to the constraint edge writing stage of the hospital department hard constraint graph. When the computer identifies multiple constraint sources between two department nodes, it uses the above constraint weight calculation method to synthesize the constraint weights of different constraint sources into the constraint weights of the same constraint edge, and writes these constraint weights and constraint types together into the hospital department hard constraint graph for subsequent constraint traversal and residual encoding.

[0047] Furthermore, constraint type labeling includes labeling constraint edges as adjacency constraint edges, medical process distance constraint edges, infection isolation constraint edges, clean and contaminated separation constraint edges, circulation constraint edges, building boundary constraint edges, access constraint edges, or evacuation constraint edges. When the same constraint edge satisfies multiple constraint types simultaneously, multiple constraint type labels are written into the same constraint edge, arranged in descending order of constraint weight. When a constraint edge connects a department node and a building space constraint node, the computer writes boundary crossing, access, or evacuation restriction labels based on the space type of the building space constraint node, used for subsequent determination of whether the candidate hospital department layout exceeds the permissible layout area or occupies an inaccessible area.

[0048] The candidate hospital department layout and hard constraint map are subjected to constraint traversal, violation calculation, and residual encoding to obtain the hospital department constraint residual vector. Spatial inclusion judgment is performed on department boundaries and building boundaries to obtain building boundary residuals. Intersection judgment is performed on spatial regions of different departments to obtain department overlap residuals. The traversable path distance between department nodes with medical process distance requirements is compared with the maximum medical process distance to obtain medical process distance residuals. Path intersection judgment is performed on clean flow lines and waste flow lines to obtain clean-waste flow line intersection residuals. The distance between infection-risk departments and general departments is compared with the minimum infection isolation distance to obtain infection isolation residuals.

[0049] Furthermore, the candidate hospital department layout includes the planar coordinates, floor number, department width, department height, and department boundary range for each department node. During the initial calculation of the hospital department constraint residual vector, the candidate hospital department layout is generated by the computer based on the department area requirements and floor space boundary matrix in the hospital layout condition encoding, forming an initial grid layout. In subsequent calculations, the candidate hospital department layout is the one output by the previous round of generative layout model. The computer traverses the constraint edges in the hospital department hard constraint graph sequentially, first reading the constraint type, and then calling the corresponding spatial judgment, path search, or flow line intersection judgment process. The initial grid layout allocates continuous grid areas sequentially within the deployable grids in the floor space boundary matrix according to the department area requirements from largest to smallest, and converts the allocated continuous grid areas into department boundary ranges.

[0050] Furthermore, the accessible path distance is calculated based on the floor space boundary matrix. The computer marks corridors, elevators, stairs, and entrances / exits as accessible grids, and areas occupied by walls, column grids, and areas that do not meet the minimum passage width as inaccessible grids. The minimum passage width is set to 1.8 meters by default, derived from the parameterized processing of hospital layout specification data in S1. When the hospital layout specification document provides different values, the value in the hospital layout specification document replaces 1.8 meters. The maximum medical process distance in the medical process distance residual is set to 80 meters by default, and the minimum infection isolation distance in the infection isolation residual is set to 30 meters by default. Both of these distance thresholds are derived from the constraint parameter vector in S1 and are overridden by the corresponding department's threshold when the hospital layout specification document provides thresholds for the corresponding departments.

[0051] Furthermore, residual coding is used to convert various constraint violations in the candidate hospital department layout into hospital department constraint residual vectors of the same dimension. Specifically, the building boundary residual is determined by the ratio of the total area of ​​all departments exceeding the building boundary to the total area of ​​all departments. The department overlap residual is determined by the ratio of the total overlapping area between different departments to the total area of ​​all departments. The medical process distance residual is determined by the excess distance of department node pairs with medical process distance requirements; the excess distance is the portion of the traversable path distance exceeding the maximum medical process distance, and is averaged according to the number of department node pairs with medical process distance requirements. The clean and dirty flow line intersection residual is determined by the number of intersections between clean and dirty flow lines and the intersection risk coefficient. The intersection risk coefficient is 1 when the intersection point is located in a regular corridor, and 2 when the intersection point is located in a waiting area, entrance / exit, or elevator lobby. The intersection risk coefficient is determined based on the spatial type of the building space constraint node where the intersection point is located. The infection isolation residual is determined based on the insufficient distance between departmental nodes requiring infection isolation. This insufficient distance is the portion where the minimum infection isolation distance exceeds the actual spatial distance, and is averaged according to the number of departmental nodes requiring infection isolation. The actual spatial distance within the same floor is determined by the shortest planar distance between the boundary of a high-risk department and the boundary of a general department. Between different floors, the distance is determined by the composite distance of the vertical spacing between floors and the planar projected distance.

