A spatial evaluation and intelligent updating method and system based on a large language model and multi-source data

Abstract

The application discloses a kind of space evaluation and intelligent updating method and system based on large language model and multi-source data, which comprises the following steps: S1, obtaining multi-source data about each space unit, thereby constructing a multi-dimensional evaluation index system, and generating a comprehensive priority index to quantify the update urgency of each space unit;S2, construct a vectorized domain knowledge base, thereby for the space unit with higher update urgency, use LLM with RAG architecture for intelligent diagnosis and scheme generation, analyze the specific reasons that cause the space unit to have higher update urgency, and according to the specific problems diagnosed and planning target, retrieve and generate adaptive update scheme.The application combines multi-source data evaluation, domain knowledge base and RAG-LLM technology to construct an AI planning assistant that understands both "spatial geography" and "human needs", and realizes intelligent diagnosis of current problems and generation of dynamically adaptive solutions.

Description

Technical Field

[0001] This invention relates to a spatial evaluation and intelligent update method and system based on large language models and multi-source data, belonging to the field of big data application technology. Background Technology

[0002] In recent years, promoting high-quality urban development and digital transformation has become an important direction for urban evolution worldwide. Spatial assessment and smart renewal are key components of urban planning and community governance, aiming to optimize the allocation of spatial resources and improve utilization efficiency and residents' quality of life through scientific assessment and intelligent technologies.

[0003] Campuses, communities, and industrial parks, as important functional spaces in cities, gather large numbers of people and engage in production and service activities, making them the most promising target areas for smart urban renewal. While traditional planning methods (such as managers' experience-based judgment) can guide the construction of physical spaces, and existing smart renewal technologies (such as the Internet of Things) can monitor and manage infrastructure, they mostly neglect the dynamic "human-centered" needs of diverse stakeholders. They cannot effectively combine information technology with actual activity scenarios, nor can they enable AI models (such as general LLM models) to understand complex spatial logic and professional planning knowledge, and they cannot effectively identify the underlying causes of spatial supply and demand mismatches. Summary of the Invention

[0004] Purpose of the invention: To address the problems existing in the prior art, this invention provides a spatial assessment and intelligent update method and system based on large language models and multi-source data. It combines multi-source data assessment, domain knowledge base and RAG-LLM technology to achieve in-depth spatial use assessment and adaptive intelligent update scheme.

[0005] Technical Solution: To achieve the above objectives, this invention provides a spatial evaluation and intelligent update method based on a large language model and multi-source data, comprising the following steps:

[0006] S1. Obtain multi-source data on each spatial unit, thereby constructing a multi-dimensional evaluation index system and generating a comprehensive priority index to quantitatively evaluate the urgency of updating each spatial unit.

[0007] S2. Construct a vectorized domain knowledge base. Then, for spatial units with higher update urgency, use the LLM of the RAG architecture for intelligent diagnosis and solution generation. Analyze the specific reasons that make the update of the spatial unit more urgent, and retrieve and generate adaptive update solutions based on the diagnosed specific problems and planning goals.

[0008] Specifically, the multi-dimensional evaluation index system includes the following evaluation indicators:

[0009] ① Space utilization intensity Hi This is used to quantify the population aggregation intensity of spatial unit i within a typical period, and its calculation method is as follows:

[0010] ,

[0011] in, and They are spatial units Average population density on weekdays and weekends;

[0012] ② User demand intensity D i This is used to quantify the user's subjective concern about the update of spatial unit i, and its calculation method is as follows:

[0013] ,

[0014] in, For social media and spatial units The frequency of relevant and effective mentions =1,...,n;

[0015] ③ Using time series volatility V i This is used to quantify the dramatic temporal changes in space use intensity over a typical period, and its calculation method is as follows:

[0016] ,

[0017] in, spatial unit Average population density across all date types spatial unit Standard deviation of population density across all date types.

[0018] Specifically, the comprehensive priority index The weighted summation is calculated by taking the standardized data and weights corresponding to each evaluation indicator. The standardized data is obtained by standardizing each evaluation indicator, and the weights are set coefficients or calculated according to a set method.

[0019] Furthermore, the minimum-maximum scaling method is used to standardize the various evaluation indicators.

[0020] Furthermore, the weights can be calculated using one or more of the following methods: entropy weighting and analytic hierarchy process.

