Urban flood resilience fine-grained evaluation method and system based on large language model

Abstract

The application relates to a city flood resilience fine evaluation method and system based on a large language model, and belongs to the technical field of city flood resilience evaluation. The application constructs a city flood resilience evaluation index system containing multiple indexes, divides a target region by using a preset resolution grid, resamples multiple-source heterogeneous data to each grid unit, and obtains standard data through preprocessing; objective weights of each index are calculated based on the standard data, subjective weights are obtained by constructing a multi-field virtual expert model for subjective weighting; the subjective and objective weights are weighted and fused to obtain a combined weight, a city flood resilience comprehensive index of each grid unit is calculated, and an evaluation result is output after grade division. The application significantly improves the knowledge dimension of subjective weighting, takes into account field knowledge and data statistical characteristics, can realize fine spatial evaluation, and can automatically run the evaluation process, thereby providing data support for city flood resilience fine management and planning decision-making.

Description

Technical Field

[0001] This application relates to the field of urban flood resilience assessment technology, and in particular to a refined assessment method and system for urban flood resilience based on a large language model. Background Technology

[0002] Current urban flood resilience assessments primarily employ traditional multi-indicator comprehensive evaluation methods such as the Analytic Hierarchy Process (AHP) and fuzzy comprehensive evaluation. These methods assess resilience by constructing an evaluation index system and conducting a comprehensive score. However, the determination of index weights in these methods heavily relies on the subjective judgment of a small number of experts, leading to problems such as a limited number of experts, difficulty in avoiding cognitive biases, and the non-reproducibility of the weighting process. This results in insufficient objectivity and repeatability of the assessment results. Although existing technologies offer combined weighting methods that integrate subjective and objective approaches, the acquisition of subjective weights still depends on traditional expert scoring, failing to fundamentally solve the aforementioned problems.

[0003] Furthermore, some existing technologies employ the Stress-State-Response (PSR) theoretical framework to construct evaluation index systems. However, configuring multiple index systems results in low computational efficiency. In practical applications, to compromise on evaluation efficiency, most use administrative divisions or coarse-resolution grids of 1000 to 5000 meters as evaluation units. This coarse spatial resolution fails to reveal the spatial differentiation characteristics of urban resilience and cannot meet the technical requirements of land spatial planning for accurate identification, precise policy implementation, and precise execution. Simultaneously, existing assessments involve multi-source heterogeneous data such as remote sensing imagery, statistical data, point-of-interest data, and meteorological data, lacking a unified spatial processing framework and relying on manual data processing and index calculation, resulting in low evaluation efficiency. Summary of the Invention

[0004] To overcome the technical problems of insufficient dimensions, accuracy and efficiency in urban flood resilience assessment in the prior art, this invention provides a refined assessment method and system for urban flood resilience based on a large language model.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A refined assessment method for urban flood resilience based on a large language model includes: Construct an urban flood resilience assessment index system that includes multiple indicators, and obtain multi-source heterogeneous data corresponding to the multiple indicators; The target area is divided into multiple grid cells using a grid with a preset resolution. The multi-source heterogeneous data is resampled into the multiple grid cells to form the original grid cell data of the multiple indicators. The original grid cell data is preprocessed to obtain the standard data of the multiple indicators. Based on the standard data of the aforementioned multiple indicators, the objective weight of each indicator is calculated. The standard data of the multiple indicators are input into the constructed multi-domain virtual expert model, and subjective weights are assigned to the multiple indicators to obtain the subjective weights of each indicator. The subjective weights and objective weights are weighted and fused to obtain the combined weights of each indicator. Based on the standard data and the combined weights, the comprehensive urban flood resilience index of each grid unit is calculated. Based on the comprehensive urban flood resilience index of each grid unit, the target area is classified into urban flood resilience levels, and the urban flood resilience assessment results of the target area are output.

[0006] Preferably, the indicator system includes a pressure layer, a state layer, and a response layer. The pressure layer includes an annual average rainfall indicator. The state layer includes indicators such as the proportion of impervious surface, river and lake water surface ratio, surface elevation, topographic slope, ecological space area ratio, landscape heterogeneity, landscape connectivity, density of permanent residents in built-up areas, and percentage of vulnerable groups in the total urban population. The response layer includes indicators such as disaster accessibility of medical facilities, density of commercial facilities, density of emergency shelters, per capita road area, and per capita GDP.

[0007] Preferably, the multi-source heterogeneous data includes natural geographic data, remote sensing interpretation data, socioeconomic data, infrastructure and POI data, meteorological data, and land use data; Among them, the original data of surface elevation and topographic slope indicators were extracted from the corresponding natural geographic data; the original data of the indicators of impervious surface ratio, river and lake water surface ratio, and ecological space area ratio were extracted from the corresponding remote sensing interpretation data; the original data of indicators of per capita GDP, built-up area resident population density, and vulnerable group ratio were extracted from the corresponding socio-economic data; the original data of indicators of medical facility accessibility during disasters, emergency shelter density, per capita road area, and commercial facility density were extracted from the corresponding infrastructure and POI data; the original data of the annual average rainfall indicator was extracted from the corresponding meteorological data; and the original data of indicators of landscape heterogeneity and landscape connectivity were extracted from the corresponding land use data.

[0008] Preferably, the original grid cell data is preprocessed to obtain standard data for multiple indicators, including: River and lake surface area ratio, land surface elevation, topographic slope, ecological space area ratio, landscape heterogeneity, landscape connectivity, accessibility of medical facilities during disasters, density of commercial facilities, density of emergency shelters, per capita road area, and per capita GDP are defined as positive indicators. These positive indicators are standardized using the first range standardization formula.

[0009] in, For the first The first grid cell Standard data for a positive indicator The raw data for the indicator. and The first The maximum and minimum values ​​of the original data for each indicator; Annual average rainfall, the proportion of impervious surface land, the density of permanent residents in built-up areas, and the proportion of vulnerable groups are defined as negative indicators. These negative indicators are then standardized using the second range standardization formula.

[0010] in, For the first The first grid cell Standard data for a negative indicator.

[0011] Preferably, the preset resolution is 30m×30m.

[0012] Preferably, the objective weights of each indicator are calculated based on the standard data of the multiple indicators, including: For each indicator, calculate the proportion of the standard data value of each grid cell to the sum of the standard data values ​​of all grid cells of that indicator, and obtain the characteristic weight of each grid cell corresponding to that indicator; Information entropy is calculated based on the aforementioned feature weights, and a difference coefficient is constructed based on the information entropy. The objective weight of each indicator is obtained by dividing the difference coefficient of each indicator by the sum of the difference coefficients of all indicators.

[0013] Preferably, the constructed multi-domain virtual expert model includes an urban planning expert model, a resilient city expert model, a hydrology and flood disaster expert model, an ecological environment expert model, and an emergency management expert model set up in parallel; By setting fixed-layer prompts, the role positioning, professional evaluation dimensions, independent judgment preferences, evaluation principles and expert consultation rules of virtual expert models in various fields are defined, and the fixed-layer prompts remain unchanged in different evaluation tasks. By setting variable-level prompts, the urban characteristics, theoretical framework, and indicator system of the input target area are defined, and these prompts are updated as the evaluation task changes.