[0052] Furthermore, the hospital department constraint residual vectors are written into the building boundary residuals, department overlap residuals, medical process distance residuals, clean / dirty flow line intersection residuals, and infection isolation residuals in a fixed order. When a certain type of constraint does not have a corresponding constraint edge or path combination, that type of residual is written as 0. When a certain type of residual is greater than the corresponding hospital department layout residual control limit, the computer will generate the department node, constraint edge, constraint type, and violation amount of that residual and write it into the residual source index table. The residual source index table and the hospital department constraint residual vectors are output together to the next step, which is used to enable the generative layout model to read the hospital department layout location, constraint relationship, and constraint type that need to be adjusted. The hospital department layout residual control limit is derived from the constraint parameter vector. When the constraint parameter vector does not provide a corresponding limit, the building boundary residual control limit, department overlap residual control limit, medical process distance residual control limit, clean / dirty flow line intersection residual control limit, and infection isolation residual control limit are all defaulted to 0, which is used to identify any hard constraint violation as a residual source.

[0053] It should be noted that the design concept of this step is to further convert the hospital layout condition coding into a hard constraint graph of hospital departments that can be traversed and calculated by a computer, and to uniformly store department nodes, building space constraint nodes, constraint edges, constraint types, and constraint weights as a graph data structure. Then, spatial inclusion, regional intersection, path distance, circulation line intersection, and isolation distance calculations are performed on candidate hospital department layouts to form hospital department constraint residual vectors of the same dimension. The technical effect is that it no longer judges the quality of a layout solely based on a single objective function or post-selection, but rather converts multiple types of hospital spatial hard constraints into indexable, computable, and feedback-enabled residual data. This allows subsequent generative layout models to read the specific violated department nodes, constraint edges, and constraint types, achieving digital constraint feedback control for the hospital layout generation process.

[0054] S3: Input the constraint residual vectors of hospital departments into the generative layout model and output the digital layout scheme of hospital departments.

[0055] The process involves inputting hospital layout condition codes, previous round candidate hospital department layouts, and hospital department constraint residual vectors, generating layout states, and implementing residual feedback control to obtain the next round of candidate hospital department layouts. The hospital layout condition codes are read through the condition coding input layer. The layout state generation layer generates candidate hospital department layouts including department identifiers, planar coordinates, floor numbers, department widths, and department heights. The hospital department constraint residual vectors are read through the residual feedback control layer. When any type of residual exceeds the corresponding hospital department layout residual control limit, the corresponding department node, constraint edge, and constraint type are written into the next round of generation conditions. The generation iteration stops when the weighted cumulative value of all types of residuals is less than the hospital department layout hard constraint energy threshold.

[0056] Furthermore, during the first generation iteration, the previous round of candidate hospital department layouts is the initial grid layout generated in S2 based on the hospital layout condition encoding. In subsequent generation iterations, the previous round of candidate hospital department layouts is the candidate hospital department layout output by the previous round of generative layout model.

[0057] Furthermore, the generative layout model includes a conditional encoding input layer, a layout state generation layer, a residual feedback control layer, and a layout state output layer. The conditional encoding input layer receives hospital layout conditional codes; the layout state generation layer receives the output of the conditional encoding input layer and the previous round of candidate hospital department layouts; the residual feedback control layer receives hospital department constraint residual vectors and a residual source index table; and the layout state output layer outputs the next round of candidate hospital department layouts. Data index consistency is maintained between layers through department identifiers and floor numbers, ensuring that each type of residual can be mapped to a specific department node and constraint edge.