[0021] Specifically, the domain knowledge base is constructed by collecting, cleaning, slicing, vectorizing, and storing domain knowledge.

[0022] Furthermore, a semantic-aware strategy is employed to slice domain knowledge.

[0023] Specifically, the intelligent diagnosis and solution generation process includes:

[0024] 2.1 For spatial units with higher urgency of updating, the multi-dimensional evaluation indicators in step S1 are used to search the constructed domain knowledge base, and the retrieved knowledge fragments guide the LLM to perform logical reasoning, thereby analyzing the specific reasons for the higher urgency of updating the spatial unit.

[0025] 2.2 Based on the specific problems diagnosed in step 2.1 and the planning goals set by the user, further retrieval is performed in the constructed domain knowledge base, and the retrieved knowledge fragments guide the LLM to generate adaptive update solutions.

[0026] Furthermore, the retrieval method includes one or more of keyword retrieval and semantic retrieval.

[0027] Furthermore, this invention also provides a spatial evaluation and intelligent update system based on a large language model and multi-source data, comprising the following modules:

[0028] The data evaluation module is used to acquire multi-source data about each spatial unit, thereby constructing a multi-dimensional evaluation index system and generating a comprehensive priority index to quantitatively evaluate the urgency of updating each spatial unit.

[0029] The knowledge construction module is used to build a vectorized domain knowledge base;

[0030] The intelligent update module is used to perform intelligent diagnosis and solution generation using the LLM of the RAG architecture for space units with higher update urgency. It analyzes the specific reasons that make the space unit more urgent to update, and retrieves and generates adaptive update solutions based on the specific problems diagnosed and planning goals.

[0031] Beneficial Effects: This invention combines multi-source data evaluation, domain knowledge base, and RAG-LLM technology to construct an AI planning assistant that simultaneously understands "spatial geography" and "human needs." Based on a closed-loop feature of "evaluation-diagnosis-generation," it dynamically analyzes the matching relationship between space use and user needs, achieving intelligent diagnosis of current problems and dynamic generation of adaptive solutions. This method effectively overcomes the shortcomings of traditional methods that separate "data-driven," "human needs," and "professional knowledge." It can be used for the dynamic evaluation and monitoring of urban functional areas such as campuses, communities, and industrial parks, discovering potential patterns of space utilization and service shortcomings, and providing scientific basis and decision support for the precise allocation and layout optimization of spatial resources. Attached Figure Description

[0032] Figure 1This is a flowchart of the spatial assessment and intelligent update method in an embodiment of the present invention;

[0033] Figure 2 This is a regional distribution map of an embodiment of the present invention;

[0034] Figure 3 , 4 These are population heat maps of different times during weekdays and weekends, respectively, according to embodiments of the present invention.

[0035] Figure 5 This is a thermal comparison chart of average population on weekdays and rest days according to an embodiment of the present invention;

[0036] Figure 6 This is a distribution chart of high-frequency keywords appearing in social media data according to embodiments of the present invention;

[0037] Figure 7 This is a diagram showing the sequence of buildings visited by students in an embodiment of the present invention;

[0038] Figure 8 This is a spatial distribution diagram of the comprehensive priority index PI in an embodiment of the present invention;

[0039] Figure 9 This is a schematic diagram of the overall scheme for implementing intelligent updates according to an embodiment of the present invention;

[0040] Figure 10 The diagram shows the renovation scheme of high-priority buildings in this embodiment of the invention, where a is a schematic diagram of the intelligent renovation of the canteen, b is a schematic diagram of the intelligent renovation of the dormitory, c is a schematic diagram of the intelligent renovation of the library, and d is a schematic diagram of the intelligent renovation of the classroom.

[0041] Figure 11 This is a planning diagram of the logistics and delivery system (drone airspace) in an embodiment of the present invention. Detailed Implementation

[0042] The preferred embodiments of the present invention will now be described in conjunction with the accompanying drawings, which will more clearly and completely illustrate the technical solution of the present invention.

[0043] Taking university campuses as an example, the smart renovation of university campuses not only involves the optimization of campus spaces and the improvement of services, but also serves as a breakthrough point for promoting the overall smart transformation of cities. Currently, the main approaches to campus renovation fall into two categories:

[0044] (1) Traditional campus planning: This method is mostly guided by the goals of managers and the experience of planners, ignoring the dynamic needs of diverse subjects such as students, teachers, and visitors, resulting in many problems such as unreasonable and inconvenient use of space by teachers and students, low quality, and disordered management.