[0014] Preferably, the standard data of the multiple indicators are input into the constructed multi-domain virtual expert model, and subjective weights are assigned to the multiple indicators to obtain the subjective weights of each indicator, including: Standard data for multiple indicators are input into the urban planning expert model, resilient city expert model, hydrology and flood disaster expert model, ecological environment expert model and emergency management expert model respectively. Each expert model outputs the initial subjective weight of each indicator and outputs the importance ranking sequence from high to low according to the initial subjective weight of each indicator. Compare the importance ranking sequence of each indicator output by each expert model, and classify the multiple indicators into consensus indicators, divergence indicators, or extreme preference indicators; For consensus indicators, the arithmetic mean of the initial subjective weights of the indicator in each expert model is taken as its subjective weight. For the divergence index, the arithmetic mean of the initial subjective weights of the index in each expert model is taken as its subjective weight, and the weight difference and compromise logic of the index in each expert model are marked when the subjective weight is output. For extreme preference indicators, the terms with the highest and lowest initial subjective weights for the indicator in each expert model are removed, and the arithmetic mean of the initial subjective weights of the indicator in the remaining expert models is taken as its subjective weight.

[0015] Preferably, the combined weights are expressed as:

[0016] Where α is the fusion coefficient of subjective weights, Let j be the subjective weight of the j-th indicator. Let be the objective weight of the j-th indicator; The comprehensive index of urban flood resilience is expressed as follows:

[0017] in, For the first The comprehensive urban flood resilience index for each grid cell ranges from [0,1]. A higher comprehensive urban flood resilience index value indicates a higher level of spatial resilience to flood disasters for that grid cell. For the first The combined weights of the indicators; For the first The first grid cell Standard data for each indicator.

[0018] A refined urban flood resilience assessment system based on a large language model, employing any of the methods described above, includes: Indicator system construction module: used to construct an urban flood resilience assessment indicator system including multiple indicators and obtain multi-source heterogeneous data corresponding to the multiple indicators; Grid cell initialization module: used to divide the target area into multiple grid cells using a grid with a preset resolution, resample the multi-source heterogeneous data into the multiple grid cells to form the original grid cell data of the multiple indicators, and preprocess the original grid cell data to obtain the standard data of the multiple indicators; Objective weight calculation module: used to calculate the objective weight of each indicator based on the standard data of the multiple indicators; Subjective weight calculation module: used to input the standard data of the multiple indicators into the constructed multi-domain virtual expert model, and to subjectively assign weights to the multiple indicators to obtain the subjective weights of each indicator; The Urban Flood Resilience Comprehensive Index Calculation Module is used to perform weighted fusion processing on the subjective weights and the objective weights to obtain the combined weights of each indicator, and calculate the urban flood resilience comprehensive index of each grid unit based on the standard data and the combined weights. Urban flood resilience assessment module: This module is used to classify the urban flood resilience level of the target area based on the comprehensive urban flood resilience index of each grid unit, and output the urban flood resilience assessment results of the target area.

[0019] Compared with the prior art, the beneficial effects of the technical solution of the present invention are: This invention provides a refined assessment method and system for urban flood resilience based on a large language model. By constructing a multi-domain virtual expert model to subjectively assign weights to multiple indicators, it effectively alleviates the problems of traditional expert scoring methods, which rely on a few experts, are susceptible to cognitive biases, and have limited assessment efficiency and dimensions. Secondly, based on the efficient assessment framework of this invention, it can support high-precision assessment of refined grid assessment units. In addition, this invention combines subjective and objective weights to obtain a combined weight, taking into account multi-domain knowledge and statistical characteristics of indicator data. This invention constructs a complete process from data preprocessing, subjective and objective weighting, weight fusion to resilience comprehensive index calculation and level classification. Each step can be deployed and executed automatically, resulting in high assessment efficiency, intuitive and traceable assessment results, and providing data support for refined urban planning decisions. Attached Figure Description

[0020] Figure 1 This is one of the flowcharts for the refined assessment method of urban flood resilience based on a large language model in Example 1; Figure 2 This is the second flowchart of the refined assessment method for urban flood resilience based on a large language model in Example 1; Figure 3 This is a graph showing the annual average rainfall data for City A in the refined assessment method of urban flood resilience based on a large language model, as described in Example 1. Figure 4This is a map showing the surface elevation data of City A in the refined assessment method of urban flood resilience based on a large language model in Example 1. Figure 5 This is a topographic slope index data map of City A in the refined assessment method of urban flood resilience based on a large language model in Example 1. Figure 6 This is a data map of the proportion of impermeable surface in City A, based on the refined assessment method of urban flood resilience based on a large language model in Example 1. Figure 7 This is a disaster accessibility index data chart of medical facilities in City A, based on the refined assessment method of urban flood resilience based on a large language model in Example 1. Figure 8 This is a data map of the emergency shelter density index of City A in the refined assessment method of urban flood resilience based on a large language model in Example 1; Figure 9 This is a graph showing the per capita road area index of City A in Example 1, based on the refined assessment method of urban flood resilience using a large language model. Figure 10 This is a data chart showing the proportion of vulnerable groups in City A, based on the refined assessment method of urban flood resilience using a large language model, as described in Example 1. Figure 11 This is a graph showing the per capita GDP data of City A in the refined assessment method of urban flood resilience based on a large language model in Example 1. Figure 12 This is a data map of the urban built-up area resident population density index of City A in Example 1, based on the refined assessment method of urban flood resilience using a large language model. Figure 13 This is a data map of the commercial facility density index of City A in Example 1, based on the refined assessment method of urban flood resilience using a large language model. Figure 14 This is a graph showing the river and lake surface area ratio data of City A in the refined assessment method of urban flood resilience based on a large language model in Example 1. Figure 15 This is a data chart of the ecological space area ratio index of City A in the refined assessment method of urban flood resilience based on a large language model in Example 1. Figure 16 This is a landscape heterogeneity index data map of City A in the refined assessment method of urban flood resilience based on a large language model in Example 1; Figure 17 This is a landscape connectivity index data map of City A in Example 1, based on the refined assessment method of urban flood resilience using a large language model. Figure 18This is a weighted ranking of various indicator combinations for City A in the refined assessment method of urban flood resilience based on a large language model, as shown in Example 1. Figure 19 This is a graph showing the flood resilience assessment results for City A using the refined urban flood resilience assessment method based on a large language model in Example 1. Figure 20 This is a verification diagram of the assessment results for City A in Example 1, which uses the refined assessment method for urban flood resilience based on a large language model. Figure 21 This is a framework diagram of the urban flood resilience refinement assessment system based on a large language model in Example 3. Detailed Implementation

[0021] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this application. To better illustrate this embodiment, some parts in the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings.

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] Example 1 This embodiment provides a refined assessment method for urban flood resilience based on a large language model. Figure 1 The flowchart of this embodiment is shown. This embodiment takes City A as the assessment object and, combined with the flood disaster characteristics of coastal megacities, conducts a refined assessment of urban flood resilience at a resolution of 30m. City A has a dense river network, high development intensity in its built-up area, and frequent extreme rainstorms, resulting in a prominent risk of urban flooding. This method can achieve high-precision, efficient, and multi-dimensional resilience assessment, providing accurate data support for the spatial layout of urban disaster prevention and mitigation.

[0024] Please combine Figure 2 , Figure 2 This illustrates the complete technical chain of the method in this embodiment. The steps are sequentially connected and can be executed automatically. The following is a detailed description of each step: Construct an urban flood resilience assessment index system that includes multiple indicators, and obtain multi-source heterogeneous data corresponding to the multiple indicators; The indicator system includes a pressure layer, a state layer, and a response layer. The pressure layer includes an annual average rainfall indicator. The state layer includes indicators such as the proportion of impervious surface, river and lake water surface ratio, surface elevation, topographic slope, ecological space area ratio, landscape heterogeneity, landscape connectivity, density of permanent residents in built-up areas, and percentage of vulnerable groups in the total urban population. The response layer includes indicators such as disaster accessibility of medical facilities, density of commercial facilities, density of emergency shelters, per capita road area, and per capita GDP.