[0058] Furthermore, the generative layout model employs a conditional generation structure. Training samples include historical hospital department layout data, candidate hospital department layouts generated by perturbing the historical data, corresponding hospital layout condition codes, and corresponding hospital department constraint residual vectors. During training, the hospital layout condition codes and perturbed candidate hospital department layouts are used as inputs, while the department coordinates, floor numbers, department widths, and department heights from the historical layout data are used as supervised outputs. The hospital department constraint residual vectors are input to the residual feedback control layer, enabling the model to learn the layout adjustment directions corresponding to different residual types.

[0059] Furthermore, in the preferred implementation, the generative layout model adopts a conditional Transformer structure, including a conditional encoding input layer, a layout state embedding layer, a multi-head attention layer, a feedforward mapping layer, a residual feedback control layer, and a layout state output layer. Specifically, the hospital layout conditional encoding output from the conditional encoding input layer is concatenated with the layout state embedding result of the previous round of candidate hospital department layouts and then input into the multi-head attention layer. The residual feedback control layer maps the hospital department constraint residual vector and the residual source index table into residual control vectors, which are then input into the layout state output layer along with the output of the multi-head attention layer. During training, the difference between the predicted department plane coordinates, floor number, department width, and department height and their corresponding values ​​in the historical hospital department layout data is used as the layout prediction loss, and the hospital department constraint residual vector corresponding to the predicted layout is used as the residual feedback loss. The weighted sum of these two values ​​is used to update the parameters of the generative layout model.

[0060] Furthermore, the residual feedback control layer reads the department nodes, constraint edges, and constraint types that generate the residuals based on the residual source index table. When the building boundary residual is greater than the hospital department layout residual control limit, the location sampling range of the corresponding department node is restricted to the layoutable area of ​​the floor space boundary matrix. When the department overlap residual is greater than the hospital department layout residual control limit, the generation probability of overlapping department nodes in the same grid area is reduced. When the medical process distance residual is greater than the hospital department layout residual control limit, the generation probability of the corresponding department node in the same floor or near adjacent traffic nodes is increased. When the clean and dirty flow line intersection residual is greater than the hospital department layout residual control limit, the generation probability of clean flow lines and dirty flow lines sharing the same access grid is reduced, and the corresponding building space constraint node is written into the next round of generation conditions. When the infection isolation residual is greater than the hospital department layout residual control limit, the location sampling exclusion range between the corresponding infection risk department node and the ordinary department node is expanded, and the corresponding department node pair is written into the next round of generation conditions.

[0061] Furthermore, the residual control limits for hospital department layout are derived from the residual types of the hospital department constraint residual vectors in S2. When the hospital layout specification data does not provide corresponding limits, the residual control limits for building boundaries, department overlap, medical process distance, clean / dirty flow line intersection, and infection isolation are all set to 0 by default, indicating that the corresponding hard constraints are not allowed to be violated. When the computer performs floating-point calculations, 0.001 is used as the numerical error tolerance; residuals less than 0.001 are treated as 0.

[0062] Furthermore, the hard constraint energy of the hospital department layout is used to determine whether to stop the generation iteration. Its calculation method involves multiplying the building boundary residual, department overlap residual, medical process distance residual, clean / dirty flow line intersection residual, and infection isolation residual by their respective residual weights, and then weighting and accumulating them. Specifically, it is expressed as follows:

[0063] in, Indicates the first The hard constraint energy of the hospital department layout corresponding to the secondary generation iteration. Indicates the first The building boundary residuals corresponding to the secondary generation iteration. Indicates the first The departmental overlap residuals corresponding to the secondary generation iteration. Indicates the first The medical process distance residual corresponding to the secondary generation iteration. Indicates the first The clean and dirty streamline cross residuals corresponding to the secondary generation iteration. Indicates the first The infection isolation residuals corresponding to the next generation iteration. , , , , These represent the weights for building boundary residuals, department overlap residuals, medical process distance residuals, clean / dirty flow line intersection residuals, and infection isolation residuals, respectively, all with a default value of 0.2. When weight configurations are provided in the hospital layout specification data, the default weights will be replaced by the weight configurations in the hospital layout specification data.