[0045] (2) Smart Campus Planning: Existing smart campus construction focuses on the construction of information infrastructure and information management platform, lacking a deep understanding of diverse human needs and multi-subject space use behaviors and problems. Information technology is not fully integrated with teaching, scientific research, social and other activity scenarios, resulting in a weak "smart" experience.

[0046] Based on this, the present invention proposes an intelligent diagnosis and generation mechanism that combines multi-source data evaluation, domain knowledge base, RAG (retrieval-enhanced generation) technology and LLM (large language model), thereby constructing an AI planning assistant that simultaneously understands "spatial geography" and "human needs".

[0047] This invention provides a spatial evaluation and intelligent update method based on a large language model and multi-source data, mainly including the following steps:

[0048] S1. Obtain multi-source data on each spatial unit, thereby constructing a multi-dimensional evaluation index system to quantitatively assess the urgency of updating each spatial unit;

[0049] Spatial analysis is the process of solving spatial problems, and acquiring spatial information is a necessary means to solve these problems. In its modern sense, spatial analysis refers to the extraction of information about geographic objects—including their location, attributes, and relationships—with the support of computer technology, in order to support specific spatial decision-making problems.

[0050] Based on this theory, this invention first uses multi-source heterogeneous datasets (including spatial, activity, and perception data) to construct a multi-dimensional evaluation index system. On the one hand, it can quantitatively evaluate the urgency of updating each spatial unit, thus serving as the basis for updating and ranking. On the other hand, it can provide a comprehensive, efficient, and intuitive evaluation basis for the matching relationship between spatial usage characteristics and user needs analysis.

[0051] Specifically, the multi-dimensional evaluation index system includes the following three core indicators:

[0052] ① Space utilization intensity H i This indicator is used to quantify spatial units. The intensity of population gathering during a typical period is based on population thermal data, and is calculated as follows:

[0053] (1),

[0054] (2),

[0055] in, spatial unit In the date type d Within a given time period, the standardized population density obtained through spatial interpolation, k=1,...,12, corresponds to 12 two-hour time periods within a day, d∈{wd, we}. First, the spatial unit is calculated according to formula (1). Average population density for each date type Then, based on the 5:2 ratio of weekdays to weekends in a typical school week, differentiated weights were assigned to the weekday (wd) and weekend (we) data, and a weighted average was calculated to obtain the comprehensive space utilization intensity. .

[0056] ② User demand intensity D i This metric is used to capture users' subjective concern about updates to a specific space. The data comes from social media data, and its calculation method is as follows:

[0057] (3),

[0058] in, For social media and spatial units The frequency of relevant and effective mentions =1,...,n. Therefore, the intensity of user demand... Through spatial units The proportion of mentions to the total frequency is calculated using a standardized method.

[0059] ③ Using time series volatility V i This indicator measures the degree of temporal variation in space use intensity within a typical cycle, reflecting use efficiency and pattern stability. The data is derived from population thermal data, and its calculation method is as follows:

[0060] (4),

[0061] (5),

[0062] in, Representative spatial unit The set of population density observations for all date types d and time periods k during the observation period, where T is the total number of observation periods. spatial unit Average population density across all date types spatial unit The standard deviation of population density across all date types was used to calculate spatial units using the coefficient variation method. Using time-series volatility (V i ).

[0063] Based on the above core indicators, a comprehensive priority index can be further calculated. To quantitatively assess the urgency of updating each spatial unit, specifically including:

[0064] 1) Standardization: To eliminate the influence of dimensions and unify the interpretation of indicators, a minimum-maximum scaling method is used. , , The original calculation results are standardized to map them to the interval [0, 1]. The calculation formula is as follows:

[0065] (6),

[0066] Will , , The original calculation results form a matrix, where Let be the original value of the i-th sample and the j-th indicator. Let be the standard value of the i-th sample and the j-th indicator. For positive indicators, the data standardization can be achieved by formula (6).

[0067] 2) Weight Calculation: The weights corresponding to each indicator can be calculated using methods such as entropy weighting and analytic hierarchy process (AHP), or they can be preset according to the actual situation. Here, to balance data objectivity and expert domain knowledge, a combination of entropy weighting (determining objective weights based on data information entropy) and AHP (determining subjective weights based on expert judgment criteria) is preferred for calculation.