[0025] In this embodiment, multiple indicator attributes are configured as positive or negative. Positive indicators indicate that the larger the indicator value, the higher the resilience level; negative indicators indicate that the smaller the indicator value, the higher the resilience level.

[0026] The urban flood resilience assessment index system constructed in this embodiment includes 15 indicators. The meaning and attributes of each indicator are as follows: Annual average rainfall directly affects surface runoff and drainage load, serving as a fundamental indicator for assessing urban flood risk. The annual average rainfall indicator is configured with a negative attribute. The proportion of impervious surface land reflects how paved surfaces of buildings and roads obstruct rainwater infiltration, significantly increasing peak runoff. The impervious surface proportion indicator is configured with a negative attribute. A higher river and lake surface ratio indicates stronger natural flood storage and drainage capacity, and a higher capacity for proactive adaptation to floods. The river and lake surface ratio indicator is configured with a positive attribute. Surface elevation reflects the natural flood avoidance advantages of high-altitude areas, making them less susceptible to waterlogging. The surface elevation indicator is configured with a positive attribute. Topographic slope affects surface runoff velocity; flat areas are prone to waterlogging, while steep slopes facilitate drainage. Topographic slope... The indicator attributes are configured as positive; the proportion of ecological space area reflects how ecological space reduces runoff and delays flood peaks through mechanisms such as plant interception, soil infiltration, and water storage in depressions, thereby enhancing the city's ability to adapt to flood disasters. The ecological space area proportion indicator attribute is configured as positive; areas with higher landscape heterogeneity have stronger buffering and regulation capabilities against flood disasters, and the urban space's adaptability and resilience can be enhanced through the complementary effects of multiple landscape types. The landscape heterogeneity indicator attribute is configured as positive; areas with higher landscape connectivity have smoother material and energy flow in the ecosystem. During flood disasters, water flow can be guided and regulated through ecological corridors, enhancing the urban space's adaptability to flood disasters. The landscape connectivity indicator attribute is configured as positive; The higher the population density of a built-up area, the more people will be affected by floods, and the greater the difficulty in organizing post-disaster evacuation, resettlement, and reconstruction. Therefore, the population density indicator for built-up areas is configured as negative. The percentage of vulnerable groups in the total urban population (the percentage of children under 14 and elderly over 65) is also important. A higher percentage of vulnerable groups leads to greater difficulty and longer recovery time after a disaster. Therefore, the percentage of vulnerable groups in the total urban population is configured as negative. Higher accessibility of medical facilities during a disaster increases the likelihood of timely treatment for injured people and strengthens the post-disaster recovery capacity of the urban spatial system. Therefore, the accessibility of medical facilities during a disaster is configured as positive. Higher density of commercial facilities also contributes to better post-disaster recovery. The better the convenience for residents to access daily necessities and basic services, the faster the recovery of social functions, and the commercial facility density index is configured as positive; the higher the density of emergency shelters, the more adequate the evacuation and temporary resettlement of residents during floods, and the faster the recovery of social functions after the disaster, and the emergency shelter density index is configured as positive; the larger the per capita road area, the stronger the transportation guarantee capacity for emergency rescue and material transportation after the disaster, which is conducive to accelerating the recovery process of the urban spatial system, and the per capita road area index is configured as positive; the higher the per capita GDP, the stronger the city's financial investment capacity in infrastructure repair, public service restoration and resettlement of disaster-stricken people, and the faster the post-disaster recovery, and the per capita GDP index is configured as positive.

[0027] The multi-source heterogeneous data includes natural geographic data, remote sensing interpretation data, socioeconomic data, infrastructure and POI data, meteorological data, and land use data. Among them, the original data of surface elevation and topographic slope indicators are extracted from the natural geographic data; the original data of the proportion of impervious surface, river and lake water surface ratio, and ecological space area ratio are extracted from the remote sensing interpretation data; the original data of per capita GDP, built-up area resident population density, and proportion of vulnerable groups are extracted from the socioeconomic data; the original data of medical facility accessibility during disasters, emergency shelter density, per capita road area, and commercial facility density are extracted from the infrastructure and POI data; the original data of annual average rainfall is extracted from the meteorological data; and the original data of landscape heterogeneity and landscape connectivity indicators are extracted from the land use data.

[0028] like Figures 3-17 As shown, natural geographic data can be obtained through Google Earth Engine and Copernicus DEM (Copernicus Digital Elevation Model). After projection transformation and slope extraction, the raw data of surface elevation and terrain slope indicators are obtained. Remote sensing interpretation data can be obtained through Landsat 8. OLI remote sensing imagery (30m resolution), after radiometric calibration and atmospheric correction, was interpreted using supervised classification to extract raw data on land cover, including the proportion of impervious surfaces, river and lake surface ratio, and ecological space area ratio. Socioeconomic data was obtained from the "A City Statistical Yearbook" and the 7th National Population Census, extracting per capita GDP, built-up area resident population density, and the proportion of vulnerable groups. Infrastructure and POI data were obtained from Baidu Maps POI and OpenStreetMap road network data. After coordinate correction and road network topology construction, indicators such as accessibility of medical facilities during disasters, density of commercial facilities, density of emergency shelters, and per capita road area were extracted. Meteorological data was obtained from the meteorological data network, acquiring annual rainfall data from various meteorological stations in A City. After spatial interpolation, a spatial distribution raster of annual average rainfall was obtained, thus yielding raw data for the annual average rainfall indicator. Land use data was obtained from land use classification data interpreted from Landsat imagery. After reclassification, Fragstats software was used to calculate raw data for landscape heterogeneity and landscape connectivity indicators.

[0029] The target area is divided into multiple grid cells using a grid with a preset resolution. The multi-source heterogeneous data is resampled into the multiple grid cells to form the original grid cell data of the multiple indicators. The original grid cell data is preprocessed to obtain the standard data of the multiple indicators. In this embodiment, a grid with a preset resolution of 30m×30m is used to divide the area of ​​City A. City A has a total area of ​​2214.60 km², an urbanization rate of 81.35% in 2024, a resident population of 10.176 million, and an average annual precipitation of 1023 mm. After division, more than two million grid cells are generated.

[0030] River and lake surface area ratio, land surface elevation, topographic slope, ecological space area ratio, landscape heterogeneity, landscape connectivity, accessibility of medical facilities during disasters, density of commercial facilities, density of emergency shelters, per capita road area, and per capita GDP are defined as positive indicators. These positive indicators are standardized using the first range standardization formula.

[0031] in, For the first The first grid cell Standard data for a positive indicator The raw data for the indicator. and The first The maximum and minimum values ​​of the original data for each indicator; Annual average rainfall, the proportion of impervious surface land, the density of permanent residents in built-up areas, and the proportion of vulnerable groups are defined as negative indicators. These negative indicators are then standardized using the second range standardization formula.

[0032] in, For the first The first grid cell Standard data for a negative indicator.

[0033] Based on the standard data of the aforementioned multiple indicators, the objective weight of each indicator is calculated. In this embodiment, the weight of the i-th evaluation unit under the j-th indicator is first calculated:

[0034] Where n is the number of grid cells, Standard data for indicators; To avoid invalid values ​​in subsequent logarithm calculations ( (In the case of =0), the standardized value is usually slightly offset, that is, let +10 -10 .