[0064] The formula for the hard constraint energy of hospital department layout is applied after each round of the generative layout model outputs the next round of candidate hospital department layouts. The computer calculates the hard constraint energy of the hospital department layout based on this formula. When all residuals are no greater than their corresponding control limits for hospital department layout residuals, and the hard constraint energy is less than 0.01, the candidate hospital department layout is deemed to meet the stopping iteration condition. The 0.01 value originates from the one-hundredth error limit after normalizing the hospital department constraint residual vector. When the hospital layout specification data provides more stringent values, the values ​​in the hospital layout specification data are used as the threshold for the hard constraint energy of the hospital department layout.

[0065] Furthermore, the next round of generation conditions is jointly generated by the hospital layout condition code, the previous round of candidate hospital department layout, the hospital department constraint residual vector, and the residual source index table. Its generation method is expressed as follows:

[0066] in, Indicates the first The next generation condition used in the next generation iteration. This indicates the hospital layout condition code. Indicates the first The candidate hospital department layout is generated through the next iteration. Indicates the first The hospital department constraint residual vector is obtained through the second generation iteration. Indicates the first The residual source index table is obtained from the secondary generation iteration. This indicates a computer data connection operation that splices data according to department identifiers, floor numbers, and constraint edge identifiers.

[0067] The next-round generation condition formula is applied to the residual feedback control layer when generating the next round of input data. The computer uses the hospital layout condition code as a global constraint, the previous round of candidate hospital department layouts as the current layout state, the hospital department constraint residual vectors as residual strength data, and the residual source index table as residual location data. These are connected to form the next-round generation condition, enabling the generative layout model to generate the next round of candidate hospital department layouts for specific department nodes, constraint edges, and constraint types.

[0068] Furthermore, stopping the generation iteration also includes determining the maximum number of iterations. The maximum number of generation iterations is set to 50 by default, derived from the training configuration of the generative layout model on hospital department layout data. When the number of generation iterations reaches 50 and the hard constraint energy of the hospital department layout is still not less than the hard constraint energy threshold, the computer outputs the candidate hospital department layout with the smallest current hard constraint energy as a candidate hospital department layout to be reviewed, and simultaneously outputs an index table of unmet residual sources. The candidate hospital department layout to be reviewed is not considered as a hospital department digital layout scheme that meets the stopping generation iteration condition.

[0069] The candidate hospital department layouts at the point where the generative layout model stops generating iterations are encapsulated, path-associated, and residuals are recorded to obtain digital layout schemes for hospital departments. The planar locations of departments in the candidate hospital department layouts are encapsulated to obtain department coordinate data. The floor numbers in the candidate hospital department layouts are encapsulated to obtain department floor data. The width, height, and boundary range of departments in the candidate hospital department layouts are encapsulated to obtain department boundary data. The adjacency relationships between departments are associated to obtain department adjacency data. The paths of patient flow, medical staff flow, material flow, and waste flow in the medical process data are associated to obtain medical process path data. The constraint residual vectors of hospital departments at the time of output are recorded to obtain constraint residual record data.

[0070] Furthermore, data encapsulation refers to the computer converting the candidate hospital department layouts at the point of stopping the generation iteration into a structured digital layout file. This structured digital layout file includes a department data table, a floor data table, a boundary data table, an adjacency data table, a path data table, and a residual record data table. The department data table uses the department identifier as the primary key and records the department's planar coordinates, floor number, width, height, and functional category. The boundary data table records the coordinate sequence of the boundary vertices for each department. The residual record data table records various residual values, the department nodes from which the residuals originate, and the constraint edges from which the residuals originate.

[0071] Furthermore, path association refers to the computer recalculating the flow paths of patients, medical staff, materials, and waste based on the floor space boundary matrix, between department coordinate data and building space constraint nodes, and binding the flow paths with the corresponding starting department identifier, target department identifier, flow type, and floor number. When a path needs to cross floors, the building space constraint nodes corresponding to elevators or stairs are written into the path data table. When multiple flow types exist within the same department, the flow path corresponding to each flow type is recorded separately.