[0068] (7),

[0069] (8),

[0070] (9),

[0071] (10)

[0072] (11),

[0073] in, Let represent the numerical proportion of the i-th sample in the j-th indicator. Let the entropy value be the value corresponding to the j-th index. The objective weight corresponding to the j-th indicator is... =1,...,n, =1,...,m, In this embodiment, there are a total of 3 indicators, so m=3; while the analytic hierarchy process (AHP) uses the 1-9 scale to compare the importance of each indicator pairwise, thereby constructing a judgment matrix A={ } m×m , To determine the importance of the j-th indicator relative to the k-th indicator, Let be the subjective weight corresponding to the j-th indicator; finally, by setting an adjustment coefficient α, a linear combination of the subjective and objective weights is obtained to obtain the final combined weight. .

[0074] 3) Weighted summation: By weighting the standardized data according to the weights corresponding to each indicator, the comprehensive priority index of each spatial unit can be calculated. As the primary basis for updating the sorting, the calculation formula is as follows:

[0075] (12),

[0076] (13)

[0077] in, These represent the weights corresponding to the three indicators. , , These represent the standard values ​​of the three indicators for the i-th sample, and formula (13) provides the indicator weights calculated in this embodiment.

[0078] Of course, in practical applications, the urgency of updating spatial units is not only reflected in the comprehensive priority index PI, but can also be analyzed through the ranking and distribution of other indicators (such as high priority index PI). Value, High Values ​​can also reflect the urgency of updates, thereby enabling a more comprehensive assessment of space use.

[0079] S2. Construct a vectorized domain knowledge base. Then, for spatial units with higher update urgency, use the LLM of the RAG architecture for intelligent diagnosis and solution generation. Analyze the specific reasons that make the update of the spatial unit more urgent, and retrieve and generate adaptive update solutions based on the diagnosed specific problems and planning goals.

[0080] To address the problems of Large Module Models (LLMs) lacking specialized domain knowledge, failing to understand complex spatial logic and professional planning knowledge, and being unable to effectively identify the underlying causes of spatial supply-demand mismatches, this invention enhances LLMs using RAG technology. Based on the aforementioned evaluation results, relevant document fragments are retrieved from the knowledge base using vectorization methods. Specific prompt word chains and retrieval mechanisms are designed based on the retrieved knowledge fragments to guide LLMs in logical reasoning, thereby achieving traceable and in-depth diagnosis of the evaluation results and generating adaptive intelligent update solutions.

[0081] Furthermore, step S2 specifically includes:

[0082] 2.1 Data Preparation Stage: This stage involves building a structured vector database (i.e., a domain knowledge base) to provide accurate external knowledge support for the subsequent generation process.

[0083] ① Knowledge Acquisition and Cleaning: First, establish a multi-level knowledge source system, including unstructured documents, such as academic literature (PDF / CAJ format) in fields such as smart campus, space optimization, and user behavior psychology, as well as relevant national / local design codes (such as the "Code for Fire Protection Design of Buildings" and the "Code for Design of Educational Buildings"), and structured question-and-answer pairs, such as organizing the "problem-strategy" correspondence in historical planning projects and constructing QA question-and-answer pairs; then, perform data cleaning, remove garbled characters, headers and footers from documents, and perform text standardization processing, thereby constructing a literature knowledge base and a question-and-answer knowledge base.

[0084] ② Knowledge Slicing: It is preferable to adopt a semantically aware slicing strategy rather than simple fixed character length cutting. For example, for laws and regulations, slicing is done by “article” to maintain the integrity of the clauses; for academic documents, slicing is done by “paragraph” or “chapter”, and a certain overlap window (such as 10-15%) is set to preserve the continuity of contextual semantics.

[0085] ③ Knowledge Vectorization and Storage: First, a pre-trained text embedding model is used to transform the segmented text blocks into high-dimensional dense vectors. Then, the vector data and its corresponding original text metadata are stored in a vector database (FAISS). Finally, an IVF (Inverted File) or HNSW (Hierarchical Navigation Small World) index is constructed to support millisecond-level approximate nearest neighbor (ANN) retrieval for large-scale data.

[0086] 2.2 Intelligent Diagnosis Stage: This stage aims to transform numerical assessment results into semantic diagnostic reports.