[0035] Calculate the information entropy of the j-th indicator. :

[0036] Where k is the adjustment coefficient, and information entropy This reflects the degree of disorder in the j-th indicator data: The closer it is to 1, the more uniform the values ​​of each evaluation unit are on this indicator, and the less information the indicator can provide. The closer it is to 0, the greater the degree of variation in the indicator value and the greater the amount of information.

[0037] Calculate the coefficient of variation for the j-th indicator:

[0038] Coefficient of difference The value of an indicator represents its information utility; the higher the value, the greater its importance in the overall evaluation.

[0039] Calculate the objective weight of the j-th indicator. :

[0040] Where m is the total number of indicators, and in this embodiment m is 15.

[0041] The standard data of the multiple indicators are input into the constructed multi-domain virtual expert model, and subjective weights are assigned to the multiple indicators to obtain the subjective weights of each indicator. Among them, the constructed multi-domain virtual expert models include urban planning expert models, resilient city expert models, hydrology and flood disaster expert models, ecological environment expert models and emergency management expert models set up in parallel. By setting fixed-layer prompts, the role positioning, professional evaluation dimensions, independent judgment preferences, evaluation principles and expert consultation rules of virtual expert models in various fields are defined, and the fixed-layer prompts remain unchanged in different evaluation tasks. By setting variable-level prompts, the urban characteristics, theoretical framework, and indicator system of the input target area are defined, and these prompts are updated as the evaluation task changes.

[0042] In this embodiment, the specific construction logic and differentiation definition of each virtual expert are as follows: Urban planning expert: Role positioning: Planning decision-maker with a core perspective on urban spatial layout, land use and infrastructure planning.

[0043] Professional assessment dimensions: urban land use structure, development intensity, blue-green space configuration, drainage network planning, and the expansion trend of impermeable surfaces.

[0044] Independent Judgment Preferences: This expert's reasoning strictly follows the principle of "prioritizing the interventionibility of planning." Their judgment preferences are as follows: (a) They prefer indicators with strong spatial planning control (such as the proportion of impermeable surfaces and the proportion of ecological space), believing that these indicators can be directly intervened through land spatial planning; (b) They pay attention to the role of the spatial organization of the built environment in mitigating flood risks; (c) They assign higher weights to infrastructure indicators reflecting the implementation of planning standards, such as per capita road area and the density of emergency shelters; (d) They believe that natural geographical conditions such as elevation and slope are fundamental constraints for planning site selection, but because they cannot be changed through planning, their weight is moderate.

[0045] Resilient city experts: Role: System resilience analyst with a core perspective on urban system resilience, adaptability, and resilience.

[0046] Professional assessment dimensions: the comprehensive ability of urban systems to withstand disturbances, absorb shocks, adapt to changes, and recover rapidly.

[0047] Independent Judgment Preferences: This expert's reasoning strictly follows the principle of "system synergy and adaptive cycle." Their judgment preferences are as follows: (a) They prefer indicators with strong systemic and synergistic characteristics, emphasizing the coupling relationship between various subsystems (natural, social, economic, and infrastructure); (b) They assign high weight to indicators reflecting ecosystem service functions, such as landscape connectivity and river / lake surface area ratio, believing that ecological networks are the "infrastructure" of urban resilience; (c) They highly value the proportion of vulnerable groups, believing that social vulnerability is a key indicator of resilience shortcomings, reflecting the "equity" dimension in resilience theory; (d) They emphasize the synergistic effect between response-layer indicators (medical accessibility, emergency shelter) and state-layer indicators.

[0048] Hydrological and Flood Disaster Expert Model: Role positioning: Disaster science researcher with a core perspective on flood formation mechanisms, hydrological processes, and disaster risk assessment.

[0049] Professional assessment dimensions: hazard of disaster-causing factors, sensitivity of disaster-prone environment, and exposure of disaster-bearing bodies.

[0050] Independent Judgment Preference: This expert's reasoning strictly follows the principle of "flood formation mechanism-driven". Their judgment preferences are as follows: (a) They assign the highest weight to average annual rainfall, believing that disaster-causing factors are the fundamental driver of flood risk; (b) They highly value surface elevation, topographic slope, and the proportion of impermeable surfaces, as these three indicators directly determine runoff paths and water accumulation distribution, and are core elements of the disaster-prone environment; (c) They assign a relatively high weight to river and lake surface area ratio, believing that natural water bodies are the core carriers of flood regulation; (d) They relatively weaken socio-economic indicators, considering them to be "disaster-bearing body attributes" rather than core elements of the flood formation mechanism.

[0051] Ecological and environmental experts: Role positioning: Ecological researcher with a core perspective on ecosystem services, landscape patterns and natural regulation functions.

[0052] Professional assessment dimensions: hydrological regulation functions of the ecosystem, such as interception, infiltration, storage, and purification, as well as the impact of landscape patterns on ecological processes.

[0053] Independent Judgment Preferences: This expert's reasoning strictly follows the principle of "prioritizing green infrastructure and ecological regulation functions." Their judgment preferences are as follows: (a) They prefer indicators that directly reflect ecosystem service capacity, such as the proportion of ecological space area, landscape heterogeneity, landscape connectivity, and river and lake surface area ratio; (b) They believe that the proportion of impermeable surface is the core driving force of ecological function degradation and assign it high weight; (c) They emphasize the value of landscape heterogeneity—diverse landscape types (woodland, grassland, wetlands, water bodies) can provide complementary hydrological regulation services; (d) They focus on landscape connectivity, believing that the connectivity of ecological corridors is key to achieving regional runoff diversion and regulation.

[0054] Emergency management expert: Role positioning: Emergency management practitioners with a core perspective of disaster emergency response, post-disaster recovery and social function reconstruction.

[0055] Professional assessment dimensions: emergency preparedness capability, response timeliness, allocation of rescue resources, and post-disaster recovery speed.

[0056] Independent Judgment Preference: This expert's reasoning strictly follows the principle of "prioritizing lifeline engineering and recovery timeliness." Their judgment preferences are as follows: (a) They pay close attention to all indicators at the response level (accessibility of medical facilities, density of emergency shelters, per capita road area, density of commercial facilities, and per capita GDP), believing that "the ability to recover quickly after a disaster" is the ultimate test of resilience; (b) They assign the highest weight to the accessibility of medical facilities during a disaster, as it directly relates to the timeliness of post-disaster life-saving efforts; (c) They emphasize per capita road area, believing that transportation is the "lifeline" of emergency rescue; (d) They focus on per capita GDP, believing that economic strength is the financial foundation for post-disaster recovery; (e) They assign relatively low weight to natural geographical and ecological indicators, considering them to be "inherent conditions" rather than manageable emergency response capabilities.

[0057] In this embodiment, the prompts for the fixed layer and the variable layer are as follows: System Prompt (Fixed Layer): You are part of the "Urban Safety Resilience Assessment Committee," a multidisciplinary team of experts including urban planning, resilient city experts, hydrology and flood disaster experts, ecological environment experts, and emergency management experts. Your role is to empower the urban flood resilience index system.

[0058] The roles of the five types of virtual experts are defined as follows. Each type of expert must strictly adhere to their respective professional perspective, evaluation dimensions, and judgment preferences during the independent reasoning phase, and must not mix roles: 1. Urban planning expert Role positioning: Planning decision-maker with a core perspective on urban spatial layout, land use and infrastructure planning.