[0072] Furthermore, the data format for the hospital department digital layout scheme includes tabular data format, JSON data format, XML data format, or spatial data format readable by CAD / BIM systems. Before output, the computer performs consistency checks on department coordinate data, department floor data, department boundary data, department adjacency data, medical process path data, and constraint residual record data. If any department identifier is missing in any data table, that department identifier is written to the output exception record. When the consistency check passes, the hospital department digital layout scheme is output.

[0073] It should be noted that the design concept of this step is to use the hospital department constraint residual vector obtained in S2 as the feedback control input of the generative layout model. This allows the model to simultaneously read the hospital layout condition code, the previous round's candidate hospital department layout, and the residual source index table during each generation round. Based on different residual types, it adjusts the sampling range of department node positions, the candidate floor range, the access grid occupancy relationship, and the inter-departmental exclusion relationship. The technical advantage lies in eliminating the post-processing method of generating candidate schemes followed by selection, as in existing technologies. Instead, it transforms hard constraint violations into calculable, locatable, and iteratively callable digital feedback signals. This enables the generative model to continuously correct the hospital department layout data during the generation process and output a structured digital layout scheme containing coordinates, boundaries, adjacency, paths, and residual records.

[0074] Example 2, an embodiment of the present invention, provides a hospital department layout generative intelligent planning system, including a layout data encoding module, a hard constraint residual calculation module, and a residual feedback generation module.

[0075] The layout data encoding module is used to acquire digital data of the hospital layout and generate hospital layout condition codes.

[0076] The hard constraint residual calculation module is used to construct a hard constraint map of hospital departments based on the hospital layout condition coding, and to calculate the constraint residual vector of hospital departments.

[0077] The residual feedback generation module is used to input the constraint residual vectors of hospital departments into the generative layout model and output the digital layout scheme of hospital departments.

Claims

1. A generative intelligent planning method for hospital department layout, characterized in that, include: Obtain digital data on hospital layout and generate hospital layout condition codes; Based on the hospital layout condition coding, a hard constraint map of hospital departments is constructed, and the constraint residual vector of hospital departments is calculated. Input the constraint residual vectors of hospital departments into the generative layout model, and output the digital layout scheme of hospital departments; Constructing a hard constraint map of hospital departments based on hospital layout condition coding involves generating nodes, identifying edge relationships, and labeling constraint types for the department attribute vectors, floor space boundary matrices, inter-department flow weight matrices, and constraint parameter vectors in the hospital layout condition coding, thereby obtaining a hard constraint map of hospital departments. Each department is generated as a department node, and building boundaries, floor boundaries, corridors, elevators and stairs are generated as building space constraint nodes; When there are adjacency requirements, medical process distance requirements, infection isolation requirements, clean and contaminated separation requirements, or flow passage requirements between two departments, constraint edges are generated between the corresponding department nodes. When there are boundary crossing, passage or evacuation restrictions between department nodes and building space constraint nodes, constraint edges are generated between department nodes and building space constraint nodes, and constraint type and constraint weight are written for each constraint edge. The calculation of hospital department constraint residual vectors includes constrained traversal, violation calculation and residual encoding of candidate hospital department layout and hospital department hard constraint map to obtain hospital department constraint residual vectors; Spatial inclusion determination is performed on the department boundary and the building boundary to obtain the building boundary residual; Intersection of different departmental spatial areas is determined to obtain departmental overlap residuals. The traversable path distance between departmental nodes with medical process distance requirements is compared with the maximum medical process distance to obtain the medical process distance residual. The path intersection of the clean flow line and the dirty flow line is determined to obtain the clean-dirty flow line intersection residual. The infection isolation residual is obtained by comparing the distance between departments at risk of infection and general departments with the minimum infection isolation distance; The process of inputting the hospital department constraint residual vector into the generative layout model includes encoding the hospital layout conditions, inputting the previous round of candidate hospital department layouts and hospital department constraint residual vectors, generating the layout state and controlling the residual feedback to obtain the next round of candidate hospital department layouts. The hospital layout condition code is read through the condition coding input layer, and the candidate hospital department layout including department identifier, plane coordinates, floor number, department width and department height is generated through the layout state generation layer. The hospital department constraint residual vector is read through the residual feedback control layer. When any type of residual exceeds the corresponding hospital department layout residual control limit, the department node, constraint edge, and constraint type corresponding to the residual will be written into the next round of generation conditions. When the weighted cumulative value of various residuals is less than the hard constraint energy threshold of the hospital department layout, the generation iteration stops.