[0087] ① Numerical conversion: For spatial units with higher urgency of updates, the various evaluation indicators in step S1 (such as...) are converted... , , The numerical values ​​and rankings are transformed into a structured natural language description (Context).

[0088] ② Data retrieval: Perform hybrid retrieval, including keyword retrieval: extract keywords from user comments (such as "queue" and "crowded") and perform precise matching in the knowledge base, and semantic retrieval: vectorize the spatial feature description of the input, and then retrieve historical cases or theoretical analyses that are most similar to its "problem pattern" from the vector database.

[0089] ③Prompt Design: Define the LLM role as "Senior Planner". Input "current situation data description" + "similar cases / theories found" into the LLM. The task is to "analyze the underlying reasons for the high PI value in this area based on the data and reference cases (such as resource misallocation, aging facilities, and single function)".

[0090] 2.3 Solution Generation Stage: This stage aims to generate specific and feasible transformation strategies.

[0091] ① Data retrieval: Based on the specific problems diagnosed in the previous stage (such as "noise interference"), as well as the planning goals input by the user (such as "low cost" and "improved interactivity") and hard constraints (such as "cannot change the main structure"), retrieve the corresponding "solutions", "design strategies" and "applicable technologies" (such as "acoustic partitioning materials" and "soundproof cabin products") from the knowledge base.

[0092] ② Chain of Reasoning (CoT): The "diagnostic conclusion" + "planning goals and hard constraints" + "retrieved solutions / design strategies / applicable technologies" are used as new context inputs to the LLM, guiding the LLM to generate solutions according to the logical path of "diagnosing the problem -> retrieved strategies -> adjusting in combination with constraints -> final solution". The LLM is required to output a standardized solution format, including "name of the transformation strategy", "specific technical means", "expected effect" and "source of reference".

[0093] ③Solution optimization: Planners can issue modification instructions to the generated solutions, and use these modification instructions as queries for a new round of retrieval, repeating the above process to achieve dynamic optimization of the solutions.

[0094] In summary, this invention integrates GIS spatial analysis (processing) , Objective data) and LLM semantic understanding (processing) Subjective perception data (such as data from RAG) can enable a more comprehensive assessment of space use, and then through RAG-enhanced LLM, a traceable and in-depth diagnosis of the assessment results can be achieved, generating adaptive smart update solutions.

[0095] Furthermore, this invention also includes a spatial evaluation and intelligent update system based on a large language model and multi-source data, which mainly consists of the following modules:

[0096] The data evaluation module is used to acquire multi-source data about each spatial unit, thereby constructing a multi-dimensional evaluation index system and generating a comprehensive priority index to quantitatively evaluate the urgency of updating each spatial unit.

[0097] The knowledge construction module is used to build a vectorized domain knowledge base;

[0098] The intelligent update module is used to perform intelligent diagnosis and solution generation using the LLM of the RAG architecture for space units with higher update urgency. It analyzes the specific reasons that make the space unit more urgent to update, and retrieves and generates adaptive update solutions based on the specific problems diagnosed and planning goals.

[0099] like Figure 1 As shown, the following will provide a study on Nanjing University's Xianlin Campus (research area as shown in the image). Figure 2 Specific embodiments (as shown):

[0100] A. Obtaining multi-source data:

[0101] First, we collected building data (ID, coordinates, name, area, etc.) and Baidu Insight population heat map data for the study area (for one consecutive week from October 21st to 27th, 2024). Figure 3 and Figure 4 (As shown); at the same time, student spatiotemporal trajectory data (54 students, 15 valid samples) and social media data (Xiaohongshu, Weibo, etc.) were collected.

[0102] B. Data Preprocessing and Fusion:

[0103] (1) Spatial matching: The GPS points in the 15 valid trajectory samples of 54 students were used to count the number of time points that fell within the outline of each building by using the point-polygon connection tool of GIS, and the buildings with high frequency of visits were identified.

[0104] (2) Heat value calculation: Perform inverse distance weighted interpolation on the scatter points of the heat map within a week to generate a raster map with a resolution of 5 meters, and calculate the average pixel value within the polygon of each building to obtain the average daily population density of each building;

[0105] (3) Text matching: Clean the collected Weibo / Xiaohongshu text. For example, identify the text "The queue at the fourth canteen is too long", extract the entity "fourth canteen", and match it with student restaurant No.456 on the map.