[0059] Professional assessment dimensions: urban land use structure, development intensity, blue-green space configuration, expansion trend of impermeable surfaces, and implementation of planning standards.

[0060] Independent assessment preferences: Adhering to the principle of "prioritizing planning interventionability." A preference is given to indicators with strong spatial planning control (such as the proportion of impervious surface and the proportion of ecological space), believing these indicators can be directly intervened in through land spatial planning; attention is paid to the role of the spatial organization of the built environment in mitigating flood risks; higher weight is given to indicators reflecting the implementation of planning standards, such as per capita road area and density of emergency shelters; natural conditions such as elevation and slope are considered fundamental constraints for planning site selection, but because they cannot be changed through planning, their weight is moderate.

[0061] 2. Resilient City Experts Role: System resilience analyst with a core perspective on urban system resilience, adaptability, and resilience.

[0062] Professional assessment dimensions: the comprehensive ability of the urban system to withstand disturbances, absorb shocks, adapt to changes, and recover quickly, as well as the synergistic relationship between various subsystems.

[0063] Independent assessment preferences: Adhering to the principle of "system synergy and adaptive cycle". A preference is placed on indicators with strong systemic and synergistic characteristics, emphasizing the coupling relationship between natural, social, economic, and infrastructure subsystems; high weight is given to indicators reflecting ecosystem service functions, such as landscape connectivity and river / lake surface area ratio, believing that ecological networks are the infrastructure of urban resilience; great importance is attached to the proportion of vulnerable groups, believing that social vulnerability is a key indicator of resilience shortcomings; and the synergistic effect between response-level and state-level indicators is emphasized.

[0064] 3. Hydrology and flood disaster experts Role positioning: Disaster science researcher with a core perspective on flood formation mechanisms, hydrological processes, and disaster risk assessment.

[0065] Professional assessment dimensions: hazard of disaster-causing factors, sensitivity of disaster-prone environment, and exposure of disaster-bearing bodies.

[0066] Independent assessment preference: Adhering to the principle of "flood formation mechanism driven by factors". Annual average rainfall is given the highest weight, as the causative factors are considered the fundamental driver of flood risk; high importance is attached to surface elevation, topographic slope, and the proportion of impermeable surfaces, as these three indicators directly determine runoff paths and water distribution; a relatively high weight is given to river and lake surface area ratio, as natural water bodies are considered the core carriers of flood regulation; socio-economic indicators are relatively weakened, as they are considered "disaster-bearing body attributes" rather than core elements of the flood formation mechanism.

[0067] 4. Ecological and environmental experts Role positioning: Ecological researcher with a core perspective on ecosystem services, landscape patterns and natural regulation functions.

[0068] Professional assessment dimensions: hydrological regulation functions of the ecosystem, such as interception, infiltration, storage, and purification, as well as the impact of landscape patterns on ecological processes.

[0069] Independent assessment preferences: Adhering to the principle of "prioritizing green infrastructure and ecological regulation functions." Prefers indicators that directly reflect ecosystem service capacity, such as the proportion of ecological space area, landscape heterogeneity, landscape connectivity, and river and lake surface area ratio; considers the proportion of impervious surface land as the core driving force of ecological function degradation, giving it high weight; emphasizes the value of landscape heterogeneity—diverse landscape types can provide complementary hydrological regulation services; and believes that landscape connectivity is key to achieving regional runoff diversion and regulation.

[0070] 5. Emergency Management Experts Role positioning: Emergency management practitioners with a core perspective of disaster emergency response, post-disaster recovery and social function reconstruction.

[0071] Professional assessment dimensions: emergency preparedness capability, response timeliness, allocation of rescue resources, and post-disaster recovery speed.

[0072] Independent assessment preferences: Adhering to the principle of "prioritizing lifeline engineering and recovery timeliness." High attention is paid to all indicators at the response level (accessibility of medical facilities, density of emergency shelters, road area per capita, density of commercial facilities, and GDP per capita), believing that "the ability to recover quickly after a disaster" is the ultimate test of resilience; the accessibility of medical facilities during a disaster is given the highest weight, as it directly relates to the timeliness of post-disaster life-saving efforts; the per capita road area is emphasized, as transportation is considered the "lifeline" of emergency rescue; relatively low weights are given to natural geographical and ecological indicators, as they are considered "inherent conditions" rather than manageable emergency capabilities.

[0073] [Core Mission] To determine the importance weights of the indicators, a committee discussion process was simulated, in which five types of experts independently assessed, compared differences, and reached a consensus, for the flood resilience assessment index system of a designated city under extreme rainstorm scenarios.

[0074] Evaluation Principles Principle 1: Mechanism First Indicators directly related to the mechanism of flood formation should be given priority.

[0075] Principle 2: Resilience Oriented The focus is on the city's ability to withstand, absorb, adapt to, and recover from disturbances.

[0076] Principle 3: Spatial Orientation The indicators should be able to reflect spatial heterogeneity.

[0077] Principle 4: Planning-oriented Indicators should have significance for planning and regulation and value for policy intervention.

[0078] Principle 5: Scientific and Rational There should be no situation where the weights of all indicators are too evenly distributed.

[0079] [Expert Consultation Mechanism] Step 1: Based on their respective roles, the five types of experts independently deduce the initial subjective weights of each indicator, and output the importance ranking sequence and judgment basis according to the initial subjective weights of each indicator from high to low.

[0080] The second step is to identify and quantify the ranking differences among different experts based on the importance ranking sequence, and define each indicator as a consensus indicator, a divergence indicator, or an extreme preference indicator.

[0081] Step 3: For consensus indicators, take the arithmetic mean of the initial subjective weights of the indicator in each expert model as its subjective weight. For the divergence index, the arithmetic mean of the initial subjective weights of the index in each expert model is taken as its subjective weight, and the weight difference and compromise logic of the index in each expert model are marked when the subjective weight is output. For extreme preference indicators, the terms with the highest and lowest initial subjective weights for the indicator in each expert model are removed, and the arithmetic mean of the initial subjective weights of the indicator in the remaining expert models is taken as its subjective weight.

[0082] Weight Constraints All weights must satisfy: The total weight should be 1; all weights should be greater than 0; two decimal places should be retained; the weight distribution should conform to the actual urban flooding patterns; simple average weighting is not allowed.

[0083] [Results Self-Check] After generating the results, please check them again: (1) Whether it conforms to the mechanism of flood formation; (2) Whether it conforms to the PSR theoretical framework; (3) Whether it conforms to the theory of resilient cities; (4) Whether there are obviously unreasonable weights; (5) Whether there is a tendency to average out.

[0084] If the self-test does not meet the above requirements, please correct it and go back to re-execute all System Prompt operations.

[0085] Finally, please output the results in standard JSON format.

[0086] User Prompt (Variable Level): [Research Subjects] City A [Urban Characteristics] Coastal megacities, dense river networks, high-density built-up environments, and frequent rainstorms and floods Theoretical Framework Pressure-State-Response (PSR): Pressure: Flood risk and stress; State: System state and buffer capacity; Response: Emergency response and recovery capacity. [Indicator System] Pressure: P1 Annual average rainfall State: S1 Surface elevation, S2 Topographic slope, S3 Resident population density of built-up area, S4 Percentage of vulnerable groups, S5 Percentage of impervious surface, S6 River and lake water surface ratio, S7 Percentage of ecological space area, S8 Landscape heterogeneity, S9 Landscape connectivity Response: R1 Accessibility of medical facilities during a disaster, R2 Density of commercial facilities, R3 Accessibility of emergency shelters, R4 Road area per capita, R5 GDP per capita [Expert Empowerment Tasks] Strictly follow the roles, principles, and negotiation mechanisms defined in the System Prompt to complete this specific authorization task: Step 1: The five types of virtual experts output the initial subjective weight of each indicator and the importance ranking of the 15 indicators based on their respective professional perspectives, and briefly explain the judgment logic of the ranking and weighting of this group.