2. The hospital department layout generative intelligent planning method as described in claim 1, characterized in that: The acquisition of hospital layout digital data includes, Field extraction, format conversion, and data verification are performed on the hospital department list, building space documents, medical process records, and hospital layout specification documents to generate department function data, building space data, medical process data, and hospital layout specification data, respectively, resulting in hospital layout digital data. Extract the department identifier, department area requirement, department function category, cleanliness / contamination level, infection risk level, and equipment dependence level from the hospital department list; Extract building boundaries, floor numbers, column grid locations, corridor locations, staircase locations, elevator locations, and entrance / exit locations from architectural space files; Extract flow data of patient flow, medical staff flow, logistics and waste flow from medical process records; Extract data on minimum departmental area, minimum corridor width, maximum medical process distance, minimum infection isolation distance, and clean / dirty flow restrictions from hospital layout standard documents.

3. The hospital department layout generative intelligent planning method as described in claim 2, characterized in that: The generated hospital layout condition code includes: The hospital layout digital data is uniformly coordinated, numerically normalized, categorized, and indexed to obtain the hospital layout condition code; By mapping the building space data to floor coordinates, the floor space boundary matrix is ​​obtained; The departmental functional data is vectorized into attributes to obtain departmental attribute vectors; The flow of medical process data between departments is statistically analyzed to obtain the inter-departmental flow weight matrix; The hospital layout standard data is parameterized to obtain a constraint parameter vector; Based on the department identifier and floor number, the floor space boundary matrix, department attribute vector, inter-department flow weight matrix, and constraint parameter vector are stored accordingly.

4. The hospital department layout generative intelligent planning method as described in claim 3, characterized in that: The output hospital department digital layout solution includes, The candidate hospital department layouts at the point where the generative layout model stops generating iterations are encapsulated, path-associated, and residual-recorded to obtain the digital layout scheme of the hospital departments. The planar locations of departments in the candidate hospital's department layout are encapsulated to obtain department coordinate data; Encapsulate the floor numbers in the department layout of the candidate hospital to obtain the department floor data; Encapsulate the width, height, and boundary range of the departments in the candidate hospital's departmental layout to obtain departmental boundary data; The adjacency relationships between departments are correlated to obtain departmental adjacency data; The pathways of patient flow, medical staff flow, logistics, and waste flow in medical process data are correlated to obtain medical process path data; Record the constraint residual vectors of hospital departments at the time of output to obtain constraint residual record data.

5. A generative intelligent planning system for hospital department layout, employing the generative intelligent planning method for hospital department layout as described in any one of claims 1 to 4, characterized in that: It includes a layout data encoding module, a hard constraint residual calculation module, and a residual feedback generation module; The layout data encoding module is used to acquire hospital layout digital data and generate hospital layout condition codes. The hard constraint residual calculation module is used to construct a hard constraint map of hospital departments based on hospital layout condition coding, and to calculate the constraint residual vector of hospital departments. The residual feedback generation module is used to input the constraint residual vectors of hospital departments into the generative layout model and output the digital layout scheme of hospital departments.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the hospital department layout generative intelligent planning method as described in any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the hospital department layout generative intelligent planning method as described in any one of claims 1 to 4.

Citation Information

Patent Citations

  • Medical space layout design optimization system based on artificial intelligence

    CN120086945A

  • Hospital building layout generation and wind environment optimization system based on deep reinforcement learning

    CN121881475A