[0106] Through the above steps, a comprehensive attribute table containing all 86 buildings on campus was generated. Each row of data includes building ID, name, average heat value on weekdays, average heat value on weekends, number of times mentioned on social media, etc., providing direct input for subsequent analysis and evaluation.

[0107] C. GIS Spatial Analysis:

[0108] Based on the above data, a preliminary analysis of space usage characteristics and user needs can be conducted, thereby providing more search directions for in-depth diagnosis of space problems and achieving a more comprehensive assessment of space usage.

[0109] like Figure 5As shown, the analysis revealed significant temporal heterogeneity in campus space use: weekdays exhibit a clear bimodal pattern (9-11 am and 3-7 pm), with high activity intensity in teaching buildings, laboratories, and libraries; rest days show a more diffuse pattern, with relatively increased activity in dormitories, canteens, and sports venues.

[0110] Through analysis Figure 6 The study found that user demand was highly concentrated on keywords such as facility configuration, unmanned delivery vehicles, smart canteens, and smart parking, reflecting a strong focus on intelligent and automated services.

[0111] Through analysis Figure 7 The study found that students’ mobility behavior is the result of the combined effects of spatial proximity, functional attractiveness and individual schedules, with some high-frequency paths (such as dormitory-teaching building-canteen-library) forming the core framework of the activity chain.

[0112] D. Smart Renewal Assessment:

[0113] Based on the multi-dimensional evaluation system described in step S1, further calculate the various evaluation indicators and comprehensive priority index for all building units, and generate [the following data] based on the calculation results. Figure 8 The results showed that the library (PI value 0.661-0.733), dormitory buildings (especially buildings 1, 2, 7, 9, and 22, PI value 0.561-0.661), as well as some teaching buildings and canteens, had the highest priority for renovation.

[0114] E. AI-assisted assessment, planning, and visualization:

[0115] With the assistance of a domain knowledge base, this invention further utilizes a RAG-enhanced large language model to intelligently diagnose and generate solutions for high-priority regions. The process of generating a visualized solution is achieved through the following steps:

[0116] (1) Input parameter construction: Integrate three types of input data: first, diagnostic context, such as: the congestion in the cafeteria is mainly due to low settlement efficiency; second, planning objectives, namely the natural language objectives input by the user (such as improving traffic efficiency and low-cost renovation); and third, constraints, namely physical or management restrictions (such as not being able to change the main structure and having to comply with fire protection regulations).

[0117] (2) Goal-oriented knowledge retrieval: Combine diagnostic problems and planning goals into query vectors, and retrieve matching solutions in question-and-answer and literature knowledge bases. For example, for queuing and efficiency, retrieve knowledge items such as AI visual settlement technology and smart food locker layout specifications.

[0118] (3) Structured cue word construction: Sending cue words to the LLM:

[0119] [Role Definition]: Senior Smart Campus Planner.

[0120] [Task Background]: Based on the [Diagnostic Report], the objective is [Planning Objective].

[0121] [Reference Knowledge]: [Technical Item 1 Found], [Technical Item 2 Found]...

[0122] [Boundary Constraints]: Must comply with [constraint conditions].

[0123] (4) Generation and optimization: LLM uses the logic of thought chain to match problems with technologies and generate structured text solutions that include strategy names, specific technical means, expected benefits and knowledge sources.

[0124] (5) Visual parameter extraction: The system automatically parses the generated "update plan text" and extracts visual entity keywords (such as "drone landing pad", "permeable pavement", "intelligent interactive screen") and spatial orientation words (such as "roof", "facade", "entrance").

[0125] (6) Input of constraints: Input the basic parameters of the target building, including the building outline, building height, and existing style images.

[0126] (7) Image generation: A generative AI model is used, with the building outline as a geometric constraint (ensuring that the generated image is still the original building structure), to transform the scheme description into a visual image, making it more intuitive and reliable.

[0127] The overall scheme generated in this embodiment is as follows: Figure 9 As shown, the specific areas to be renovated include:

[0128] E1, Library (High PI Value):

[0129] Based on the assessment results (high PI value) and combined with high-frequency words such as "poor seat reservation system", "seat grabbing", "serious seat hogging", and "loud noise" in social media data, RAG-LLM diagnosed two main problems: the seat management system has a poor user experience and loopholes, resulting in serious seat hogging and low efficiency; and the spatial functional zoning is unclear, with collaborative discussion areas and quiet study areas interfering with each other.