[0087] Step 2: Compare the initial subjective weights and importance ranking sequences output by the five virtual experts, and identify and list all consensus indicators, divergence indicators, and extreme preference indicators.

[0088] Step 3: According to the corresponding fusion rules given by System Prompt, calculate the subjective weights of each indicator for consensus indicator, divergence indicator, and extreme preference indicator.

[0089] Step 4: Perform a self-check on the subjective weights of all indicators. If there are any cases that do not meet the check conditions, iterate and correct them and roll back to re-execute the entire process until the check passes. Step 5: For each indicator, briefly explain the reasons for the formation of the subjective weight of that indicator.

[0090] Step 6: Output the results.

[0091] Output Format Please strictly adhere to the following JSON format for output: Json {"weights": { "Average Annual Rainfall": {"weight":, "reason":}"Surface Elevation": {"weight":, "reason":}"Topographic Slope": {"weight":, "reason":}"Resident Population Density in Built-up Area": ​​{"weight":, "reason":}"Percentage of Vulnerable Groups": {"weight":, "reason":}"Percentage of Impervious Surface Area": ​​{"weight":, "reason":}"River and Lake Water Surface Ratio": {"weight":, "reason":}"Percentage of Ecological Space Area": ​​{"weight":, "reason":}"Landscape Heterogeneity": {"weight":, "reason":}"Landscape Connectivity": {"weight":, "reason":}"GDP per Capita": {"weight":, "reason":}"Road Area per Capita": {"weight":, "reason":}"Accessibility of Medical Facilities During Disasters": {"weight":, "reason":}"Accessibility of Emergency Shelters": {"weight":, "reason":}"Density of Commercial Facilities": {"weight":, "reason":}}} .

[0092] The standard data of the multiple indicators are input into the constructed multi-domain virtual expert model, and subjective weights are assigned to each indicator to obtain the subjective weights of each indicator, including: Standard data for multiple indicators are input into the urban planning expert model, resilient city expert model, hydrology and flood disaster expert model, ecological environment expert model and emergency management expert model respectively. Each expert model outputs the initial subjective weight of each indicator and outputs the importance ranking sequence from high to low according to the initial subjective weight of each indicator. Compare the importance ranking sequence of each indicator output by each expert model, and classify the multiple indicators into consensus indicators, divergence indicators, or extreme preference indicators; For consensus indicators, the arithmetic mean of the initial subjective weights of the indicator in each expert model is taken as its subjective weight. For the divergence index, the arithmetic mean of the initial subjective weights of the index in each expert model is taken as its subjective weight, and the weight difference and compromise logic of the index in each expert model are marked when the subjective weight is output. For extreme preference indicators, the terms with the highest and lowest initial subjective weights for the indicator in each expert model are removed, and the arithmetic mean of the initial subjective weights of the indicator in the remaining expert models is taken as its subjective weight.

[0093] After completing the differentiated role modeling of the above five types of virtual experts, the multi-domain virtual expert model achieves subjective empowerment through the following process: In the multi-domain virtual expert model, five types of virtual experts independently receive the same input information (city characteristics, indicator system, evaluation task) and make independent judgments under the constraints of their respective role definitions and judgment preferences.

[0094] Each type of virtual expert independently ranks the importance of 15 indicators; Each type of expert independently outputs a summary of the logical basis for the ranking, clearly stating the focus of their professional perspective; Each type of expert's reasoning process does not interfere with or refer to each other, ensuring the purity of independent reasoning.

[0095] A multi-domain virtual expert model automatically compares the index rankings and weight assignments of five types of experts, identifying and quantifying the following differences: Consensus indicators: Indicators with small differences in the ranking of the five types of experts (Spearman correlation coefficient > 0.8) and all assigned high weights reflect interdisciplinary consensus; Disagreement index: An index in which different experts make significant differences in ranking (Spearman correlation coefficient < 0.5), reflecting differences in disciplinary perspectives; Extreme preference index: An index that is weighted by a single expert by more than two standard deviations above the mean of other experts, reflecting the unique concerns in that field.

[0096] The system simulates the expert committee's "discussion-consultation-consensus" decision-making process: The system is based on the independent ranking results of five types of experts and adopts an equal-weighted comprehensive strategy—each type of expert has equal weight (each accounting for 1 / 5), reflecting the academic democratic principle of equal dialogue among multiple disciplines; For consensus indicators, the arithmetic mean of the weights of the five types of experts is taken directly; For divergent indicators, the system automatically generates a "dispute explanation" in the reason field of the JSON while outputting the weights, indicating the weight differences and compromise logic of the indicator from different expert perspectives; For extreme preference indicators, the truncated mean method is used: after removing the highest and lowest weights, the mean of the remaining three weights is taken, which preserves their unique concerns while avoiding the excessive dominance of a single perspective.

[0097] Step 4: Weight Output The final output consists of the subjective weights of 15 indicators, ensuring that: the total weight = 1; all weights are greater than 0; two decimal places are retained; the weight distribution conforms to the urban flooding pattern, and all indicator weights must not be averaged (i.e., automatic regeneration is triggered when the standard deviation of all weights from the mean is <0.005).

[0098] Compared with traditional expert empowerment methods, this invention has the following advantages: the number of experts is unlimited, and an infinite number of experts can be simulated; cognitive biases can be avoided through multi-role independent reasoning and cross-validation; the empowerment results can provide detailed reasoning chains and explanatory texts, with good interpretability.

[0099] In this embodiment, after the multi-domain virtual expert model outputs subjective weights, it also includes quality verification, which includes: Multiple rounds of sampling: Set Temperature=0.1 to perform multiple rounds of sampling, take the average weight as the final subjective weight, and evaluate the stability of the weight; Expert rule constraints: By strictly constraining the judgment preferences of five types of virtual experts, each expert is forced to output a differentiated weight distribution that conforms to their professional logic, thus avoiding the tendency of weight convergence and averaging from a technical mechanism. Indicator logic verification: Verify whether the weights are consistent with the flood mechanism; Consistency verification: Spearman correlation coefficient is used to test the reliability of AI weighting. A threshold of 0.7 is used to determine whether consistency is passed. If it is passed, it is confirmed that the weighting results of the multi-domain virtual expert model have the equivalence of human experts.

[0100] The subjective weights and objective weights are weighted and fused to obtain the combined weights of each indicator. Based on the standard data and the combined weights, the comprehensive urban flood resilience index of each grid unit is calculated. The combined weights are expressed as follows:

[0101] Where α is the fusion coefficient of subjective weights, Let j be the subjective weight of the j-th indicator. Let be the objective weight of the j-th indicator; In this embodiment, the combined weight of City A is as follows: Figure 18 As shown in Table 1, Table 1 Weight Allocation Table for Each Indicator

[0102] The comprehensive index of urban flood resilience is expressed as follows:

[0103] in, For the first The comprehensive urban flood resilience index for each grid cell ranges from [0,1]. A higher comprehensive urban flood resilience index value indicates a higher level of spatial resilience to flood disasters for that grid cell. For the first The combined weights of the indicators; For the first The first grid cell Standard data for each indicator.

[0104] Based on the comprehensive urban flood resilience index of each grid unit, the target area is classified into urban flood resilience levels, and the urban flood resilience assessment results of the target area are output.

[0105] In this embodiment, as shown in Table 2, the Natural Breaks Jenks method is used to classify the comprehensive urban flood resilience index. The Natural Breaks Jenks method is an optimal classification method based on the internal distribution characteristics of the data, which determines the classification breakpoints by minimizing within-group differences and maximizing between-group differences.

[0106] The comprehensive index of urban flood resilience is divided into five levels, from Level 1 to Level 5, corresponding to high resilience, relatively high resilience, medium resilience, relatively low resilience, and low resilience, respectively.

[0107] Table 2. Urban Flood Resilience Comprehensive Index Classification Table

[0108] Figure 19 The results of the urban flood resilience assessment of City A are shown. High-resilience areas are mainly concentrated in the ecological barrier zone in the southwest and the central urban area; the southwest relies on the natural water storage capacity of high vegetation cover and topographic slope; and the central urban area benefits from high-standard engineering flood control facilities and a well-developed emergency response network.

[0109] The medium toughness zone is mainly concentrated in the central region, serving as a buffer zone between the high and low toughness zones.

[0110] The low-resilience areas exhibit a coexistence of concentrated, contiguous areas and fragmented distributions; the concentrated, contiguous areas are mainly distributed in the north, while the fragmented distribution is concentrated in the southeast. To verify the reliability of the assessment results, this embodiment overlays the identified low-resilience areas with the distribution of flood-prone areas in City A in recent years. The disaster-affected area data is based on a list of flood-prone areas published online, and spatial information is obtained through geocoding and coordinate transformation, ultimately identifying 30 disaster-affected areas in City A. Figure 20 As shown, the overlay analysis results indicate that the disaster-affected points are highly concentrated in medium and low toughness grids, with high spatial overlap and good spatial identification of the model. This verification result confirms the accuracy of the assessment of this invention from an empirical perspective.

[0111] This embodiment can accurately identify the spatial heterogeneity of resilience at the micro-plot and community scales—for example, the method in this embodiment can identify the intermingling distribution of patches with high and low resilience values, rather than simply classifying the entire area into the same resilience level. This enhanced level of precision provides direct technical support for "accurate identification, precise policy implementation, and precise execution" in territorial spatial planning.

[0112] Example 2 This embodiment also provides a web-based tool that integrates the urban flood resilience refinement assessment method based on a large language model described in Embodiment 1 into the tool, enabling the assessment process to run online and facilitating planning and disaster prevention personnel to quickly conduct urban flood resilience refinement assessments.

[0113] The tool includes a pressure layer, a state layer, and a response layer. It incorporates core algorithms such as multi-source data spatialization preprocessing, data normalization, objective weighting using entropy method, subjective weighting by virtual experts from multiple domains, weighted fusion of subjective and objective weights, weighted comprehensive evaluation, and resilience level classification. It supports fully automated processing from raw data import to flood resilience assessment result output.

[0114] Data import module: Users upload the raw data of each indicator according to the specified format (GeoTIFF raster format is supported). Automatic calculation module: sequentially completes the steps of grid resampling and spatial matching, index standardization, objective weight calculation using entropy method, subjective weighting of multiple virtual expert models, fusion of subjective and objective weights, calculation of comprehensive flood resilience index of grid unit, and classification of resilience level; Result output module: automatically generates raster map of comprehensive flood resilience index, spatial distribution map of resilience level, and statistical table of index weights, supporting one-click download.

[0115] The tool supports one-click generation of GIS Urban Flood Resilience Index (RCI) raster maps. Users only need to upload the required multi-source indicator data, and the system can automatically complete data spatial alignment, indicator standardization, subjective and objective weight calculation, weight fusion, and comprehensive index calculation. The tool has a built-in two-layer prompt-driven multi-domain virtual expert weighting logic, which can reproduce the comprehensive judgment effect of multidisciplinary experts without the need for offline human expert scoring. It outputs weight results that take into account domain knowledge and data statistical characteristics, and quickly completes the entire set of assessment calculations for the target area.

[0116] The tool supports downloading results data to local GIS software for personalized rendering and in-depth analysis. The tool's built-in algorithm framework has good scalability; when changing the study area or the type of disaster assessment, the algorithm framework can be reused simply by modifying the corresponding indicator system and prompt word configuration.

[0117] Example 3 like Figure 21 As shown, this embodiment provides a refined assessment system for urban flood resilience based on a large language model, including: Indicator system construction module 301: used to construct an urban flood resilience assessment indicator system including multiple indicators and obtain multi-source heterogeneous data corresponding to the multiple indicators; Grid cell initialization module 302: used to divide the target area into multiple grid cells using a grid with a preset resolution, resample the multi-source heterogeneous data into the multiple grid cells to form the original grid cell data of the multiple indicators, and preprocess the original grid cell data to obtain the standard data of the multiple indicators; Objective weight calculation module 303: used to calculate the objective weight of each indicator based on the standard data of the multiple indicators; Subjective weight calculation module 304: used to input the standard data of the multiple indicators into the constructed multi-domain virtual expert model, and to subjectively assign weights to the multiple indicators to obtain the subjective weights of each indicator; Urban flood resilience comprehensive index calculation module 305: used to perform weighted fusion processing of the subjective weight and the objective weight to obtain the combined weight of each indicator, and calculate the urban flood resilience comprehensive index of each grid unit based on the standard data and the combined weight; Urban flood resilience assessment module 306: This module is used to classify the urban flood resilience level of the target area based on the comprehensive urban flood resilience index of each grid unit, and output the urban flood resilience assessment results of the target area.

[0118] In the specific implementation process, the indicator system construction module 301 first constructs an urban flood resilience assessment indicator system containing multiple indicators and acquires multi-source heterogeneous data corresponding to each indicator. Next, the grid unit initialization module 302 divides the target area into grids with a preset resolution, resamples and maps the multi-source heterogeneous data to each grid unit to obtain the original grid unit data, and outputs the standard data for each indicator after preprocessing. Subsequently, the objective weight calculation module 303 calculates the objective weights of each indicator based on the standard data. The subjective weight calculation module 304 inputs the standard data into the constructed multi-domain virtual expert model to complete the subjective weighting of each indicator, obtaining the subjective weights. The urban flood resilience comprehensive index calculation module 305 performs weighted fusion of subjective and objective weights to obtain the combined weights, and calculates the urban flood resilience comprehensive index for each grid unit based on the standard data. Finally, the urban flood resilience assessment module 306 completes the flood resilience level classification of the target area based on the comprehensive index of the grid units and outputs the corresponding urban flood resilience assessment results. The various modules of this system cooperate with each other to fully realize the entire process of multi-source data processing, subjective and objective weighting, comprehensive index calculation, and result output, enabling efficient and refined assessment of urban flood resilience.