[0130] To address the above issues, RAG-LLM proposed a solution (such as...). Figure 10(as shown in c) includes: deploying a real-time seat monitoring and intelligent reservation platform based on sensor / visual recognition, integrating check-in, departure timekeeping, and anti-seat-hogging reminder functions (smart space management); refined acoustic zoning, dividing the area into an absolute quiet zone, a collaborative discussion zone, and a leisure communication zone, and adding individual cubicles (space function optimization); and deploying an APP-based indoor navigation and information display screen (guidance optimization).

[0131] E2, Dormitory Area (High PI Value):

[0132] Based on the assessment results (high PI value) and combined with social media data such as "outdated facilities", "lack of public areas", "lack of refrigerators", "far from teaching buildings", and "inconvenient transportation", the problems diagnosed by RAG-LLM are: outdated facilities, lack of public service space, and excessively long commuting distance for some dormitories.

[0133] To address the above issues, RAG-LLM proposed a solution (such as...). Figure 10 (as shown in b) includes: piloting rapid renovation of modular integrated bathrooms (facility upgrade); embedding 24-hour public living corners with smart charging, shared refrigerators, and self-service laundry facilities within buildings (service integration); utilizing the elevated floors of building clusters to transform them into shared living rooms, including shared kitchens and study / discussion areas (space regeneration); and deploying a smart energy and environmental management system, including IAQ sensors and smart meters (smart management).

[0134] E3, Cafeteria (High PI value):

[0135] Based on the assessment results (high PI value), combined with the high demand for "Smart canteen" in social media data, as well as complaints such as "poor food quality" and "long waiting times", the RAG-LLM diagnosed the following problems: the quality of catering services needs to be improved and the operational efficiency (especially in checkout and queuing) is low.

[0136] To address the above issues, RAG-LLM proposed a solution (such as...). Figure 10 (as shown in a) includes: deploying AI-powered visual intelligent checkout counters (intelligent checkout); establishing a big data operation optimization platform to predict demand and analyze waste (data-driven operation); upgrading online ordering and setting up intelligent pickup lockers and drone take-off and landing points (integrated ordering); and providing real-time queuing information for each food stall on the campus APP (information visualization).

[0137] E4. Teaching and Learning Spaces (High PI Value):

[0138] Based on the assessment results (some teaching buildings have high PI values), combined with the demand for "Smart teaching buildings" in social media data, as well as issues such as "insufficient power outlets in some classrooms" and "long distances between teaching buildings", the problems diagnosed by RAG-LLM are: single-function teaching spaces, insufficient informal learning spaces, and insufficient configuration of some facilities (such as power supply).

[0139] To address the above issues, RAG-LLM proposed a solution (such as...). Figure 10 (d) includes: selecting some teaching buildings (such as ID9) for pilot renovations, adding small maker spaces, data labs, and VR experience areas (classroom smart upgrades); adding seats, discussion tables, and power outlets to spacious corridors and ground floor lobbies in teaching buildings, transforming them into informal learning and exchange points (revitalizing public spaces); and evaluating and renovating inefficient professional buildings (such as ID 80, 81, 82) into professional-themed learning and discussion rooms (inefficient space regeneration).

[0140] E5. Logistics and Distribution Space Renewal (High) value):

[0141] Social media data shows that unmanned delivery vehicles are the most frequent demand for food delivery. Based on this, RAG-LLM diagnosed that the demand for "last 100 meters" logistics delivery on campus (especially food delivery) is extremely strong, but the existing delivery methods have problems of low efficiency and difficulty in management.

[0142] To address the above issues, RAG-LLM proposed a solution (such as...). Figure 11 As shown, this includes: constructing a collaborative delivery system (multimodal delivery) of "unmanned vehicles (ground backbone) + drones (rapid air response)"; planning first-level (ground-air hub), second-level (smart express cabinet station) and third-level (instant delivery point) nodes, thereby forming a three-level logistics node; dividing the vertical airspace into three levels (L1: 0-30m, L2: 30-60m, L3: 60-120m) according to U-Space recommendations, and planning flight corridors (airspace planning).