[0119] The same or similar labels correspond to the same or similar parts; The terms used to describe positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this application. Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A refined assessment method for urban flood resilience based on a large language model, characterized in that, include: Construct an urban flood resilience assessment index system that includes multiple indicators, and obtain multi-source heterogeneous data corresponding to the multiple indicators; The target area is divided into multiple grid cells using a grid with a preset resolution. The multi-source heterogeneous data is resampled into the multiple grid cells to form the original grid cell data of the multiple indicators. The original grid cell data is preprocessed to obtain the standard data of the multiple indicators. Based on the standard data of the aforementioned multiple indicators, the objective weight of each indicator is calculated. The standard data of the multiple indicators are input into the constructed multi-domain virtual expert model, and subjective weights are assigned to the multiple indicators to obtain the subjective weights of each indicator. The subjective weights and objective weights are weighted and fused to obtain the combined weights of each indicator. Based on the standard data and the combined weights, the comprehensive urban flood resilience index of each grid unit is calculated. Based on the comprehensive urban flood resilience index of each grid unit, the target area is classified into urban flood resilience levels, and the urban flood resilience assessment results of the target area are output.

2. The refined assessment method for urban flood resilience based on a large language model according to claim 1, characterized in that, The indicator system includes a pressure layer, a state layer, and a response layer. The pressure layer includes an annual average rainfall indicator. The state layer includes indicators such as the proportion of impervious surface, river and lake water surface ratio, surface elevation, topographic slope, ecological space area ratio, landscape heterogeneity, landscape connectivity, density of permanent residents in built-up areas, and percentage of vulnerable groups in the total urban population. The response layer includes indicators such as disaster accessibility of medical facilities, density of commercial facilities, density of emergency shelters, per capita road area, and per capita GDP.

3. The refined assessment method for urban flood resilience based on a large language model according to claim 2, characterized in that, The multi-source heterogeneous data includes natural geographic data, remote sensing interpretation data, socio-economic data, infrastructure and POI data, meteorological data, and land use data; Among them, the original data of surface elevation and topographic slope indicators were extracted from the corresponding natural geographic data; the original data of the indicators of impervious surface ratio, river and lake water surface ratio, and ecological space area ratio were extracted from the corresponding remote sensing interpretation data; the original data of indicators of per capita GDP, built-up area resident population density, and vulnerable group ratio were extracted from the corresponding socio-economic data; the original data of indicators of medical facility accessibility during disasters, emergency shelter density, per capita road area, and commercial facility density were extracted from the corresponding infrastructure and POI data; the original data of the annual average rainfall indicator was extracted from the corresponding meteorological data; and the original data of indicators of landscape heterogeneity and landscape connectivity were extracted from the corresponding land use data.

4. The refined assessment method for urban flood resilience based on a large language model according to claim 3, characterized in that, The original grid cell data is preprocessed to obtain standard data for multiple indicators, including: River and lake surface area ratio, land surface elevation, topographic slope, ecological space area ratio, landscape heterogeneity, landscape connectivity, accessibility of medical facilities during disasters, density of commercial facilities, density of emergency shelters, per capita road area, and per capita GDP are defined as positive indicators. These positive indicators are standardized using the first range standardization formula. in, For the first The first grid cell Standard data for a positive indicator The raw data for the indicator. and The first The maximum and minimum values ​​of the original data for each indicator; Annual average rainfall, the proportion of impervious surface land, the density of permanent residents in built-up areas, and the proportion of vulnerable groups are defined as negative indicators. These negative indicators are then standardized using the second range standardization formula. in, For the first The first grid cell Standard data for a negative indicator.

5. The refined assessment method for urban flood resilience based on a large language model according to claim 1, characterized in that, The preset resolution is 30m×30m.

6. The refined assessment method for urban flood resilience based on a large language model according to claim 1, characterized in that, Based on the standard data of the aforementioned multiple indicators, the objective weights of each indicator are calculated, including: For each indicator, calculate the proportion of the standard data value of each grid cell to the sum of the standard data values ​​of all grid cells of that indicator, and obtain the characteristic weight of each grid cell corresponding to that indicator; Information entropy is calculated based on the aforementioned feature weights, and a difference coefficient is constructed based on the information entropy. The objective weight of each indicator is obtained by dividing the difference coefficient of each indicator by the sum of the difference coefficients of all indicators.

7. The refined assessment method for urban flood resilience based on a large language model according to claim 1, characterized in that, The constructed multi-domain virtual expert models include parallel-set expert models for urban planning, resilient cities, hydrology and flood disasters, ecological environment, and emergency management. By setting fixed-layer prompts, the role positioning, professional evaluation dimensions, independent judgment preferences, evaluation principles and expert consultation rules of virtual expert models in various fields are defined, and the fixed-layer prompts remain unchanged in different evaluation tasks. By setting variable-level prompts, the urban characteristics, theoretical framework, and indicator system of the input target area are defined, and these prompts are updated as the evaluation task changes.

8. The refined assessment method for urban flood resilience based on a large language model according to claim 7, characterized in that, The standard data of the multiple indicators are input into the constructed multi-domain virtual expert model, and subjective weights are assigned to each indicator to obtain the subjective weights of each indicator, including: Standard data for multiple indicators are input into the urban planning expert model, resilient city expert model, hydrology and flood disaster expert model, ecological environment expert model and emergency management expert model respectively. Each expert model outputs the initial subjective weight of each indicator and outputs the importance ranking sequence from high to low according to the initial subjective weight of each indicator. Compare the importance ranking sequence of each indicator output by each expert model, and classify the multiple indicators into consensus indicators, divergence indicators, or extreme preference indicators; For consensus indicators, the arithmetic mean of the initial subjective weights of the indicator in each expert model is taken as its subjective weight. For the divergence index, the arithmetic mean of the initial subjective weights of the index in each expert model is taken as its subjective weight, and the weight difference and compromise logic of the index in each expert model are marked when the subjective weight is output. For extreme preference indicators, the terms with the highest and lowest initial subjective weights for the indicator in each expert model are removed, and the arithmetic mean of the initial subjective weights of the indicator in the remaining expert models is taken as its subjective weight.

9. The refined assessment method for urban flood resilience based on a large language model according to claim 1, characterized in that, The combined weights are expressed as follows: Where α is the fusion coefficient of subjective weights, Let j be the subjective weight of the j-th indicator. Let be the objective weight of the j-th indicator; The comprehensive index of urban flood resilience is expressed as follows: in, For the first The comprehensive urban flood resilience index for each grid cell ranges from [0,1]. A higher comprehensive urban flood resilience index value indicates a higher level of spatial resilience to flood disasters for that grid cell. For the first The combined weights of the indicators; For the first The first grid cell Standard data for each indicator.

10. A refined assessment system for urban flood resilience based on a large language model, employing the method described in any one of claims 1-9, characterized in that, include: Indicator system construction module: used to construct an urban flood resilience assessment indicator system including multiple indicators and obtain multi-source heterogeneous data corresponding to the multiple indicators; Grid cell initialization module: used to divide the target area into multiple grid cells using a grid with a preset resolution, resample the multi-source heterogeneous data into the multiple grid cells to form the original grid cell data of the multiple indicators, and preprocess the original grid cell data to obtain the standard data of the multiple indicators; Objective weight calculation module: used to calculate the objective weight of each indicator based on the standard data of the multiple indicators; Subjective weight calculation module: used to input the standard data of the multiple indicators into the constructed multi-domain virtual expert model, and to subjectively assign weights to the multiple indicators to obtain the subjective weights of each indicator; The Urban Flood Resilience Comprehensive Index Calculation Module is used to perform weighted fusion processing on the subjective weights and the objective weights to obtain the combined weights of each indicator, and calculate the urban flood resilience comprehensive index of each grid unit based on the standard data and the combined weights. Urban flood resilience assessment module: This module is used to classify the urban flood resilience level of the target area based on the comprehensive urban flood resilience index of each grid unit, and output the urban flood resilience assessment results of the target area.

Images