[0143] This invention addresses the practical needs of future urban spatial intelligent upgrading and campus planning. First, it constructs a spatial intelligent upgrading assessment model based on multi-source data (including spatial data, human activity data, and user perception data). This model identifies key areas and directions for upgrading. Then, it enhances the LLM (Limited Learning Model) with an integrated domain knowledge base, enabling intelligent diagnosis of current problems and the generation of dynamically adaptive solutions. This method overcomes the limitations of traditional planning methods, such as manager-oriented approaches, lack of accurate data, and the disconnect between AI technology and spatial geographic cognition. It provides a replicable technical path and methodological tools for the intelligent transformation of campuses and other functional areas of the city.

[0144] The above-described specific embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Various modifications, substitutions, and improvements made by those skilled in the art to the technical solutions of the present invention based on the provided textual description and drawings, without departing from the design concept and spirit of the present invention, should all fall within the scope of protection of the present invention.

Claims

1. A spatial evaluation and intelligent update method based on large language models and multi-source data, characterized in that, Includes the following steps: S1. Obtain multi-source data on each spatial unit, thereby constructing a multi-dimensional evaluation index system and generating a comprehensive priority index to quantitatively evaluate the urgency of updating each spatial unit. S2. Construct a vectorized domain knowledge base. Then, for spatial units with higher update urgency, use the LLM of the RAG architecture for intelligent diagnosis and solution generation. Analyze the specific reasons that make the update of the spatial unit more urgent, and retrieve and generate adaptive update solutions based on the diagnosed specific problems and planning goals.

2. The spatial assessment and intelligent updating method according to claim 1, characterized in that, The multi-dimensional evaluation index system includes the following evaluation indicators: ① Space utilization intensity H i This is used to quantify the population aggregation intensity of spatial unit i within a typical period, and its calculation method is as follows: ， in, and They are spatial units Average population density on weekdays and weekends; ② User demand intensity D i This is used to quantify the user's subjective concern about the update of spatial unit i, and its calculation method is as follows: ， in, For social media and spatial units The frequency of relevant and effective mentions =1,...,n; ③ Using time series volatility V i This is used to quantify the dramatic temporal changes in space use intensity over a typical period, and its calculation method is as follows: ， in, spatial unit Average population density across all date types spatial unit Standard deviation of population density across all date types.

3. The spatial assessment and intelligent updating method according to claim 1, characterized in that, The comprehensive priority index The weighted summation is calculated by taking the standardized data and weights corresponding to each evaluation indicator. The standardized data is obtained by standardizing each evaluation indicator, and the weights are set coefficients or calculated according to a set method.

4. The spatial assessment and intelligent updating method according to claim 3, characterized in that, The minimum-maximum scaling method was used to standardize the various evaluation indicators.

5. The spatial assessment and intelligent updating method according to claim 3, characterized in that, The weights are calculated using one or more of the following methods: entropy weighting and analytic hierarchy process.

6. The spatial assessment and intelligent updating method according to claim 1, characterized in that, The domain knowledge base is constructed by collecting, cleaning, slicing, vectorizing, and storing domain knowledge.

7. The spatial assessment and intelligent updating method according to claim 6, characterized in that, A semantically aware strategy is used to slice domain knowledge.

8. The spatial assessment and intelligent updating method according to claim 1, characterized in that, The intelligent diagnosis and solution generation process includes: 2.1 For spatial units with higher urgency of updating, the multi-dimensional evaluation indicators in step S1 are used to search the constructed domain knowledge base, and the retrieved knowledge fragments guide the LLM to perform logical reasoning, thereby analyzing the specific reasons for the higher urgency of updating the spatial unit. 2.2 Based on the specific problems diagnosed in step 2.1 and the established planning goals, further searches are conducted in the constructed domain knowledge base, and the retrieved knowledge fragments guide the LLM to generate adaptive update solutions.

9. The spatial assessment and intelligent updating method according to claim 8, characterized in that, The retrieval methods include one or more of keyword retrieval and semantic retrieval.

10. A spatial assessment and intelligent update system based on a large language model and multi-source data, characterized in that, Includes the following modules: The data evaluation module is used to acquire multi-source data about each spatial unit, thereby constructing a multi-dimensional evaluation index system and generating a comprehensive priority index to quantitatively evaluate the urgency of updating each spatial unit. The knowledge construction module is used to build a vectorized domain knowledge base; The intelligent update module is used to perform intelligent diagnosis and solution generation using the LLM of the RAG architecture for space units with higher update urgency. It analyzes the specific reasons that make the space unit more urgent to update, and retrieves and generates adaptive update solutions based on the specific problems diagnosed and planning goals.

Images